Available online at www.sciencedirect.com
Chemical Engineering and Processing 47 (2008) 1744–1764
Two-stage stochastic programming with fixed recourse via scenario planning with economic and operational risk management for petroleum refinery planning under uncertainty Cheng Seong Khor a,c , Ali Elkamel a,∗ , Kumaraswamy Ponnambalam b , Peter L. Douglas a a Department of Chemical Engineering, University of Waterloo, Ontario N2L 3G1, Canada Department of Systems Design Engineering, University of Waterloo, Ontario N2L 3G1, Canada c Chemical Engineering Programme, Universiti Teknologi PETRONAS, Bandar Seri Iskandar, Malaysia b
Received 4 May 2007; received in revised form 21 September 2007; accepted 24 September 2007 Available online 2 October 2007
Abstract This work proposes a hybrid of stochastic programming (SP) approaches for an optimal midterm refinery planning that addresses three sources of uncertainties: prices of crude oil and saleable products, demands, and yields. An SP technique that utilizes compensating slack variables is employed to explicitly account for constraints’ violations to increase model tractability. Four approaches are considered to ensure solution and model robustness: (1) the Markowitz’s mean-variance (MV) model to handle randomness in the objective function coefficients by minimizing the variance (economic risk) of the expected value of the random coefficients; (2): the two-stage SP with fixed recourse approach to deal with randomness in the RHS and LHS coefficients of the constraints by minimizing the expected recourse costs due to constraints’ violations; (3) incorporation of the MV model within the framework developed in (2) to formulate a mean–risk model that minimizes both the expectation and the operational risk measure of variance of the recourse costs; and (4) reformulation of the model in (3) by adopting mean-absolute deviation (MAD) as the measure of operational risk imposed by the recourse costs for a novel refinery planning application. A representative numerical example is illustrated. © 2007 Elsevier B.V. All rights reserved. Keywords: Two-stage stochastic programming; Refinery planning; Optimization under uncertainty; Scenario analysis; Mean-variance; Mean-absolute deviation (MAD)
1. Introduction It is a well-recognized problem that chemical process systems are subject to uncertainties presented by random events such as raw material variations, demand fluctuations, and equipment failures. The present work is intended to contribute towards mitigating this challenge by utilizing stochastic programming (SP) methods and analyses that are typically employed in computational finance applications, which have been demonstrated to be useful for screening alternatives on the basis of the expected value of economic criteria as well as the economic and operational risks involved [1,2]. Several approaches have been reported in the literature addressing the problem of production planning under uncertainty. Extensive reviews surveying various issues in this area are available, for example, by [3–6]. In general, planning models can be broadly categorized into three temporal classifications based on the addressed time horizons [7,8], namely (1) strategic (long-term, e.g. [9,10]); (2) tactical (medium- or midterm, e.g. [11,12]), and (3) operational (short-term, e.g. [13,14]). A discussion of their features and characteristics from a practical perspective is provided by [15]. The focus of this work is on the midterm tactical planning of petroleum refineries.
∗
Corresponding author. Tel.: +1 519 888 4567; fax: +1 519 746 4979. E-mail addresses:
[email protected] (C.S. Khor),
[email protected] (A. Elkamel),
[email protected] (K. Ponnambalam),
[email protected] (P.L. Douglas). 0255-2701/$ – see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.cep.2007.09.016
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
1745
Problems of design and planning of chemical processes and plants under uncertainty have been effectively treated in the process systems engineering (PSE) literature using the well-known approach of two-stage stochastic programming (SP) with recourse model. Under this framework, the problem is posed as one of optimizing an objective function that conventionally consists of two terms (or stages). The first corresponds to the “here-and-now” decisions of the global or planning variables, whose fixed values are selected ahead of, and thus independent of, the realization of the uncertain events. The second term represents and quantifies the expected value of the “wait-and-see” decisions due to the production variables, whose flexible values will be adjusted to achieve feasibility during operation, in response to revelation of the specific values of the uncertain parameters [16,17]. Further, variabilities due to production shortfalls and surpluses are accounted for by appending an additional second-stage term to the objective function, giving rise to the notion of operational risk [18] that results in a mean-risk structure of the model [19,20]. The presence of uncertainty is translated into the stochastic nature of the recourse costs associated with the second-stage decisions. Hence, the goal in the two-stage modelling approach to planning decision under uncertainty is to commit initially to the planning variables in such a way that the sum of the first-stage costs and the expected value plus deviations of the typically more expensive random second-stage recourse costs is minimized [21]. Approaches differ primarily in the way the expected value and its deviation terms are evaluated. 2. Problem statement The midterm refinery production planning problem addressed in this paper can be stated as follows. It is assumed that the physical resources of the plant are fixed and that the associated prices, costs, and demands are externally imposed [22]. The objective is to determine the optimal planning by computing the amount of materials that are processed at each time in each unit, in the face of three major uncertainties that are considered simultaneously, namely (1) market demand for products; (2) prices of crude oil and the saleable products; and (3) product (or production) yields of crude oil from chemical reactions in the primary crude distillation unit. A hybrid of stochastic programming techniques is applied within the framework of the classical two-stage stochastic program with fixed recourse to reformulate a deterministic planning problem. This approach is accomplished by adopting the mean-variance (E–V or MV) portfolio optimization model of Markowitz [23,24] in handling risk arising from variations in both profit and the recourse penalty costs due to violations of the stochastic constraints. A numerical study based on the deterministic refinery planning model of [25,26] is utilized to demonstrate the implementation of the proposed approaches without loss of generality. The single-objective linear programming (LP) model is first solved deterministically and is then reformulated with the addition of stochastic dimension according to principles and approaches outlined under the general model development. 3. General formulation of the deterministic midterm refinery planning model The basic framework for the deterministic planning model is mainly based on models formulated by [11,27], apart from those specific to refinery planning as proposed by [28–30]. Consider the production planning problem of a typical refinery operation with a network of M continuous processes and N materials as shown in Fig. 1. Let j ∈ J index the set of continuous processes whereas i ∈ I index the set of materials. These products are produced during n time periods indexed by t ∈ T to meet a prespecified level of demand during each period. Given also are the prices and availabilities of materials as well as investment and operating cost data over a time period. A typical aggregated mixed-integer linear planning model consists of the following sets of constraints and objective function.
Fig. 1. A network of processes and materials of a typical oil refinery operation (based on [28]).
1746
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
(a) Production capacity constraints: xj,t = xj,t−1 + CEj,t
∀j ∈ J
yj,t CELj,t ≤ CEj,t ≤ yj,t CEU j,t where yj,t =
(1) ∀j ∈ J, ∀t ∈ T
1
if there is an expansion
0
otherwise
(2)
(3)
(b) Demand constraints: Si,t + Li,t = di,t ,
∀i ∈ I , ∀t ∈ T
(4)
L U di,t ≤ Si,t ≤ di,t ,
∀i ∈ I , ∀t ∈ T
(5)
(c) Availability constraints: ∀i ∈ I , ∀t ∈ T
(6)
∀i ∈ I
(7)
∀i ∈ I , ∀t ∈ T
(8)
pLt ≤ Pt ≤ pU t , (d) Inventory requirements:
fmin f fmax Ii,t ≤ Ii,t ≤ Ii,t s f Ii,t = Ii,t+1 ,
This constraint is needed because a certain level of inventory must be maintained at all times to ensure material availability, in s , the starting inventory of material addition to the amount of materials purchased and/or produced. Eq. (8) simply states that Ii,t+1 f f = I f denotes i in period t + 1 is the same as Ii,t , the inventory of material i at the end of the preceding period t (if t = 1, then Ii,t i,1 the initial inventory). (e) Material balances: s f + bi,j xj,t − Si,t − Ii,t = 0, ∀i ∈ I , ∀t ∈ T (9) Pt + Ii,t j∈J
Objective function: a profit maximization function over the time horizon is considered as the difference between the revenue due to product sales and the overall costs, with the latter consisting of the cost of raw materials, operating cost, investment cost, and inventory cost: f − ˜ s ˜ i,t Ii,t i ∈ I γi,t Si,t + i∈Iγ i ∈ I λi,t Pi,t − i ∈ I λi,t Ii,t − j ∈ J Cj,t xj,t (10) max profit z0 = − i ∈ I hi,t Hi,t − j ∈ J (αj,t CEj,t + βj,t yj,t ) − (rt Rt + ot Ot ) t∈T 4. General formulation of the stochastic midterm refinery planning models In spite of the resulting exponential increase in the problem size with the number of uncertain parameters, the scenario analysis approach has been considerably used in the literature and has been proven to provide reliable and practical results for optimization under uncertainty [12,31]. Hence, in this work, it is adopted for describing uncertainty in the stochastic parameters. Representative scenarios are constructed to model uncertainty in the random variables of prices, demands, and yields within the two-stage stochastic programming (SP) framework. 4.1. Approach 1: Risk Model I The first approach adopts the classical Markowitz’s MV model to handle randomness in the objective function coefficients of prices, in which the expected profit is maximized while an appended term representing the magnitude of operational risk due to variability or dispersion in price, as measured by variance, is minimized [32]. The model can be variably formulated as minimizing risk (i.e., variance) subject to a lower bound constraint on the target profit (i.e., the mean return). Malcolm and Zenios [33] present an application of the MV approach by adopting the robust optimization framework proposed by Mulvey et al. [34] to the problem of capacity expansion of power systems. The problem is formulated as a large-scale nonlinear program with variance of the scenario-dependent costs included in the objective function. Another application using variance is employed by [35], also within a robust optimization framework of [34], for investment in the long-range capacity expansion of chemical process networks under uncertain demands.
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
1747
4.1.1. Sampling methodology by scenario generation for the recourse model under price uncertainty A collection of scenarios is generated that best captures and describes the trend of raw material prices of the different types of crude oil and the sales values (prices) of the saleable refining products for a representative period of time based on available historical data. A probability p s , with index s denoting the sth scenario, is assigned to each scenario to reflect the likelihood of each scenario being realized with s ∈ S ps = 1. 4.1.2. Expectation of the objective function To represent the different scenarios accounting for uncertainty in prices, the price-related random objective function coefficients comprising: (1) λi,t for the costs of different types of crude oil that can be handled by the crude distillation unit of a refinery and (2) γ i,t for the sales prices of the refined products, are added with an index s subscript, each with an associated probability ps . For ease of reference, both groups of price (or cost) parameters are redefined as the parameter ci,s,t or cicr ,s,t ; the difference between the two is in the use of the index icr (and the corresponding set of Icr ) to refer to “products” that are actually crude oils, as distinguished from the index i that is used to indicate the saleable products. Since the objective function given by Eq. (10) is linear, it is straightforward to show that the expectation of the random objective function with random price coefficients is given by: f − ˜ s ˜ i,t Ii,t i∈I s ∈ S ps ci,s,t Si,t + i∈Iγ i ∈ Icr s ∈ S ps cicr ,s,t Pt − i ∈ I λi,t Ii,t − i ∈ I hi,t Hi,t (11) E[z0 ] = − j ∈ J (αj,t CEj,t + βj,t yj,t ) + rt Rt + ot Ot t∈T Consideration of the expected value of profit alone as the objective function, which is characteristic of the classical stochastic linear programs introduced by [36,37], is obviously inappropriate for moderate and high-risk decisions under uncertainty since most decision makers are risk averse in facing important decisions. As stressed by [34], the expected value objective ignores both the risk attribute of the decision maker and the distribution of the objective values. Hence, variance of each of the random price coefficients can be adopted as a viable risk measure of the objective function, which is the second major component of the MV approach adopted in Risk Model I. 4.1.3. Variance of the objective function Variance for the expected value of the objective function (10) is derived as: 2 V (z0 ) = Si,t V (ci,s,t ) + Pi2 ,t V (ci ,s,t ) t∈T i∈I
(12)
t ∈ T i ∈ I
Since the above derivation does not explicitly evaluate variances of the random price coefficients as given by V(ci,s,t ) and V (ci ,s,t ), we consider the following alternative definition for variance from [23] that yields: V (z0 ) = ps1 (zs1 − E[z0 ])2 + ps2 (zs2 − E[z0 ])2 + · · · + psω (zsω − E[z0 ])2
(13)
The objective function for the stochastic model is now given by: max z1 = E[z0 ] − θ1 V (z0 ) s.t. constraints (1) − (8)
(14)
The model is subject to the same set of constraints as the deterministic model, with θ 1 as the risk tradeoff parameter (or simply termed as the risk factor) associated with risk reduction for the expected profit. θ 1 is varied over the entire range of (0, ∞) to generate a set of feasible decisions that have maximum return for a given level of risk, which is equivalent to the “efficient frontier” portfolios advocated by Markowitz [24] for investment applications. It is noteworthy that from a modelling approach perspective, θ 1 is also a scaling factor since the expectation operator and the variance are of different dimensions. If it is desirable to obtain a term that is dimensionally consistent with the expected value term, then the standard deviation of z0 may be considered, instead of the variance, as the risk measure (in which standard deviation is simply the square root of variance). As well, θ 1 represents the weight or weighting factor for the variance term in a multiobjective optimization setting that consists of the components mean and variance. However, the primary difficulty in executing model (14) is in determining a suitable set of values for θ 1 that caters to decision makers who are either risk-prone or risk-averse. An approach to circumvent this problem is available, as highlighted by [38–40], in which the variance (or the standard deviation) of the objective function is minimized as follows: max z1 = −V (z0 ) √ (or max z1 = − V (z0 ))
(15)
while adding the inequality constraint for the mean of the objective function that stipulates a certain target value for the desired profit to be achieved: E[z0 ] = Target profit
(16)
1748
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
Thus, the final form of Risk Model I is given by: max z1 = −V (z0 ) s.t. E[z0 ] ≥ Target profit
(RM1)
constraints (1) − (10) To determine a suitable range for the target profit value, a test value is assumed and the corresponding solution is computed. Then, the test value is increased or decreased, with the solution computed each time to investigate and establish the range of target values that ensures solution feasibility. 4.2. Approach 2: Expectation models I and II In Approach 2, the MV model developed in Approach 1 is incorporated within a two-stage SP with fixed recourse framework to handle randomness in the right-hand side (RHS) and left-hand side (LHS) coefficients of the related constraints. 4.2.1. Modelling demand uncertainty Uncertainty in market demand introduces randomness in constraints for production requirements of intermediates and saleable products as given by Eq. (4). The sampling methodology employed for scenario construction is similar to the case of price uncertainty in Approach 1, involving the generation of representative scenarios of demand uncertainty for N number of products with the associated probabilities that indicate their comparative frequency of occurrence. One of the main consequences of uncertainty within the context of decision-making is the possibility of infeasibility in the future. The two-stage recourse modelling framework provides the liberty of addressing this issue by postponing some decisions into the second stage; however, this comes at the expense of the use of corresponding penalties in the objective function [40]. Decisions that can be delayed until after information about the uncertain data is available almost definitely offer an opportunity to adjust and adapt to the new information received. There is generally value associated with delaying a decision, when it is possible to do so, until after additional information is obtained [41]. In devising the appropriate penalty functions, we resort to the introduction of compensating slack variables in the probabilistic constraints to eliminate the possibility of second-stage infeasibility [42]. Additionally, the recourse-based modelling philosophy requires the decision maker to impute a price as a penalty to remedial activities that are undertaken in response to uncertainty. For applications in production planning, these can be assumed as standard fixed costs. However, according to [40], under some circumstances, it may be more appropriate to accept the possibility of infeasibility, provided that the probability of this event is restricted below a given threshold. This is addressed in the subsequent approaches by appending an appropriate risk measure to the objective function. Compensating slack variables accounting for shortfall and/or surplus in production are introduced in the stochastic constraints with the following results: (1) inequality constraints are replaced with equality constraints; (2) numerical feasibility of the stochastic constraints can be ensured for all events; and (3) penalties for feasibility violations can be added to the objective function [42]. Since a probability can be assigned to each realization of the stochastic parameter vector (i.e., to each scenario), the probability of feasible operation can be measured. Assigning penalties to the objective function is adopted from the approach suggested by [43–46], in which a cost is assigned to the violation of any of the constraints. In this work, a non-negative second-stage recourse slack variable z+ i,s quantifies the shortfall in production, which is penalized in the objective function according to the cost of purchasing this makeup product from the open market. Likewise, for overproduction (surplus) with respect to market demands, the recourse slack variable z− i,s is penalized based on the inventory cost for storing the excess of production. The expected values of the recourse penalty costs of ci+ and ci− for infeasibility due to shortfall and surplus of production, respectively, are minimized in the objective function in an effort to maximize the expected profit. Thus, the expected recourse penalty for the second-stage costs due to uncertainty in the demand for product i for all considered scenarios is given by: − − Es,demand = ps (ci+ z+ (17) i,s + ci zi,s ) i∈Is∈S
To ensure that the original information structure associated with the decision process sequence is honoured, for each of the products whose demand is uncertain, the number of new constraints to be added to the stochastic model counterpart, in replacement of the original deterministic constraint, corresponds to the number of scenarios. Herein lies a demonstration of the fact that the size of a recourse model increases exponentially since the total number of scenarios grows exponentially with the number of random parameters [41]. In general, the new constraints take the form of: − Si,t + z+ i,s − zi,s = di,t,s ,
∀i ∈ IP , ∀t ∈ T, ∀s ∈ S
(18)
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
1749
4.2.2. Modelling yield uncertainty Uncertainty in product yields introduces randomness in the material balances that are given by Eq. (9). The scenario construction to model yield uncertainty of products k from material i is similar to the approach for modelling demand uncertainty. Note that in order to ensure that the material balances are satisfied, the summation of yields must be equal to unity. The non-negative second-stage + − recourse slack variables yi,k,s and yi,k,s represent shortage and excess in yields, respectively, with their corresponding fixed unit + − . Thus, the expected recourse penalty for the second-stage costs due to yield uncertainty recourse penalty costs given by qi,k and qi,k is: + + − − Es,yield = ps (qi,k yi,k,s + qi,k yi,k,s ) (19) i ∈ I s ∈ Sk ∈ K
Ns new constraints to represent the Ns number of scenarios dealing with yield uncertainty are introduced for each product whose yield is uncertain [41], with the general form of the new constraints given by: + − s f + bi,j xj,t + yi,k,s − yi,k,s − Si,t − Ii,t = 0, i ∈ I, k ∈ K, s ∈ S (20) Pt + Ii,t j∈J
Table 2 in the numerical example in Section 6 of this article presents the scenario formulation to model uncertainties in prices, demands, and yields simultaneously. The two major assumptions that enable the combination of the sub-scenarios for each of the uncertain parameters of prices, demands, and yields are that: (1) the uncertain parameters in each scenario are highly-correlated; and (2) each of the random variables (or equivalently, each of the scenarios) are assumed to be independent of one another. These assumptions lead to two implications: (1) they obviate the need to construct a joint probability distribution function (in the sampling methodology) that encompass scenarios of all the possible combinations of the three random variables (this means that, for instance, the possibility of a scenario in which prices are “average” with demand being “above average” and yield being “below average” is not considered); (2) the covariance term in the MV model becomes equal to variance [47]. The corresponding expected recourse penalty for the second-stage costs is given by: − − + + − − Es = Es,demand + Es,yield = ps [(ci+ z+ p s ξs (21) i,s + ci zi,s ) + (qi yi,s + qi yi,s )] = i∈Is∈S
i∈Is∈S
− − + + − − where ξi,s = (ci+ z+ i,s + ci zi,s ) + (qi yi,k,s + qi yi,k,s ). Thus, Expectation Model I is formulated as:
max z2 = z1 − Es = E[z0 ] − θ1 V (z0 ) − Es s.t.
(EM1)
deterministic and stochastic constraints (1) − (3), (6) − (8), (18), and (20)
As remarked in Approach 1, a potential complication with Expectation Model I lies in computing a suitable range of values for the operational risk factor θ 1 . Therefore, an alternative formulation of minimizing variance while adding a target profit constraint is employed for Expectation Model II: max z2 = −V (z0 ) − Es s.t.
E[z0 ] ≥ Target profit
(EM2)
deterministic and stochastic constraints (1) − (3), (6) − (8), (18), and (20) 4.3. Approach 3: Risk Model II The goal of Approach 3 is to append an operational risk term to the mean-risk model formulation in Approach 2 to account for the significance of both financial risk (as considered by Approach 1) as well as operational risk in decision-making. Variance for the various expected recourse penalty for the second-stage costs Vs is derived as: Vs =
ps (ξs − Es ) = 2
s∈S
⇒ Vs =
s∈S
s∈S
ps
ps ξs −
ps ξs
s ∈ S
− − (ci+ z+ i,s + ci zi,s ) i∈I
2
+ − + qi− yi,k,s ) +(qi+ yi,k,s
−
i ∈ I s ∈ S
ps
− − (ci+ z+ i,s + ci zi,s )
2
(22)
+ − − +(qi+ yi,k,s + qi yi,k,s )
Note that the index s and the corresponding set S is used to denote scenarios for the evaluation of the inner expectation term to distinguish them from the original index s used to represent the scenarios. Vs is weighted by the operational risk factor θ 2 ∈ (0, ∞).
1750
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
The formulation of Risk Model II is as follows: max z3 = z2 − θ2 Vs = E[z0 ] − θ1 V (z0 ) − Es − θ2 Vs s.t. deterministic and stochastic constraints (1) − (3), (7), (8), (9), (24), and (26)
(RM2)
4.4. Approach 4: Risk Model III In their pioneering work, Konno and Yamazaki [48] propose a large-scale portfolio optimization model based on mean-absolute deviation (MAD). This serves as an alternative measure of risk to the standard Markowitz’s MV approach, which models risk by the variance of the rate of return of a portfolio, leading to a nonlinear convex quadratic programming (QP) problem. Although both measures are almost equivalent from a mathematical point-of-view, they are substantially different computationally in a few perspectives as highlighted by [49,50]. In essence, the use of MAD is due to its computationally-attractive linear property [51] further demonstrated that MAD is an authentic measure of risk in view of its compatibility with von Neumann’s principle of maximization of expected utility (MEU) under risk aversion; a result corroborated by [52]. This substantiates the solid economic foundation of the theoretical properties of MAD [50]. Therefore, in this approach, we develop Risk Model III as a reformulation of Risk Model II by employing the mean-absolute deviation, in place of variance, as the measure of operational risk imposed by the recourse costs to handle the same three factors of uncertainty (prices, demands, and yields). To the best of our knowledge, this is the first such application of MAD, a widely-used metric in the area of system identification and process control, for risk management in refinery planning. The L1 risk of the absolute deviation function is given by [48]: ⎡ ⎤ ⎤ ⎡ n n (23) W(x) = E ⎣ Rj xj − E ⎣ Rj xj ⎦ ⎦ j=1 j=1 Thus, the corresponding mean-absolute deviation of the expected penalty costs is formulated as: Ws = ps |ξs − Es | = ps ξ s − ps ξs s∈S s ∈ S +s ∈+S + + (ci zi,s + ci− z− (ci zi,s + ci− z− i,s ) i,s ) − ⇒ Ws = ps p s + + − − + + − − +(qi yi,k,s + qi yi,k,s ) +(qi yi,s + qi yi,s ) i∈I i ∈ I s ∈ S s∈S
(24)
This nonlinear function can be linearized by implementing the transformation procedure outlined by [48] and revisited in [53], in which W must satisfy the following conditions: + + − − (ci+ z+ (ci zi,s + ci− z− i,s + ci zi,s ) i,s ) ps Ws ≥ − ps (25) − + + − − − +(qi+ yi,s + qi− yi,s +(qi+ yi,k,s ) + qi yi,k,s ) i∈I i ∈ I s ∈ S s∈S + + − − (ci+ z+ (ci zi,s + ci− z− i,s + ci zi,s ) i,s ) ps ps − (26) Ws ≥ + − + − − + qi− yi,s ) +(qi+ yi,k,s +(qi+ yi,s + qi yi,k,s ) i∈I i ∈ I s ∈ S s∈S and Ws ≥ 0
(27)
Similar to Risk Model II, the adoption of MAD is weighted by the operational risk factor θ 3 (0 < θ 3 < ∞) in Risk Model III, to give the following formulation: max z4 = z2 − θ3 Ws = E[z0 ] − θ1 V (z0 ) − Es − θ3 Ws s.t. deterministic and stochastic constraints (1) − (3), (6) − (8), (18), and (20) MAD linearization conditions (25) − (27)
(RM3)
5. Analysis of the computational results of the stochastic model formulations In the context of production planning, robustness can generally be defined as a measure of the resilience of the planning model to respond in the face of parameter uncertainty and unplanned disruptive events [54]. To investigate and interpret the behaviour and overall robustness of the proposed multiobjective optimization models in this work, we carry out a series of rigorous computational experiments to establish the effectiveness of the stochastic models in hedging against uncertainties posed by randomness in prices,
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
1751
demands, and yields. This is executed by adopting an analysis similar to the application of the robust optimization approach of [34] to the capacity expansion planning problems by [33,35] for an electrical power generation systems and a petrochemicals processing complex, respectively. Two performance metrics that have been previously utilitized in the optimization literature are considered to quantitatively measure and account for characteristics of planning under simultaneous uncertainty in prices, demands, and yields. The two metrics are: (1) the concepts of solution robustness and model robustness, again adopted from [34] and (2) the coefficient of variation Cv . Additionally, we consult many of the useful suggestions and guidelines that are offered in the classical paper by [55] on issues and techniques concerning test problems for computational experiments. 5.1. Solution robustness and model robustness It is desirable to demonstrate that proposed stochastic formulations provides a robust solution as well as a robust model. According to [34], a robust solution remains close to optimality for all scenarios of the input data while a robust model remains almost feasible for all the data of the scenarios. In refinery planning, model robustness or model feasibility is as essential as solution optimality. For example, in mitigating demand uncertainty, model feasibility is represented by an optimal solution that has almost no shortfalls or surpluses in production as reflected by the expected total unmet demand and total excess production, respectively, both of which should be kept to a minimum; in the former case, to gain customer demand satisfaction while in the latter, to improve inventory management. A tradeoff exists between solution optimality and model- and solution-robustness. To investigate these trends, the following parameters are analyzed from the raw computational results of the refinery production rates for the models: • • • •
the expected deviation in profit as measured by variance V(z0 ); the expected total unmet demand (i.e., production shortfall); the expected total excess production (i.e., production surplus); and the expected recourse penalty costs Es .
5.2. Coefficient of variation To interpret the solutions obtained from the stochastic models, we propose to investigate their corresponding coefficient of variation Cv . Cv for a set of values is defined as the ratio of the standard deviation to the expected value or mean and is usually expressed in percentage. It is calculated as: √ standard deviation V σ Cv = × 100% = × 100% = × 100% (28) mean μ E Statistically, Cv is a measure of reliability, or evaluated from the opposite but equivalent perspective, it is also indicative of the degree of uncertainty. It is alternatively interpreted as the inverse ratio of data to noise in the data in signal-processing-related applications. Thus, it is apparent that a small value of Cv is desirable as it signifies a small degree of noise or variability (e.g., in a data set) and hence, reflects low uncertainty. It follows that in the realm of stochastic optimization, Cv can be purposefully employed to investigate, denote, and compare and contrast the relative uncertainty in the models being studied. In a risk minimization model, as the expected value is reduced, the variability in the expected value (for example, as measured by variance or standard deviation) is reduced as well. The ratio of this change can be captured and described by Cv . Consequently, a comparison of the relative merit of models in terms of their robustness can be represented by their respective values of Cv , in the sense that a model with a lower Cv is favoured since there is less uncertainty associated with it, thus contributing to its reliability; this is in tandem with the original definition of Cv as a measure of reliability. In fact, Markowitz [23] advocates that the use of Cv as a measure of risk would equally ensure that the outcome of a decision-making process still lies in the set of efficient portfolios for the case of operational investments. In a data set of normally distributed demands, if the Cv of demand is given as a case problem parameter, the standard deviation is computed by the multiplication of Cv with the deterministic demand [56]. Hence, increasing values of Cv result in increasing fluctuations in the demand and this is again undesirable. Computation of Cv is based on the objective function of the formulated model. Table 1 displays the expressions to compute Cv for the proposed stochastic model formulations. Note that Cv for the deterministic case of each stochastic model should be equal to zero by virtue of its standard deviation assuming a value of zero since it is based on the expected value solution. 6. Numerical example We demonstrate the implementation of the proposed stochastic model formulations on the refinery planning linear programming (LP) model of [25,26]. The original single-objective LP model is first solved deterministically and is then reformulated with the addition of the stochastic dimension according to the four proposed formulations. The complete scenario representation of the prices, demands, and yields is provided in Table 2.
1752
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
Table 1 Determination of the coefficient of variation Cv for the deterministic and stochastic models Approach
Model
Deterministic
Objective function
Coefficient of variation Cv =
cT x
Cv = 0 √
1
Risk Model I
max z1 = E[z0 ] − θ 1 V(z0 ) or max z1 = −V(z0 )
2
Expectation Models I and II
I: max z2 = E[z0 ] − θ 1 V(z0 ) − Es II: max z2 = −V(z0 ) − Es
3
Risk Model II
max z3 = E[z0 ] − θ 1 V(z0 ) − Es − θ 2 Vs
4
Risk Model III (MAD)
max z4 = E[z0 ] − θ 1 V(z0 ) − Es − θ 3 Ws
√ σ V = μ E
V (z0 ) E[z √ 0] V (z0 ) Cv = E[z0 ] − Es √ V (z0 ) + Vs Cv = √E[z0 ] − Es V (z0 ) + W(ps ) Cv = E[z0 ] − Es Cv =
The deterministic objective function of the LP model is given by: maximize z = −8.0x1 + 18.5x2 + 8.0x3 + 12.5x4 + 14.5x5 + 6.0x6 − 1.5x14
(29)
in which the negative coefficients denote the purchasing and operating costs while the positive coefficients are the sales prices of products. If c is the row vector of the price (or cost) and x is the column vector of production flowrate, then the objective function can be generally written as: T random z=c x= ci,s xi , i = {1, 2, 3, 4, 5, 6, 14} ∈ Iprice ⊆ I, s = {1, 2, 3} ∈ S (30) s∈S
i∈I
Table 2 Complete scenario formulation for the refinery production planning problem under uncertainty in commodity prices, market demands for products, and product yields Product type (i)
Price uncertainty Crude oil (1) Gasoline (2) Naphtha (3) Jet fuel (4) Heating oil (5) Fuel oil (6) Cracker feed (14)
Variance of Price V(ci,s ) (($/t)2 )
Objective function coefficient of price, ci,s ($/t) Scenario 1 (above average)
Scenario 2 (average (expected value/mean))
Scenario 3 (below average)
−8.8 20.35 8.8 13.75 15.95 6.6 −1.65
−8.0 18.5 8.0 12.5 14.5 6.0 −1.5
−7.2 16.65 7.2 11.25 13.05 5.4 −1.35
Right-hand-side coefficient of constraints for production requirement (t/day)
Penalty cost incurred per unit ($/t) + Shortfall in Production (ci,s )
Demand uncertainty Gasoline (2) Naphtha (3) Jet fuel (4) Heating oil (5) Fuel oil (6)
2835 1155 2415 1785 9975
2700 1100 2300 1700 9500
2565 1045 2185 1615 9025
25 17 5 6 10
Left-hand side coefficient of mass balances for fixed yields (unit less)
Yield uncertainty Naphtha (7) Jet fuel (4) Gas oil (8) Cracker feed (9) Residuum (10) Probability ps
−0.1365 −0.1575 −0.231 −0.21 −0.265 0.35
−0.13 −0.15 −0.22 −0.20 −0.30 0.45
0.352 1.882 375 0.352 0.859 375 1.156 375 0.198 0.012 375
− Surplus in Production (ci,s )
20 13 4 5 8 Penalty cost incurred per unit ($/U)
−0.1235 −0.1425 −0.209 −0.19 −0.335 0.20
+ Yield decrement (qi,k,s )
− Yield increment (qi,k,s )
5 5 5 5 5
3 4 3 3 3
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
1753
6.1. Approach 1: Risk Model I Since our main objective is to demonstrate the methodological validity of the proposed mathematical programming tool without claiming that the model captures all detailed aspects of the problem, three coefficients of variation depicting three corresponding scenarios are considered to be representative of the uncertainty in the objective coefficients of prices, based on historical data. The three scenarios represent: (1) the “above average” or optimistic scenario denoting a representative 10 percent positive deviation from the mean value; (2) the “average” or realistic scenario that takes on the expected values or mean; and (3) the “below average” or pessimistic scenario, denoting a representative 10 percent negative deviation from the mean value. The formulation for Risk Model I is as follows: 2 2 max z1 = −V (z0 ) = ps (ci,s − c¯ i,s ) xi s∈S i∈I ps ci,s xi ≥ Target profit value s.t. E[z0 ] = s∈S
(RM1 )
i∈I
random ⊆ I, s = {1, 2, 3} ∈ S i = {1, 2, 3, 4, 5, 6, 14} ∈ Iprice
deterministic constraints (first stage) in the LP model. As the main focus of this paper is on the risk-incorporated models of Risk Models II and III, the computational results for Risk Model I is not presented here. 6.2. Approach 2: The expectation models I and II For simplicity of demonstration of this stochastic approach, it is assumed that there is no alternative source of production; hence, if there is a shortfall in production, the demand is actually lost. Thus, the corresponding model considers the case where the in-house production of the refinery has to be anticipated at the beginning of the planning horizon, that is, the production variables x are fixed, no vendor production is allowed, and all unmet demand is lost. A five (5) percent standard deviation from the mean value of market demand for the saleable products in the LP model is assumed to be reasonable based on statistical analyses of the available historical data. To be consistent, the three scenarios assumed for price uncertainty with their corresponding probabilities are similarly applied to describe uncertainty in the product demands as shown in Table 2, alongside the corresponding penalty costs incurred due to the unit production shortfalls or surpluses for these products. To ensure that the original information structure associated with the decision process sequence is honoured, three new constraints to model the corresponding three scenarios generated for each product with uncertain demand are added to the stochastic model in place of the original deterministic constraint. Altogether, this sums up to 3 × 5 = 15 new constraints in place of the five constraints in the LP model. On the other hand, to be consistent with the cases of prices and demands uncertainty, three representative scenarios are also considered for modelling yield uncertainty for the LHS coefficient of fixed yields from the primary distillation unit (PDU) in the LP model. Each scenario corresponds to the depiction of “average product yield”, “above average product yield”, and “below average product yield”, with a 5% deviation from the mean value of yield assumed to be reasonable based on the available historical data. To ensure satisfaction of the material balances, yields for the bottom product of PDU, i.e., the residuum (or residual) is determined by subtracting the summation of yields for the other four products from unity. This does not distort the physics of the problem as the yield of residuum is relatively negligible anyway in a typical atmospheric distillation unit. The penalty costs incurred per unit of shortages or excesses of crude oil yields are also shown in Table 2. Thus, Expectation Model I is formulated as follows: + + ) (ci zi,s + ci− z− 2 i,s 2 ¯ i,s ) xi − max z2 = ps Ci,s xi − θ1 ps (Ci,s − C ps + − +(qi+ yi,s + qi− yi,s ) i∈Is∈S i∈Is∈S s∈S i∈I s.t.
deterministic constraints (first stage) in the LP model,
(EM1 )
stochastic constraints (second stage): − xi + z+ i,s − zi,s = di,s ,
∀i ∈ I, ∀s ∈ S
+ − Ti x1 + xi + yi,k,s − yi,k,s = 0,
i ∈ I, k ∈ K, s ∈ S
random i = {1, 2, 3, 4, 5, 6, 14} ∈ Iprice ⊆ I, s = {1, 2, 3} ∈ S.
(31) (32)
1754
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
The alternative Expectation Model II is expressed as the following: 2 − − + + − − 2 ¯ i,s ) xi − ps (Ci,s − C ps [(ci+ z+ max z2 = − i,s + ci zi,s ) + (qi yi,s + qi yi,s )] i∈I
s∈S
s.t.
E[z0 ] =
s∈S
i∈Is∈S
ps Ci,s xi
≥ target profit value,
(EM2 )
i∈I
deterministic constraints in the LP model, stochastic constraints (31) and (32) random ⊆ I, s = {1, 2, 3} ∈ S. i = {1, 2, 3, 4, 5, 6, 14} ∈ Iprice
As in the case of Risk Model I, the computational results for Expectation Models I and II are not presented here as the emphasis of this paper is on the risk-based models of Risk Models II and III. It is noteworthy that although theoretically, the recourse model solution is, in general, likely to be more representative and more robust with more scenarios considered, this consideration is at the expense of being computationally expensive due to the “curse of dimensionality” caused by an exponential growth in the problem size as previously emphasized. Furthermore, with a substantially large number of scenarios taken into account (for instance [28] considers 600 scenarios in their model), more “noise” or disturbances are likely to be present in the data. Additionally, employing penalty functions in modelling violations of constraints is also largely restricted in that many new non-negative compensating slack variables need to be added. 6.3. Computational results and discussion for approach 3: Risk Model II The formulation of Risk Model II for the numerical example is given by the following: + + (ci zi,s + ci− z− i,s ) 2 max z3 = ps Ci xi − θ1 xi V (Ci ) − ps + − + qi− yi,s ) +(qi+ yi,s i∈Is∈S i∈I i∈Is∈S ⎫2 ⎧ + − ⎪ ⎪ (ci+ z+ + ci− z− ) + (qi+ yi,k,s + qi− yi,k,s ) ⎪ ⎪ i,s i,s ⎬ ⎨ i∈I −θ2 ps − − + + − − ⎪ ⎪− ps [(ci+ z+ s∈S ⎪ ⎭ ⎩ i,s + ci zi,s ) + (qi yi,k,s + qi yi,k,s )] ⎪
(RM2 )
i ∈ I s ∈ S
s.t. deterministic constraints in the LP model, stochastic constraints (31) and (32), random i = {1, 2, 3, 4, 5, 6, 14} ∈ Iprice ⊆ I, s = {1, 2, 3} ∈ S.
Tables 3–5 tabulate the computational results for the implementation of Risk Model II on GAMS [57] over a range of values of the operational risk parameter θ 2 with respect to the recourse penalty costs, for three representative cases of θ 1 = 1E–10, 1E–7, and 1.55E–5, respectively. An example of the detailed results is presented in Table 6 for θ 2 = 50 of the first case. Starting values of the first-stage deterministic decision variables have been initialized to the optimal solutions of the deterministic model. Fig. 2 depicts the corresponding efficient frontier plot for Risk Model II while Fig. 3 provides the plot of the expected profit for different levels of risk. A number of different parameters are of interest in observing the trends that contribute to robustness in both the model and the computed solution. Fig. 3 shows that smaller values of θ 1 correspond to higher expected profit. With increasingly larger θ 1 , the declining expected profit becomes almost constant as it plateaus at the value of $81,770. The converse is true as well in which increasingly smaller θ 1 result in rising expected profit that eventually becomes roughly constant at the value of $79,730. Although the pair of increasing θ 2 with fixed value of θ 1 corresponds to decreasing expected profit, it generally leads to a reduction in expected production shortfalls and surpluses as well, thus reflecting high model feasibility. Therefore, a suitable operating range of θ 2 values ought to be selected to tradeoff for achieving optimality between expected profit and expected production feasibility. Increasing θ 2 also reduces the expected deviation in the recourse penalty costs under different realized scenarios. This in turn translates to increased solution robustness. Hence, it primarily depends on the policy adopted by the decision maker, as characterized by the values of the factors θ 1 and θ 2 chosen, in reflecting whether these tradeoffs are acceptable based on the desired degree of model robustness and solution robustness, as advocated by [35]. In general, the coefficients of variation decrease with smaller values of θ 2 . This is definitely a desirable behaviour since it indicates that for higher expected profits, there is diminishing uncertainty in the model, thus signifying model and solution robustness. It is
Operational risk factor θ 2
1E−6 1E−4 1E−3 1E−2 1E−1 1 5 10 50 100 1000 5000 10000 10500 10716 10717
Optimal objective value 27710 26900 25670 25510 25490 25490 25490 25490 25490 25490 25490 25490 25490 25490 25490
Expected variation in profit V(z0 )(E+8)
Expected total unmet demand/ production shortfall
1.787 1.787 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782
2575 2575 2467 2539 2518 2385 2508 2566 2556 2519 2519 2385 2542 2409 2508
Expected total excess production/production surplus
Expected recourse penalty costs Es
Expected variation σ = in recourse penalty costs Vs
√ V (z0 ) + Vs
Expected profit E[z0 ]
26820 54060 8.167E+6 13670 81770 26820 54060 8.167E+6 13670 81770 26710 53840 2.047E+5 13360 79740 26780 54200 2047 13350 79730 26760 54230 20.47 13350 79730 26600 54230 0.205 13350 79730 26750 54230 0.008 13350 79730 26810 54230 0.002 13350 79730 26800 54230 8.189E−5 13350 79730 26800 54230 2.047E−5 13350 79730 26800 54240 0 13350 79730 26630 54240 0 13350 79730 26780 54240 0 13350 79730 26650 54240 0 13350 79730 26750 54240 0 13350 79730 (GAMS output: Infeasible solution. There are no superbasic variables.)
Note: Trial solutions for θ 1 < 0.000001 are not considered since improvement in expected profit is not anticipated based on trends in computed values.
μ = E[z0 ] − Es
Cv =
σ μ
Stochastic Deterministic 27720 27720 25900 25530 25490 25490 25490 25490 25490 25490 25490 25490 25490 25490 25490
0.4932 0.4932 0.5157 0.5228 0.5236 0.5236 0.5237 0.5237 0.5237 0.5237 0.5237 0.5237 0.5237 0.5237 0.5237
(infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible)
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
Table 3 Representative computational results for Risk Model II for θ 1 = 1E−10
1755
1756
Operational risk factor θ 2
1E−6 1E−4 1E−3 1E−2 1E−1 1 5 10 50 1E2 1E3 5E3 1E4 1.05E4 10716 10717
Optimal objective value 27690 26880 25680 25490 25470 25470 25470 25470 25470 25470 25470 25470 25470 25470 25470
Expected variation in profit V(z0 )(E+8)
Expected total unmet demand/production shortfall
1.787 1.787 1.787 1.787 1.787 1.787 1.787 17.82 17.82 17.82 17.82 17.82 17.82 17.82 17.82
2575 2575 2467 2565 2385 2456 2519 2508 2385 2385 2508 2456 2475 2409 2584
Expected total excess production/production surplus
Expected recourse penalty costs Es
Expected variation in recourse penalty costs Vs
σ=
√ V (z0 ) + Vs
Expected profit E[z0 ]
26820 54060 8.167E+6 13670 81770 26820 54060 8.167E+6 13670 81770 26710 53840 2.047E+5 13360 79740 26800 54200 2047 13350 79730 26620 54230 20.47 13350 79730 26700 54230 0.205 13350 79730 26760 54230 0.008 13350 79730 26750 54230 0.002 13350 79730 26630 54230 8.189E−5 13350 79730 26630 54230 2.047E−5 13350 79730 26750 54240 0 13350 79730 26700 54240 0 13350 79730 26720 54240 0 13350 79730 26650 54240 0 13350 79730 26820 54240 0 13350 79730 (GAMS output: Infeasible solution. There are no superbasic variables.)
Note: Trial solutions for θ 1 < 0.000001 are not considered since improvement in expected profit is not anticipated based on trends in computed values.
μ = E[z0 ] − Es
27720 27720 25900 25530 25490 25490 25490 25490 25490 25490 25490 25490 25490 25490 25490
Cv =
σ μ
Stochastic
Deterministic
0.4932 0.4932 0.5157 0.5228 0.5236 0.5236 0.5237 0.5237 0.5237 0.5237 0.5237 0.5237 0.5237 0.5237 0.5237
(infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible)
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
Table 4 Representative computational results for Risk Model II for θ 1 = 1E−7
Operational risk factor θ 2
1E−6 1E−4 2E−4 5E−4 1E−3 1E−2 1E−1 1 10 50 1E2 5E2 1E3 5E3 1E4 1.2E4 1.25E4
Optimal objective value 24940 24130 23600 23140 22930 22750 22730 22730 22730 22730 22730 22730 22730 22730 22730 22730
Expected variation in profit V(z0 )(E+8)
Expected total unmet demand/production shortfall
1.787 1.787 1.784 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782 1.782
2575 2575 2435 2363 2493 2524 2531 2579 2580 2409 2591 2551 2532 2409 2561 2597
Expected total excess production/production surplus
Expected recourse penalty costs Es
Expected variation in recourse penalty costs Vs
σ=
√ V (z0 ) + Vs
Expected profit E[z0 ]
μ = E[z0 ]−Es
26820 54060 8.167E+6 13670 81770 27720 26820 54060 8.167E+6 13670 81770 27720 26670 53590 3.010E+6 13470 80570 26970 26600 53450 8.187E+5 13380 79760 26310 26730 53840 2.047E+5 13360 79740 25900 26760 54200 2047 13350 79730 25530 26770 54230 20.470 13350 79730 25490 26820 54230 0.205 13350 79730 25490 26820 54230 0.002 13350 79730 25490 26650 54230 8.187E−5 13350 79730 25490 26830 54230 2.047E−5 13350 79730 25490 26790 54230 0 13350 79730 25490 26770 54240 0 13350 79730 25490 26650 54240 0 13350 79730 25490 26800 54240 0 13350 79730 25490 26840 54240 0 13350 79730 25490 (GAMS output: Infeasible solution. A free variable exceeds the allowable range.)
Note: Trial solutions for θ 1 < 0.000001 are not considered since improvement in expected profit is not anticipated based on trends in computed values.
Cv =
σ μ
Stochastic
Deterministic
0.4932 0.4932 0.4993 0.5085 0.5157 0.5228 0.5236 0.5236 0.5237 0.5237 0.5237 0.5237 0.5237 0.5237 0.5237 0.5237
(infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible)
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
Table 5 Representative computational results for Risk Model II for θ 1 = 1.55E−5
1757
1758
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
Table 6 Detailed computational results for Risk Model II for θ 1 = 1E−10, θ 2 = 50 First-stage variable
Stochastic solution
Product (i)
− Production shortfall z+ ij or surplus zij (t/d)
Scenario 1 z+ i1
Scenario 2 z− i1
z− i2
z+ i3
z− i3
700.0 0 0 0 0
0 55.00 1338 1898 237.5
565.0 0 0 0 0
0 110.0 1453 1983 712.5
205.0 1388 1335 1350 975.0
0 0 0 0 0
302.5 1500 1500 1500 450.0
x1 x2 x3 x4 x5 x6 x7
15000 2000 1155 3638 3598 9738 2155
x8
4635
Production yields LHS coefficients randomness
x9 x10 x11 x12 x13 x14
4350 5475 1000 2698 1937 2500
Naphtha (7) Jet Fuel (4) Gas Oil (8) Cracker Feed (9) Residuum (10)
0 0 0 0 0
107.5 1275 1170 1200 1500
0 0 0 0 0
x15 x16 x17 x18 x19 x20
1850 1000 1375 899.4 475.6 125.0
E (penalty costs) Etotal
18980 54230
24410
10850
Expected profit z ($/day)
Demands RHS coefficients randomness Gasoline (2) 835.0 0 Naphtha (3) 0 0 Jet Fuel (4) 0 1223 Heating Oil (5) 0 1813 Fuel Oil (6) 237.5 0
Scenario 3
z+ i2
79730
also observed that for values of θ 2 approximately greater than or equal to 2, the coefficient of variation remain at a static value of 0.5237, therefore indicating overall stability and minimal degree of uncertainty in the model. Note also that the computational times for the model solutions are not reported since current technological prowess has enabled solutions that are within reasonably short times for the problem size handled in the numerical study. 6.3.1. Limitations of approach 3 Since variance is a symmetric risk measure, profits both below and above the target levels (or in general, the downside risk and upside variations, respectively) are penalized equally, when it is actually desirable to only penalize the former instance [48,58,59]. In other words, constraining or minimizing the variance of key performance metrics to achieve robustness, which in this case, are the profit and the recourse penalty costs, may result in models that overcompensate for uncertainty, as highlighted by [60].
Fig. 2. The efficient frontier plot of expected profit vs. both economic and operational risks (as measured by standard deviation) due to variations in profit and recourse penalty costs, respectively, for Risk Model II.
Risk factors θ1 , θ3
0, 0 1E−6, 1E−4 1E−5, 1E−3 1E−4, 1E−3 1E−4, 1E−2 2E−4, 1E−2 3E−4, 1E−2 5E−4, 1E−2 6E−4, 1E−2 7E−4, 1E−2 8E−4, 1E−2 9E−4, 1E−2 1E−3, 1E−2 1E−2, 0.1 2E−2, 0.1 1, 1 40, 20 1E2, 10 1E3, 1E2 1E4, 1E2
Optimal objective value 27720 27580 26380 14360 14340 2802 −6832 −25310 −34150 −42800 −49890 −56170 −62410 −62500 −1248000 −6.237E7 −2.495E9 −6.237E9 −6.237E10
V (z0 ) +W(ps )
Expected profit E[z0 ]a
Expected total unmet demand/ production shortfall
13.35 13.35 13.35 13.35 13.35 9.634 9.634 8.837 8.837 7.966 6.392 6.237 6.237 6.237 6.237 6.237 6.237 6.237 6.237
2575 26820 54060 2618 11560 81770 2575 26820 54060 2618 11560 81770 2575 26820 54060 2618 11560 81770 2575 26820 54060 2618 11560 81770 2575 26820 54060 2618 11560 81770 2575 17800 40230 2628 9815 62330 2575 17800 40230 2628 9815 62330 3610 16370 42200 4253 9400 61110 3610 16370 42200 4253 9400 61110 5556 15470 47550 4256 8926 60560 9424 13670 58200 4262 7995 59450 9836 13470 59330 4263 7898 59330 9836 13470 59330 4263 7898 59330 9836 13470 59330 4263 7898 59330 9836 13470 59330 4263 7898 59330 9836 13470 59330 4263 7898 59330 9836 13470 59330 4263 7898 59330 9836 13470 59330 4263 7898 59330 9836 13470 59330 4263 7898 59330 (GAMS output: Infeasible solution. A free variable exceeds the allowable range.)
Expected total excess production/production surplus
Expected recourse penalty costs Es
Deviation in recourse penalty costs W(ps )
σ=
Expected deviation between profit V(z0 )(E+7)
μ = |E[z0 ] − Es |
27720 27720 27720 27720 27720 22100 22100 18920 18920 13000 1252 0 0 0 0 0 0 0 0
Cv =
σ μ
Stochastic
Deterministic
0.4169 0.4169 0.4169 0.4169 0.4169 0.4442 0.4442 0.4970 0.4970 0.6863 6.387 →∞ →∞ →∞ →∞ →∞ →∞ →∞ →∞
(infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible) (infeasible)
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
Table 7 Representative computational results for Risk Model III for a selected representative range of values of θ 1 and θ 3
1759
1760
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
Fig. 3. Plot of expected profit for different levels of risk as represented by the economic risk factor θ 1 and the operational risk factor θ 2 (with θ 1 and θ 2 on logarithmic scales due to wide range of values) by employing variance as the risk measure for Risk Model II.
Moreover [32] emphasizes that another problem with employing variance as a risk measure is that the points on the MV efficient frontier may be stochastically dominated by other feasible solutions. A solution xI is stochastically dominated by another solution xII if for every scenario, the profit generated by xII is at least as large as the profit given by xI and yields a strictly greater profit for at least one scenario; a condition known as Pareto optimality in the multiobjective optimization literature [58]. This observation has been confirmed, among others, by [61], which explains that this trend is because the mean–risk approach, in general, may produce inferior conclusions when typical dispersion statistics (such as variance) are employed as the risk measure. These authors further advocate the crucial importance of constructing mean-risk models that are in agreement with stochastic dominance relations since the latter is a fundamental concept in decision theory and economics, as highlighted by [62]. Additionally, incorporating variance of the recourse function as part of the objective function could potentially cause the problem to lose convexity. Thus, without employing the techniques of global optimization, one is liable to get trapped in local optima. In some models, incorporating the variance of the recourse function into the objective function leads to poor design or planning in the
Fig. 4. The efficient frontier plot of expected profit versus both the economic and operational risks imposed by variations in profit (measured by variance) and the recourse penalty costs (measured by mean-absolute deviation), respectively, for Risk Model III.
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
1761
sense that the design or plan is unnecessarily expensive even though the variance is small [63]. Furthermore [40] remarks that this approach has its roots in the Markowitz’s model that is based on assumptions such as normally distributed returns, which may not necessarily hold in some (or possibly most) applications. To handle these shortcomings [58] proposes a multiobjective optimization approach of simultaneous profit maximization and risk minimization for each profit target, which has been extensively applied in subsequent work by Bagajewicz and his co-workers (for examples, see [28,64,65]). On the other hand [32] accounts for the expected downside risk in solving a real-world problem in the automobile industry. In this approach, a decision maker sets a target value for the desired profit; hence, the risk associated with a decision is measured by the failure to meet the target profit. A more general approach to ensure robustness is proposed by [60], which can be tailored to various types of constraints to be imposed on the system and on specific suitable performance metrics. Other potentially more representative risk measures should also be considered with [66] providing a recent review of a wide choice of risk measures applicable within a two-stage stochastic programming framework. In this work, we choose to adopt the mean-absolute deviation (MAD) as the measure of operational risk imposed by the modelling approach of recourse costs for a novel refining industry application in an effort to remedy the disadvantages of using variance.
6.4. Computational results and discussion for approach 4: Risk Model III Risk Model III for the numerical example is formulated as follows: max z4 = E(z0 ) − θ1 V (z0 ) −
− − + + − − ps [(ci+ z+ is + ci zis ) + (qi yis + qi yis )]
∈Is∈S i+ + + − − (ci z+ (ci zi,s + ci− z− i,s + ci zi,s ) i,s ) ps p − − θ3 s + − + − − + qi− yi,s ) +(qi+ yi,k,s +(qi+ yi,s + qi yi,k,s ) s∈S
i∈I
i∈Is ∈S
s.t. deterministic constraints in the LP model, stochastic constraints (31) and (32), MAD linearization conditions (25)–(27) From Table 7 and the corresponding efficient frontier plot in Fig. 4, similar trends with Risk Model II (and also the expected value models) are observed in which decreasing values of θ 1 corresponds to higher expected profit until a certain constant profit value is attained ($81,770). The converse is also true in which a constant profit of $59,330 is reached in the initially declining expected profit for increasing values of θ 1 . One of the reasons the pair of decreasing values of θ 1 with fixed value of θ 3 leads to increasing expected profit is attributable to the general decrement in production shortfalls but increasing production surpluses, in which the fixed penalty cost for the latter is lower than the former (as is typically the case). It certainly augurs well to select a lower operating value of θ 1 to achieve both high model feasibility as well as increased profit. Moreover, lower values of θ 1 correspond to decreasing variation in the recourse penalty costs, which implies solution robustness. This argument is further strengthened by the reduced values of the coefficient of variation, which indicates less uncertainty in the model on a whole. In general, this again proves that a proper selection of the operating range of θ 1 and θ 3 is crucial in varying the tradeoffs between the desired degree of model robustness and solution robustness, to ultimately obtain optimality between expected profit and expected production feasibility (see Table 7 to validate these observations and inductions).
7. Conclusions In this work, we focus on a systematic methodology for developing explicit yet robust stochastic programming models for midterm refinery planning problems by simultaneously accounting for uncertainties in commodity prices, product demands, and product yields. In addition, we incorporate the importance of economic and operational risk in decision-making under uncertainty in the proposed models by explicitly considering the tradeoffs between expected profit and variation in both profit and penalty costs arising from recourse actions. The concept of mean-absolute deviation is advocated as a risk management tool for refinery planning in comparison with the theoretically- and computationally-inferior traditional risk metric of variance.
Acknowledgements The first author is grateful to Universiti Teknologi PETRONAS for financial support in funding his graduate studies at the University of Waterloo, during which this project was undertaken. The second author would like to thank NSERC (Canada) for funding assistance.
1762
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
Appendix A
Sets and indices set of materials or products i I J set of processes j K set of products with yield uncertainty k S set of scenarios s T set of time periods t Deterministic parameters bi,j stoichiometric coefficient for material i in process j Cj,t operating cost of process j in period t L , dU di,t,s , di,t,s i,t,s demand for product i in period t per realization of scenario s, with its corresponding constant lower (superscript L) and upper (superscript U) bounds fmin , I fmax minimum and maximum required amount of inventory for material i at the end of period t Ii,t i,t hi,t unit cost of subcontracting or outsourcing the production of product type i in period t amount of crude oil purchased in period t Pt pLt , pU lower and upper bounds of the availability of crude oil during period t t rt , ot cost per man-hour of regular and overtime labour in period t, respectively αj,t variable-size cost coefficient for the investment cost of capacity expansion of process j in period t βj,t fixed-cost charge for the investment cost of capacity expansion of process j in period t γ i,t unit sales price of product type i in period t γ˜ i,t value of the final inventory of material i in period t unit purchase price of crude oil type i in period t λi,t λ˜ i,t value of the starting inventory of material i in period t θ 1 , θ 2 , θ 3 risk factors or weighting factors (weights) for multiobjective optimization procedure Stochastic parameters demand for product i in time period t per realization of scenario s di,s,t probability of scenario s ps γ i,s,t unit sales price of product type i in period t per realization of scenario s λt unit purchase price of crude oil in period t per realization of scenario s Recourse parameters fixed penalty cost per unit demand di,s of underproduction (shortfall) of product i per realization of scenario s (also the cost ci+ of lost demand) ci− fixed penalty cost per unit demand di,s of overproduction (surplus) of product i per realization of scenario s (also the cost of inventory to store production surplus) + qi,k fixed unit penalty cost for shortage in yields from material i for product k − fixed unit penalty cost for excess in yields from material i for product k qi,k Deterministic variables (first-stage decision variables) L , CEU capacity expansion of the plant for process j that is installed in period t, with its corresponding constant lower CEj,t , CEj,t j,t (superscript L) and upper (superscript U) bounds Hi,t amount of product type i to be subcontracted or outsourced in period t s ,I f Ii,t i,t initial and final amount of inventory of material i in period t Li,t amount of lost demand for product i in period t Pt amount of crude oil purchased in period t Rt , Ot regular and overtime working or production hours in period t, respectively Si,t amount of product i sold in period t xj,t production capacity of process j during period t xj,t−1 production capacity of process j during period t − 1 yj,t binary decision variable that equals one (1) if there is an expansion for process j at the beginning of period t, and zero (0) otherwise
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
1763
Stochastic recourse variables (second-stage decision variables) + yi,k,s amount of shortage in yields from material i for product type k per realization of scenario s − amount of excess in yields from material i for product type k per realization of scenario s yi,k,s + amount of unsatisfied demand for product i due to underproduction per realization of scenario s zi,s amount of excess product i due to overproduction per realization of scenario s z− i,s References [1] J.M. Mulvey, D.P. Rosenbaum, B. Shetty, Strategic operational risk management and operations research, Eur. J. Operat. Res. 97 (1997) 1–16. [2] S.A. Zenios, Operational Optimization, Cambridge University Press, Cambridge, 1993. [3] G.E. Applequist, O. Samikoglu, J. Pekny, G.V. Reklaitis, Issues in the use, design, and evolution of process scheduling and planning systems, ISA Trans. 36 (2) (1997) 81–121. [4] N. Shah, Single- and multisite planning and scheduling: current status and future challenges, in: J.F. Pekny, G.E. Blau (Eds.), AIChE Symposium Series: Proceedings of the Third International Conference on Foundations of Computer-Aided Process Operations, vol. 320(94), Snowbird, Utah, USA, July 5–10, 1998, American Institute of Chemical Engineers (AIChE), New York, 1998, pp. 75–90. [5] L. Cheng, E. Subrahmanian, A.W. Westerberg, Multiobjective decision processes under uncertainty: applications, problem formulations, and solution strategies, Ind. Eng. Chem. Res. 44 (2005) 2405–2415. [6] C.A. M´endez, J. Cerda, I.E. Grossmann, I. Harjunkoski, M. Fahl, State-of-the-art review of optimization methods for short-term scheduling of batch processes, Comput. Chem. Eng. 30 (2006) 913–946. [7] A. Gupta, C.D. Maranas, A hierarchical Lagrangean relaxation procedure for solving midterm planning problems, Ind. Eng. Chem. Res. 38 (1999) 1937–1947. [8] C.J. Vidal, M. Goetschalckx, Strategic production–distribution models: a critical review with emphasis on global supply chain models, Eur. J. Operat. Res. 98 (1997) 1–18. [9] N.V. Sahinidis, I.E. Grossmann, R.E. Fornari, M. Chathrathi, Optimization model for long range planning in chemical industry, Comput. Chem. Eng. 13 (9) (1989) 1049–1063. [10] L.C. Norton, I.E. Grossmann, Strategic planning model for complete process flexibility, Ind. Eng. Chem. Res. 33 (1994) 69–76. [11] C.M. McDonald, I.A. Karimi, Planning and scheduling of parallel semicontinuous processes. 1. Production planning, Ind. Eng. Chem. Res. 36 (1997) 2691–2700. [12] A. Gupta, C.D. Maranas, A two-stage modeling and solution framework for multisite midterm planning under demand uncertainty, Ind. Eng. Chem. Res. 39 (2000) 3799–3813. [13] N. Shah, C.C. Pantelides, R.W.H. Sargent, A general algorithm for short-term scheduling of batch operations—II. Computational issues, Comput. Chem. Eng. 17 (2) (1993) 229–244. [14] I.A. Karimi, C.M. McDonald, Planning and scheduling of parallel semicontinuous processes. 2. Short term scheduling, Ind. Eng. Chem. Res. 36 (1997) 2701–2714. [15] D.E. Shobrys, D.C. White, Planning, scheduling and control systems: why can they not work together, Comput. Chem. Eng. 24 (2000) 163–173. [16] J. Acevedo, E.N. Pistikopoulos, Stochastic optimization based algorithms for process synthesis under uncertainty, Comput. Chem. Eng. 22 (1998) 647–671. [17] I.E. Grossmann, K.P. Halemane, R.E. Swaney, Optimization strategies for flexible chemical processes, Comput. Chem. Eng. 7 (4) (1983) 439–462. [18] W.R. Pulleyblank, Mathematical sciences in the nineties, IBM J. Res. Dev. 47 (1) (2003) 89–96. [19] A. Ruszczy´nski, A. Shapiro, Optimization of risk measures, in: G. Calafiore, F. Dabbene (Eds.), Probabilistic and Randomized Methods for Design under Uncertainty, Springer-Verlag London, London, UK, 2006. [20] W. Ogryczak, A. Ruszczy´nski, On consistency of stochastic dominance and mean–semideviation models, Math. Program. B 89 (2001) 217–232. [21] C. Swamy, D.B. Shmoys, Algorithms column: approximation algorithms for two-stage stochastic optimization problems, ACM SIGACT News 37 (1) (2006) 1–16. [22] G.V. Reklaitis, Review of scheduling of process operations, in: AIChE Symposium Series: Selected Topics on Computer-Aided Process Design and Analysis, vol. 214(78), 1982, pp. 119–133. [23] H.M. Markowitz, Portfolio selection, J. Finance 7 (1) (1952) 77–91. [24] H.M. Markowitz, Portfolio Selection: Efficient Diversification of Investments, Coyles Foundation Monograph. New Haven & London: Yale University Press, 1959 (this has more recently been reprinted as: H.M. Markowitz, Portfolio Selection. Blackwell Publishers Inc., Oxford, UK, 1991). [25] V. Ravi, P.J. Reddy, Fuzzy linear fractional goal programming applied to refinery operations planning, Fuzzy Sets Syst. 96 (1998) 173–182. [26] D.H. Allen, Linear programming models for plant operations planning, Br. Chem. Eng. 16 (1971) 685–691. [27] M.G. Ierapetritou, E.N. Pistikopoulos, Novel optimization approach of stochastic planning models, Ind. Eng. Chem. Res. 33 (1994) 1930–1942. [28] A. Pongsakdi, P. Rangsunvigit, K. Siemanond, M.J. Bagajewicz, Operational risk management in the planning of refinery operations, Int. J. Prod. Econ. 103 (2006) 64–86. [29] M.A.H. Dempster, N. Hicks Pedron, E.A. Medova, J.E. Scott, A. Sembos, Planning logistics operations in the oil industry, J. Operat. Res. Soc. 51 (11) (2000) 1271–1288. [30] L.F. Escudero, F.J. Quintana, J. Salmeron, CORO, a modeling and an algorithmic framework for oil supply, transformation, and distribution optimization under uncertainty, Eur. J. Operat. Res. 114 (3) (1999) 638–656. [31] A. Gupta, C.D. Maranas, Managing demand uncertainty in supply chain planning, Comput. Chem. Eng. 27 (2003) 1219–1227. [32] G.D. Eppen, R.K. Martin, L. Schrage, A scenario approach to capacity planning, Operat. Res. 37 (4) (1989) 517–527. [33] S.A. Malcolm, S.A. Zenios, Robust optimization of power systems capacity expansion under uncertainty, J. Operat. Res. Soc. 45 (9) (1994) 1040–1049. [34] J.M. Mulvey, R.J. Vanderbei, S.A. Zenios, Robust optimization of large-scale systems, Operat. Res. 43 (2) (1995) 264–281. [35] J.K. Bok, H. Lee, S. Park, Robust investment model for long-range capacity expansion of chemical processing networks under uncertain demand forecast scenarios, Comput. Chem. Eng. 22 (1998) 1037–1049. [36] G.B. Dantzig, Linear programming under uncertainty, Manage. Sci. 1 (3/4) (1955) 197–206. [37] E. Beale, On minimizing a convex function subject to linear inequalities, J. R. Stat. Soc., Series B (Method.) 17 (2) (1955) 173–184. [38] P. Terwiesch, M. Agarwal, D.W.T. Rippin, Batch unit optimization with imperfect modelling: a survey, J. Process Contr. 4 (1994) 238–257. [39] Gotoh, Jun-ya, Yuichi Takano, Newsvendor solutions via conditional value-at-risk minimization, Eur. J. Operat. Res. 179 (2007) 80–96. [40] S. Sen, J.L. Higle, An introductory tutorial on stochastic linear programming models, Interfaces 29 (1999) 33–61.
1764
C.S. Khor et al. / Chemical Engineering and Processing 47 (2008) 1744–1764
[41] J.L. Higle, Stochastic programming: optimization when uncertainty matters, Chapter 1, The INFORMS Tutorials in Operations Research (OR) Series, Presented at INFORMS Annual Meeting at New Orleans and San Francisco, November 13–16, 2005, tucson.sie.arizona.edu/faculty/higle/images/pdf/HigleInforms05.pdf (accessed on November 11, 2005). [42] R.L. Clay, I.E. Grossmann, A disaggregation algorithm for the optimization of stochastic planning models, Comput. Chem. Eng. 21 (1997) 751–774. [43] M.A.H. Dempster, Introduction to stochastic programming, in: M.A.H. Dempster (Ed.), Stochastic Programming, Academic Press, New York, 1980, pp. 3–59. [44] W.H. Evers, A new model for stochastic linear programming, Manage. Sci. 13 (1967) 680–693. [45] R.J.-B. Wets, Solving stochastic programs with simple recourse, Stochastics 10 (3–4) (1983) 219–242. [46] P. Kall, S.W. Wallace, Stochastic Programming, John Wiley & Sons, New York, 1994. [47] F.P. Bernardo, E.N. Pistikopoulos, P.M. Saraiva, Integration and computational issues in stochastic design and planning optimization problems, Indust. Eng. Chem. Res. 38 (1999) 3056–3068. [48] H. Konno, H. Yamazaki, Mean-absolute deviation portfolio optimization model and its applications to Tokyo Stock Market, Manage. Sci. 37 (5) (1991) 519–531. [49] H. Konno, A. Wijayanayake, Portfolio optimization under D.C. transaction costs and minimal transaction unit constraints, J. Global Optim. 22 (2002) 137–154. [50] H. Konno, T. Koshizuka, Mean-absolute deviation model, IIE Trans. 37 (2005) 893–900. [51] W. Ogryczak, A. Ruszczy´nski, From stochastic dominance to mean–risk models: semideaviations as risk measures, Eur. J. Operat. Res. 116 (1999) 33–50. [52] Speranza, M. Grazia, A heuristic algorithm for a portfolio optimization model applied to the Milan stock market, Comput. Operat. Res. 23 (5) (1996) 433–441. [53] C. Papahristodoulou, E. Dotzauer, Optimal portfolios using linear programming models, J. Operat. Res. Soc. 55 (2004) 1169–1177. [54] J.P. Vin, M.G. Ierapetritou, Robust short-term scheduling of multiproduct batch plants under demand uncertainty, Indust. Eng. Chem. Res. 40 (2001) 4543–4554. [55] R.L. Rardin, B.W. Lin, Test problems for computational experiments—issues and techniques, in: J.M. Mulvey (Ed.), Proceedings of a Conference on Evaluating Mathematical Programming Techniques, National Bureau of Standards at Boulder, Colorado on January 5–6, 1981, Springer-Verlag, Berlin, Heidelberg, New York, 1982. [56] J.Y. Jung, G. Blau, J.F. Pekny, G.V. Reklaitis, D. Eversdyk, A simulation-based optimization approach to supply chain management under demand uncertainty, Comput. Chem. Eng. 28 (2004) 2087–2106. [57] A. Brooke, D. Kendrick, A. Meeraus, R. Raman, GAMS—A User’s Guide, GAMS Development Corporation, Washington DC, USA, 1998, http://www. gams.com/docs/gams/GAMSUsersGuide.pdf (accessed December 9, 2005). [58] A. Barbaro, M.J. Bagajewicz, Managing operational risk in planning under uncertainty, AIChE J. 50 (5) (2004) 963–989. [59] Y. Simaan, Estimation of risk in portfolio selection: the mean-variance model and the mean-absolute deviation, Manage. Sci. 43 (1997) 1437–1446. [60] N.J. Samsatli, L.G. Papageorgiou, N. Shah, Robustness metrics for dynamic optimization models under parameter uncertainty, AIChE J. 44 (9) (1998) 1993–1998. [61] W. Ogryczak, A. Ruszczy´nski, Dual stochastic dominance and related mean-risk models, SIAM J. Optim. 13 (1) (2002) 60–78. [62] D. Dentcheva, A. Ruszczy´nski, Optimization with stochastic dominance constraints, SIAM J. Optim. 14 (2) (2003) 548–566. [63] S. Sen, Personal communication (via e-mail), November 29, 2005. [64] A. Aseeri, M.J. Bagajewicz, New measures and procedures to manage operational risk with applications to the planning of gas commercialization in Asia, Comput. Chem. Eng. 28 (2004) 2791–2821. [65] A. Bonfill, M. Bagajewicz, A. Espuna, L. Puigjaner, Risk management in the scheduling of batch plants under uncertain market demand, Ind. Eng. Chem. Res. 43 (2004) 741–750. [66] T.K. Kristoffersen, Deviation measures in linear two-stage stochastic programming, Math. Methods Operat. Res. 62 (2) (2005) 255–274.