Integration of Demand Dynamics and Investment Descisions on DER ...

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. IEEE TRANSACTIONS ON SMART GRID

1

Integration of Demand Dynamics and Investment Decisions on Distributed Energy Resources Farbod Farzan, Farnaz Farzan, Member, IEEE, Mohsen A. Jafari, Member, IEEE, and Jie Gong

Abstract—This research directly couples investment decisions on distributed energy resources (DERs) with demand side management (DSM) strategies that can be adopted over time by residential communities. The formulation also takes into account factors that usually contribute to long-term market variations and short-term operational volatilities. This paper uses a high resolution adaptive model (Hi-RAM) for short-term load calculations on a premise that expected dynamical effects due to DSM strategies and their interactions cannot effectively be captured by traditional forecast models. The integration of Hi-RAM into investment formulation allows for certain What-If investment scenario analysis on the use of advanced technologies, new plug-ins, and consumer response to power price fluctuations, and more. To demonstrate the significance of coupling of DER investment decisions and DSM strategies, three scenarios are presented. The base scenario (Mode I) results can be reproduced using traditional forecast models and is included for baseline analysis and benchmarking. In Mode II, new load patterns for plug-in electric vehicles are introduced. Mode III is an extended version of Mode II, where smart devices are used within households. The comparison of investment decisions from the three modes clearly demonstrate that interaction effects of DSM strategies matter. Index Terms—Capacity planning, demand dynamics, distributed energy resources (DERs), plug-ins, price responsive demand.

N OMENCLATURE

Variables AI BGen C CHPGH CHPGHW CHPGHNW D φG φG_L φG_S GP Hd Pd PB EPG PLLj ROI SoC SP SOG ϑCAP

Cash invested in other alternatives in period ($). Boiler hourly heat generation (kW). Storage charge (kW). CHP total hourly electricity generation (kW). CHP hourly wasted heat generation (kW). CHP hourly nonwasted heat generation (kW). Storage discharge (kW). φ total hourly power generation (kW). φ hourly power generation serves load (kW). φ hourly power generation to storage (kW). Gas price (GP) ($/kWh). Heat demand. Power demand. Total financial charges on borrowed funds ($). Hourly electricity purchased from macro grid (kWh). Partial load levels of CHP. Return of cash invested on alternative investment. Storage state of charge (kW). Electricity spot price (SP). Annual on-site generation operational savings ($). Accumulated installed capacity (kW).

Index φ i j r t ϑ y

Index of power generator assets; φ ∈ {WT, PV, GF, CHP}. Index of household. Index of combined heat and power (CHP) regions. Index of household (HH) residents. Index of time (hr). Portfolio assets; ϑ ∈ {PV, WT, GF, CHP, Boiler, ST}. Index of time (year).

Manuscript received December 1, 2014; revised March 26, 2015; accepted April 19, 2015. Paper no. TSG-01184-2014. F. Farzan and J. Gong are with the Department of Civil and Environmental Engineering, Rutgers University, Piscataway, NJ 08901 USA (e-mail: [email protected]; [email protected]). F. Farzan is with the Det Norske Veritas-Germanischer Lloyd Energy Advisory, Chalfont, PA, USA (e-mail: [email protected]). M. A. Jafari is with the Department of Industrial and Systems Engineering, Rutgers University, Piscataway, NJ 08854 USA (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TSG.2015.2426151

Parameters AIC Beff Blimit CapExϑ CHPLLjeff CϑO&M DF Egrid FC GFHR HPR IRR LRP MLST Mult STeff STPR

Initial available cash ($). Boiler efficiency. Maximum borrowing limit ($). Unit capacity cost of ϑ ($/kw). CHP electric efficiency associated to load level. ϑ nonfuel operational and maintenance cost. Annual discount rate. Grid average heat rate (mmbtu/kwh). Finance charge rate. Gas-fired (GF) heat rate (mmbtu/kwh). Heat to power ratio. Investment rate of return. Land rental price ($/Acre). Minimum percentage of storage level. Transmission costs multiplier. Electricity storage efficiency. Electricity storage charge/discharge power rate.

c 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. 1949-3053 See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 2

IEEE TRANSACTIONS ON SMART GRID

I. I NTRODUCTION HE VALUE of a distributed energy resources (DERs) portfolio depends on its projected return on investment and the potential growth in its operating income while serving underlying demand. For a DER, the investment payoff is directly linked to the operation of the physical assets, and return on investment depends on how these operations are utilized overtime [1], [2]. Depending on when and how much investment is made, long-term value of a DER could be assessed. Investment decisions are usually influenced by fuel costs, the price of technology, state incentives, and parameters such as finance charge rates/terms [3]. Economic benefits of a DER could be enhanced by contributing strategies such as demand side management (DSM) [4], [5]. In this context, DSM can be regarded as a means of not only reducing energy costs but also generating revenue by reducing load on the grid [6]–[8]. In the presence of DSM, energy demand becomes highly dynamic and difficult to predict (e.g., demand can be curtailed/shifted over time as a result of customer response to electricity prices). Integration of DSM as a resource necessitates inclusion of such dynamics into investment modeling. In this paper, we investigate investment decisions on DER by taking into account the specifics of HH behavioral and physical characteristics, which happen to be the main load drivers in residential communities. Investment planning also takes into account factors that usually contribute to longterm market variations, and short-term operational volatilities. This paper uses a bottom-up demand model. The bottom up model [high resolution adaptive model (Hi-RAM)] is capable of capturing behavioral patterns of end users and lends itself to certain What-If analysis on the use of advanced technologies, new plug-ins, and consumer response to price fluctuations according to their demographic characteristics. This would allow investors to examine return on investment as a function of technology and behavioral pattern changes over time; such analysis cannot be carried out using stationary forecast models, which are solely obtained on the basis of historical data.

T

II. R ELATED S TUDIES Planning, operation, and control of DER are extensively studied in literature. Focusing on operation and control of DER, Costa and Kariniotakis [10] applied a stochastic optimization framework to optimize the operation of a DER configuration, where they observed significantly different results versus that of deterministic case. Farzan and Jafari [9] proposed a two-stage optimization model to optimize the hourly operation of a DER aiming at optimizing one-day-ahead plans and daily operations. By considering a trade-off between revenue and risk, Beraldi and Conforti [11] and Beraldi et al. [12] proposed a two-stage stochastic optimization model where weighted mean-risk sum is the objective function of the problem. Commercially available software systems also exist, such as hybrid optimization model for electric renewables [13], which enables to analyze the combination of renewable and conventional energy resources.

By incorporating investment into operation and control problem, El-Khattam et al. [14] studied the capacity investment in order to optimize the sizing and siting for DER. Their objective function includes investment and operating costs. Cost-benefit analysis is conducted in order to obtain both optimal size and site. References [15] and [16] applied nonlinear optimization approach toward the problem of optimal portfolio sizing considering static and stochastic demand, respectively. Farzan et al. [17] investigate investment decisions on DER by combining short-term operational volatility and long-term market variations. Using DER in conjugation with DSM provides both economic and environmental advantages [18]. Centolella [19] described how price responsive demand can be effectively integrated into wholesale power markets. Cecati et al. [20] proposed active energy management system endowed with an optimal power flow integrating DSM and active management schemes for the optimization of a smart grid in a competitive power market. Nordman [21] presented the relationship between networks and consumer electronics. They discussed capabilities to control the temperature and volume of heat, cooling, or ventilation, and lighting patterns depending on what the occupant is doing. However, models for capturing such dynamics were not provided. Roos and Lane [22] proposed optimal load scheduling strategy to minimize the electricity costs of an industrial end user under real time pricing. An analytical approach was followed to describe the potential electricity cost savings to an industrial end user under real time pricing through intelligent demand management. Calloway and Brice [23] proposed a short-term physically-based load model of the demand of a group of air conditioners as a function of outside temperature, and air conditioning system parameters. The formulation provides the optimal schedule of electricity usage given the predetermined electricity price schedule. Hubert and Grijalva [24] discussed the critical role that energy optimization algorithms will play at residential level to effectively achieve benefits of energy saving via participation of informed customers. They presented a vision of a home electrical system which consists of renewable generation assets, electric vehicle, and sets of controllable and uncontrollable appliances. Also, Xiong et al. [25] recommended an algorithm for power usage of home appliances where real-time price signals are sent from utility to customer end. Finally, Costanzo et al. [28] proposed a layered architecture for load management system in smart buildings with emphasis on optimal energy consumption management. The proposed framework enables investigation of long-term investment decisions on DER considering near-future scenarios such as large-scale penetration of plug-in electric vehicles (PEV), and smart grid enabling technologies (e.g., DSM). Unlike studies such as [18] and [28] that assume expected measures for responded demand, in this paper curtailed/shifted demand will be directly calculated thanks to end-use models in Hi-RAM. Hi-RAM breaks down the complex interactions between the network participants by modelling each end-use. Moreover, it not only investigates DER operational costs in conjugation with DSM as in [26] and [27] but also suggests

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. FARZAN et al.: INTEGRATION OF DEMAND DYNAMICS AND INVESTMENT DECISIONS ON DERs

capacity planning solutions for a DER portfolio. Through a number of numerical experiments, we demonstrate that the inclusion of underlying dynamics significantly impact investment decisions. For this demonstration we consider three modes of investment, where Mode I simulates a baseline model which works similar to existing models in the literature. The other two modes are variations of our proposed investment model.

3

Fig. 1.

WT capacity factor.

Fig. 2.

CHP electrical efficiency at each region.

III. P ROBLEM S TATEMENT The problem of interest is to formulate DER investment decisions (e.g., sizing, configuration, and timing) by directly taking into account DSM solutions and interactions between these solutions. Our hypothesis is that distribution (expected value and variance of) savings from DSM strategies will significantly influence the investment decisions. DSM strategies may include a host of options, including but not limited to: 1) adoption of advanced control and scheduling for EV charging; and 2) smart home technologies (e.g., with ability to respond to price signals). Our main premise is that savings from DSM solutions evolve over time and depend on the interaction between these solutions. Furthermore, such dynamics can hardly be captured through forecast models that are solely built on the basis of observed historical data. Real life data collection of such effects and their interactions would usually require extensive experiments and observations, and would be expensive if not infeasible. On the other hand, the use of dynamic bottom up models that work on the basis of atomic level data and closely capture behaviors and patterns under various DSM strategies can be much more cost effective and practical. In essence the simulated data from a bottom up model is used as surrogate for observed data. In practice, one can use combined framework where historical data is used primarily and augmented by secondary data from such bottom up simulations. The proposed solution to the above problem constitutes the integration of the following submodels. 1) A dynamic model that accurately captures load behavior under a discrete set of configurations and design choices that pertain to behavioral patterns and technological solutions that are adopted at individual HH level. We will briefly discuss this model later in this paper (see Section III-C). 2) A stochastic investment model in the form of mixed integer non-linear programming (MINLP) that optimizes cash flow over a horizon (CFY ) and projected cash flow Y ). beyond the planning horizon (CF Cash flows include operational savings due to DER investment, finance payments, and alternative investment. It is assumed that each year the investor has opportunity to invest on other activities. Also, limited funds are available to be borrowed in each year in order to invest on DER assets. Monte Carlo technique is applied for solving stochastic MINLP problem. This paper will focus on the investment model and, in particular, it will demonstrate how to incorporate the bottom-up dynamical model of demand into investment decisions. Despite earlier works (see [3], [9], [17], [18], [30]) where power

consumption is calculated from load forecast models obtained from historical data, this paper uses a bottom-up demand model. Furthermore, in comparison to [3], [9], and [17] which considered GF, photovoltaic cells (PV), wind turbine (WT), electric battery storage (ST), and purchase form the grid in DER portfolio, in this paper two new resources are added into the mix. Those are CHP and DSM. Moreover, unlike [3], [9], and [17], where DER cost savings function was calculated separately and was fed to a stochastic longterm investment model, we will solve operation and investment optimization problems together combined. Objective function for the investment problems is formulated as Y = SOGY + ROIY + BY − PBY − SOAY max CFY + CF Y + ROI Y+1 − P − AIY + SOG BY+1 . A. Operational Constraints In this section, formulation for each asset generation is described. GF is assumed to generate power according to its capacity and specific heat rate as in [3] and [9]. For WT and PV generation, capacity factor approach is applied, where WT and PV are assumed to generate power according to their capacity factor (CFφ|φ ∈ {WT, PV}). For WT it is assumed that capacity factor follows (CFWT) based on Fig. 1 where the ratio is the function of hourly wind speed. It is also assumed that for wind speeds below 4 m/s and above 25 m/s, the WT does not function [29]. PV capacity factor (CFPV) is assumed to be a ratio of hourly solar intensity to the intensity which PV generates its maximum power. This gives both WT and PV hourly generation as below φGy,t = φCAPy × CFφy,t , φ ∈ {WT, PV}.

(1)

CHP consumes gas and generates both power and heat. Since CHP electrical efficiency is highly dependent on its load level [30], piecewise linear electrical efficiency algorithm is applied for CHPG formulation. Three load regions with specific efficiency are assumed (PLLj , CHPLLj eff ) as in Fig. 2 in each “y,” according to accumulated installed capacity (φCAP|φ = CHP) the allowable load for each region follows: LLjy = PLLj × CHPCAPy , j = 1 . . . 3.

(2)



Two sets of binary constraints as in (3) and (4) along with binary decision variable for CHP status (CHPOS ) are introduced which force the unit to be shut down if the load level is below LL1. “M” refers to a large number PLL1 × CHPCAPy − PRLL1y,t ≤ M × 1 − CHPOSy,t (3) PLL1 × CHPCAPy − PRrLL1y,t > −M × CHPOSy,t . (4)

each year (DEROCy ) follows: ⎡

D T GPy ⎣ ϑGy,t × DEROCy = + CϑO&M ϑ ∈ ϑeff d=1 t=1

× {GF, Boiler} + φGy,t φ∈{WT,PV}

GPy × CφO&M ϑGy,t × + CϑO&M ϑeff

× ϑ ∈ {GF, Boiler} + φGy,t × CφO&M + EPGy,t

Total hourly electricity generation (CHPG) calculates as in (5) with respect to unit status and electricity generation at each region (PRLLj) CHPGy,t =

J

PRLLjy,t

j=1

⎧ ⎨ = CHPCAPy × (PLL1 ) × CHPOSy,t , j = 1 × ≤ CHPCAPy × CHPOSy,t ⎩ j = 1. × PLLj − PLLj−1 , (5) The amount of hourly generated heat (CHPGHW +CHPGHNW ) is constrained as CHPGHWy,t + CHPGHNWy,t = HPR × CHPGy,t .

(6)

After defining all generation constraint, now we move on to electricity storage. At each time period, the available energy kept in storage (SoC) is conserved by SoCy,t

Dy,t = SoCy,t−1 + Cy,t × STeff − , t = 1. STeff

(7)

Cy,t and Dy,t are charging and discharging quantities from storage during each hour. Storage charge and discharge are constrained by maximum charge/discharge power rate of the device Cy,t × STeff +

Dy,t ≤ STPR × t. STeff

(8)

Moreover, the amount of energy stored in the device is constrained between min/max of allowable storage level. This will constrain SoC at each time step with respect to accumulated capacity at “y”

φ∈{WT,PV}

⎤

× SPy,t ⎦ +

RLφy .

(11)

φ∈{WT,PV}

RLφ’s are calculated as: φlkw × LRPy × φCAPy . For calculating the fuel-related CHP operation costs, slope cost (CostSlopej) is defined as below ⎧ 1 ⎪ j=1 ⎨ GPy × CHPLLj , eff PLLj−1 PLLj CostSlopejy = − ⎪ ⎩ GP × CHPLLjeff CHPLLj−1eff , j = 1. y PLLj −PLLj−1 In (11), SP is assumed to be a random function of natural GP, grid average heat rate (Egrid ), a multiplier (Mult) which accounts for transmission costs, and hourly profile of electricity price as a percentage of peak price. Adopted from [31], Ornstein–Uhlenbeck Brownian motion with mean reverting drift is used to model natural GP. With knowing underlying price values along with both power (Pd) and heat demand (Hd), the annual operational cost without DER (NoDEROCy ) calculates as below

D T GPy Pdy,t × SPy,t + Hdy,t × . (12) NoDEROCy = Beff d=1 t=1

Using (11) and (12), one could calculate savings due to on-site generation (SOG) as follow: SOGy = NoDEROCy −DEROCy .

MLST × STCAPy ≤ SoCy,t ≤ (1 − MLST) × STCAPy . (9) In (10), allowable energy for charging the device (C) is constrained based on available portion of generation which is not served the load (φG_S) Cy,t = φG_Sy,t = φGy,t − φGLy,t . (10) φ

B. DER Operational Saving Cost of operating DER is the cost of both fuel and nonfuel on-site generation plus if any, the cost of purchasing power from grid (EPG). It is assumed that if PV or WT is installed there would be associated costs for renting the needed l and (RLφ|φ ∈ {WT, PV}). In this order, DER operational cost at

C. Serving Dynamic Demand Both power and heat demand should be met as the main constraint of any energy system [32]. Recognizing its importance, the summation of hourly power generation which serves load (φG_L), discharge from ST (D) and electricity purchased from macro grid (EPG) should be equal or greater than power demand at each time step (13). This also holds for heat φG_Ly,t + Dy,t + EPGy,t ≥ Pdy,t (13) φ

BGeny,t + CHPGHNWy,t = Hdy,t .

(14)

Pd and Hd derive from stochastic bottom-up adaptive model (Hi-RAM). At HH level, model takes into HH=i ), residents’ account HH physical characteristics (XPh


5

HH=i ), and ambient variables (Z HH=i ). demographics (YDemo Amb Considering “I” HHs this gives

Pdy,t , Hdy,t ∼

I HH=i HH=i HH=i f XPh , YDemo , ZAmb

(15)

i=1

where HH=i XPh ∈ {HHSize} HH=i YDemo ∈ {Age, Gender, Empolyement Status} HH=i ZAmb ∈ {Ambient Temperature, Solar Intensity}.

The embedded dynamics of the model works on the basis of Markovian stochastic model for simulating human activities. Given “R” residents in the ith HH, sequence of activities will be simulated where transition probabilities (T) are in form of: HH=i TR=r Sn Sm (t) ∼ g(YDemo ). Transition probabilities are calibrated using American time-use survey data, publicly available in U.S. Department of Labor website (http://www.bls.gov/tus/). Simulated activities then will be converted to its associated load (e.g., watching TV, dishwashing, etc.). Herein, we call these “loads associated to specific activities.” “Base-load” is the other load category which consists of: 1) heating, ventilation and air conditioning (HVAC); 2) lighting; and 3) cold appliances load. HVAC model works on basis of a Newton’s law of heating and cooling. The lighting energy use model takes into account perception of the natural light level (solar intensity) within a building and number of people who are active. At each time interval “t,” latter calculates from the simulated activities of residents. This follows as: r=R / {away, sleeping} 1, if Sr (t) ∈ 1n(t), 1n(t) := 0, else

Fig. 3.

Integration of demand dynamics to capacity planning.

cash flow beyond the horizon at horizon year [9]. Cash flow constraints (16)–(18) are adopted from [9] and [17] CFY = SOGY + ROIY + BY − PBY − SOAY − AIY .

ROIY which is the return of cash invested on an alternative investment other than DER in the previous period (AIY−1 ), calculates as: ROIY = AIY−1 × (1 + IRR). AI is constrained based on the availability of cash, where at the beginning of horizon (y = 1) it cannot exceed initial available cash (IAC), and in following years it follows as: AIy ≤ SOGy−1 + ROIy , y = 1.

r=1

where S refers to states (activities). Moreover, cold appliances load follows probabilistic Bernoulli distribution. Knowing number of HHs under the study, both power and heat demands are obtained from (15). Two What-if scenarios are defined. These scenarios are denoted by Modes II and III. In Mode II, new load patterns from PEV are introduced. PEV charging schedule is based on occupancy pattern simulated by Hi-RAM and takes into consideration the driving habits of individuals. Charging schedules are not necessarily averse of price. Mode III is the upgraded version of Mode II where smart devices are used within premises. This enables the implementation of price responsive DSM strategies via programmable communication devices (PCD). PCD enables end-use models to reflect the function of: 1) intelligent thermostats that learn temperature preference of occupants; 2) price responsive thermostats with precooling capabilities; 3) smart electric plugs with the capability of altering the switch status; and 4) automated dimmer switch with the capability of altering lighting levels based on the real time market price (see [33]). Conceptual framework that demonstrates the integration of bottom-up demand model to capacity planning (investment problem) is depicted in Fig. 3. D. Cash Flow Constraints Investment decisions will be made in order to maximizing the end of horizon cash flow plus the projected value of any

(16)

(17)

At each period, it is assumed that borrowed fund cannot exceed the limit (By ≤ Blim). Moreover, Funds borrowed are restricted to be used to purchase on-site resources and cannot be invested on analternative option. Therefore, total DER investment in each period is constrained by funds available through borrowing and SOAy SOAy + By = CCϑy (18) ϑ

where CCϑ is the cost associated to purchase incremental capacity of ϑ(ACϑ) : CCϑy = CapExϑy × ACϑy . Having purchased ACϑ at each year, accumulated capacity of each asset could be calculated as follows: y=1 ACϑy , ϑCAPy = ϑCAPy−1 + ACϑy , y = 1. Fund borrowed in period x would entitle the borrower to payment flows in coming periods. Total payment in each period would be PBy =

y−1 x=1

FC × By y = x + 1, . . . , x + FT. 1 − (1 + FC)−FT

Y includes perpetual savings from the DER (SOG Y ), the CF Y+1 ), and the return on cash invested in the last period (ROI


Fig. 4.


Hourly solar intensity (top) and wind speed (bottom).

remainder of finance charges (P BY ). These follow as: Y = SOG

Fig. 5.

Power demand Mode I.

Fig. 6.

Power demand Mode II.

Fig. 7.

Power demand Mode III.

∞

SOGY+1 1 SOGY+1 = 1 + DF (1 + DF)s DF s=0

1 Y+1 = ROIY+1 ROCI (1 + DF) Y+FT PBt P BY = . (1 + DF)t t=Y+1

Nonnegativity constraints are also considered for variables which cannot take negative values: ACϑy , ϑCAPy , SOAy , By , AIy ≥ 0. IV. R ESULTS AND D ISCUSSION In this section, daily DER operation along with investment decisions will be presented for three modes of electricity demand. Results corresponding to Mode I (base) are reproducible using time series forecast models and are provided here as a benchmark to evaluate the results from Mode II (new end-use consumption in form of PEVs) and Mode III (customer’s response to price signals and other DSM strategies). Community of 40 HHs is selected; note that there are typically 40 HHs connected to a primary distribution feeder. HHs are selected from residential energy consumption survey (RECS) data. RECS feeds our demand model with unique information about characteristics of HH members (e.g., age, gender, education level, and employment status) and appliances within premises (see [33] for detailed information regarding application of RECS in bottom-up models). It should be noted that our model is not limited to any particular HH and this selection is just for the purpose of demonstration. Both power and heat load will be supplied by either DER or macro-grid. Hourly power demand medians with corresponding 25th and 75th percentiles for Modes I–III are demonstrated in Figs. 5–7, respectively. Heat demand is same for all three different lectricity modes (Fig. 8). Planning horizon is assumed to be five years. Availability of renewables such as solar intensity and wind speed are illustrated in Fig. 4. Operational and financial parameters are provided in Table I. Moreover, unit capacity costs are presented in Table II (values are adopted from [37]). A. Mode I: Daily DER Operation and Investment Decisions Average daily operation in horizon year for Mode I is demonstrated in Fig. 9 along with cumulative installed

capacities (see Table III). One may observe that generation from renewable sources namely PV and WT follow the availability of their corresponding resources (Fig. 4). Moreover, it could be seen that CHP generation is negligible during early hours (EH) of the day (between 1st and 6th hour period). As depicted in Fig. 8, heat demand during EH is in the minimum level except for 6th period. This sudden spike in heat demand with considering relatively low power demand (see 6th hour in Fig. 5) limits the potential of CHP. In such situation according to (3) and (4) the unit will be shut down (CHPOS = 0) since: PLL1 × CHPCAPy ≤ PRLL1y,t . Poor coordination between power and heat level in this time period, leads to highest EPG level throughout the day. Fig. 10 illustrates average sources which contribute to charge electricity storage rather


7

TABLE III AVERAGE C UMULATIVE I NSTALLED C APACITY ( K W) A MONG A LL M ODES

Fig. 8.

Heat demand across all models. TABLE I O PERATIONAL AND F INANCIAL PARAMETERS

TABLE II U NIT C APACITY C OST ($/ K W)

Fig. 9.

Fig. 10.

Mode I, state of charge and resources.

Fig. 11.

PEV recharging.

Mode I, average hourly operation (horizon year).

than directly serving the load. This type of analysis is feasible due splitting mechanisms (φG_L, φG_S), as in (10). As it can be seen major contribution is from renewable resources. This indicates that, existence of electricity storage helps to accommodate excess generation from intermittent renewable resources and to participate into price arbitrage operations. B. Mode II: PEV Emergence and Investment Decisions In Table III, average cumulative capacity installments are demonstrated for all three modes. As it can be seen, by considering the emergence of PEV (Mode II) investment decisions may differ compare to Mode I. In this Mode II, PEV recharging schedule is based on HHs occupancy pattern. This means charging is an option when they are at home.

This mostly happens in early hours of the day, and peak hours through end of the day. Simulated by Hi-RAM, the frequency of PEV recharging is depicted as in Fig. 11. The positive correlation between recharging pattern and wind availability (see Fig. 4), suggested more capacity installment of WT in Mode II compare to Mode I (see Table III). Moreover, higher peak values in Mode II as a result of PEV recharges, ended-up in more CHP installment compare to Mode I. One may notice that by incorporating such demand dynamics (e.g., large PEV penetration to communities) into capacity planning problem, more sustainable strategy than Mode I could be suggested that can hedge against possible future supply shortcomings.



Fig. 12. Average end uses consumption Mode II (red) versus Mode III (blue).

Fig. 13.

Cash flow at end of horizon + beyond horizon.

C. Mode III: DSM as Resource In this mode, three end-uses namely lighting, PEV charging, and space cooling are responding to SP. Average profile of end-uses consumption in both Modes II and III are depicted in Fig. 12. Cash flows at the end of horizon plus beyond the Y ) in different Modes are illustrated in horizon (CFY + CF Fig. 13. One may observe that addition of price responsive DSM into portfolio (Mode III) results in more cash flow than both Modes I and II. To better envision the reason behind this, financial activities during planning horizon are provided for both Modes II and III in Fig. 14 (negative values indicate outflows). As it can be seen when DSM is available, more operational savings will be achieved. This will expand the opportunity to invest on other activities according to C.19. By comparing the configuration of installed assets in Modes II and III, one finds more investment on PV and less investment on CHP. The latter is mainly due to the curtailment of peak load. The former would lead to more installation of electricity storage. Higher unit cost of PV relative to CHP (see Table I) leads to more funds to procure resources in Mode III compared to Mode II. However, more operational savings would be obtained in Mode III since production cost of PV is less than CHP. Another reason relates to price arbitrage

Fig. 14.

Financial activities in Modes II and III.

operations. More electricity generation from PV leads to more contribution of this resource to recharge electricity storage (recharging with cheaper electricity). Mode III results suggest


that the addition of DSM into the portfolio optimization results in not only higher operational savings but also in less environmental burdens.

V. C ONCLUSION This research intends to formulate DER investment decisions by directly taking into account savings from DSM strategies. Hi-RAM that is capable of calculating DSM dynamical impacts is coupled with long-term formulation of a capacity planning problem. It was argued that traditional forecast models obtained solely from historical data fail to capture interaction effects of DSM strategies and, thus may lead to investment decisions that do not truly reflect optimal conditions. It is observed that emergence of PEV load varies the investment decisions. In particular more investment in WTs in Mode II than Mode I due to the coordination of charging habits and wind availability. Liu et al. [26] observed similar patterns by using transportation data. One may observe that existence of DSM (e.g., smart charging) reduces the operational costs (see Figs. 13 and 14). Investigation of such operational cost savings are observed in [26] and [27]. This framework not only investigates operational cost savings as a result of DSM existence, but also suggests optimal configuration for DER portfolio in host of DSM. It is demonstrated that existence of DSM flattens the load profile (peak reduction and valley filling). This will avoid further investment in generation assets (particularly in dispatchable resources) in order to meet peak capacity. Proposed investment strategy in Mode II could be considered as a “more sustainable solution” where it can hedge against possible future supply shortcomings. In Mode III, decisions are even smarter than Mode II, where existence of price responsive demand would lead to utilization of more renewables and revenue. Coordination of our results with existing literature is interesting since the simulated data from a bottom up model is used as surrogate for historical domain specific data (e.g., electric meter, transportation, and demand response data). Unlike the existing literature, this framework not only investigates load dynamics in the presence of advanced technologies, new plug-ins, and consumer response, but also suggests DER capacity planning solutions. This is important particularly for new communities where historical data is not available/sufficient. Furthermore, case of study can be easily expanded/changed in Hi-RAM. One direction for the future work is the addition of industrial and commercial load to this framework. Other direction would be participation of daily activities and consequent loads into demand response programs. One could derive willingness-to-shift functions for so-called delay-able activities. Investigating the effect of different climate on investment decisions is one another direction for future work.

R EFERENCES [1] H. Liang and W. Zhuang, “Stochastic information management in microgrid operations,” IEEE Commun. Surveys Tuts., vol. 16, no. 3, pp. 1746–1770, Nov. 2014.

9

[2] A. Bhattacharya, “Optimal microgrid energy storage strategies in the presence of renewables,” in Proc. Ind. Syst. Eng. Research Conf., Montreal, QC, Canada, 2014, pp. 1–8. [3] F. Farzan et al., “Microgrids for fun and profit: The economics of installation investments and operations,” IEEE Power Energy Mag., vol. 11, no. 4, pp. 52–58, Jul./Aug. 2013. [4] U.S. DOE. (Feb. 2006). Benefits of Demand Response in Electricity Markets and Recommendations for Achieving Them. [Online]. Available: http://eetd.idi.gov [5] R. Poudineh and T. Jamasb, “Distributed generation, storage, demand response and energy efficiency as alternatives to grid capacity enhancement,” Energy Policy, vol. 67, pp. 222–231, Apr. 2014. [6] T. Hoff, “Identifying distributed generation and demand-side management opportunities,” Energy, vol. 17, no. 4, pp. 89–105, 1996. [7] M. Albadi and F. El-Saadany, “Demand response in electricity markets: An overview,” IEEE Power Eng. Soc. Gen. Meeting, Tampa, FL, USA, 2007, pp. 1–5. [8] C. Woo and L. Greening, “Demand response resources: The U.S. and international experience,” Energy, vol. 35, no. 4, pp. 1515–1866, 2010. [9] F. Farzan, M. A. Jafari, R. Masiello, and Y. Lu, “Toward optimal dayahead scheduling and operation control of microgrids under uncertainty,” IEEE Trans. Smart Grid, vol. 6, no. 2, pp. 499–507, Mar. 2015. [10] L. M. Costa and G. Kariniotakis, “A stochastic dynamic programming model for optimal use of local energy resources in a market environment,” in Proc. IEEE Power Tech, Lausanne, Switzerland, 2007, pp. 449–454. [11] P. Beraldi, D. Conforti, and A. Violi, “SICOpt: Solution approach for nonlinear integer stochastic programming problems,” J. Optim. Theory Appl., vol. 143, no. 1, pp. 17–36, 2009. [12] P. Beraldi, D. Conforti, and A. Violi, “A two-stage stochastic programming model for electric energy products,” Comput. Oper. Res., vol. 35, no. 10, pp. 3360–3370, 2008. [13] (Oct. 11, 2014). Energy Analysis. [Online]. Available: http://www.nrel.gov/analysis/models_tools.html [14] W. El-Khattam, K. Bhattacharya, and Y. Hegazy, “Optimal investment planning for distributed generation in a competitive electricity market,” IEEE Trans. Power Syst., vol. 19, no. 3, pp. 1674–1684, Aug. 2004. [15] T. Logenthiran and D. Srinivasan, “Optimal capacity and sizing of distributed energy resources for distributed power systems,” Renew. Sustain. Energy, vol. 4, no. 5, 2012, Art. ID 053119. [16] S. V. B. Bruno, and C. Sagastizábal, “Optimization of real asset portfolio using a coherent risk measure: Application to oil and energy industries,” Optim. Eng., vol. 12, nos. 1–2, pp. 257–275, 2011. [17] F. Farzan, F. Farzan, K. Gharieh, M. A. Jafari, and R. Masiello, “Closing the loop between short-term operational volatilities and long-term investment risks in microgrids,” in Proc. IEEE ISGT Conf., Washington, DC, USA, Feb. 2014, pp. 1–5. [18] A. Siddiqui, M. Stadler, C. Marnay, and J. Lai, “Optimal control of distributed energy resources and demand response under uncertainty,” in Proc. Int. Assoc. Energy Econ., Rio de Janeiro, Brazil, Jun. 2010, pp. 1–17. [19] P. Centolella, “The integration of price responsive demand into regional transmission organization wholesale power markets and system operations,” Energy, vol. 35, no. 4, pp. 1568–1574, 2010. [20] C. Cecati, C. Citro, A. Piccolo, and P. Siano, “Smart grids operation with distributed generation and demand side management,” in Modeling and Control of Sustainable Power Systems. Berlin, Germany: Springer-Verlag, 2012, pp. 27–46. [21] B. Nordman, “Networks in buildings,” in Proc. ACEEE Sum. Study Energy Efficien. Build., Pacific Grove, CA, USA, 2008, pp. 1–14. [22] J. G. Roos and I. E. Lane, “Industrial power DR analysis for one part real-time pricing,” IEEE Trans. Power Syst., vol. 13, no. 1, pp. 159–164, Feb. 1998. [23] T. M. Calloway and C. W. Brice, “Physically-based model of demand with applications to load management assessment and load forecasting,” IEEE Trans. Power App. Syst., vol. PAS-101, no. 12, pp. 4625–4631, Dec. 1982. [24] T. Hubert and S. Grijalva, “Realizing smart-grid benefits requires energy optimization algorithms at residential level,” in Proc. IEEE Innov. Smart Grid Technol. (ISGT), Anaheim, CA, USA, 2011, pp. 1–8. [25] G. Xiong, C. Chen, S. Kishore, and A. Yener, “Smart (in-home) power scheduling for demand response on the smart-grid,” in Proc. Innov. Smart Grid Technol. (ISGT), Anaheim, CA, USA, 2011, pp. 1–7. [26] C. Liu, J. Wang, A. Botterud, Y. Zhou, and A. Vyas, “Assessment of impacts of PHEV charging patterns on wind-thermal scheduling by stochastic unit commitment,” IEEE Trans. Smart Grid, vol. 3, no. 2, pp. 675–683, Jun. 2012.


[27] J. Ma, J. Deng, L. Song, and Z. Han, “Incentive mechanism for demand side management in smart grid using auction,” IEEE Trans. Smart Grid, vol. 5, no. 3, pp. 1379–1388, May 2014. [28] G. T. Costanzo, G. Zhu, M. F. Anjos, and G. Savard, “A system architecture for autonomous demand side load management in smart buildings,” IEEE Trans. Smart Grid, vol. 3, no. 4, pp. 2157–2165, Dec. 2012. [29] (Jul. 11, 2014). Outstanding Efficiency, Siemens Wind Turbine SWT-2.3-93. [Online]. Available: http://www.energy.siemens.com/ hq/pool/hq/power-generation/wind-power/E50001-W310-A102-V64 A00_WS_SWT-2.3-93_US.pdf [30] Environmental Protection Agency. (2008). Technology Characterization: Fuel Cells. [Online]. Available: http://www.epa.gov/chp/documents/ catalog_chptech_fuel_cells.pdf [31] F. Farzan, K. Mahani, M. A. Jafari, F. Farzan, and W. Katzenstein, “A simple approximate approach to optimal sizing and investment timing of distributed energy resources,” in Proc. 18th Real Opt. Group Conf., Medellin, Colombia, 2014, pp. 1–16. [32] R. Walawalkar, S. Fernands, N. Thakur, and K. Chevva, “Evolution and current status of demand response (DR) in electricity markets: Insights from PJM and NYISO,” Energy, vol. 35, no. 4, pp. 1553–1560, 2010. [33] F. Farzan, M. A. Jafari, J. Gong, F. Farzan, and A. Stryker, “A multi-scale adaptive model of residential energy demand,” Appl. Energy, vol. 150, pp. 258–273, Jul. 2015. [34] (Jul. 13, 2014). Air Pollution Control Technology Fact Sheet, EPA-452/F-03-009. [Online]. Available: http://www.epa.gov/ttn/ catc/dir1/fscr.pdf [35] R. Poore and C. Walford, “Development of an operations and maintenance cost model to identify cost of energy savings for low wind speed turbines,” Nat. Renew. Energy Lab., Golden, CO, USA, Tech. Rep. NREL/SR-500-40581, 2008. [36] (Jul. 10, 2014). Catalog of CHP Technologies, U.S. Environmental Protection Agency. [Online]. Available: http://www.epa.gov/chp/documents/catalog_ chptech_full.pdf [37] (2013). Updated Capital Cost Estimates for Utility Scale Electricity Generating Plants, U.S. Energy Information Administration. [Online]. Available: http://www.eia.gov/forecasts/capitalcost/pdf/ updated_capcost.pdf


Mohsen A. Jafari (M’98) received the M.S. and Ph.D. degrees from Syracuse University, Syracuse, NY, USA, in 1981 and 1985, respectively. He is a Professor and the Chair with the Department of Industrial and Systems Engineering, Rutgers University, Piscataway, NJ, USA, and the Principal with the Rutgers Center for Advanced Infrastructure and Transportation, where he oversees the Transportation Safety Resource Center and Information Management Group. He recently started the Laboratory for Sustainable Systems, Rutgers University. He was a Consultant to several Fortune 500 companies, and national and international government agencies. He was a Principal/Co-Principal to over $18 million of research and development funding from the U.S., international government agencies, and industry. His current research interests include control and optimization of large complex systems in transportation and energy applications. He has authored more than 70 technical articles, more than 60 conference papers, and more than 100 invited and contributed presentations, and holds three patents. He actively collaborates with universities and national laboratories in U.S. and abroad. Dr. Jafari was a recipient of the IEEE Excellence Award in Service and Research, the SAP Curriculum Award, and received the Transportation Safety Award twice.

Farbod Farzan received the B.S. degree in chemical engineering from the University of Tehran, Tehran, Iran, in 2008, and the M.S. degree in environmental engineering from Rutgers University, Piscataway, NJ, USA, in 2011, respectively, where he is currently pursuing the Ph.D. degree with the Department of Civil and Environmental Engineering. He is the member of the Laboratory on Energy Sustainability and Systems Research Group, Rutgers University, with the focus on development and analysis of sustainable and smart energy solutions. His current research interests include smart-grid data analytics, building automation and energy modeling, demand side management, building performance analysis, and investment strategies for distributed energy resources.

Farnaz Farzan (M’14) received the B.S. degree in mechanical engineering from the University of Tehran, Tehran, Iran, in 2005, and the M.S. and Ph.D. degrees in mechanical and aerospace engineering and industrial and systems engineering from Rutgers University, Piscataway, NJ, USA, in 2008 and 2013, respectively. She is a Senior Consultant with DNV GL, Chalfont, PA, USA, where she is experienced in stochastic optimization in investment and operation of distributed energy resources, and power-market simulation. Her current research interests include system-level impacts and benefits of storage on power system, the behavior of dispatchable demand response and dynamic price response on wholesale markets, and the impact of visibility and control of distributed energy resources on power systems.

Jie Gong received the B.S. degree from Shanghai Jiao Tong University, Shanghai, China, and the M.S. degree from Texas Tech University, Lubbock, TX, USA, in 1999 and 2005, respectively, and the Ph.D. degree from the University of Texas at Austin, Austin, TX, in 2009, all in civil engineering. He was an Assistant Professor with the Department of Civil and Environmental Engineering, Rutgers University, Piscataway, NJ, USA. He was a Principal/Co-Principal Investigator for research projects funded by the National Science Foundation, the U.S. Department of Housing and Urban Development, the U.S. Department of Transportation, the New Jersey Department of Transportation, the New Jersey Board of Public Utility, and the Illinois Department of Transportation. His research was supported by industry organizations and companies such as Bentley Systems, Exton, PA, USA, and the St. Louis Masonry Institute, St. Louis, MO, USA. His current research interests include community resilience and sustainability with a focus on sensing, building data modeling, and analytics for high-performance building systems.