Paper No. 210
MODE CHOICE MODELLING FOR WORK TRIPS IN AN INDIAN METROPOLITAN CITY USING SOFT COMPUTING TECHNIQUE Ashu.S.Kedia1, Krishna Saw2, B.K. Katti3 P.G. Student, Civil Engineering Department, SVNIT, Surat-395007, Gujarat, India,
[email protected], 2 Transport Planner, Institute of Urban Transport (India), New Delhi,
[email protected] 3 Visiting Professor, Civil Engineering Department, SVNIT, Surat-395007, Gujarat, India,
[email protected] 1
Abstract Travel mode choice decision plays a vital role in determination of mode usage by the urban trip maker and have significant bearing on urban transportation planning and traffic operations. Auto pollution and congestion mitigation strategies are strongly associated with mode choice and composition of the traffic stream moving on the road network. Most of the mode choice models are based on the principle of random utility maximization derived from econometric theory. Since 1970 different Logit models have emerged as important tools of mode choice analysis. However they do suffer from certain drawbacks through the linear property and the synergy effect of the utility functions and may not address the correlations among the explanatory variables effectively. Mostly the crisp inputs are used in Logit modelling and thereby the uncertainty prevailing in human decisions in respect of urban travelers in their mode choice is not addressed in the process. However, Fuzzy logic bypasses the binary crisp derivations of the inputs and accepts multivalued inputs in linguistic expressions. Therefore, the attempt here is to opt for Fuzzy Logic System in building a mode choice model for Work Trips which has major share by covering various income groups of Indian society. It is further tested with sensitivity analysis considering an important zone of Surat city in South-Gujarat region as the study area. Keywords: fuzzy logic, mode choice, rule base, soft computing, urban travel 1. Introduction Four – Stage Modelling is a primary tool for forecasting urban travel demand and performance of a transportation system. It performs conditional prediction of the travel demand in order to estimate the likely transportation consequences of several transport policies considered for implementation. Mode choice modelling and prediction relate closely to transportation system policies and congestion mitigation strategies. Logit Model is a traditional approach adopted for mode choice analysis on the basis of crisp input variables such as travel cost, travel time, comfort and convenience etc. The decision process of trip maker for choosing a mode involves human judgements which are not precisely captured by Logit models. This can be overcome by using artificial intelligence techniques like fuzzy logic for modeling mode choice behavior. Fuzzy logic bypasses the concept of crisp values of input variables and it accepts the input values in linguistic terms. Keeping this in view, mode choice modelling has been carried by employing Fuzzy Rule Based System (FRBS) to address the uncertainty prevailing in human linguistic expressions for the attributes. FRBS comprises of several If-then rules which closely resemble human behaviour for selection of travel mode. 2. Literature Support In the recent past, researchers have sought to apply Fuzzy Logic for mode choice modelling. Fuzzy Logic theory was initiated by L.A. Zadeh in 1965. (Mamdani and Assilian, 1975) adopted Fuzzy Logic in linguistic synthesis. In late 1980s a group of Japanese authors made significant contributions to Fuzzy Set theory applications in Traffic and Transportation. (Wang and Mendel, 1992) developed a general method to generate fuzzy rules from numerical data. (Teodorovic, 1999) offered better representation of fuzzy logic in utility of travel mode rather than considering it as a deterministic concept. (Holland, 2000) determined modal split using a fuzzy inference system which he adopted to determine the preference for a given set of perceived journey characteristics. Logit models with fuzzy
Paper No. 210 reasoning-based utility functions can describe a human decision with vagueness explicitly (Mizutani and Akiyama, 2001). Fuzzy Based Models have upper edge over the traditional Logit models to address the uncertainty lying in the choice making behaviour of travellers (Errampalli, 2008). Fuzzy Logic based mode choice models are advocated for better prediction over the traditional MNL model by (Dell’Orco and Ottomanelli, 2012) and (Pulugurta, Arun and Errampalli, 2013). Though the fuzzy logic based model development is a time consuming process, it offers a significant flexibility in evaluating any kind of sensitive policy (Kumar, Sarkar and Errampalli, 2013). 3. Research Approach The major tasks of the methodology are summarized in four basic stages as under in Figure 1. Definition of Problem & Delineation of Study Area
Data Base Development & Analysis
Mode Choice Model Development & Validation
Study Application
Figure 1: Basic Research Stages
Development of Fuzzy Rule Based Mode Choice Model FRB-MCM-W is dealt in third stage followed by its validation. 4. Study Area & Work Trip Characteristics 4.1 Study Area and Database Athwa zone (Figure 2) of fast growing metropolitan Surat city is the study area of the present project. The city covers 326 sq. km. area and possesses a population of 4.6 million (Census 2011). There has been a significant growth in vehicular population in the last decade to cross the mark of 1.8 million vehicles. Per Capita Income is also quite high compared to other cities in the state, resulting into a high private vehicular ownership.
Figure 2: Study area – Athwa zone of Surat city (Source: S.M.C)
Athwa Zone consists of 22 numbers of census wards to cover 87,000 households in 98 sq.km area. The wards in inner fringe area are clubbed to form eight sub-zones for the study purpose and Home Interview surveys were carried with pre-designed questionnaire to capture travellers‟ choice decisions. Nearly 650 filtered samples of 900
Paper No. 210 samples covering all the sections of society are considered for modelling purpose. The questionnaires were found defective in extracting certain data in initial period. Such observations needed the necessary filtrations for lack of desired information and were filtered out at the time of analysis. Here modal choice is restricted to three modes of auto-rickshaw, two-wheeler and car in absence of desired city bus services. The survey data of three major attributes on income levels, in-vehicle travel time and travel comfort and convenience provides the platform for developing the model. Thus, Travel Cost of the trip is considered indirectly through the income groups. 4.2 Mode Share for Work Trips The mode share observations for the three major income groups LIG, MIG and HIG are as shown in Figure 3, wherein the maximum number of work trips are found to be commuted by two-wheeler for MIG and HIG and it is highest by Auto-rickshaw mode for LIG. The share of non-motorized modes such as walking and bicycle are very poor. Very poor share is in city bus service.
60
Percent
50 40 30
LIG
20
MIG
10
HIG
0 2W
Auto Rickshaw
Bus
Car
Cycle
Walk
Vehicle Type Figure 3: Observed modal share for work trips
5. Development of Fuzzy Rule Based Model (FRB-MCM-W) 5.1 Model Structure Three fundamental sequential stages in Fuzzy Logic Modeling are: Fuzzifying the crisp inputs, Fuzzy Inference System using “If-Then” rules and Defuzzification to get the crisp answer (Figure 4). These stages form the sequential operation structure of the model. CRISP INPUT
FUZZIFICATION
FUZZY INFERENCE ENGINE
RULE BASES
DEFUZZIFICATION
CRISP OUTPUT
Figure 4: Fuzzy Model Operation Structure
The model inputs are income level (HHI), in-vehicle travel time (TT) and comfort & convenience level (C&C) and output variable pertains to three modes auto-rickshaw, two-wheeler and car. Bicycle and walking modes are not considered here for the work trips for their higher trip lengths and similar is the case for the buses for their very poor service. Comfort here refers to the mode characteristics while convenience refers to door to door service availability by the mode. 5.2 Framing of Membership Functions
Paper No. 210 Membership Function formation is an important step of the fuzzification process of the crisp or linguistic inputs. Prior to fuzzification, the inputs and outputs are to be categorized to suit the requirements. Normally Triangular Membership Functions are preferred for the input and output fuzzy numbers so as to derive the fuzzy membership values in the range of 0-1 to reflect on degree of association. Here, income level is categorized into five fuzzy sets and travel time and Comfort and convenience levels are categorized in three fuzzy sets. Comfort & convenience are related to the aspects of vehicle-in comfort against the environmental impacts and the flexibility of service in terms of door to door movements. The commuters were provided three options of A, B and C meaning poor, fair and good level of comfort & convenience respectively for all three modes so that they can rate as per their experience or expectations with the influence factors as mentioned earlier. Instead of trip length which is on average 5.5 km for the work trips, travel time is preferred to have a good response from the travellers during the survey. Output of the model i.e. chosen mode is also categorized into three fuzzy sets viz. auto-rickshaw, two-wheeler, and car in view of the poor transit trips in absence of the effective public transport systems. Moreover, the urban work trips by bicycle and walking are quite meagre for longer trips and are ignored here. These variables are provided ranges for their Triangular membership as shown in Table 1a to Table 1d. Table 1a: MF of Input Variable: HH Income
Membership Function Very Low Low Medium High Very High
Shape Triangular Triangular Triangular Triangular Triangular
Range [0 0 15] [10 20 30] [20 30 40] [35 42.5 50] [45 95 95]
Table 1b: MF of Input Variable: Travel Time
Membership Function Low Medium High
Shape Triangular Triangular Triangular
Range [0 0 15] [5 20 40] [30 75 75]
Table 1c: MF of Input Variable: Comfort & Convenience level
Membership Function Low Medium High
Shape Triangular Triangular Triangular
Range [40 40 62.5] [55 62.5 70] [67.5 90 90]
Table 1d: MF of Output Variable: Travel Mode
Membership Function Auto-rickshaw Two-wheeler Car
Shape Triangular Triangular Triangular
Range [0 0 1] [0.5 1.25 2] [1.5 3 3]
Typical MATLAB snapshot of Mamdani Fuzzy Inference System is shown in Figure 5. The snapshots showing membership functions of input variables are shown in Figure 6a to Figure 6c respectively and the same for output variable is shown in Figure 6d.
Paper No. 210
Figure 5: Typical MATLAB Snapshot of Mamdani FIS
Figure 6a: MFs of Input Variable (HHI)
Figure 6b: MFs of Input Variable (TT)
Figure 6c: MFs of Input Variable (C&C)
Figure 6d: MFs of Output Variable (Mode)
5.3 Building of ‘If-then’ Rules Basic concept in fuzzy logic is that of a fuzzy if-then rule, called a Fuzzy Rule Base or Fuzzy Inference System. Here, the fuzzy „if-then‟ rules are developed and the typical example is “if household income is very low and travel time is low and comfort and convenience level is low then likely travel mode is Auto-Rickshaw”. 9 numbers of rules were framed for each income set as shown in Figure 7. Hence, a total of 45such rules are framed to constitute the rule base by different combinations among variables for five income categories.
Paper No. 210
Comfort & Convenience Level Travel Time Low Household Income Very Low
Travel Mode
Low
Medium
Medium
Auto-rickshaw, Two- Wheeler, Car
High High Figure 7: Schematic Diagram of Fuzzy Rules Formation
5.4 Defuzzification Process The fuzzy numbers are further defuzzified by Centre of Gravity method of defuzzification to get the result back in crisp format. The output is in terms of chosen travel mode. The rule viewer window is presented in Figure 8. The typical output value of 1.16 is for the three specified inputs of income (22,000), travel time (13 min) and comfort & convenience level (60 %).
Figure 8: Typical MATLAB Snapshot of Rule Viewer Window
Nearly 72 % of the data has been employed for calibration purpose and remaining for validation purpose. Fuzzification, Inference Process and defuzzification followed for the work are briefed below. The Mamdani type fuzzy inference system has been considered for modelling.
Paper No. 210 6. Analysis and Discussion
General Observations
Field observed and model predicted results are presented in Table 2. Accuracy level of 74% and 75% while calibrating and validating the dataset respectively has been achieved through the developed model. The developed model is predicting the auto-rickshaw mode with limited accuracy, and can be refined with certain changes in rule base. Table 2: Cross Classification Table for Calibration Dataset (Work Trips)
Observed (No. of Trips)
Predicted (No. of Trips) Two-wheeler
Auto-rickshaw
Car
Total
Two-wheeler
162
50
0
212
Auto-rickshaw
36
11
0
47
Car
4
0
86
90
Total
202
61
86
349
%
57.88
17.48
24.64
100
Prediction Accuracy observed for the validation dataset is 75.91 % which is very closer to predicted one. The comparison of observed and predicted modal split is shown in Figure 9 and has been found satisfactory. 70 60 Share %
50 40 30
Observed
20
Predicted
10 0 2W
Auto-Rickshaw
Car
Mode Figure 9: Comparison of Observed and Predicted Modal Split
Income Based Analysis
As the vehicle ownership is a function of household income, the set of available alternatives for travel mode varies among different socio-economic classes of the society. Table 3 shows the observed and predicted mode usage pattern for different income groups considered in this study. The developed FRB-MCM-W is able to precisely predict the travel modes for MIG and HIG, whereas it is having some discrepancy in case of LIG for which, lower know how of LIG people towards perception of travel parameters can be considered as the possible reason. It is interesting to note the meagre share of auto-rickshaws by the high income group for their work trips compared to other income groups as the ownership of vehicles is very high for the HIG.
Paper No. 210 Table 3: Comparison of Observed and Predicted Mode Choice Behaviour
Modal Share (%) Mode
LIG
MIG
HIG
Two-wheeler
53.0 (40.9)
70.0 (59.9)
54.5 (58.2)
Auto-rickshaw
43.9 (59.1)
13.0 (24.6)
4.7 (0.9)
Car
3.0 (0)
16.9 (15.5)
40.8 (40.8)
Total
100
100
100
Note - Values in brackets are Predicted values
7. Sensitivity Analysis
Effect of Reduction in Comfort and Convenience level of Car for work trips
Off-Street Parking is a facility consisting of dedicated parking spaces in terms of open spaces or multi-storeyed buildings. These dedicated parking spaces will make the car user to park it away from the destination and hence it will lead to discomfort in terms of long walking from its activity/service place to the parking spot carrying often the luggage through shopping activity. Secondly, if on-street parking is controlled, the car user has to search the space for his parking elsewhere resulting into dropping of car usage and rather move by two-wheelers or three-wheelers. The situation is further aggravated if highly congested traffic prevails en-route during peak periods on the main corridors to result in significant convenience level. The impact of considerable reduction in comfort and convenience level due to off-street parking and congested corridors can be realized in terms of modal split. The share of Car is likely to decrease from 40.8 % to 6.1% for HIG and from 15.5 % to 2.4 % for MIG as shown in Table 4 by higher level of discomfort and traffic congestion.
Table 4: Comparison of Modal Split after off-street Parking Port (%)
Mode
LIG
MIG
HIG
Two-wheeler Auto-rickshaw
39.4 (40.9) 60.6 (59.1)
72.9 (59.9) 24.6 (24.6)
93.0 (58.2) 0.9 (0.9)
Car Total
0.0 (0) 100
2.49 (15.5) 100
6.1 (40.8) 100
Note – Values in bracket are modal share before off-street parking
8. Conclusion Mode choice in fact is a subjective decision of the traveler for his socio-economic, modal, trip and network characteristics. Logit modelling in this regard is a popular conventional approach based on crisp inputs. In reality, it depends on the judgement by the traveler perceived and expressed in linguistic terms, wherein the boundaries are not fixed and uncertainty prevails. Hence Fuzzy Rule Based mode choice model is considered as the better option to address the shortcomings in crisp based modelling. FRB –MCM-W developed here for work trips addresses the uncertainty pertaining to travellers‟ decision process. Categorization of the various variables incorporated in the development of model and their ranges play important role in enhancing accuracy levels. The developed model finds applications in formulation of operation strategies and effective management of traffic flow on urban roads with due considerations for various income categories for fast growing Indian cities. The focus of the research is on modal choice behaviour for work trips in the presence of very meagre five percent share of city bus services during the study period which is not the case in majority of studies of similar kind. Acknowledgements
Paper No. 210 We thank all the respondents of the Household Interview Surveys conducted in Surat city for giving their precious feedbacks without which it was not possible to conduct this study. Also, we are thankful to Dr. G.J. Joshi for facilitating the research study being the Section Head at P.G. Section in Transportation Engineering & Planning, S.V. National Institute of Technology, Surat, India. References 1.
2. 3. 4. 5.
6. 7.
Holland, R. (2000). Fuzzy logic model of mode choice. In Proceedings of Seminar of the European Transport Conference 2000, Held at Homerton College, Cambridge, UK, 11-13 September, Transport Modelling. Volume P-445. Kumar, M., Sarkar, P., Errampalli, M., (2013). Development of Fuzzy Logic Based Mode Choice Model Considering Various Public Transport Policy Options. IJTTE, 3(4), pp. 408-425. Mamdani, E. H., & Assilian, S. (1975). An experiment in linguistic synthesis with a fuzzy logic controller. International journal of man-machine studies, 7(1), 1-13. Mizutani, K., & Akiyama, T. (2000, July). A Logit Model for Modal Choice with a Fuzzy Logic Utility Function. In Traffic and Transportation Studies (pp. 311:318). ASCE. Pulugurta, S., Arun, A., & Errampalli, M. (2013). Use of Artificial Intelligence for Mode Choice Analysis and Comparison with Traditional Multinomial Logit Model. Procedia-Social and Behavioral Sciences, 104, 583592. Teodorović, D. (1999). Fuzzy logic systems for transportation engineering: the state of the art. Transportation Research Part A: Policy and Practice, 33(5), 337-364. Wang, L. X., & Mendel, J. M. (1992). Generating fuzzy rules by learning from examples. Systems, Man and Cybernetics, IEEE Transactions on, 22(6), 1414-1427.