MODELING TOLL ROAD RAMP-UP PERIOD USING AGENT-BASED SIMULATION WITH REINFORCEMENT LEARNING APPROACH(STUDY CASE: SOREANG-PASIRÂ KOJAÂ TOLLÂ ROAD)

• Main Author: Indra Dharmawan, Weka
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/74851
• Institution: Institut Teknologi Bandung

There are several issues that have received less attention but have a significant impact on toll road infrastructure projects, especially those with private sector investment. One of them is the ramp-up period, as the delay of a new toll road infrastructure service in building the demand of its potential users due to the learning and adapting process for regional travelers due to not knowing and feeling the benefits of what is offered when used. Traffic volumes during the ramp-up period experienced significant growth and tended to fluctuate. However, over time, it will be different when entering a steady state phase with reduced growth and fluctuations in traffic volume, indicating that toll road infrastructure has reached its demand equilibrium and is starting to enter the market maturity phase. Conventional models that are still aggregation in nature have not been able to anticipate the problems of the ramp-up period. Therefore, a disaggregate model approach is considered more realistic to predict the potential volume of toll road traffic during the ramp-up period. The learning curve in the form of a traffic volume growth graph is considered as an illustration of the learning and adapting process in the decision-making attitude of travelers to determine route choices. The research methodology begins with a literature study to identify parameters that affect the learning behavior and decision-making attitude of travelers during the route selection process. An inventory of the elements involved in the model was conducted to assist in making research instruments and survey needs. The revealed preference (RP) and stated preference (SP) survey process of travelers was conducted in the study area (Bandung City, Bandung Regency, and Cimahi City). Model development uses agent-based simulation with a reinforcement learning (RL) approach to represent the learning process in the form of reward or punishment for the route, the decision-making process for alternative route options is approached by a discrete selection model (DCM) in the form of a logitmultinomial. The existence of limited information and uncertainty conditions on the travel route to be chosen by the travel agent, allows irrational decision-making to be modeled based on the theory of bounded rationality (BR). The model that has been built is simulated on a hypothetical small network. After being considered good enough, it is implemented in a case study of real network problems in the field. The final stage of the research methodology is to draw conclusions and suggestions, which hopefully can be used as a basis for further research. Simulated on a hypothetical network of 2 routes and expanded to 3 routes, it is understood that there are 3 parameters that affect the shape of the learning curve output and convergence rate as a determinant of toll road ramp-up duration, namely learning rate (?), logit model attribute coefficients (? ), and bounded rationality (?). In the model, the learning rate parameter functions to update the perceived value of alternative route attributes (in this case travel time and cost) from previous travel experiences, while the attribute coefficients of the logit model and bounded rationality, have an effect in determining and updating the probability value of alternative route choices. The model results all indicate the existence of user equilibrium conditions. The application of the model to the case study, obtained the prediction results of the largest Soroja Toll Road traffic volume is at the Pasir Koja-Margaasih Toll Gate which is dominated by groups of workers, freight transportation groups, other travel groups, student groups, and business groups. This is influenced by land use with access to the West Margaasih and East Margaasih toll gates, namely the Kutawaringin industrial area and dense residential areas (including Taman Kopo Indah Housing). The road segment at the Soreang-Kutawaringin Toll Gate is the lowest, due to only being served by one Toll Gate, namely the West Kutawaringin Toll Gate as access to the Si Jalak Harupat stadium. The movement of toll road traffic in the Margaasih-Katapang Toll Gate segment can be considered as trough traffic, whose movement comes from the Pasir Koja Toll Gate to the Soreang Toll Gate. The magnitude of the learning level parameter (?) affects the convergence time, the agent traveler group of workers (? = 0.65) is reached around the 250th day, while the business people and others group (? = 0.45) around the 500th day, in contrast to the student group (? = 0.10) around the 1,000th day, and the freight transportation group (? = 0.45) but still shows an increase in traffic volume (in the Pasir Koja-Margaasih Toll Gate segment). The results of the sensitivity test of the model to toll rates in the Soroja Toll Road case study, predicted traffic volume when the toll rate was increased to Rp. 5,000.00 decreased by 51% from the original free. For an increase in the toll tariff to Rp. 8,500.00, users of the Soroja Toll Road became 29%, then the tariff was increased again to Rp. 17,500.00 the traffic volume decreased to 13%. However, it appears that freight transportation is less sensitive to tariff increases than passenger cars, based on the survey results, generally freight cars using the Soroja Toll Road are fruit and vegetable carriers that must immediately arrive at the location (market), in addition, toll fees are borne by the company (generally for non-fruit and vegetable freight transportation). Travelers using passenger cars, who are most sensitive to toll tariffs are groups of students or college students, this is possible because transportation costs are still burdened by their parents. The ramp-up period of the model results based on the convergence approach using the gradient (ratio) method with the tolerance value h assumed to be 0.95 < h < 1.00 and H = 12 months, the length of the ramp-up of the Soroja Toll Road model results is 16 + 12 = 28 months (2 years, 4 months), while for field data assumed to be 0.85 < h < 1.00 and H = 12 months, the length of the ramp-up period is 18 + 12 = 30 months (2.5 years). Determining the predicted ramp-up length of the Soroja Toll Road is rather difficult to compare with the actual data due to the abnormal traffic volume that occurred during the COVID-19 outbreak in March 2020 to February 2022. However, visually it can be seen that the ramp-up condition occurs during the 30th month, where traffic growth has started to slope. The validation of the simulation model that has been built to predict the traffic volume of the Soroja Toll Road with actual data in this dissertation research shows that the determination coefficient test obtained the determination value (R2) is 86.80% (model competence is considered strong), while the test results using the mean absolute percentage error (MAPE) are 30.54%, which indicates that the model competence is still declared feasible.