Enhancing urban traffic management through shared autonomous electric vehicles and dynamic simulation Authors: Jingfa Ma, Hu Liu, Lingxiao Chen Published: 2024 ABSTRACT In the face of rapid urbanization and the increasing number of vehicles, urban centers are struggling with traffic congestion. This study presents a dynamic travel strategy using the MATSim platform to schedule urban travel, incorporating a model for shared autonomous electric vehicles. The model is evaluated using a baseline scenario for Shanghai, exploring the effects of vehicle range, charging capabilities, and power supply strategies on the uptake of shared autonomous electric vehicles. Results indicate that enhancements in vehicle range and charging efficiency slightly decrease the use of autonomous vehicles by 2.5%, as the existing vehicle specifications already meet daily travel needs in Shanghai. Additionally, the transition from traditional charging stations to a battery-swapping system does not significantly alter overall travel behavior of shared autonomous electric vehicles. These findings provide insights into the deployment of intelligent traffic systems to alleviate urban traffic congestion. Introduction Transportation systems are crucial for modern urban life, yet rapid urban population growth has led to increased motor vehicle numbers, exacerbating challenges like traffic congestion [1]. Shanghai, a major economic hub and one of China’s most densely populated cities, exemplifies these issues with severe congestion. According to the "2023 China Urban Transportation Report," Shanghai is now the third most congested city in China, with its congestion index rising to 1.834 from the previous year. This persistent congestion not only disrupts urban mobility but also impedes broader urban development. To address these challenges, intelligent transportation systems (ITS) are essential. ITS integrate advanced technologies, including sensors, computers, electronics, and sophisticated communication and management strategies, to enhance the safety and efficiency of urban transportation networks [2]. In recent years, research on shared transportation systems has delved deeper into alleviating urban traffic pressure and enhancing transport efficiency. For instance, studies have shown that integrating the location-routing problem of green logistics with eco-friendly packaging can effectively reduce the environmental impact of transportation [3]. The multi-depot vehicle routing problem has been optimized further through smart recycling pricing and transportation resource sharing, improving resource utilization and logistics efficiency [4]. Additionally, research on the truck-drone hybrid routing problem, which considers time-dependent travel times on roads, has proposed new combinations of transportation modes to boost efficiency [5]. The two-tier multi-depot multi-period location-routing problem significantly enhances the overall efficiency of logistics systems by simultaneously handling pick-up and delivery tasks [6].Given its high travel demand and congestion levels, Shanghai is a critical case study for implementing such systems to alleviate traffic issues and support ongoing rapid development. The rise of mobile internet, smartphones, and innovative automotive technologies has facilitated the emergence of shared autonomous driving. This modern transportation mode provides membership-based access to autonomous vehicles for short durations, managed by third-party organizations. Users incur time-based charges and use digital platforms for reservations, vehicle collection, and payments. Shared autonomous driving offers various operational models: round-trip, where vehicles are returned to their pickup points; one-way, allowing drop-off at different locations but posing fleet management challenges; and the increasingly popular point-to-point service, enabling departures and arrivals at user-defined locations. Enhanced by advanced data processing technologies, the point-to-point model aims to improve vehicle distribution efficiency across the network. In the realm of global autonomous driving vehicles, initial research and testing of autonomous driving technology were conducted by Google on traditional fuel-powered vehicles in 2010. Due to the rapid development of electric vehicles, evolving autonomous driving technologies have been increasingly applied to electric vehicles. With advancements in technology, infrastructure, and public policy, from the provision of partial autonomous driving functions in 2014 to the support of full autonomous driving capabilities by 2024, autonomous driving technology in electric vehicles has continuously improved. Consequently, an increasing number of electric vehicles are adopting advanced autonomous driving technologies. The application of autonomous driving technology in point-to-point mode shared electric vehicles has sparked considerable discussion and interest. Given the nascent stage of shared autonomous vehicles, the intelligent transportation sector is navigating uncharted territory, raising crucial inquiries about vehicle selection, charging methods (whether to use charging stations or battery swaps), and optimal placement of charging stations to enhance user engagement and minimize expenses. The traditional method of trial and error, though prevalent, falls short in efficiently scaling autonomous vehicle services for point-to-point sharing in China. This necessitates a refined model to accurately forecast the dynamics of supply and demand for these vehicles. The complexity of shared autonomous vehicle systems, encompassing elements like "user, location, parking, vehicle, and energy," presents challenges due to the interconnected feedback between supply and demand. For example, the future vehicle availability at a location could be influenced by various factors including the charging needs and expected arrival times of vehicles en route, which may be affected by traffic congestion. Consequently, the development of a shared autonomous vehicle framework demands a comprehensive understanding of spatial and temporal vehicle data. Traditional mathematical models may not fully capture this complexity, whereas simulation techniques can effectively represent the actions and interactions of distinct agents or events, fulfilling the shared autonomous vehicle model requirements. In light of this, our study proposes a dynamic travel strategy that uses an adaptive scheduling method to optimize vehicle dispatching and user demand matching during simulation iterations. This is implemented using the MATSim platform, an enhanced open-source multi-agent traffic simulation framework tailored for dynamic travel scenarios, to build a shared autonomous driving model. The aim is to enhance intelligent transportation and alleviate traffic congestion.By integrating travel, road, and public transportation data, we constructed and evaluated a foundational scenario for Shanghai.This study examines how variables such as vehicle range, charging speed, and power options affect user demand.The goal is to lay a scientific foundation for the development and design of autonomous vehicle sharing systems, ultimately promoting efficient urban travel planning and advancing autonomous driving technology to mitigate traffic congestion in Shanghai. The structure of this paper includes: Introduction: Discussion on research background and real-world needs. Introduction: Discussion on research background and real-world needs. Related Review: Review of relevant research achievements, specifying the key focus and significance of this study. Related Review: Review of relevant research achievements, specifying the key focus and significance of this study. Dynamic Travel Planning Optimization: Convergence optimization of MATSim platform for dynamic travel. Dynamic Travel Planning Optimization: Convergence optimization of MATSim platform for dynamic travel. Autonomous Driving Shared Car Module Constructiont: Detailed description of the development of the shared autonomous driving model, including the design of baseline scenarios and simulated situations. Autonomous Driving Shared Car Module Constructiont: Detailed description of the development of the shared autonomous driving model, including the design of baseline scenarios and simulated situations. Results: Presentation of model validation and scenario results. Results: Presentation of model validation and scenario results. Finally, conclusions and future prospects are discussed. Literature review When facing the challenge of traffic congestion, one of its core issues is the imbalance between supply and demand. The increase in traffic demand directly affects the continuous traffic flow on roads in densely populated areas, which are not only the main arteries for inter-city traffic but also critical points directly affected by congestion within cities. Various factors contribute to traffic congestion, including accidents, adverse weather conditions, slow-moving heavy vehicles, sudden vehicle breakdowns, traffic violations, congestion at intersections, and infrastructure conditions [7]. In response to these challenges, the application of intelligent transportation systems has been proposed, which includes the application of dynamic travel and autonomous driving technologies. The former can reasonably plan traffic behaviors and activities based on travelers’ activity plans, while the latter can avoid traffic congestion by dynamically planning vehicle operations in real-time, thereby improving travel efficiency and alleviating urban congestion. This study has several limitations. Firstly, the simulations rely on data concerning vehicle range, charging speed, and user behavior, which may not fully capture the complexity of real-world scenarios. Secondly, the analysis is constrained by available data, which may not encompass all variables influencing traffic patterns and vehicle usage. Despite these limitations, this study makes a significant contribution to the field of intelligent transportation systems. The study introduces a dynamic travel planning model using MATSim to optimize urban mobility and facilitate the application of shared autonomous electric vehicles. The research provides valuable insights into how vehicle range and charging infrastructure affect the adoption and usage of autonomous driving vehicles. Furthermore, the findings highlight the potential of battery swapping systems as an alternative to traditional charging stations. These contributions lay the groundwork for future research, urban traffic management, and practical applications of autonomous driving technology. Application of dynamic travel [8] Provided system predictions via mobile devices to passengers of stochastic multi-service public transit networks, including transit travel times and arrival times at stops, utilizing a dynamic optimal strategy search approach [9]. Studies have shown that informing drivers in advance about real-time traffic conditions and potential delays is crucial. In situations of traffic congestion, effectively influencing drivers to change their original route selections through dynamic travel information applications using variable message signs is considered an effective approach. A predictive model based on historical traffic data and current traffic information was proposed, presenting a data-driven traffic path planning algorithm aimed at finding the shortest travel time [10]. In recent years, methods for addressing traffic congestion have focused on optimizing path algorithms, controlling traffic signals, and reducing vehicle density to increase average vehicle speeds and thus mitigate the impacts of traffic congestion. Dynamic travel management is applied to single public transit networks and subway networks based on timetable-based road networks, or used in predictive models to find the shortest travel time traffic paths. Although there have been studies on dynamic travel and traffic management, few have directly applied dynamic travel planning to address traffic congestion with shared autonomous vehicles. This paper utilizes the MATSim platform as a supporting tool, which stands for Multi-Agent Transport Simulation. This open-source software, based on agent-based simulation methods, can simulate the travel behaviors of a large number of individuals within a city or region, including mode choice, path selection, and time distribution, thereby predicting traffic flow, congestion situations, and travel time distribution, among other traffic indicators. Support for autonomous driving implementation In the realm of shared autonomous vehicle simulation studies, researchers primarily deploy two kinds of models: agent-based and event-based. Agent-based models are adept at depicting the behavior and interaction of autonomous entities within a framework. For instance, Heilig and colleagues applied this approach to estimate the demand for shared autonomous vehicles around Stuttgart over a period of one week [11]. Zhang and team adopted it to examine the dynamics within integrated multimodal transport systems, analyzing how vehicle and charging system characteristics interact [12]. Wang and his group explored the benefits of agent-based simulations for enhancing shared mobility services [13]. Meanwhile, Creutz and associates utilized MATSim, Meanwhile, Creutz and his colleagues studied the appeal of shared autonomous vehicles using MATSim (Multi-Agent Traffic Simulation), an agent-based microscopic traffic simulation tool capable of modeling individual behavior in large-scale transportation systems; they gradually refined the simulation framework to include features such as station capacity and advanced reservation mechanisms, and established the base scenario for their experiments in Zurich, to model the appeal of shared autonomous vehicles, progressively refining their simulation framework to include features like station capacity and advanced reservation mechanisms, setting up a foundational scenario in Zurich for their experiments. This area of study has attracted extensive examination into the dynamics of supply and demand for shared autonomous vehicles, covering aspects such as the variation in membership and fleet sizes, optimization of parking space, adjustments in shared vehicle pricing, vehicle redistribution, and the competitive landscape among providers. On the contrary, discrete event models map out sequences of individual occurrences over time, mirroring the ongoing shifts within a system’s state [14]. Jefferies and team employed such a model to evaluate electric bus systems, focusing on scheduling and the charging infrastructure [15]. Similarly, Wang and colleagues utilized an event-based approach to model how the demand for shared autonomous vehicles and parking spaces might evolve [16]. Though both model types have been scrutinized, discrete event models stand out for their ability to capture the dynamic nature of events over time, reflecting systemic changes. Yet, few have managed to incorporate specific functionalities crucial for simulating shared autonomous vehicle scenarios, such as: Allowing autonomous vehicles to compete with other transport modes directly, with algorithms that help agents choose more efficient travel options. Allowing autonomous vehicles to compete with other transport modes directly, with algorithms that help agents choose more efficient travel options. Adjusting autonomous vehicle travel times based on current traffic conditions. Adjusting autonomous vehicle travel times based on current traffic conditions. Enabling simulations to dynamically adjust routes and departure times based on real-time data. Enabling simulations to dynamically adjust routes and departure times based on real-time data. Providing detailed spatial and temporal output, essential for precise modeling and analysis of transportation networks. Providing detailed spatial and temporal output, essential for precise modeling and analysis of transportation networks. Advantages of MATSim Implemented in Java as an open-source simulation platform, MATSim excels at depicting traffic jams, modeling intricate supply-demand dynamics, facilitating agent interactions, and supporting agent learning and adaptation. It delivers outputs with detailed geographical and chronological precision. This suite of features establishes MATSim as the go-to software for analyzing dynamic travel behaviors and the intricacies of shared autonomous vehicle simulations. In MATSim, shared transportation systems are positioned to compete against traditional transport methods, leveraging a sophisticated scoring system to guide agents toward the most efficient travel options. In MATSim, shared transportation systems are positioned to compete against traditional transport methods, leveraging a sophisticated scoring system to guide agents toward the most efficient travel options. It models the impact of traffic congestion on the duration of shared transportation journeys. It models the impact of traffic congestion on the duration of shared transportation journeys. MATSim boasts the functionality to adjust travel paths and scheduling in real-time, responding dynamically to changing conditions. MATSim boasts the functionality to adjust travel paths and scheduling in real-time, responding dynamically to changing conditions. It offers outputs that are rich in spatial and temporal details, enhancing the depth and utility of simulations. It offers outputs that are rich in spatial and temporal details, enhancing the depth and utility of simulations. Modeling methods Problem statement With rapid urbanization and an increasing number of vehicles, especially in megacities like Shanghai, traffic congestion is becoming increasingly severe. Congestion not only disrupts residents’ daily travel but also negatively impacts the city’s economic development.Current research primarily focuses on sharing conventional fuel vehicles, while electric autonomous vehicles face additional challenges due to their range and charging time limitations.Efficiently integrating vehicle sharing and autonomous driving technologies to enhance traffic efficiency and reduce congestion is an urgent issue needing resolution.The widespread adoption of electric vehicles requires effective charging infrastructure.However, traditional charging station models may face spatial and temporal limitations in urban areas. The effectiveness and feasibility of battery swapping systems as an alternative solution require further exploration and validation.Addressing these issues, this study aims to develop a dynamic travel planning model to optimize intelligent transportation systems, enhance the efficiency of shared autonomous electric vehicle applications, and ultimately alleviate traffic congestion. Optimization of convergence criteria for dynamic travel planning algorithms MATSim dynamically simulates agents’ travel behavior by following a detailed process that includes initialization, plan generation, iterative simulation, evaluation, and optimization. The objective is to optimize travel paths and activity scheduling, reflecting real-time traffic conditions and agent interactions. The MATSim Simulation Process is illustrated in Fig 1. Initialization: Input data such as city maps, road networks, agent locations, and travel demands are prepared. Initialization: Input data such as city maps, road networks, agent locations, and travel demands are prepared. Plan Generation: Agents’ travel demands are converted into detailed activity and movement plans. Plan Generation: Agents’ travel demands are converted into detailed activity and movement plans. Iterative Simulation: Agents execute their plans in a simulated environment, interacting with other agents and simulating traffic flow over a set period. Iterative Simulation: Agents execute their plans in a simulated environment, interacting with other agents and simulating traffic flow over a set period. Evaluation: Agents’ behavior and simulation results are evaluated to score and optimize future plans. Evaluation: Agents’ behavior and simulation results are evaluated to score and optimize future plans. Optimization: Plans are adjusted based on evaluation scores to improve travel efficiency and path selection. Optimization: Plans are adjusted based on evaluation scores to improve travel efficiency and path selection. Five main steps of MATSim. This cycle continues until a termination condition is met, such as a preset time or traffic stability, allowing MATSim to assess the impact of various traffic strategies and policies.However, a high number of iterations will also lead to increased model complexity and increased computing time and resource requirements.The questions are as follows: Extended Computation Time: Simulating multiple plans for each agent results in a significant computational workload, especially with numerous travel plans. Extended Computation Time: Simulating multiple plans for each agent results in a significant computational workload, especially with numerous travel plans. Increased Model Complexity: As the number of travel plans grows, maintaining and analyzing the model becomes more challenging. Increased Model Complexity: As the number of travel plans grows, maintaining and analyzing the model becomes more challenging. Higher Demand for Computational Resources: More travel plans require additional computational resources, which may be limited. Higher Demand for Computational Resources: More travel plans require additional computational resources, which may be limited. One of the main factors affecting the number of iterations in MATSim is the number of plans generated by the agents. Currently, MATSim offers two plan selection settings: Unlimited Plans: This setting allows for the generation of an unlimited number of travel plans. While it comprehensively covers all possible travel scenarios, it results in excessive computational demands and increased model complexity. Unlimited Plans: This setting allows for the generation of an unlimited number of travel plans. While it comprehensively covers all possible travel scenarios, it results in excessive computational demands and increased model complexity. Limited to 5 Plans: In this setting, each agent retains only the top 5 highest-scoring plans. This reduces the computational burden by limiting the number of plans, but it may lead to the elimination of potentially optimal plans. Limited to 5 Plans: In this setting, each agent retains only the top 5 highest-scoring plans. This reduces the computational burden by limiting the number of plans, but it may lead to the elimination of potentially optimal plans. To address the aforementioned issues, this paper proposes a method for dynamically adjusting the number of plans. Initially, a higher limit on the number of plans is set at the beginning of the iterations. As the iterations progress and the system stabilizes, the limit on the number of plans is gradually reduced. This approach maximizes the retention of high-scoring plans and minimizes the risk of deleting globally optimal plans due to a fixed limit. Model optimization Through this improved approach, the main innovation lies in setting convergence conditions. In the computation of the mean, the method involves iterating through each plan in the plan list, summing their scores, and then dividing by the number of plans to obtain the average. Subsequently, in computing the standard deviation, the method calculates the square of the difference between each plan’s score and the mean, sums these squared values, divides by the number of plans, and finally takes the square root to obtain the standard deviation. Next, the percentage change between the current mean and the mean from the previous iteration is calculated and compared against a preset threshold to determine if convergence has been reached. Similarly, the percentage change between the current standard deviation and the standard deviation from the previous iteration is calculated and compared against the threshold. This method assesses convergence by comparing the rate of change between the current mean and the previous iteration’s mean, as well as between the current standard deviation and the previous iteration’s standard deviation. mean=∑Scoreii (1) mean is the average score of the current iteration." i represents the ith iteration (i = 1, 2, 3…) Scorei is the plan score of the ith iteration. std=∑(Scorei−mean)2i (2) std is the standard deviation of the current iteration. |mean−prevMean|prevMean