An evolutionary algorithm for an agent-based fleet simulation focused on electric vehicles

This paper presents a multi-agent simulation for fleet applications with a primary focus on electric vehicles. The overall system design is characterized by a distributed and decentralized structure. Basic idea is to use an evolutionary algorithm in combination with a Monte Carlo method to optimize behavioral patterns of agents. Within an iterative process, agent plans are executed, scored and modified - if necessary. In doing so, a set of activity schedules is created that is compatible with given constraints, such as range restrictions of electric cars or space limitations at charging stations. In so doing agents compete with each other for limited resources within a single iteration step. The proposed model makes use of a map-based approach that takes time variant traffic conditions into account. With help of speed profiles and a longitudinal vehicle dynamic model, travel time and energy consumption calculations are carried out. As the fleet and the charging infrastructure configuration are input parameters of the proposed model, it is possible to perform case studies for several electrification scenarios. In total two different scenarios are explored. First the proposed simulation model is validated with help of measured fleet data. Second the suitability of electric cars for a taxi use case is assessed. Within the first simulation study, the daily mileage is reflected with an average variation of 7.8 % and waiting times with an average deviation of 0.05 %. The second electrification scenario provides transparent indications for choosing both an appropriate electric vehicle concept and charging infrastructure configuration.     «

This paper presents a multi-agent simulation for fleet applications with a primary focus on electric vehicles. The overall system design is characterized by a distributed and decentralized structure. Basic idea is to use an evolutionary algorithm in combination with a Monte Carlo method to optimize behavioral patterns of agents. Within an iterative process, agent plans are executed, scored and modified - if necessary. In doing so, a set of activity schedules is created that is compatible with gi...     »