This work presents a novel approach to the simulation-based optimisation for Autonomous Transportation Systems (ATS) with the use of the proposed parallel genetic algorithm. The system being developed uses GPUs for the implementation of a massive agent-based model of Autonomous Vehicle (AV) behaviour in an Artificial Multi-Connected Road Network (AMСRN) consisting of the “Manhattan Grid” and the “Circular Motion Area” that are crossed. A new parallel Real-Coded Genetic Algorithm with a Scalable Nonuniform Mutation (RCGA-SNUM) is developed. The proposed algorithm (RCGA-SNUM) has been examined with the use of known test instances and compared with parallel RCGAs used with other mutation operators (e.g., standard mutation, Power Mutation (PM), mutation with Dynamic Rates (DMR), Scalable Uniform Mutation (SUM), etc.). As a result, RCGA-SNUM demonstrates superiority in solving large-scale optimisation problems when decision variables have wide feasible ranges and multiple local extrema are observed. Following this, RCGA-SNUM is applied to minimising the number of potential traffic accidents in the AMСRN.

Представлен подход к исследованию эффектов сегрегации с использованием разработанной мультисекторной модели ограниченного соседства. Предложена модель эволюционной динамики сообщества, состоящего из местного (коренные жители) и внешнего населения (мигрантов), взаимодействующих в искусственной социально-экономической системе, в которой выделены ключевые секторы экономики: добыча сырья (первичный сектор, привлекающей преимущественно мигрантов), производственный сектор (вторичный сектор, привлекающий преимущественно коренных жителей) и сфера низкотехнологичных и высокотехнологичных услуг (третичный и четверичный сектора экономики, привлекающие мигрантов и коренных жителей, соответственно). Формирование рабочих мест в данных секторах экономики осуществляется централизованно с использованием ранее предложенного алгоритма нечeткой кластеризации. Выполнены имитационные эксперименты и исследованы эффекты сегрегации, обусловленные стремлением агентов к поиску наиболее предпочтительных рабочих мест в условиях ограниченного соседства при различных сценарных условиях. Используя предложенный генетический алгоритм, решена важная оптимизационная задача по максимизации темпов роста ВВП и минимизации уровня сегрегации населения.

The article examines a system for controlling the ecological modernization dynamics of enterprises developed with the help of simulation modelling methods and implemented using the example of the Republic of Armenia (RA). The system has been developed for strategic decision-making directed at modernization of enterprises of RA, their transformation from an initial non-ecological state towards the state of ecologically pure manufacturing.The main feature of the software developed is an original agent-based model describing the dynamics of the ecological-economics system. The system has been implemented using the AnyLogic platform. This model is integrated with a multidimensional data warehouse, genetic optimizing algorithm (modifi ed for the bi-objective optimization problem of an ecological-economics system), a subsystem of simulation results visualization (Graphs, Google Maps) and other software modules designed with use of the Java technologies.The target functionalities of the bi-objective optimization problem of the ecological-economics system are minimized integrated (accumulated) volume of total emissions into the atmosphere and maximized integrated (averaged) index of industrial production of the agent's population. The problem was formulated and solved for the fi rst time. Moreover, values of objectives are calculated by means of simulation, as the result of activity of all agent-enterprises in a population and taking into account their internal interaction. The 270 enterprises of RA which are the main stationary sources of emissions of harmful substances were selected for the research. In addition, there is a generalized agent-consumer and the agent-government completing ecological regulation through the mechanisms of penalties, subsidies and rates of emissions fees. The simulation core is the developed algorithm of behavior for each agent-enterprise providing the mechanism of agent transition from an initial non-ecological state towards other possible states. At the same time, control of the evolutionary dynamics of agents is implemented with the help of the suggested genetic algorithm. As a result, the system we developed makes it possible to search Pareto-optimal decisions for a bi-objective optimization problem of the agent-based ecological-economics system.

No abstract

A new approach to modeling the spatial dynamics of unmanned ground vehicles (UV) and conventional vehicles (CV) using the FLAME GPU supercomputer agent-based simulation platform is presented. A new simulation model of an artificial road network (ARN) of the "Manhattan Grid" type is proposed, within the framework of which the spatial dynamics of the UV and CV ensemble is studied under various scenario conditions. The effects of "turbulence" and "crush" (traffic congestion) resulting from intensive and dense traffic of vehicles are investigated.