Research And Implementation Of Vehicle Swarm Intelligence Learning Method For Travel Efficiency

The perception ability,computing ability and communication ability of autonomous driving vehicles are limited.It is unable to obtain a wider range of traffic situation information and deploy the multi-vehicle cooperative algorithm with high computational complexity.With the development of edge computing,Internet of vehicles and 5G communication technology,the computing,communication and perception capabilities of road traffic network have been greatly improved.Vehicles can obtain more abundant traffic situation information and complete some complex tasks by using these capabilities of road traffic network.Therefore,this paper proposes a multi-vehicle cooperation framework based on swarm intelligence to build the computing model of swarm intelligence in edge cloud.Based on this framework,two algorithms are proposed to improve the efficiency and safety of road traffic network from macro and micro perspectives respectively.Multi-vehicle path planning determines the distribution of traffic flow in the entire road network,so it improves the efficiency of the road traffic network from a macro point of view.The current method of using global traffic situation information to plan the driving path for a single vehicle does not consider the interaction between multiple vehicles,which leads to the phenomenon of vehicles crowding on the road.The multi-vehicle path planning algorithm based on centralized controller can not be applied in the actual road network because of its high computational complexity and may damage the interests of the vehicles themselves.In this paper,a population game based on best response dynamics is proposed to model multi-vehicle cooperative path planning problem which ensures the autonomous decision of vehicles and the maximization of their own interests.Simulation experiments show that the proposed path planning algorithm can greatly improve the throughput of the road network,reduce the travel time for vehicles to reach the destination,and make the traffic flow on the road in a uniform state.The cooperative control of multiple vehicles at the intersection plans the driving behavior of vehicles at the intersection,and ensures that vehicles can pass through the intersection quickly without collision,thus improving the operation efficiency and safety of the road network from a micro level..At present,the intersection control based on no signal lights takes the vehicle as the instruction execution node to execute the instructions issued by the edge.This method damages the autonomy of the vehicle,and can not cope with the complex and changeable traffic situation,and there are safety risks.In this paper,a neural network model of multi-vehicle cooperative control based on multi-agent reinforcement learning is proposed,and the global traffic situation information is used to coordinate the training process of each agent,and finally the result of cooperative crossing can be achieved.Simulation experiments show that the multi-agent reinforcement learning algorithm used in this paper enables vehicles to learn a good cooperative strategy to pass the intersection and achieve a very low probability of vehicle collision.