Research On Path Planning Of Complex Parking Lot Based On Q-learning

With the increasing share of urban vehicles,it is difficult to solve the problem of urban parking.Traditional parking lots have the problems of insufficient parking spaces and low level of intelligence,while urban development requires the continuous improvement of social public infrastructure such as parking lots.Compared with the parking lots in traditional business districts,the volume of high-rise buildings in urban shopping malls is constantly expanding.In order to solve the problem of insufficient parking spaces and improve the utilization rate of space,the construction of parking lots is developing in a threedimensional way,which means that the environment in parking lots is more complicated,drivers have some difficulties in finding parking spaces,and the requirements for parking lot path planning are also becoming higher,If the current parking lot is full,vehicle parking needs to consider the surrounding parking lot and road traffic conditions.As an important part of the intelligent parking lot,path planning can improve the problems of the current parking lot.Through the path planning,vehicles can be guided to drive to the destination with the optimal path.To a certain extent,the driving time for vehicles to find parking spaces can be reduced,the driving efficiency of vehicles can be improved,and the economic consumption can be saved.Traditional path planning algorithms often need to process the environmental information first,so it is difficult to ensure the real-time performance of the algorithm.Reinforcement learning can make use of the interaction between agent and environment to carry out self-learning and give the optimal strategy.Therefore,this paper proposes the following improvement schemes based on the reinforcement learning algorithm on the parking path planning problem:(1)In order to improve the convergence speed of the path planning algorithm for multiple areas in the parking lot,the SL-Q(λ)algorithm based on action selection optimization is proposed.The Q-learning algorithm is combined with the qualification trace to update the Q value calculation formula,and the exploration probability and the number of successful explorations Correlation makes the value of the exploration probability dynamic,and conducts simulation experiments in a multi-region parking lot environment to verify the feasibility of the algorithm.(2)In order to solve the dimensional disaster problem caused by the multi-storey parking environment,the DEDQN algorithm research based on the optimization of the experience pool is proposed,the target network formula is improved,and the experience pool for storing effective samples that can reach the final target point and explore the final position close to the target point is added.A simulation experiment is carried out in a multilevel constrained parking lot environment to verify the feasibility of the algorithm.(3)In order to solve the problem of multi-vehicle parking in a parking lot and the need to find nearby parking lots when the current parking spaces are full,a path planning method based on multi-agents is proposed,which uses a combination of centralized learning and distributed learning.Joint state,joint action and reward function,use the DEDQN algorithm to experiment,experiment in the multi-storey parking environment for the feasibility of multi-agents to find multiple targets,and test the multi-agents in multiple parking environments to find the optimal parking plan feasibility.This paper studies the problems that may arise in complex parking lots,and verifies the feasibility and stability of the algorithm by building an environment based on real parking lots.Finally,other optimizable directions are proposed to provide references for subsequent research.