Research On Parking Space Prediction Methods Based On Wavelet Neural Network And Cuckoo Search Algorithm

With the rapid development of country's economy,people's lives are becoming more and more affluent,more and more private cars have entered thousands of ordinary families,which make the problem of city traffic more and more serious.Providing parking information reasonably for drivers can reduce time of finding available parking spaces,relieve traffic pressure,reduce air pollution and improve the utilization of parking spaces.Therefore,predicting real-time free parking spaces accurately has become an important research direction to solve the above problems.The provision of forecast information can avoid the occurrence of situations that a driver finds a parking space when he starts,but the parking space is full when he arrives,which also can alleviate congestion in some areas and idleness in other areas.At present,neural networks and their derived algorithms are commonly used for predicting the parking information.Firstly,the original data is reconstructed by means of phase space reconstruction,etc.to obtain training data embodying the internal relationship of the data,which is called data preprocessing.In the selection of neural networks,shallow networks such as BP neural network,wavelet neural network,Elman network,etc.are commonly used.At the same time,prediction models are constructed by combining other optimization algorithms,such as particle swarm optimization(PSO),genetic algorithm(GA),and cuckoo search algorithm(CS)and so on.This paper analyzes and summarizes these methods,and improves them on the basis of their advantages so as to obtain a fast and accurate prediction model.In order to optimize the above prediction model further,this paper continues to propose an improved CS algorithm for optimizing wavelet neural networks.The improved CS algorithm changes the traditional updating and evaluation strategy to the group updating and evaluation strategy that is proposed on the basis of the dimension-by-dimension updating and evaluation strategy.To the problems that the number of dimensions of nest is large and the fitness function runs for a long time,the improved CS algorithm not only retains the advantage of rapid convergence of the dimension-by-dimension updating and evaluation strategy,but also enhances the inter-dimension relationships of the nest,reduces the time complexity and improves the overall running time.The experimental results show that the proposed CS algorithm outperforms the original CS algorithm and the CS algorithm with the dimension-by-dimension updating and evaluation strategy in terms of overall running time and prediction accuracy of the model.