Research On Prediction Algorithm Of Free Parking Stalls Based On Genetic Algorithm And Ensemble Learning

With the rapid development of China's comprehensive national strength,proud progress has been made in all aspects of the construction of the motherland.The single car ownership shows a rapid growth trend every year.By September 2018,the number of motor vehicles in China had reached 265 million,and the accompanying traffic problems were also increasing.The problem of parking is more difficult.Parking lot as a supporting service develops slowly,and the more unbalanced the development,the more vicious circle.To solve these problems with the development of existing technology,we can look at them from two perspectives:virtual and real.From the realistic point of view,the parking lot construction is increased,but there are many factors involved in the urban construction.Although the problems are in existence,the priority of solving them is not high in general.So from the virtual aspect,we can make real-time and effective prediction of the spare parking spaces in the existing parking lots,provide reference for drivers to travel,and relieve the above problems by sharing information under the existing conditions.The prediction of spare parking spaces in parking lots studied in this paper is essentially a practical application of time series prediction.According to the forecast demand,it can be divided into two forecasting methods: short-term and multi-step.The former uses existing data sets to predict the spare parking spaces at a certain time point in the future.The exponential smoothing method and its derivative algorithm are widely used.The latter is based on historical data sets to predict the number of spare parking spaces in the future.Neural network correlation algorithm is a common method to solve this kind of problem.This paper analyses the short-term and multi-step forecasting methods of spare parking spaces nowadays.Through the analysis of their principles and applications,it is found that there are obvious shortcomings in the methods and some areas with strong regularity.Two improved algorithms are proposed.Because of the advantages of the model,the improved algorithm gets rid of the constraints of short-term and multi-step prediction requirements,and has better compatibility.Moreover,the accuracy and robustness of the improved algorithm are obviously improved compared with the traditional one.Aiming at the problems of low accuracy and poor scalability of current parking lot prediction algorithms,this paper first proposes a RNN combination algorithm based onensemble learning algorithm and GRU structure.Firstly,by training the data set,we select the optimal combination form of the base learner of ensemble learning.Then,we divide the parking space data into different steps and input them into the combination model to train the model.Finally,we compare the optimal structure with the traditional network.The experimental results show that the algorithm improves the prediction accuracy and robustness,effectively prevents over-fitting and local minimization,and effectively solves the problem of multi-scene implementation of the unified prediction model.In order to further optimize the model and improve the accuracy and extensibility of the algorithm in prediction.In this chapter,a RNN parking space prediction algorithm based on particle swarm optimization and genetic algorithm is proposed.The algorithm optimizes the model and model parameters respectively.Firstly,in order to avoid the problem of stacking algorithm in practical application,the part of ensemble learning algorithm is optimized.The combination framework of random forest algorithm and gradient lifting algorithm in ensemble learning algorithm is used,and the number of regressors is optimized by setting a search space.In the aspect of neural network,the hidden layer search space,the neuron search space,the network type search space and the Batch-size search space are set up for the hidden layer of RNN.The neural network structure of the prediction model is optimized by using the advantage of genetic algorithm in information sharing ability.Finally,the PSO algorithm is used to search the optimal solution for many parameters of the model,which is easy to calculate and fast.Finally,the optimized prediction model is obtained.