Research On Medium And Long-Term Prediction Method Of Urban Parking Space Based On Long Short-Term Memory And Attention Mechanism

With the rapid development of China's economy and the increasing number of motor vehicles,"parking difficulty" has become one of the main problems perplexing major cities.How to use information and intelligent means to improve the utilization rate of parking space,optimize the prediction effect of parking occupancy,and better use the parking forecast in parking guidance information system,has become the key problem of the whole parking industry.In this paper,a multivariable neural network prediction model including weather conditions,working days and holidays is constructed for berth occupancy prediction.In order to effectively predict and model the berth occupancy,two neural network berth occupancy prediction models are designed in this paper.(1)Considering the complexity of time series forecasting problem,in order to obtain accurate and stable prediction effect,this paper models the berth occupancy problem based on Long Short-Term Memory(LSTM).As a variant of the Recurrent Neural Network(RNN),LSTM improves the long-term memory ability of the network by changing the internal structure of the network,and solves the problem of gradient disappearance in RNN.When dealing with the sequence to sequence problem,it can effectively explore the nonlinear relationship between the sequences,and the berth occupancy prediction conforms to the sequence to sequence characteristics.Therefore,LSTM is suitable for the berth occupancy prediction in this paper.Whether the LSTM neural network can learn the trend change rule implied in the sequence from the training samples determines the prediction effect of berth occupancy.In this paper,the effects of the target step size of input prediction and the data sample size for training on the prediction accuracy and stability of the model are investigated.This paper comprehensively considers the relevant factors affecting the berth occupancy rate,and obtains the correlation between the three types of characteristics and the berth occupancy rate through the correlation analysis,and takes the three types of characteristics and berth occupancy data as the input variables of the network.Moreover,this paper will input multiple sets of data sets into the model for testing,and analyze the prediction results of the model.Compared with other prediction models such as RNN,the prediction accuracy and stability of LSTM are greatly improved with the increase of prediction step size(2)Only LSTM is used to train and forecast the berth occupancy prediction model.With the increase of prediction step size,the prediction accuracy of the model will decrease.In order to further optimize the above model,this paper further proposes a prediction model of seq2seq berth occupancy based on attention mechanism.The output of the input sequence is preserved by Bidirectional Long Short-Term Memory(BiLSTM)encoder,and the time pattern information of multivariable is obtained by Convolutional Neural Networks(CNN),and the relevant variables are weighted and context information is stored.By training the model,the input sequence is associated with the model output according to the learning feature information,and the high correlation sequence is assigned higher learning weight.Then,the output sequence is used to extract the useful features with high correlation.Through the test of experimental data,the root mean square error and mean absolute error were used to evaluate the prediction results.The model still maintains accuracy and stability while increasing the prediction step sizeIn general,this paper solves the key problem of medium and long-term prediction of parking occupancy in parking guidance information system,and verifies the effectiveness and feasibility of the proposed scheme through experiments.