Research On The Prediction Method Of The Number Of Vacnt Parking Spaces Based On Artificial Intelligence

At present,many parking assistant tools,such as the smart phone APPs can display the real-time vacant parking space availability.However,the number of vacant parking spaces changes as the cars enter or leave the parking lot,which obsoletes the parking space information previously acquired.Therefore,people are more willing to learn the parking spaces available at sometime in the future instead of real-time information.A precise vacant parking space availability prediction is beneficial to better plan people's trips and improve the utilization of parking facilities.In this paper,a novel prediction model for the number of vacant parking spaces after a specific period of time is proposed based on Support Vector Regression(SVR)with Fruit Fly Optimization Algorithm(FOA).In the proposed model,the SVR parameters are initialized as the fruit fly population,and FOA is utilized to search the optimal parameters for SVR.Sufficient experiments within various scenarios,i.e.,predicting the vacant parking space availability in parking lots with various capacities after various periods of time,have been conducted to verify the effectiveness of the proposed FOA-SVR prediction model.Three other commonly used prediction models,i.e.back propagation(BP)neural network,extreme learning machine and wavelet neural network,are used as the comparison models.The experimental results show that the proposed FOA-SVR method has higher accuracy and stability in all the prediction scenarios.Besides,this paper proposes a novel iterative multi-step Long Short-Term Memory Recurrent Neural Network(LSTM-NN)model to predict the number of vacant parking space.The key parameters of the model are optimized by grid search method.The prediction model is then benchmarked on five other well-known ones,i.e.,Gated Recurrent Units Neural Network(GRU-NN),and Stacked Autoencoder(SAE),Support Vector Regression(SVR),Back Propagation Neural Network(BPNN)and K-Nearest Neighbor(KNN),whose key parameters are sufficiently optimized as well.Sufficient experiments within two parking lots with various capacities and traffic flows have been conducted to evaluate the models' performances on short-term predictions and longterm predictions.Experiments results show that the proposed iterative multi-step LSTM-NN model outperforms all the other benchmark models,especially in the commercial parking lot with large traffic flow.