Prediction Technology Of Urban Available Parking Space Based On Deep Learning

In recent years,China is accelerating the construction and development of new smart cities,and has promised the world to achieve the goal of carbon peak within a period of time in the future.Parking space prediction can not only promote the construction of smart cities,but also reduce carbon emissions and accelerate the realization of carbon peak.By predicting the availability of parking Spaces,car owners can know in advance whether there are available parking Spaces when they arrive at the parking lot within a certain period of time,thus reducing carbon emissions and reducing traffic pressure.Aiming at the low prediction accuracy of medium and long term available parking Spaces,a prediction algorithm of medium and long term available parking Spaces based on improved SRU model was proposed.The algorithm predicted the number of medium and long term available parking Spaces by combining historical parking data and historical weather information.The improvement method includes:establishing weather input gate,and increasing the periodic influence of SRU model on historical parking data by introducing weather input gate.The Tanh activation function is introduced,the weather input gate and the historical data of the input are nonlinear transformed,and the peephole is further connected to improve the prediction ability of the model.The simulation results show that the prediction accuracy of the improved SRU model is 4.77%,5.05% and 1.18% higher than that of the LSTM network when the step size is 1 hour,2 hours and 3 hours respectively.In order to further satisfy the owner’s short-term parking demand,this article design a short-term parking occupancy rate prediction algorithm,the algorithm for most parking prediction algorithm to forecast the context information only,ignoring the adjacent parking lot to predict the impact of parking,because a parking lot parking state has high correlation with adjacent parking lot parking,Aimed at this problem put forward a kind of short-term parking share forecasting algorithm based on ResNet-LSTM,the model USES data is Melbourne public data sets,through history parking data,weather information and location information to predict parking occupancy rate,for a parking space sparse block using k-means clustering algorithm will disperse neighborhood parking space,By comparing k-means algorithm with GMM algorithm,k-means clustering effect is better.Experimental comparison of different algorithms shows that the average absolute error of 5-minute prediction of the proposed RESNET-LSTM algorithm is 1.4% higher than that of LSTM network and 0.68% higher than that of ResNet algorithm.