User Behavior Analysis And Residual Parking Number Forecast Based On Big Parking Data

At present,with the development of transportation informatization and the application of modern city guidance system,it has accumulated a large amount of data on dynamic traffic and State change of parking lot.As the number of remaining cars constantly updated,the number of remaining cars is not consistent with the number of valid parking spaces in the parking lot,resulting in the user to find a parking lot again.The main work to relieve the dynamic traffic pressure is to mine the effective characteristics and rules of parking user behavior from the parking lot data,and forecast the number of remaining vehicles when the user arrives at the parking lot,to assist the user to select the best parking lot.This paper first analyzes the present situation and development trend of parking management system both at home and abroad.Combined with the domestic parking management needs and parking characteristics,this paper studies the characteristics of user behavior changes based on large parking data from the micro and macro perspective.Based on the data of single parking lot,the behavior characteristics of parking users and the change of the number of parking spaces are studied.Based on the data of multiple parking lots,the spatiotemporal and Variation characteristics of parking lots are studied.Meanwhile,the relationship between parking time and weather is analyzed combining with the weather and real-time parking data.The results show that from the microscopic point of view,there is a periodic change of days and weeks in the number of parking spaces in the parking lot and the traffic flow,while bad weather and holidays have some influence on the parking behavior of users.From the macroscopic point of view,the number of remaining vehicles in the same type of parking lots and the traffic flow are similar in time,and different types of parking spaces show aggregation in space,and are positively correlated with population density.Based on the dynamic traffic flow and state data of Wuhan parking lot in 2016,the forecasting characteristics of the remaining parking spaces were constructed from the vertical and horizontal directions,including the number of remaining parking spaces in the historical time,the working day type,the vehicle residence time,the parking vehicle flowrate and weather conditions,and then the remaining parking spaces can be predicted.According to the real-time characteristics of the remaining parking spaces in the parking lot,the short-term and long-term forecasting models of the remaining parking spaces based on the LSTM network are proposed and the data is normalized.Then,the model learning rate,sample number,the number of hidden layer nodes and other parameters are adjusted.Three parking lots were selected to evaluate the prediction results.Finally,through the comparison of ARIMA time series prediction model and the prediction model proposed in this paper based on the LSTM network,it is concluded that the LSTM network model has a high accuracy in predicting the number of remaining vehicles,which provides a scientific reference for the user to choose the parking lot.