Research On The Short-term And Long-term Prediction Method Of Off-street Parking Demand

With the introduction and deployment of the transportation power strategy,while the country vigorously develops the transportation construction,it also causes the reform of the residents' travel structure.The proportion of self-driving travel increases,and the usage rate of vehicles increases greatly,so the parking problem has become one of the traffic problems in various regions.Inadequate planning of existing parking spaces in cities,unscientific standards for parking facilities,and inadequate parking management are highlighted.The parking problem has become an important issue in the development of urban transportation.It is generally believed that smart parking solutions can effectively solve the parking problem currently facing.Parking demand prediction is a prerequisite for implementing a smart parking solution.It is a key step for parking guidance systems and parking reservations systems.Effective and accurate prediction of the number of vehicles in a parking lot can help travelers make travel decisions and help managers make scientific and systematic parking lot management scheme as well as reduce the increase of road traffic flow caused by parking problems.In order to accurately predict parking demand,this paper mainly does the following:(1)Analyzing the time series of parking quantity,and using the hierarchical clustering analysis method to determine the time characteristics of parking data.The results show that parking data has obvious regular characteristics on working days and non.working days.The time series is divided into working days and non.working days for forecasting respectively;Chi.square test method is used to determine the rules of parking arrival and departure processes in the parking lot.(2)Using the research results in Chapter 2,when the laws of parking arrival and departure are known,based on the Markov birth and death process,a regular formula for the number of parking spaces in a parking lot with time is established,that is,the number of parking lots is expressed as parking.A function of arrival rate,probability of each car leaving in the parking lot,time,and initial number of stops.In the prediction process,the parking trend of the day is pre.classified to increase the accuracy of the model parameter calibration,and the curve parameters are used to calibrate the model parameters.(3)Based on the research results in Chapter 3,based on the principle of not wasting forecast resources and being able to effectively sense the change in the number of parking occupation,further calculate the time interval and time period of LSTM network parking demand prediction,and use the obtained time series data separately.Long.term and short.term memory neural networks are used to make predictions,and the particle swarm optimization algorithm is used to optimize the initialization parameters of the neural network to obtain a parking demand prediction method based on a combination model of particle swarm and LSTM networks.