Prediction Research For Available Parking Space Based On PW-LSTM And SAW Model

With the sharp growth of vehicle retention in the China,in order to alleviate the increasingly prominent parking problem in large cities,the research and development of city-level parking guidance systems is imperative.The prediction of available parking space(APS)is a very important intelligent technical means,which is of great significance for the driver to choose parking lots and driving routes reasonably.However,the prediction is inseparable from the support of data.It is actually difficult for the parking guidance system to collect all parking lots' data of real-time parking space and influencing factors without huge economic and time investment.In terms of short-term forecasting,the model often leads to poor prediction accuracy due to the lack of understanding of external influencing factors.In the medium and long-term forecast,as the prediction step size or period increases,the model will have error accumulation problems.Some researches in the above two issues facing short-term and medium-long term forecasts in this paper:(1)The sequence data of APS contains basic characteristics such as trend,periodicity and chaos.It is also affected by random factors such as weather environment on the basis of trend changes.Therefore,the nonlinear modeling ability of the prediction model is very important.In order to reveal the nonlinear dynamics in the process of short-term forecasting for APS,this paper proposes a long short term memory network model PW-LSTM based on phase space reconstruction combined with weather environment.The method maps the time series of APS into the two-dimensional phase space according to the time delay and embedding dimension,and then trains the LSTM model combined with the weather features to predict the future APS in the short-term.(2)Short-term forecasting is inseparable from the support of real-time data.However,it is necessary for parking lots without real-time data to combine medium and long term prediction techniques.This paper proposes hybrid prediction model SAW based on non-stationary stochastic process and wavelet neural network.The historical series of APS with periodic characteristics is taken as the research object in the paper.Firstly,the time series of APS is statistically analyzed for its non-stationary process based on the central limit theorem and the large number theorem.Then,the proposed model SAW is tranined to predict the APS in a certain period of time in the future effectively combined with WNN(Wavelet Neural Networks).In a word,two key technologies of parking guidance system,the short-term and medium-long-term forecasting of APS time series are realized in this paper.And the feasibility and effectiveness of the proposed methods are verified by experiments.The experimental data proves that the series data of APS based on phase space reconstruction is more in line with its nonlinear dynamic characteristics,which is beneficial to the learning and training of the model.In addition,the introduction of the related weather factors significantly improves the accuracy of the short-term prediction model.In the medium and long-term forecasting,compared with the WNN and Lyapunov exponential method,SAW has lower computational complexity,relatively more accurate prediction effect,and effectively solves the error accumulation problem caused by the loss of real-time data support in the multi-step long-term prediction.Accurate prediction of APS realize urban parking real-time induction rapidly with low cost,and alleviate parking difficulty.It has broad market prospects,good economic benefits and strong social benefits.