Model For Multi-Step Ahead Available Parking Spaces Forecasting Based On Wavelet Transform

With the increase of the number of automobiles, the city parking demand occurred a rapid growth, and the city automobile parking had become a very serious traffic problem. And parking spaces forecasting is important for drivers to choose parking lots and driving routes reasonably.This paper proposes a new methodology for multi-step ahead APS forecasting which combines wavelet transform (WT), artificial neural network (ANN) and forecasting strategies based on the changing characteristics of APS. Firstly, APS time series were decomposed and reconstituted by wavelet transform, Then, combining artificial neural network, five strategies for multi-step ahead time series forecasting consisting of Recursive strategy, Direct strategy, Multi-Input Multi-Output (MIMO) strategy, DIRMO strategy (combination of Direct and MIMO strategies) and newly proposed RECMO strategy which is the combination of Recursive and MIMO strategies were used to forecast the reconstructed time series respectively. Finally, integrating the predicting results from the several reconstructed time series produces the final forecasted APS.The following three findings appear to be consistently supported by the experimental results. Applying wavelet transform to multi-step ahead available parking spaces forecasting can effectively improve forecasting accuracy, the forecasting results of DIRMO and RECMO strategies are more accurate than other strategies, and RECMO strategy saves up model training time compared to DIRMO strategy, consuming the least training time among five forecasting strategies.