Real-time multistep prediction of public parking spaces based on Fourier transform–least squares support vector regression

Abstract

Multistep prediction of public parking spaces in the parking guidance and information system and parking reservation system has great benefits for intelligent parking. This study analyzes the C0 complexity of parking space occupancy time series from the frequency domain aspect. Results show that regular components account for the vast majority of parking space occupancy time series and can be considered a “quasiperiodic” series, which provides the theoretical basis for multistep prediction. This study combines the idea of Fourier transform (FT) and a machine learning method least squares support vector regression (LSSVR) together and proposes the Fourier transform–least squares support vector regression (FT–LSSVR) multistep prediction algorithm. As taking consideration of a predicting step threshold, this method has the power to predict single-step and multistep public parking spaces. Verification on two typical public parking lots in Hangzhou shows the great performance of FT–LSSVR. The prediction accuracy of proposed FT–LSSVR immensely outperforms the traditional LSSVR prediction after considering the step threshold. Moreover, the proposed method did not add the computational time complexity compared with the traditional LSSVR prediction. Thus, the proposed method is more suitable for real-time systems for its high prediction accuracy and less complex calculation.