A Prediction Method for Bike-sharing Time Series Based on K-means Clustering and Gaussian Process Regression

Real-time usage prediction for shared bikes has important significance in intelligent transportation and travel management. In this paper, a time series prediction method for shared bikes based on K-means clustering and Gaussian process regression (GPR) is proposed. The method transforms the shared bikes time series into multiple subsequences through sliding window method to ensure that the time information is not lost in the GPR prediction process. By applying K-means clustering to select inducing points, data sparsity is achieved, and the selected points are used in the training of GPR model to achieve real-time usage prediction. Experimental results indicate that the proposed method effectively captures short-term trends and periodic characteristics in the time series, and significantly reduce the computational complexity compared with the traditional GPR model, while improving the training efficiency and generalization ability of the model.