CNN-BiLSTM Based Demand Prediction for Shared Bicycles

Abstract

In the dynamic landscape of bicycle-sharing enterprises, precise demand forecasting emerges as a cornerstone for balancing supply and demand. This study builds upon existing research by further investigating the spatial and temporal dynamics of shared bicycle usage in Shanghai, offering new perspectives and deepening our understanding of this field. Our innovative approach combines the feature extraction of Convolutional Neural Network (CNN) with the skillful time-series analysis of Bidirectional Long and Short-Term Memory Network (BiLSTM), culminating in the novel CNN-BiLSTM model.This model's distinctiveness lies in its dual capacity to harness spatial characteristics and temporal patterns in predicting shared bicycle demand. Utilizing a comprehensive correlation analysis, we identify and incorporate key factors influencing bicycle usage into the model, enhancing its predictive accuracy. Benchmarked against conventional models like CNN, LSTM, and BiLSTM, the CNN-BiLSTM model demonstrates superior performance with the lowest RMSE and MAE, coupled with the highest $\mathbf{R}^{2}$ value. These metrics underscore our model's enhanced predictive capabilities. Our findings hold significant implications for bicycle-sharing enterprises. By accurately forecasting demand, this research offers a strategic tool for optimizing business operations and enhancing management efficiency, thereby contributing valuable insights to the domain of shared transportation logistics.