Research on the Application of Sharing Bicycle Demand Prediction in Urban Planning

This research applies several deep learning models for shared bike demand prediction. By selecting and analyzing a dataset from Kaggle, this research pays attention to identifying factors that have significant impact on bike sharing demands and utilizing models. The methodology includes data description and preprocessing, while the Root Mean Squared Error (RMSE) and the Mean Absolute Error (MAE) are used to appraise the model performance, with cross-validation ensuring robustness. Experimental results demonstrate that deep-learning models, especially Transformer and Long-Short Time Memory (LSTM), outperform traditional approaches in both prediction accuracy and adaptability to dynamic rental fluctuations. SHAP value analysis reveals strong nonlinear effects of temperature and time features, while attention weight visualization highlights the Transformer models ability to detect significant events. Furthermore, deep learning models exhibit greater stability under extreme weather conditions and respond more effectively to sudden demand surges during holidays. These findings contribute to the optimization for bike-sharing operations and development of data-driven urban mobility solutions, which will enhance predictive capabilities in complex urban environments.