Research on Construction and Contribution Analysis of Demand Forecasting Model Based on GBDT-LightGBM Algorithm

To construct a reasonable demand forecasting model and reduce the shortcomings of existing algorithms in this type of forecasting. In this paper, we construct a short-term demand forecasting model based on Gradient Boosting Decision Tree (GBDT) - The Light Gradient Boosting Machine (LightGBM) using an integration algorithm based on GBDT as the base learner, to solve the demand for short-term data forecasting. The optimal model is obtained through iteration by varying the amount of integration of GBDT and setting MAE, RMSE, and R2 test parameters. In this paper, the daily bike-sharing data of Washington, D.C. from 2011 to 2012 is used as the research object. The results show that the model achieves good results in experimental prediction, where MAE is 90.7788, RMSE is 125.5685, and R2 is 0.6680. It proves the robustness and generality of the model and provides some basis and methodology for solving short-term predictions.