Online Boosting Regression Method and Its Application to Load Forecasting in Energy-Intensive Enterprise

Focusing on the electricity load forecasting in energy-intensive enterprises, we propose an online forecasting algorithm based on the boosting method. The algorithm converts a regression problem such as load forecasting to a binary classification problem via geometric operation, and then it obtains a timely adaptive ensemble max-margin separating plane using an online smooth boosting algorithm. We prove that such a plane is theoretically identical to a regression function for the original load forecasting problem, and also prove that the proposed algorithm has convergence for the training set. Simulation results show that the proposed algorithm improves real-time tracking capability by implementing online updating, and it improves the accuracy and stability of load forecasting through the boosting algorithm. The forecasting error is less than 3%.