Hierarchical parameter optimization based support vector regression for power load forecasting
Zeyu Wang, Xiaojun Zhou, Jituo Tian, Tingwen Huang
Abstract
Power load forecasting is an important task of smart grid, which is of great significance to the sustainable development of society. In this paper, a hybrid support vector regression (HSVR) is raised for the medium and long term load forecasting. To further improve prediction accuracy, the coupling and interdependent relationship between hyperparameters and model parameters in the optimization process is focused. A hierarchical optimization method based on nested strategy and state transition algorithm (STA) is proposed to find optimal parameters. The effectiveness of the proposed hierarchical optimization method is confirmed on several benchmarks, and the resulting hierarchical optimization method based SVR is also successfully applied to a real industrial power load forecasting problem in China.