Relevance Vector Machine with compounded kernels for regression and classification in power systems forecasting

The electric power load forecasting and the power systems Transient Stability Assessment (TSA) are classic and basic forecasting problems of regression and classification. Now the artificial intelligent technology is one of the most popular methods to solve them. The paper utilizes the linearly compound principle to construct multiple kernel functions to enhance the Relevance Vector Machine (RVM) learning model forecasting accuracy. With these compounded kernel functions, the RVM learning models are used for the electric power load forecasting and the power systems TSA. The results show, all the compounded kernels RVM models give the better accuracy than the single kernel ones, no matter in regression pattern or classification pattern. Besides of these, the probabilistic forecasting results also are given, based on the exclusive probability character of the RVM learning models.