A Short-Term Load Forecasting Method via Model Selection Based on Random Forest

Short-term load forecasting(STLF) is an essential module of energy management system, which is of great signifi-cance to the economic dispatch and operation stability in smart grid. There is a large collection of methods developed for STLF, but it is still challenging to provide high precision STLF under different weather conditions which are the main factors affecting power generation load, especially for distributed photovoltaic power generation load. A short-term load forecasting method via model selection based on random forest is proposed in this paper to realize reliable and accurate daily power generation load forecasting under different conditions. We first perform clustering analysis on the raw data through K-means. In particular, we consider both weighted meteorological factors and historical load to improve clustering performance. Secondly, we establish a model pool consisting of state-of-the-art machine learning(ML) models which is selected from four alternative ML models, and each model is the best model for each cluster. Then, we train a random forest based on each set of data and its optimal model label. In the prediction stage, random forest is utilized to directly select an appropriate model from model pool to obtain the final prediction load. The performance of the proposed method is validated on real generation load of practical scenarios. The result indicates the superiority and advantages of the model selection based STLF method compared with the single model methods, and the mean absolute error(MAE), root mean square error(RMSE) and mean absolute percentage error(MAPE) are reduced by 118.5054(KW), 10.43% and 2.08%, respectively.