Multi-dimensional Short-Term Load Forecasting Based on XGBoost and Fireworks Algorithm

Nowadays, most load forecasting of power system only takes advantage of sequential characteristics of the load itself. In fact, meteorological conditions, population and geographical factors will also have a significant impact on shortterm load, thus reducing the accuracy of load forecasting. In this paper, we introduce XGBoost into load forecasting, making full use of its parallelism, anti-overfitting, second-order Taylor expansion and other characteristics, and modeling with multidimensional factors. The results show that the XGBoost model considering only sequential characteristics works fine in predicting the trend of short-term load, while the multidimensional model combined with temperature, humidity and rainfall factors has a significant improvement in prediction accuracy.