A noval space-time feature extraction approach for load forecasting

Precisely load forecasting is one of the most important points in power system researches. Especially in distribution network, an accurate load forecasting can significantly facilitate the power system control. Therefore this paper presents a novel load forecasting approach combining the features of the load time-sequenced with spatial distribution characteristics, which enables the analysis of the dynamic behaviors of the power system operation. Because of the outlier tolerance benefitting from the space-time domain analysis, the presented approach is able to supply reliable and efficient predicted load forecasting results. In order to evaluate the performances of the approach, a practical dataset of a real distribution system from Northeast China has been employed. The experimental results show that the precision of the load forecasting can be effectively improved by the proposed method, which outperforms weighted daily periodicity model algorithms in prediction accuracy and robustness.