Ultra-short-term load forecasting using robust exponentially weighted method in distribution networks

Ultra-short-term forecasting results of the loads of distribution transformers are one of the main sources of the pseudo measurements in state estimation programs for distribution networks, and the forecasting accuracy seriously affects the state estimation results. This paper describes a robust exponentially weighted load forecast model to improve the forecasting accuracy. Firstly, a load change rate estimating method based on the trend similarity of the load curve segment is proposed to improve the accuracy for inflection point of load curve. Then, an exponentially weighted model combined with the Huber ψ -function is introduced, which is robust for bad data. Finally, these two algorithms are combined. The combined method has been tested for a real distribution networks and the results show this method has good prediction precision especially for the inflection point of load curve, and has the ability of automatic compression of bad data.