Short-term load forecasting based on time series reconstruction and support vector regression

The load curve presents certain randomness for reasons such as human social activities, the load of electric vehicle charging and discharging and so on, which covers up the regularity of load sequence. This paper proposes an approach which can restore the original feature of the loads. In this approach, firstly, the bad data is excluded; Secondly, the characteristics of the time series are extracted by the time series reconstruction; At last, the loads are forecasted by support vector regression and the forecasting sequences are taken as training samples of the next prediction. After repeat predictions, the final series are obtained then restored, and precise prediction result is obtained. This paper uses the load data of a city in Jiang Su. Compared with traditional support vector regression method, this approach ensures a higher prediction accuracy.