Application of Online SVR in Very Short-Term Load Forecasting

摘要

Very short-term load forecasting (VSTLF) plays a very important role in an efficient planning and operation of power systems. In order to overcome the disadvantages of inefficiency in online settings existed in support vector machines, online SVR is first introduced into VSTLF and used to construct a 5-minute load forecasting model in this paper. The proposed method can efficiently update the trained forecasting model whenever a sample is added to or removed from the training set without retraining the whole training data from scratch every time. Moreover, it in fact forms a negative feedback with the use of the rolling window technique, which greatly improves the prediction accuracy. The effectiveness of the proposed method is validated by the simulation results on the real-world dataset obtained from NYISO website.