An improved hybrid model for short term power load prediction

Accurate and stable power load prediction is useful for electric power enterprises. However, accurate and stable power load prediction becomes very difficult. In order to improve prediction accuracy and stability, an improved hybrid model based on variational mode decomposition (VMD) optimized by the cuckoo search algorithm (CSA), seasonal autoregressive integrated moving average (SARIMA) and deep belief network (DBN) is put foreword for short term power load prediction. First, the original power load is decomposed into several regular and random sub-series by VMD-CSA. Second, the regular sub-series is predicted by SARIMA, and the random sub-series is predicted by DBN. Third, the final prediction result is the sum of each sub-series prediction result. The validity of the proposed model is verified by using power load from three different markets. Experimental results show that the proposed model has more accurate and stable results.