Short-term electricity price forecasting based on period-decoupled price sequence

The paper presents a period-decoupled price forecasting method. The period-decoupled price sequence has simpler features compared with the chronological price sequence, therefore it is more suitable to be analyzed and modeled. The correlation coefficient with electricity price sequence is determined as the standard to select the influence factors. System load rate is selected to be an important factor in price forecasting, which is the ratio of the system load to the available system capacity, and indicates the relation of supply and demand. The wavelet analysis and neural network are used as the price forecasting tool. The accuracy of price forecasting with different influence factors and different neural networks is studied. Furthermore, the forecasting methods based on the period-decoupled price sequence and the chronological price sequence are compared. The historical data from New England market is used in the case study to forecast the day-ahead system marginal price in the fourth quarter of 2002 continuously. The numerical results show that the forecasting method based on the period-decoupled price sequence can decrease mean absolute percentage error up to 3 percent compared with the forecasting method based on chronological price sequence.