Long-Term Power Load Forecasting via Sine Neural Network Using Multiple Functions Decomposition Method

The rapid development of economy and technology has led to the increase of electricity demand recently. Thus, grasping the trend and change rules of power load is beneficial to the reasonable planning of power system and dispatching of power resources. As there are certain regularities in the general power load data, we find that, in the general framework and form of sine neural network (SNN), the multiple sine functions decomposition method, also in this paper specifically termed sine functions fitting (SFF), is a kind of effective method for power load prediction tasks based on the analysis of periodic change of power load and long-term change trend characteristics. Considering the fact that the more recent period of power load data has a stronger correlation to future power load, we further propose sine functions fitting with relearning (SFF-R). The SFF-R method can improve the forecasting accuracy of the future power load by strengthening the learning of the power load data in the recent period. Through comparative experiments, it can be concluded that the SFF-R method is superior in predicting future power load trend. The SFF-R method not only achieves good prediction performance, but also has lower computational complexity for practical applications. Finally, the SFF-R method, as an improved method of multiple functions decomposition, is used to predict the trend of power load of twelve countries in the next two years.