Short-Term Power Load Forecasting via Recurrent Neural Network with Similar Day Selection

With the increase of the penetration rate of renewable energy generation, the smart grid system needs to be more intelligent, flexible and interactive. Load forecasting, especially the short-term load forecasting for power users, is playing an important role in the future grid planning and operation. Aiming at dealing the strong randomness and low prediction accuracy of short-term load forecasting, we proposed a short-term load forecasting model for smart grid via recurrent neural network with similar day selection. Firstly, the fuzzy clustering method was used to preprocess the sample data to generate the similar day set, and then an improved grey correlation analysis method was used to select the similar days from this set. The recurrent neural network was trained after the similar day selection. Our model can avoid the recurrent neural network falling into the local optimum and leading to a low prediction accuracy. Finally, we implemented our model in a real scenario with historical data of a certain area in Guangdong Province, the experimental results show that the proposed method has higher prediction accuracy and stability, and enjoy a good application prospect.