Day ahead Residential load forecasting with neural memory network

Abstract

Day ahead forecasting of residential load (DAFRL) plays a key role in power system. To facilitate precise DAFRL, a neural memory network-based forecasting model was proposed. Firstly, the residential users with similar electricity power consumption mode are to be classified via the K-Means clustering. Thereafter, load data is de-noised with wavelet. A Neural Memory Network (NMN) model is developed to perform combination forecasting of residential users in the end. The discrete Fourier transform is employed to decompose the memory state into multi frequency components, thereafter to implement combination forecast of day ahead load with these frequency components. Various features, mean square error (MSE), Root mean square error (RMSE), Mean absolute error (MAE) included, were calculated to evaluate the capacity of NMN. Compared with the long short-term memory (LSTM), numerical simulation result indicates the proposed approach do better.