Short Term Power Load Forecasting by AP Frequency Clustering and LSTM

With the expansion of customer base in many regions, the electricity consumption patterns of customers become more diverse and complex. However, the correlation between users’ power consumption patterns, which is considered helpful for load forecasting, is often neglected in the traditional forecasting technology. For this purpose, we proposed a new short-term power load forecasting method called FFT-AP-LSTM by integrating the Fast Fourier Transform (FFT), the Affinity Propagation (AP) algorithm, and the Long Short-Term Memory (LSTM) model. In this method, to determine the similarity of users’ power consumption patterns, FFT is firstly employed to analyze users’ loads and extract frequency domain characteristics. Then, the AP algorithm is adopted to classify all users according to frequency domain characteristics. Finally, we analyze each class using LSTM to determine the relevant model for load forecasting. The experimental results on actual data sets show that the proposed method outperforms the common time series prediction algorithms.