Short-Term PV Power Prediction Method Based on Weather Classification and TCN-Attention

For the purpose of improving the prediction accuracy of short-term PV power under different weather conditions, a short-term PV power prediction method based on weather classification combined with temporal convolutional network-attention mechanism combination model is proposed. Firstly, Spearman and Kendall correlation coefficients are used to select the main meteorological factors affecting PV power, including total solar irradiance, direct irradiance, global horizontal irradiance, temperature and relative humidity. Then, the K-means++ algorithm was employed to classify the historical PV data into three types: sunny, cloudy and rainy days. Finally, based on the temporal feature extraction ability of time-convolutional network and the ability of attention mechanism to highlight the key features, the combined time-convolutional network-attention mechanism model is established. Simulation results show that the proposed method is able to improve the accuracy of short-term PV power prediction under different weather conditions with strong adaptability.