Short-Term Prediction of Photovoltaic Power Based on TCN-LSTM-Attention Model and Kmeans++

Accurate photovoltaic prediction is helpful for the scheduling and stable operation of power systems. Due to the influence of various factors such as seasons, geography, and atmospheric conditions, the intensity of solar radiation has obvious discontinuity and uncertainty, which leads to changes in the power generation capacity of photovoltaic power generation systems, posing a challenge to the stability of power systems. At the same time, facing the complex and ever-changing fluctuations in photovoltaic power, due to the limitations of a single network model, it is difficult to accurately characterize the photovoltaic output characteristics. Therefore, this paper proposes a TCN-LSTM-Attention model based on Kmeans++ for short-term photovoltaic output prediction. Firstly, by preprocessing historical photovoltaic output and meteorological data, a high-quality dataset is obtained. Secondly, the Kmeans++ algorithm is used for large-scale weather classification, model training and prediction are carried out under each cluster. Compared with the TCN, LSTM, and TCN-Attention models, the proposed TCN-LSTM-Attention model improves the predictive performance under different weather conditions and delves deeper into the relationship between radiation and photovoltaic output.