Short Term Photovoltaic Power Prediction Based on TCN-SAM-BiLSTM

Short-term photovoltaic power prediction is of great significance for the safe operation of photovoltaic grid-connected systems. Aiming at the problems of noise interference and unstable prediction effects of single models in photovoltaic power prediction, this paper proposes a short-term photovoltaic power prediction model based on Temporal Convolutional Network (TCN), Spatial Attention Mechanism (SAM), and Bidirectional Long Short-Term Memory (BiLSTM). Firstly, the maximum information coefficient is used to filter out key feature variables, ensuring the quality and correlation of input data. Then, the TCN is used to extract time series features, and the SAM is introduced to weight the photovoltaic power features, thereby enhancing the model's ability to recognize the trend of photovoltaic power changes. Finally, by combining the bidirectional dependency capture capability of BiLSTM, short-term prediction of photovoltaic power generation can be achieved. The experimental results show that compared with traditional single prediction models, the proposed prediction method significantly improves the accuracy of photovoltaic power generation prediction, with a reduction in Mean Absolute Error (MAE) by up to 2.327, Root Mean Square Error (RMSE) by up to 1.858, and an increase in the coefficient of determination (R
²
) by up to 0.0832.