Short-Term Photovoltaic Power Prediction Based on Similar-Day Method Using LSTM-TCN Model

Short-Term Photovoltaic Power Forecasting (STPPF) has become an essential tool for power system scheduling with the increasing adoption of renewable energy sources. This manuscript introduces a LSTM-TCN photovoltaic power forecasting model based on similar day clustering. The grey relational analysis method is firstly employed to identify the key meteorological factors affecting photovoltaic power generation. Then the K-means algorithm is utilized to cluster the meteorological data. Finally, the model utilizes morphological similarity distance to select days with similar meteorological conditions for photovoltaic power forecasting. The model integrates the strengths of Long Short-Term Memory networks (LSTM) and Temporal Convolutional Networks (TCN) to enhance the accuracy of time series data prediction. LSTM is to process time series data and capture long-term dependencies, while TCN extracts features through dilated convolution structures. A case study using data from the Yulara solar system in Australia validated the model's performance under various seasons and weather conditions. The forecasting performance was assessed by using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The results indicate that the LSTM-TCN model outperforms the single model under sunny or cloudy weather conditions in both seasons. It demonstrates that the proposed model exhibits strong accuracy and significant application potential in the field of photovoltaic power forecasting.