Photovoltaic Power Prediction Based on Feature Fusion and LightGBM

Solar energy, as a secure and reliable clean energy source, stands as one of the primary green energy solutions globally. Therefore, accurate load forecasting plays a crucial role in maintaining the balance of supply and demand in the power grid and ensuring the smooth operation of the electricity market. The current mainstream research algorithms include time series analysis, support vector machine studies, and artificial neural network research. However, these methods often suffer from drawbacks such as high requirements for historical data and lengthy training times. In response to these challenges, this study proposes a photovoltaic power prediction model based on feature fusion and the LightGBM algorithm. Firstly, considering the periodic variations between photovoltaic power generation and time, several lag and rolling statistical features are constructed based on the original data. Secondly, a LightGBM model is established and trained using data before and after feature fusion. Lastly, the trained model is employed to predict test data, yielding the final predictions. Experimental results using photovoltaic power generation data show that after feature fusion, both MSE and MAE values decrease, and R² is closer to 1. Thus, the proposed model demonstrates an effective prediction of photovoltaic power generation.