Integrating a hybrid data processing strategy into an optimized light gradient boosting machine for photovoltaic power forecasting

As smart photovoltaic distribution networks and large-scale photovoltaic integration continue to evolve, accurate short-term photovoltaic power forecasting has become crucial for ensuring the safe, stable, and economic operation of the system. To address the challenges posed by multiple meteorological influencing factors and the volatility of photovoltaic power generation, this study proposes a hybrid prediction model that integrates an optimized light gradient boosting machine with a data processing strategy. Firstly, a mixed data processing strategy is adopted, which utilizes support vector machine to filter factors and reduce the dimensionality of the model. On this basis, Gaussian mixture model clustering is employed to partition different types of databases, resulting in different data feature libraries for light gradient boosting machine prediction models. Finally, the optimal hyperparameters of the model can be obtained through the gray wolf optimizer algorithm module for predictive evaluation. The proposed hybrid model is applied to perform prediction on a real-world dataset provided by the European Centre for Medium Range Weather Forecasts. Compared with 11 popular machine learning algorithms, our model demonstrates a decrease in mean absolute error, mean square error, and root mean square error on the test dataset by 12.35–70.53%, 24.58–91.69%, and 13.18–71.17%, respectively. Furthermore, the hybrid model also exhibits good interpretability and strong generalization performance.