An Incremental Learning Method for PV Generation Prediction Based on MMD Data Screening

Photovoltaic (PV) generation has garnered significant attention due to its straightforward approach to harnessing energy. However, the inherent volatility in PV power generation poses substantial challenges to the stability and reliability of the power grid. This study introduces an incremental learning method based on Maximum Mean Discrepancy data screening (IL- MMDS) to address the declining accuracy of prediction models in dynamic environments. Incremental learning helps to mitigate the issue of catastrophic forgetting in neural networks by enabling real-time model updates, allowing the model to adapt to evolving data. The proposed method and applies the Maximum Mean Discrepancy strategy to select samples that most effectively enhance the model's predictive performance and utilizes an experience replay buffer to store representative historical data. Experimental results demonstrate that compared to four existing baseline methods, the proposed IL- MMDS method delivers superior performance in PV power generation forecasting, particularly in handling complex environmental changes, resulting in higher prediction accuracy and more stable performance.