KNN Short-term Photovoltaic Power Prediction Algorithm Based on Interval Division

To enhance the dispatchability of photovoltaic power generation and improve equipment status observability, this study investigates short-term PV power forecasting. By collecting environmental parameters to construct datasets and analyzing existing prediction methods, we propose an improved K-Nearest Neighbor (KNN) prediction model. The model introduces an interval partitioning mechanism that determines output intervals based on real-time samples' proximity to historical data in feature space. It then performs weighted calculations using the K closest historical sample power values within each interval to generate final predictions. This approach features a simple structure and high computational efficiency. Experimental results demonstrate that the model achieves an average absolute percentage error (MAPE) of 7.83% and a root mean square error (RMSE) of 0.5053, with precision meeting engineering application requirements. These findings provide effective technical support for intelligent operation of photovoltaic power stations.