Random forest solar power forecast based on classification optimization

Highlights: • An optimized hybrid method called PCA-K-means-HGWO-RF is proposed to forecast PV power. • Optimizing model parameters and input variables are the main ways to improve model accuracy. • A better combination of Principal Component Analysis and k-means can achieve better clustering results. • The Differential Evolution Grey Wolf Optimize algorithm is used to determine random forest model parameters. • The proposed method has the best performance compared with other methods. -- Abstract: With the rapid development of the photovoltaic industry, the share of photovoltaic power generation in the power trading market is growing. The intermittent and uncontrollable characteristics of photovoltaic power generation have a huge impact on the stability of the power system. To reduce the occurrence of such conditions, it is necessary to improve the prediction accuracy of photovoltaic power generation. However, in the traditional modeling process, the accuracy of the model is often poor due to excessive noise in the original data or improper parameter adjustment. In this paper, Principal Component Analysis and K-means clustering algorithm combined with random forest algorithm optimized by Differential Evolution Grey Wolf Optimizer are used to model the photovoltaic power generation in three regions. Principal Component Analysis and K-means clustering are used to obtain the hourly point features similar to the predicted time points, and then the input data is filtered to reduce the noise data interference. At the same time, the popular optimization algorithm quickly selects the Random Forest parameters, which greatly avoids the artificial filtering factors and causes the error to be generated. Through the establishment of comparative experiments, it is found that the recommended model has higher prediction accuracy and robustness.