Research on the Short-Term Power Interval Prediction Method for Distributed Power Sources in Distribution Networks Based on Quantile Random Forests

With the widespread application of Distributed Generation (DG) in new energy power systems, accurate prediction of its power output has become a key issue. To address this problem, this study proposes a short-term power interval forecasting method for distributed power generation in distribution networks based on Quantile Random Forests. Initially, in-depth processing was performed on historical photovoltaic and wind power active power data, as well as meteorological data. This included the imputation of missing values via KNN, the handling of outliers, and normalization, as well as the selection of major influencing factors on the output of distributed power through correlation analysis. Subsequently, we employed advanced technologies such as quantile regression, random forests, and confidence intervals to construct a Quantile Random Forest interval forecasting model tailored to distributed power in distribution networks. After the model's construction, adjustments were made to the model parameters, followed by cross-validation and further debugging and optimization. Lastly, the key meteorological information for future prediction was input into the optimized model for forecasting. The forecasting method in this study takes full advantage of the Quantile Random Forest's strengths in dealing with non-linear, high-dimensional data, and handling missing data, enhancing prediction accuracy and reducing prediction bias, providing significant guidance for actual network operation and dispatching. Future research will continue to deepen the understanding and application of the Quantile Random Forest model in hopes of improving forecasting results and optimizing grid operation.