A study on improved random forest-based anomaly detection of regional tariff data under distributed photovoltaic access

• Improved Random Forest with LDA-PCA optimization for PV-integrated tariff anomaly detection. • Hybrid sampling (0.6 oversampling + undersampling) achieves G-mean >0.927 in imbalanced data. • Waveform entropy <0.11, centroid >0.907 ensures precise feature extraction. • Voting-based classification enhances accuracy in regional anomaly identification. • Validated in low/high-latitude regions for reliable PV-grid tariff stability. To ensure the reliability of the measurement and calculation results for electric energy and electricity charges, it is necessary to accurately determine the anomalous data in the regional electricity pricing datasets and consequently propose an anomaly detection method for regional electricity price data based on an improved random forest algorithm under distributed photovoltaic access. This method analyzes the importance of regional electricity price data and the formative factors of anomalous data according to the regional electricity market framework. Linear discriminant theory is incorporated to optimize the principal component analysis method for extracting characteristics of regional electricity price data. Subsequently, decision trees are constructed, incorporating oversampling factors and random undersampling techniques to enhance the random forest model. The model determines the training set and test set through mixed sampling. After delineating the range of anomalous regional electricity price data, the final classification of such anomalies is determined through voting mechanisms. Test results demonstrate the method's strong application performance, with waveform entropy results remaining below 0.11, centroid results exceeding 0.907, and effective feature extraction outcomes. With an oversampling factor of 0.6 and G-mean values consistently above 0.927, the method reliably completes anomaly detection for electricity price data across both low and high latitude regions.