基于数据驱动与改进随机森林的配电网多元源荷异常数据检测研究

Research on Detection of Multiple Source Load Anomaly Data in Distribution Networks Based on Data Driven and Improved Random Forest

为进一步提升配电网日常运行的稳定性,提出一种基于改进随机森林算法的配电网异常数据识别分类方法。其中,以随机森林算法作为基础的异常数据识别分类方法,并引入改进的SMOTE算法和Relief F算法分别对随机森林算法的采样过程和特征选择过程进行优化,进一步提升其识别分类性能。仿真结果表明,在单节点的异常数据识别分类测试中,与传统的决策树算法、前馈神经网络BPNN以及支持向量机SVM相比,改进的随机森林算法具有更高的识别分类精度,准确率、精确度、召回率分别达到了99.40%、98.97%、98.47%,同时算法所需的运行时间也更短;在多节点异常数据的识别分类测试中,基于改进随机森林算法的异常数据识别分类方法具有较高的识别精度,准确率和召回率均稳定在97%以上,与其他方法相比,该方法还具有更好的稳定性。综上,构建的基于改进随机森林算法的配电网异常数据识别分类方法性能良好,能够应用于实际的配电网日常维护管理,提升配电网的运行稳定性,可行性较高。

To further improve the stability of daily operation of distribution networks,a distribution network anomaly data recognition and classification method based on improved random forest algorithm is proposed.Among them,an anomaly data recognition and classification method based on the random forest algorithm is introduced,and improved SMOTE algorithm and Relief F algorithm are introduced to optimize the sampling process and feature selection process of the random forest algorithm,further improving its recognition and classification performance.The simulation results show that in the single node abnormal data recognition and classification test,compared with traditional decision tree algorithms,feedforward neural network BPNN,and support vector machine SVM,the improved random forest algorithm has higher recognition and classification accuracy,with accuracy,accuracy,and recall rates of 99.40%,98.97%,and 98.47%,respectively.At the same time,the algorithm requires shorter running time;In the recognition and classification testing of multi node abnormal data,the abnormal data recognition and classification method based on the improved random forest algorithm has high recognition accuracy,with accuracy and recall rates stable at over 97%.Compared with other methods,this method also has better stability.In summary,the constructed distribution network anomaly data recognition and classification method based on improved random forest algorithm has good performance and can be applied to practical daily maintenance and management of distribution networks,improving the operational stability of distribution networks,and has high feasibility.