电能表计量性能评价及预测方法研究

Research on Evaluation and Prediction Method of Energy Meter Metering Performance

为解决对电能表进行评价的时间和误差特征维度单一等问题,提出一种电能表计量性能评价及预测方法。首先,利用电能表拆回检测及其首次检定多时间维度的误差数据,提出利用K-means聚类算法对电能表计量性能进行评价的方法,从均值、方差、平均偏移量、最大偏移量误差进行多特征维度评价分析。然后,提出一种利用电能表基本信息特征对其评价类别进行预测的方法,采用集成学习Stacking模型进行预测,以支持向量机(SVM)、K近邻(KNN)、梯度提升决策树(GBDT)和极端梯度提升(XGBoost)相异模型为Stacking模型基分类器,并利用合成少数类过采样技术(SMOTE)处理类别不平衡问题。对比结果表明,平均偏移量维度对其计量性能评价效果最好,同时Stacking模型预测准确率优于单个模型,SMOTE采样后准确率提升2.5%,最终预测准确率达到88.5%。该结果验证了利用电能表基本信息特征进行评价类别预测的有效性。

To solve the problems of evaluating the time and error characteristics of energy meters with a single dimension,an evaluation and prediction method of energy meter metering performance is proposed.Firstly,using the error data in multiple time dimensions of the electric energy meter dismantling and testing and its first inspection,electric energy meter metering performance evaluation method by using K-means clustering algorithm is proposed,to evaluate and analyze in multiple feature dimensions from the mean,variance,average offset,and maximum offset error.Then,a method is proposed for predicting the evaluation categories of energy meters using their basic information features,using integrated learning Stacking model for prediction,with support vector machine(SVM),K-nearest neighbors(KNN),gradient boosting decision tree(GBDT)and extreme gradient boosting(XGBoost)phase difference models as Stacking model-based classifiers,and using synthetic minority oversampling technique(SMOTE)to deal with the category imbalance problem.The comporison results show that the average offset dimension has the best effect on its metering performance evaluation,while the prediction accuracy of Stacking model is better than that of individual models,and the accuracy of SMOTE sampling is improved by 2.5%,and the final prediction accuracy reaches 88.5%.The result verifies the effectiveness of using the basic information characteristics of energy meters for evaluation category prediction.