基于随机森林理论的配电变压器重过载预测

Heavy Overload Forecasting of Distribution Transformers Based on Random Forest Theory

针对使用传统分类器预测配变重过载会因为重过载样本率较低而带来的总正确率很高,重过载预测正确率却很低这一问题,将重抽样与随机森林理论引入分类模型中,构建重抽样-随机森林分类器对配变重过载进行预测。首先将观测中重过载样本和正常样本按照一定的比例进行抽样形成新的子样本,重复上述过程获得大量的新子样本。接着根据随机森林理论构建一系列的分类器,并用新子样本对分类器进行训练,得到分类模型。最终的预测结果由所有分类器预测结果的众数所决定。对山东省某市的配变进行重过载预测,并将上述方法与传统分类器进行比较。结果表明,新方法在预测配变日重过载类型、重过载开始与结束时间、重过载严重程度方面有较高的准确率。

In view of problem that prediction of heavily overloaded distribution transformers with traditional classifier brings higher overall correct rate and lower heavy overload correct rate due to low sampling rate of heavy overload, a resampling and random forest regression method is introduced to classification model to predict heavily overloaded distribution transformers with resampling and random forest regression classification model. Firstly, heavy overload samples and normal samples from observation were randomly sampled to form a new sub sample according to a certain ratio, thus large number of new sub samples were obtained by repeating above process. Then, according to random forest theory, a series of classifiers are constructed, and the classification model is trained with sub samples. Final prediction is decided by all classifiers＇ prediction results. In comparison of performances between above method and traditional classifiers in prediction of heavily overloaded distribution transformers in Shandong Province, results show that the new method has higher accuracy in predicting heavy overload types, starting and ending times and heavy overload severity.