基于随机森林理论的配电变压器重过载预测

发布时间：2018-04-30 07:03

本文选题：配变重过载 + 样本比率　；参考：《电网技术》2017年08期

【摘要】：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。
[Abstract]:In order to solve the problem that using traditional classifier to predict the overloading rate of distribution variable heavy load, the total correct rate is very high, but the correct rate of heavy overload prediction is very low because of the low overloading sample rate, so the resampling and random forest theory are introduced into the classification model. A resampling-random forest classifier is constructed to predict the distribution overloading. First, the overloaded and normal samples are sampled in a certain proportion to form a new subsample, and a large number of new sub-samples are obtained by repeating the above process. Then a series of classifiers are constructed according to the stochastic forest theory. The classifier is trained with new subsamples and the classification model is obtained. The final prediction results are determined by the modes of all classifiers. The overloading prediction of distribution transformer in a certain city of Shandong Province is carried out, and the above method is compared with the traditional classifier. The results show that the new method has a high accuracy in predicting the type of daily overload, the start and end time of heavy overload, and the severity of heavy overload.
【作者单位】：北京交通大学电子信息工程学院;中国电力科学研究院;国网山东省电力公司;
【基金】：国家电网公司科技项目资助(XX71-14-036)~~
【分类号】：TM421

【参考文献】