灰狼算法优化 SVR 的 10kV 配网线损率预测研究-《计算机技术与发展》

文章信息/Info

Title:: Research on Line Loss Rate Prediction of 10kV Distribution Network Based on SVR Optimized by Gray Wolf Algorithm

作者:: 杨正宇^1* ; 沈志强² ; 郑成源³; 1. 云南电力试验研究院(集团)有限公司,云南昆明 650217;
2. 云南电网有限责任公司临沧供电局,云南临沧 677000;
3. 云南电网有限责任公司电力科学研究院,云南昆明 650217

Author(s):: YANG Zheng-yu1* ; SHEN Zhi-qiang2 ; ZHENG Cheng-yuan3; 1. Yunnan Electric Power Test & Research Institute Group Co. ,Ltd. ,Kunming 650217,China;
2. Lincang Power Supply Bureau of Yunnan Power Grid Co. ,Ltd. ,Lincang 677000,China;
3. Electric Power Research Institute of Yunnan Power Grid Co. ,Ltd. ,Kunming 650217,China

Keywords:: gray wolf algorithm; 10kV distribution network; Mahalanobis distance; principal component analysis; line loss rate

摘要:: 有效控制线损率不仅能为电力企业带来经济效益,而且能进一步提高一次能源的利用率。为了实现对 10 kV 配电网线损率的准确预测,提出基于灰狼算法( Gray Wolf Optimizer) 优化支持向量机回归( Support Vector Regression) 的 10 kV配电网线损率预测方法;采用基于马氏距离的异常值检验及主成分分析法(Principal Components Analysis)对原始数据进行预处理,保证数据的清洁性的同时剔除原始数据中的冗余信息。利用 GWO 算法强搜索性的特点与 SVR 进行结合建立模型,通过与原始 SVR、ABC-SVR、BP 神经网络模型的预测结果进行比较,GWO-SVR 模型的预测精度最高,其均方根误差(RMSE) 和平均绝对误差( MAE) 分别为 0. 233 2 和 0. 195 8,最大相对误差为 14. 4% ,并且该模型具有最快的运算速度。

Abstract:: Effective control of line loss rate can not only bring economic benefits to power enterprises,but also improve the utilization rateof primary energy. In order to achieve accurate prediction of 10kV distribution network line loss rate, a Support Vector Regressionprediction method based on Gray Wolf Optimizer was proposed. The outlier test based on Mahalanobis distance and PrincipalComponents Analysis are used to preprocess the original data to ensure the cleanliness of the data and eliminate the redundant informationin the original data. The strong search ability of GWO algorithm was combined with SVR to establish the model. Compared with the prediction results of original SVR,ABC-SVR and BP neural network models,the prediction accuracy of GWO-SVR model was the highest,and its root mean square error ( RMSE) and mean absolute error ( MAE) were 0. 233 2 and 0. 195 8, respectively. The maximumrelative error is 14. 4% ,and this model has the fastest computing speed.