基于动态分组的支持向量机窃电识别方法

Power Stealing Identification Method by Support Vector Machine Based on Dynamic Grouping

针对电力用户的窃电诊断问题,以真实的、海量的用户用电数据为基础,基于机器学习技术,对样本数据进行数据清洗和特征选择。首先,将电力用户按照用电特性结合窃电相关的多个因素进行动态的分组,保证窃电分析的针对性和适应性;然后,使用基于高斯核函数的支持向量机算法构建分类模型,并对分类结果进行可视化,通过五折交叉验证对该算法进行仿真计算,分析模型的可靠性及稳定性;最后,通过开展现场核查,进一步验证该方法的有效性。

Aiming at the problem of theft and leakage diagnosis for electricity consumers,based on real and massive user electricity consumption data,and machine learning technology,data cleaning and feature selection are carried out on the sample data.Power users are dynamically grouped based on their electricity consumption characteristics and multiple factors related to power stealing,ensuring the pertinence and adaptability of theft and leakage analysis.Build a classification model using support vector machines algorithm based on Gaussian kernel function,visualize the classification results,and simulate the algorithm through five fold cross validation to analyze the reliability and stability of the model.Finally,the effectiveness of this method is further verified through on-site verification.