基于半监督学习时间序列回归算法的窃电量估算方法

Estimation method of stolen electricity amount based on time series regression algorithm with semi‑supervised learning

窃电用户窃电量的精确估算对挽回电力企业的经济损失和依法处置窃电用户具有重要的实际意义。为实现用户窃电量精确估算,通过对时间序列回归算法进行优化改进,提出了一种新的窃电量估算方法。该方法以用户当前用电量时间序列样本和历史同期时间序列样本为基础,通过引入最大均值差异(maximum mean difference,MMD)得到基于最大均值差异-最小二乘支持向量回归(MMD-least square support vector regression,MMD-LSSVR)的半监督学习回归算法,以提高回归的准确度和数据的利用度、降低时间复杂度;同时,通过引入交叉变异人工蜂群算法(artificial bee colony based on crossover mutation,CMABC)对算法关键参数进行最佳适应度约束,以提高估算结果精度和收敛速度。在此基础上,提出了基于MMD-CMABC-LSSVR的窃电量估算方法。算法验证结果表明,采用所提方法,其估算电量与实际用电量的相对误差仅为2%,精度远优于传统方法;实际应用案例表明,所提方法可有效恢复窃电时间区段内窃电用户负荷和计量曲线,并精确估算出窃电量。所提方法为反窃电稽查工作提供了一种新的有效手段,具有良好的应用前景。

Accurate estimation of the stolen electricity amount is of great practical significance to recover the economic losses of power enterprises and punish electricity stealing users according to law.In order to realize the accurate estimation of stolen electricity amount,a new estimation method of stolen power based on optimizing and improving the time series regression algorithm is proposed in this paper.This method is based on the user's current electricity consumption time series samples and historical time series samples of the same period.By introducing the maximum mean difference(MMD),a semi-supervised learning regression algorithm based on MMD-LSSVR(MMD-least square support vector regression)is developed to improve the accuracy of regression and the data utilization,and reduce the time complexity.Meanwhile,the optimal fitness constraint is carried out on the key parameters of the algorithm by introducing artificial bee colony based on crossover mutation(CMABC)algorithm to improve the accuracy and convergence speed of the estimation results.On this basis,a method for estimation of electricity stolen based on MMD-CMABC-LSSVR is proposed.The algorithm verification results show that the relative error between the estimated electricity amount and the actual electricity amount is only 2%,and the accuracy of the method proposed in this paper is far better than that of the traditional methods.The applications of actual cases show that the proposed method in this paper can effectively restore the user's load and metering curve during the stealing time period and accurately estimate the amount of electric power stealing.This method provides a new and effective means for the anti-stealing-power inspection work,and has a good engineering application prospect.