李骁 赵曦 王兆军 任大为 刘丽君 刘志美

摘 要：当前敏感台区反窃电监测方法在面对连续监测状况时，监测数据中含有大量噪声数据和无用数据，导致对窃电行为的诊断依据不足，为解决该问题，提出基于递归小波神经网络的敏感台区反窃电监测方法。利用A/D采集电路和互感器采集用户用电数据，通过数据预处理剔除噪声数据和无用数据，同时使数据归一化。在此基础上，提取用电数据特征，并从电流、电压、功率因数、电量四个方面确定窃电行为判别指标，将数据与特征值输入至递归小波神经网络中，结合判别指标输出精准的窃电行为判别结果，实现敏感台区反窃电监测。实验结果表明，研究方法能够准确捕捉到功率因数的变化，窃电行为判别时间短，敏感台区反窃电监测效果更优。

关键词：递归小波神经网络;敏感台区;反窃电监测;判别指标;数据归一化

中图分类号：TM73      文献标识码：A

Abstract：The current anti-electric theft monitoring methods in sensitive stations face continuous monitoring conditions， and the monitoring data contains a large amount of noise data and useless data， resulting in insufficient diagnosis basis for electric theft behavior. To solve this problem， a recursive wavelet neural network is proposed. Anti-theft monitoring method in sensitive station area. Use A/D acquisition circuit and transformer to collect user electricity data， remove noise data and useless data through data preprocessing， and normalize the data at the same time. On this basis， the characteristics of electricity consumption data are extracted， and the identification indicators of electricity theft behavior are determined from four aspects of current， voltage， power factor， and electricity. The data and feature values are input into the recursive wavelet neural network， and the identification indicators are combined to output accurate. The result of discrimination of electricity theft conducts anti-electricity theft monitoring in sensitive stations. The experimental results show that the research method can accurately capture the power factor changes， the time for discriminating power theft behavior is short， and the anti-power theft monitoring effect is better in the sensitive area.

Key words：recursive wavelet neural network; sensitive station area; anti stealing monitoring; discrimination index; data normalization

近几年，违规用电行为越来越多，并且逐渐向着多样化、隐藏化方向发展，现场检查存在人力不足、行动盲目，执行力差等问题，造成了严重的社会不良影响[1-2]。对此，有学者研究出了相关的敏感台区反窃电监测方法，通过对台区用户数据采集和智能分析，降低并确定查窃范围，极大地提高了查处窃电行为的效率[3-4]。当前使用的敏感台区反窃电监测方法多数是基于用电信息采集大数据的监测方法或基于动态模拟技术的监测方法[5]。基于用电信息……