一种基于小波变换能量与神经网络结合的串联型故障电弧辨识方法

发布时间：2018-10-18 17:34

【摘要】：针对交流串联型故障电弧发生时回路电流幅值较小、传统线路保护装置不能有效检测的问题,提出一种基于小波变换能量与神经网络结合且适用于多种典型负载的串联型低压交流故障电弧辨识方法。利用自制的电弧发生装置模拟产生低压交流故障电弧,获取了6种典型家用负载情况下电路正常运行及产生串联型故障电弧时回路的电流信号。对采集的信号进行小波分解,将各层细节信号能量的平均值和标准差输入BP神经网络后构成小波神经网络,实现对不同负载测试样本的辨识。采用粒子群优化算法计算神经网络训练初始值,利用自适应学习率方法提高了训练速度。算法输出结果含义明确,输入层特征量选取合理。实验结果表明,采用该方法进行故障电弧辨识的准确率达到95%以上。
[Abstract]:Aiming at the problem that the current amplitude of the circuit is small when the AC series fault arc occurs, the traditional line protection device can not detect effectively. A low voltage AC fault arc identification method based on wavelet transform energy and neural network is proposed, which is suitable for a variety of typical loads. The low voltage AC fault arc is simulated by a self-made arc generator, and the current signals of the circuit running normally and producing series fault arc under six typical household loads are obtained. The collected signal is decomposed by wavelet transform, the average energy and standard deviation of each layer of detail signal are input into BP neural network to form wavelet neural network, and the identification of different load test samples is realized. Particle swarm optimization (PSO) algorithm is used to calculate the initial training value of neural network and the adaptive learning rate method is used to improve the training speed. The output result of the algorithm is clear and the input level feature is reasonable. The experimental results show that the accuracy of fault arc identification is over 95%.
【作者单位】：上海交通大学电子信息与电气工程学院;
【基金】：上海市“科技创新行动计划”2009年度社会发展领域重点科技资助项目(09231202600)
【分类号】：TM501.2

【参考文献】