基于有限脉冲反应和径向基神经网络的触电信号识别

针对农村低压电网剩余电流保护与动作技术中，如何检测总泄漏电流中人体触电支路电流的难题，该文利用严格线性相位与任意幅度特性的FIR（finite impulse response）数字滤波技术和具有自适应性与最佳逼近特性的RBF（radial basis function）神经网络有机结合，提出一种基于FIR数字滤波的RBF神经网络作为触电电流信号的检测方法。首先，采用FIR数字滤波器选定合适的窗函数和滤波阶数，对触电试验获得的总泄漏电流及触电电流进行滤波预处理；然后，将预处理后的信号波形作为样本集，选定适合的RBF函数，建立从总泄漏电流中提取触电电流波形的3层RBF神经网络模型。仿真试验结果表明：该方法速度快且稳定，检测值与实际值的平均相对误差为3.76%，具有良好的适应性和实用性，对于研制新一代剩余电流保护动作装置具有重要意义。

Recognition of electric shock signal based on FIR filtering and RBF neural networks

Abstract: Residual current protection device (RCD) has been widely used in low-voltage, rural power grids because it plays a very important role in avoiding physical shock, equipment damage, and electrical fires, etc, caused by leakage. At present, a setting value of leakage current can often be used as a key action for RCD. However, the electric shock current signal of the human body cannot be detected, and when unexpected current values close to or more than the setting value emerge, this will lead to the malfunction of RCD. In order to overcome the shortcomings above, we present a new recognition method for electric shock signal using digital filter technology and radial basis neural network. The method has three main stages. First, total leakage current and electric short current has been pre-processed using the finite impulse response digital filtering, which was designed by choosing suitable window functions and filter order. Second, the pre-processed signals are trained to create a three-level radial basis neural network. Last, the electric short current can be recognized by inputting the filtered total leakage current signal to the radial basis neural network, thus establishing the detection model. Experiments showed that the proposed method achieves an average relative error of 3.76% between detected value and actual value. The robustness, adaptability, and practicality of the proposed method also have been proven by the results. The proposed method made a definite practical significance for developing a new device of residual current protection.