【摘要】 为有效提取触电故障特征,实现从剩余电流中分离出触电电流,提出了一种基于变分模态分解（Variational Mode Decomposition,VMD）与长短期记忆神经网络（Long Short Term Memory,LSTM）相结合的触电电流提取方法 .利用果蝇优化算法对VMD参数[K,α]寻优获得最优参数组合[6,280],以VMD分解的剩余电流最佳模态分量突变特性为依据,定义相邻周期电流幅值和的增长率η₁、η₂作为判别触电事故的特征量;以6层模态分量信号重构触电信号,构建基于LSTM网络的触电电流检测模型. 240组触电信号研究结果表明:当η₁、η₂至少一个满足大于1%时,均发生触电,否则无触电事故发生;相比于VMD-BP、VMD-RBF检测模型,VMD-LSTM检测模型提取的触电电流与真实触电电流的相关系数平均值分别提高了6.2%、2.3%,均方根误差平均值分别降低了36.8%、27.1%,提出的方法具有更高的检测精度.研究结果为研发基于生物体触电电流动作的剩余电流保护装置提供了参考.更多还原

【Abstract】 In order to effectively extract the characteristics of electric shock faults and realize the separation of electric shock current from residual current,this paper proposes a new method of electric shock current extraction based on the combination of Variational Mode Decomposition（VMD）and Long Short Term Memory（LSTM）. The parameters of VMD[K,α]were optimized by fruit fly optimization algorithm to obtain the optimal combination of parameters[6,280]. Based on the mutation characteristics of the optimal modal component of the residual current decomposed by VMD,the growth rates η₁ and η₂ of the sum of the current amplitudes in adjacent periods are defined as the characteristic quantities for judging electric shock accident. The electric shock current signal is reconstructed from the six layers modal component signal,and the electric shock current detection model based on LSTM network is constructed. The results of 240 groups of electric shock signals show that,when at least one of η₁ and η₂ is greater than 1%,electric shock occurs,otherwise no electric shock accident occurs. Compared with the VMD-BP and VMD-RBF detection models,the average correlation coefficients of the VMD-LSTM detection model are increased by 6.2% and 2.3%,respectively,and the average root mean square error is reduced by 36.8% and 27.1%,respectively.The method proposed in this paper has higher detection accuracy. The research results provide a certain reference for the development of residual current protection devices based on the electric shock current of the biological body.更多还原