基于CDAE和TCN/BLSTM模型的电能质量扰动分类方法

Power Quality Disturbance Classification Method Based on CDAE and TCN/BLSTM Models

随着新型电力系统中电能质量扰动(PQDs)愈加复杂,为提升PQDs分类准确率并增强算法的噪声鲁棒性,将卷积降噪自编码器(CDAE)、时域卷积网络(TCN)与双向长短期记忆(BLSTM)相结合,提出一种基于CDAE和TCN/BLSTM模型的电能质量扰动分类方法。首先,通过CDAE以原始信号为目标重构含噪信号;然后,利用TCN和BLSTM并行挖掘扰动的抽象和时序特征;最后,特征合并层融合两种特征并完成分类。仿真结果表明,该方法可有效分类强噪声下的20类PQDs信号且平均准确率达99.23%,相比于其他主流的分类方法,所提方法具有更好的分类效果和抗噪性能。

With the increasing complexity of power quality disturbances(PQDs)in new power systems,in order to improve the accuracy of PQDs classification and enhance the noise robustness of the algorithm,the paper proposes a power quality disturbance classification method based on CDAE and TCN/BLSTM models by combining convolutional denoising auto-encoder(CDAE),temporal convolutional network(TCN),and bidirectional long short-term memory(BLSTM).Firstly,CDAE is used to reconstruct noisy signals aiming at the original signal.Then the parallel abstract and temporal feature mining of the disturbance is done with TCN and BLSTM.Finally,the two features are fused by feature merging layer and the classification is completed.The simulation results show that this method can effectively classify 20 types of PQD signals under strong noise,with an average accuracy of 99.23%.Compared with other mainstream classification methods,the proposed method has better classification and anti-noise performance.