基于竞争神经网络的逆变器状态识别方法研究

Research on inverter state recognition method based on competitive neural network

针对地铁车辆辅助逆变电路中电容软故障无明显征兆,难以识别的问题,本文提出一种基于时域参数的样本特征提取,结合模糊聚类方法建立全体样本特征的模糊相似矩阵,并采用竞争神经网络模型对滤波电容进行状态分类。实际应用时,首先建立MATLAB电路模型,分别对该电路的不同故障状态和正常状态进行仿真分析,提取输出电压信号进行时域分析得到信号特征向量,作为特征样本;其次,利用全部状态的特征样本求得模糊相似矩阵,并将其引入竞争神经网络对故障进行分类。实验表明,此方法能简单有效检测区分软硬故障,实现滤波电容状态识别,分类正确率达到93.75%。

Aiming at the problem that the capacitor soft fault in the auxiliary inverter circuit of the subway vehicle has no obvious symptoms and is difficult to identify, a method of extracting sample features based on time-domain parameters is proposed in this paper, which combines fuzzy clustering method to establish a fuzzy similarity matrix of all sample features and use a competitive neural network model to classify the state of filter capacitors. In actual application, the MATLAB circuit model is first established, and the different fault states and normal states of the circuit are simulated and analyzed respectively. The output voltage signal is extracted and analyzed in the time domain to obtain the signal feature vector as a feature sample. The feature samples of all states are used to obtain fuzzy similarity matrix which will be to introduced into a competitive neural network to classify faults. Experiments show that this method can easily and effectively detect soft and hard faults and realize filter capacitor state recognition leading to the 93.75% classification accuracy rate.