针对电力电子电路中器件的参数故障诊断问题，提出一种基于模糊推理的分类器融合诊断方法。采用神经网络和支持向量机分类器作为模糊推理输入的两种子分类器，设计了对应的隶属度函数对2种分类器输出进行模糊化处理；根据预先处理好的模糊变换矩阵进行计算，最终得到系统的融合输出。通过对一个Cuk电路的仿真实验和分析表明，所得诊断方法是有效的，优于任意单个子分类器的诊断精度，该方法对于随机噪声具有较好的鲁棒性。

Aiming at the parametric faults diagnosis of the power electronic circuits, a method based on classifiers fusion was proposed to diagnose the circuit by using the fuzzy inference process. The artificial neural network (ANN) and the support vector machines classifier (SVC) were regarded as two sub-classifiers of the input of the inference system. The membership functions were designed to process the output of the sub-classifiers, and the fuzzy inference process functioned by computing the fuzzy transformation matrix. The simulation results from diagnosing a Cuk circuit reveal that the performance of the fusion process is valid and superior to anyone of the sub-classifiers and furthermore, the fusion method is robust to random noise.

国家自然科学基金项目(60374008, 60501022, 90505013)；航空科学基金项目(04I52068, 2006ZD52044)。