Abstract: For the question caused by traditional support vector machine algorithm in analog circuit fault diagnosis, the way using support vector machine algorithm of hybrid kernel function (HSVM) and particle swarm optimization (PSO) is proposed. First, after analyzing the transient circuit under test, and writing down the output voltage, wavelet package is used to extract the output voltage feature; second, we use PSO to optimize the kernel weight and structure parameters of HSVM; last, the trained model is used to diagnose the fault. This method not only reduces the randomness of parameters selection, but also the accuracy of simulation result is improved 5%. The effectiveness is proved during the process of fault diagnosis in high-pass filter analog circuit.

Key words: support vector machine of hybrid kernel function, particle swarm optimization, wavelet package, analog circuit, fault diagnosis