SVM with optimized parameters and its application to electronic system fault diagnosis

More and more electronic systems become the subsystems of large and complex system. So their safe and reliability are most important for the whole system's reliability, and more emphasis has been laid on the accurate fault diagnosis of the electronic system. In this paper, based on the characteristic of non-linear, complexity, strong interference and diversity showed in electronic system fault diagnosis, we propose a method of u sing an optimal support vector machine (SVM) to diagnose the electronic system fault. In the optimal SVM, we improve Chaos Particle Swarm Optimization (CPSO) algorithm to achieve the parameter optimization, which could improve the efficiency of choosing parameter and avoid the deficiency of speed and accuracy in choosing parameter. In the improved algorithm, the chaotic sequence is used to initiate individual position, which strengthens the diversity of searching. And an effective method that identifies premature stagnation is embedded to the PSO algorithm, so once premature stagnation happens, the center of all individual best places will replace the global best place. So the colonial diversity is increased, and the premature convergence is avoided to some degree. The fault diagnosis simulation result of electronic system shows the fault detection rate reaches 98.2% and the average fault recognition rate is over 97% with this SVM, and the improved CPSO is an effective method for the parameter optimization of SVM.