Faults Diagnose In Traction Control Unit Of Subway Vehicle Based SVM Parameter Optimization

Nowadays, with the development of automation technology, subway vehicle is faster, more punctual and more convenient. It has been the main transport of people in many cities. So the subway vehicle should maintain stable operating state for our security. The traction control unit(TCU) is one of the most important units of subway vehicle, if something wrong happened with TCU, it should be done faults diagnose immediately, and it is important to give the maintenance advice. Some faults diagnose work has been done in this paper, and the work details are as follows:(1) Select the multi-class support vector machine as the basic troubleshooter. It is very difficult to obtain the precise mathematical model as the structure of TCU is very complex. The multi-class support vector machine(SVM) can be performed when the system model is unknown, it has great classification performance for small sample and high-dimensional pattern recognition problem, so it is suitable for the TCU faults diagnose. SVM overcomes the defects of neural network, it needn’t too much sample, and will converge to the global optimal solution, avoid falling into local optimum effectively.(2) SVM parameters optimization with improved particle swarm algorithm. The parameters C and σ are very important to the performance of multi-class SVM. However, random test method is time-consuming and undesirable. This paper show the improve particle swarm optimization(IPSO) to choose the two parameters. Chaos initialization, adaptive inertia weight has been applied in basic PSO. In the optimization process, premature stagnation timer has been used to determine whether the population gets into premature stagnation, and help the swarm escape from local optima.(3) Fault diagnosis verification based on run data of one city subway line 6. This paper sum up nine kinds of faults types of TCU and eleven variables of TCU according to the failure frequency and failure overlapping relationship. Overall accuracy of fault diagnosis is 93.07%. The experimental results show that the IPSO-SVM classifier is suitable for faults diagnose of TCU of metro.