基于SVM的故障预报中的并行优化仿真研究

在基于支持向量机的故障预报过程中,故障样本数据的不相关变量会影响支持向量机的性能;加权支持向量机中直接选择加权系数存在很多不足;支持向量机参数主要凭人的经验或通过多次实验获得,还没有一个确定而有效的方法.针对这三种问题,提出了采用改进的人工鱼群算法将特征选择、加权系数、支持向量机参数进行并行优化的方法,并将此方法应用于船舶动力装置冷凝器的故障预报中.仿真结果表明:相对于单独优化,并行优化能够在更短的时间内进行最有效的故障特征提取,并且提高支持向量机的性能;相对于遗传算法,改进人工鱼群算法能够以更快的速度达到最终的优化结果.

Research on Parallel Optimization Simulation of Fault Prediction Based on SVM

In the fault prediction based on Support Vector Machine(SVM),irrelevant variables in the fault samples spoil the performance of SVM;selecting the weighting factors in weighted SVM directly has many disadvantages;SVM parameters are mostly selected artificially or obtained through experiment time after time,a certain and effective method has not been found. Aiming at the three problems,a method jointly optimizing the feature selection,the weighting factors and the SVM parameters with a modified Artificial Fish Swarm Algorithm(AFSA) was proposed. This method is used in the fault prediction of condensator in naval vessel propulsion plant. The experimental results show that the parallel optimization method can select the best fault features in shorter time and improve the performance of SVM than the separate optimization methods,and MAFSA can get the final result faster than genetic algorithm(GA).