改进支持向量机对污水处理厂运行状况的故障诊断

Fault Diagnosis of WWTP Based on Improved Support Vector Machines

针对污水处理厂运行时故障数据不平衡性和代价敏感等特点,构造风险泛函RWLOO(α)来改进支持向量机(Support vector machine,SVM);并用遗传算法(GA)对风险泛函求全局最优.在GA对RWLOO(α)寻优过程中,SVM的几个参数以及核函数同时进行最优化.结果表明:用改进的SVM对污水处理厂的故障数据进行分类时,比未经改进的SVM错分类率低16.5%.

Because of the characteristics of the abnormal data in waste water treatment plant （WWTP）, such as the unbalanced distribution and cost sensitiveness of the fault classes data, a risk functional RwLoo （a） with weight coefficient based on leave-one-out errors was presented, and then Genetic Algorithms （GA） was used to globally optimize the risk functional RwLoo（ a ）. In the optimization algorithm, the kernel function and some parameters of support vector machine （SVM） were optimized synchronously. The improved SVM was used to classify the dataset of WWTP, and the results have indicated that the misciassification rate of the improved SVM is 16.5 % lower.