FINNIM: Iterative Imputation of Missing Values in Dissolved Gas Analysis Dataset
Dissolved gas analysis (DGA) is essential to the fault diagnosis of oil-immersed power transformer. After thoroughly analyzing the gas production mechanism of power transformer faults, it has been found that there are no explicit mapping functions between the single fault of power transformer and the content of gas. To handle this problem, a multi-class classification model for power transformer fault diagnosis based on least squares support vector machines (LS-SVMs) is presented. Appropriate parameters are very crucial to the learning performance and generalization ability of LS-SVMs. However, the determination of LS-SVMs parameters, more dependent on experience, has always been a problem in research field. To overcome this problem, bacterial colony chemotaxis (BCC) algorithm is firstly introduced to select the LS-SVMs hyper-parameters in this paper. Finally, based on the concentration distribution of some typical fault gases, the proposed method is applied to recognize the faults, and ulteriorly a comparison with IEC three-ratio method, BP neural network (BPNN) and the model optimized by grid search is made in order to evaluate the method properly. Experimental results show that recognition rate of LS-SVMs with BCC is 18.52, 14.82 and 3.71 percents higher than that of IEC three-ratio method and BPNN and LS-SVMs with grid search, respectively. So the effectiveness and practicability of the proposed method is proved.