Weighted Elitist Teaching-Learning-Based Optimization For Support Vector Machine In Fault Diagnosis Of Chemical Process

Support vector machine is a familiar technique in the field of machine learning which has already been applied to fault diagnosis widely. In this paper, weight elitist Teaching-learning-based optimization is proposed to optimize the parameters of SVM, which helps to identify faults in Tennessee Eastman Process. Completed work is summarized as follows:1. To improve the performance of TLBO, an ameliorated strategy to change the weight of elitist TLBO is proposed. Inertia weight of elite learners’and ordinary learners’are reduced separately based on their different properties in elitist TLBO. Compared with several optimization algorithms, the result implies weighted elitist TLBO has a better performance of searching the optimal solution in test function.2. The penalty parameter and kernel function parameter in SVM decide the performance of classifiers. In this paper, the selection of appropriate parameters in the training process of SVM is equivalent to the optimization problem related to fitness function value. After the calculation of optimal solution, weighted elitist TLBO provides the optimal parameters in SVM.3. The SVM optimized by the proposed weighted elitist TLBO is applied for fault diagnosis in Tennessee Eastman process. Compared against the original TLBO and some other popular and powerful optimization algorithms for SVM in TE fault diagnosis, it shows that weighted elitist TLBO-SVM has a better performance.