Fault diagnosis method for water quality monitoring and control equipment in aquaculture based on multiple SVM combined with D-S evidence theory

Abstract

A water quality monitoring and control (WQMC) system is an important tool designed to maintain good water quality in aquaculture. A variety of events could cause WQMC equipment to malfunction. Such a problem would in turn result in the generation of unreliable monitoring information and reduced water quality control. The high-dimensional dataset generated by the WQMC equipment makes fault identification difficult. To solve the problem of fault identification, this paper reduced the effects of autocorrelation on the variables and determined feature space based on dynamic principal component analysis (DPCA). Based on cross-correlation analysis, 24 support vector machine (SVM) classifiers were developed for the multi-SVM model. A vote procedure was proposed to identify fault types and conflicts. To solve the conflicts and reduce incorrect diagnosis results, an amendment based on D-S theory determined which fault contributed more to causing the corresponding symptoms. Experimental results showed that the single-pass accuracy of the multi-SVM model in random sample tests varied from 90% to 94% and the combined method could effectively solve the conflicts. Furthermore, the fault identification accuracy could be improved by 3–5%. Incorrect fault diagnosis results remained, and the successful amendment ratio required further improvement. However, the proposed method was helpful for the maintenance and management of WQMC equipment.