PSO based LS-SVM approach for fault prediction of primary air fan

The primary air fan is one of the most important auxiliary equipment of the thermal power plant and the online monitoring and fault prediction can assist in guaranteeing the reliable and stable operation of power generation. The performance degradation and deterioration can be proactively detected and restrained before the fatal failure occurs, so as to promote the system maintenance with reduced costs. With the recognition that the operating conditions vary over time and the operational variables are often strongly cross-coupled in the power plant, this paper presents a PSO based Least-Square Support Vector Machines (LS-SVM) approach to predict the vibration of primary air fan with significantly reduced complexity as well as improved accuracy, which can be adopted for further potential fault diagnosis. Through collecting the information of operational states of primary fan at different measurement locations, this work aims to predict the fan vibration at different operational conditions, and hence to further identify the anomalies and performance degradation of the fan. The suggested solution is evaluated through a set of simulation experiments based on the field measurements from Hequ power plant by using the BP neural network as the comparison benchmark, and the numerical results verify the effectiveness with expected performance.