Investigation on flow instrument fast fault detection based on LSSVM predictor

Aimed at the issue of real time fault diagnosis in Flow Totalizer, a new type predictor based on Least Squares Support Vector Machine (LSSVM) was put forward. By comparing predictive value with flow meter output value, fault diagnosis was carried out. Predictive errors and the speed of prediction were considered in this algorithm, by dealing with compromise between them, the samples were selected and a higher predicting speed was ensured. The analysis and simulation results showed that the relative higher speed could be acquired by training LSSVM with the selected-samples and reducing the precision of predictor. So this type of predictor was more suitable in real time fault detection.