Sensor fault detection with online sparse least squares support vector machine

In this paper, we present the theory of online sparse least squares support vector machine (OS-LSSVM) for prediction and propose a predictor with OS-LSSVM to detect sensor fault. The principle of the predictor and its online algorithm are introduced. Compared with the traditional least squares support vector machine (LSSVM), OS-LSSVM has an advantage on training speed owing to the online training algorithm based on the base vector set. The real-time output data of sensor is employed as the training vector to establish the regression model. This method is compared with the LSSVM predictor in the experiment. Three typical faults of sensors are investigated and the simulation result indicates that the OS-LSSVM predictor can diagnose sensor fault accurately and rapidly, thus it is especially suitable for online sensor fault detection.