An RVM Fuzzy Model Identification Method and Its Application to Fault Prediction
Hu Chang · Acta Automatica Sinica · 2011
For a dynamic system with complexity, morbidity and nonlinearity, it is significant and difficult to establish a fault prediction model accurately in general. Instead, to construct a suitable fuzzy model may be an effective alternative. In this paper, the inherent relationship between relevance vector machine (RVM) and fuzzy inference system (FIS) is investigated firstly, then the uniformly approximating capability of FIS based on RVM is proved. Next, a fuzzy model identification method based on RVM and gradient descent (GD) algorithm is presented as well. Finally, a new fault prediction algorithm is given on the basis of the presented fuzzy model identification method. The simulation studies illustrate that the presented fuzzy modeling method can generate a compacter model and achieve higher prediction accuracy as well. Based on the new fault prediction algorithm, the system fault can be predicted correctly.