Research on the health prediction of system based on relevance vector machine and ant colony algorithm

With the rapid development of modern industrial society, the function of the equipment is more and more perfect, and its structure is also more complicated. Therefore, how to find fault more timely and accurately, and predict remaining useful life of circuit and the failure of industrial production and is an important issue. The prediction model of Relevance Vector Machine is a probabilistic sparse model based on Bayesian framework, it can be used for regression and classification model analysis. However, the kernel function parameters have a great impact on the performance of RVM. The data was putting into a higher dimension space when linear inseparable data is more and more complex. The linear mixed kernel function was constructed with Gaussian kernel function and polyno­mial kernel function for the approximation ability and better generalization ability of model. In order to improve the effect of classification and regression, and to avoid introducing too many parameters, the new method of optimizing the kernel parameters is proposed. Ant colony algorithm is superior in four aspects, such as comprehensive computation, complexity, stability and precision, which is used to optimize the RVM kernel parameters. Thus, accurate prediction of the circuit is realized. Finally, the effectiveness of the method is verified by the data of the lithiuM BATTERY.