Reliability Forecasting Of The System Based On The Least Squares Support Vector Regression

With the social transformation, the development of culture and the rapid progress of science and technology, reliability of engine system has attracted more and more attention, how to improve the reliability of engine system has gradually become a very important scientific problem, there are many ways to assess the safety of system, the reliability prediction is one of the most general methods. Soft computing techniques, including neural network and support vector machine, have been widely applied to system reliability prediction with their abilities to deal with high non-linearity, but they all have some problems, the most important one is that, simply using support vector machine or neural network methods did not consider the uncertainty and stochastic of reliability of time series data. Therefore, how to establish the reasonable reliability prediction model is a very important significance. To solve this problem, this paper will complete the following tasks:1 Acquaint the current lack of reliability prediction method based only neural network and support vector machine by reading a lot of related references, propose the iterated nonlinear filters algorithm based on least square support vector machine prediction method, this method is reasonably considered the uncertainty of reliability data in time series.2 Establish the mathematical prediction model of the engine system reliability. In order to use the iterative nonlinear filtering algorithm, establish the engine failure and reliability of time series forecasting model based on the state space model of least square support vector regression model, including state transition equation and observation equation.3 To verify the proposed methods, use IEKF and IUKF algorithms in the proposed model on MATLAB simulation platform, obtain the prediction results, and finally compare the prediction results with the other prediction models, including Auto Regressive(AR)、Radial Basis Function Neural Networks( RBF-NNs)、 Multilayer Perception Neural Networks(MLP-NNs)in the references, the final results shows that, the models the paper proposed has better prediction errors.