Study on nonlinear multifunctional sensor signal reconstruction method based on LS-SVM

In this paper, the nonlinear multifunctional sensor signal reconstruction method based on the least squares support vector machine (LS-SVM) is proposed. Different from the reconstruction methods with empirical risk minimization, the support vector machine (SVM) is a new machine learning method based on structural risk minimization, which is applicable to the case of small sample size calibration data, and can efficiently restrain overfitting and improve generalization capability. With SVM as a basis, the LS-SVM involves equality constraints instead of inequality constraints, so the solving process of the quadratic programming problem can be greatly simplified. In this study, L-fold cross validation is adopted to optimize the adjustable parameters. The reconstruction of input signals of a multifunctional sensor was carried out in two situations of different nonlinearities for which the reconstruction accuracies were 0.154% and 1.146%, respectively. The experimental results demonstrate the high reliability and high stability of the proposed LS-SVM reconstruction method, as well as the feasibility.