Prediction Of Fiber Sensing Signal Based On Multi-Scale Kernel Optimization

Fiber Bragg Grating(FBG)sensor have good resistance to harsh environments.So that it can be widely used in various areas such as large structure and perimeter security.Therefore,it is necessary to analyze the FBG sensing data.Relevance Vector Machine(RVM)is an important learning method aiming to fit the target data in the field of machine learning.It has attracted much attention because of its sparsity,global optimality and the ability to solve nonlinear problems by using kernel functions.In this paper,biased wavelet kernel is constructed based on the method of RVM and characteristics of the biased wavelet.With the selection method of kernel function,we propose an adaptive method to select the biased wavelet kernel.Based on the diversity of data,a hybrid RVM algorithm is proposed.First of all,the training data set is divided into several parts according to the features of data.And the adaptive method is used to filter the biased wavelet kernels for each part.Then RVM models are trained by these kernels.And the algorithm is applied to the Fiber Bragg Grating Temperature Sensing System to predict the sensing data.The main work is as follows:(1)A set of biased wavelet kernel functions are constructed according to the characteristics of biased wavelet.Unlike other kernels,biased wavelet have adjustable nonzero mean which make the kernel of RVM more flexible.With the method of Kernel Target Alignment(KTA),the biased parameter is selected to improve the performance of RVM model.In the standard dataset,the feasibility of the method is verified by the related tests.(2)Based on the features of data,the diversity of data and the factors which influence the diversity are investigated.The data is segmented by using the factors.And different biased wavelet kernel functions are constructed by the method of KTA to highlight the diversity of data.(3)Using the hypothesis test,it is found that the variance is the factor influencing the prediction performance of the sparse probability model.And the variance is used as criteria to build a multi-predictor setup.In the standard dataset,the hybrid algorithm showed an increased prediction accuracy compared to using only individual original RVM.(4)Based on the understanding of the operation principle of FBG sensor,the influence of the temperature change on the wavelength of the FBG sensor is obtained.Then the hybrid RVM algorithm was applied to the temperature data through the FBG sensor.And a prediction system based on the FBG temperature sensor was established.