Research And Application Of Incremental Learning Algorithm Based On Relevance Vector Machine

Relevance vector regression(RVR)is an important learning method aiming to fit the target data in the field of machine learning,and has attracted much attention because of itssparsity,global optimality and the ability to solve nonlinear problems by using kernel functions.In this paper,we study the theory and method about using online incremental relevance vector machine for data prediction,based on relevant vector machine and incremental learning,we build relevant vector machine sparse probabilistic model,and propose incremental precision adaptive online learning mechanisms,which effectively improve the accuracy of relevance vector machine and online learning prediction in real-time and reliability.The main work is as follows:(1)An incremental learning algorithm of “L” relevance vector regression(LIRVR)is proposed,based on incremental learning of relevance vector regression(IRVR).The algorithm mainly aims at reducing the negative effects caused by the loss of relevance vector in IRVR,by preserving the relevance vectors samples and eliminating the non-relevance vector samples,enhances the role of relevance vector in the process of incremental learning,and reduces the time and space complexity of algorithm.In the artificial dataset,the feasibility of the algorithm is verified in accuracy and efficiency of prediction by the related tests.(2)In the process of incremental learning.By studying the characteristics of each sample in the dataset,we can fully mine the distribution information and error information of samples.And on the basis of this,we can put forward the local density factor and error factor.Then the sample characteristics are embedded into the LIRVR algorithm,and an incremental learning algorithm of relevance vector regression based on the feature of samples(SCBIRVR)is proposed.This algorithm takes into account the characteristics of the samples,which ensures the validity of the algorithm.(3)Compared with others sensor,the Fiber Grating Sensor has the advantages of high precision,transmission without electricity,anti-electromagnetic interference and so on.Applying SCBIRVR algorithm to the high-speed railway safety monitoring system based on Fiber Bragg grating,can not only monitor the running status of high-speed railway in real time,and obtain the reliable operation data,but also predict the health of high-speed railway using the model of the algorithm after learning at the same time.The use of algorithmplays a warning role,and achieves good results.