Research And Application Of Relevance Vector Machine Theory In Building Electrical Fault Diagnosis

As an integral part of smart buildings,building electrical system plays a vital role in people’s daily life.At the same time,the normal operation of the building electrical system will provide guarantee for the development of smart buildings.It can be seen that the fault diagnosis of building electrical system has become a key link in the design and development of building electrical system.At present,the traditional building electrical fault diagnosis method is still in the development stage from manual detection to data-driven diagnosis.The traditional fault diagnosis method is not only low in accuracy,but also low in efficiency.Therefore,the study of intelligent fault diagnosis methods for building electrical systems will further promote the development of intelligent buildings.However,the failure of building electrical system often leads to the loss of manpower and material resources,which means that it is difficult to obtain the failure data of building electrical system.In this paper,the relevance vector machine model based on Bayesian model is studied for the small sample of building electrical system fault data.A fault diagnosis method based on complementary ensemble empirical mode decomposition,correlation coefficient theory and multi-kernel relevance vector machine is proposed.A fault diagnosis method based on main wavelet energy and swarm intelligence algorithm to optimize relevance vector machine is proposed,and a design of fault diagnosis interface for building electrical system based on Tkinter module of Python platform is proposed.The specific research is as follows:(1)A fault diagnosis method based on complementary integration empirical mode decomposition,correlation coefficient theory and multi-kernel relevance vector machine is proposed.Firstly,the complementary ensemble empirical mode decomposition method is studied,which decomposes the original signal into multiple effective intrinsic mode components and constructs feature vectors based on the magnitude of the correlation coefficient.Secondly,the multi-kernel model optimization relevance vector machine model is studied.Finally,the proposed method was validated by simulating fault data on the building electrical experimental platform MA2067.The experimental results show that the average fault diagnosis accuracy of this method can reach 96.75%.Compared with BP neural network and other traditional classifiers,multi-kernel relevance vector machine provides higher accuracy of fault diagnosis,which indicates that multi-kernel relevance vector machine has good classification ability for small sample data.Among them,the complementary ensemble empirical mode decomposition method effectively removes the noise redundancy in the original signal,and the multi-kernel relevance vector machine model effectively improves the computational efficiency and accuracy of the single-kernel model.(2)A fault diagnosis method based on main wavelet energy and swarm intelligence algorithm to optimize correlation vector machine is proposed.Firstly,the wavelet packet energy feature extraction is studied to extract the wavelet packet energy entropy of the original signal.Then the principal component analysis method is studied to reduce the dimension of multi-dimensional energy.Then the relevance vector machine is optimized by grey wolf algorithm and quantum genetic algorithm.Finally,the proposed method was validated by simulating fault data on the building electrical experimental platform MA2067.The experimental results show that the average fault diagnosis accuracy of this method can reach 97.5%.This method has higher fault diagnosis accuracy compared to the previous method.This method effectively solves the problem of the dependence of the multidimensional features extracted by the previous method on the multi-kernel model.Among them,wavelet packet energy method extracts features of original signal effectively,principal component analysis method reduces dimensions of features effectively,and relevance vector machine optimized by swarm intelligence algorithm provides more stable fault diagnosis results.(3)The design of fault diagnosis interface of building electrical system based on Tkinter module of Python platform is studied to simplify the process of fault diagnosis of building electrical system.Users can not understand the composition of building electrical system and its diagnostic principle and algorithm,and can also complete the fault diagnosis of building electrical system with software.The design of this page further promotes the application and implementation of the proposed algorithm,providing research ideas for achieving building intelligence.