Application Research On D-S Evidence Theory In Thermal Power Unit

Thermal power unit is composed of three main equipments of boiler, stea m turbine and electric motor. Its structure is complex and its parameters are numerous and they interact with each other. Some of the equipments are extremely prone to failure, once it breaks down, the failure will bring unforeseen effects and losses to the power plant and it will influence the safe and economic operation of thermal power unit. Rely on the mechanism analysis will increase the difficulty, while just rely o n personal experience will cause greater error and increase the subjectivity.We can make full use of the data generated by thermal power unit, and apply it to the information fusion algorithm to improve reliability and validity of fault diagnosis and pattern recognition.D-S evidence theory is a kind of artificial intelligence algorithm processing uncertain information. At present, the theory of D-S evidence theory is becoming more and more mature, the application is more and more extensive, and it is a simple and reliable information fusion algorithm. But there are two problems in its application. Firstly, the method of constructing confidence function distribution is more subjective. Secondly, manual calculation is heavy. To solve these two problems, this paper proposes a method of using historical fault samples of thermal power unit. Taking transformer oil fault diagnosis as an examp le, the probability distribution of historical samples is judged, and then the probability density is calculated by using the logarithmic probability distribution formula, and then the reliability function is obtained by normalized processing. Finally, the fusion results are got by using the combination rule.The MATLAB program are applied to the fault diagnosis of coal mill and diagnosis of coal type in order to reduce the difficulty of calculation. The fusio n diagnosis results show that the solution of t he two problems can not only improve the reliability, but also greatly reduce the computational amount, and provide reference for further research.