Transformer Fault Diagnosis And Research Based On Bayesian Networks

As the core equipment in the power system,the transformer plays a vital role.Whether the operation state of the transformer is normal or not directly determines the operation state of the power system.With the intelligent development of power grid construction,the voltage level of transformer is higher and higher,and the required capacity is also larger and larger.Once the transformer fails,it will cause the operation stagnation of the power system and lead to power failure,resulting in economic losses.If the situation is serious,it can cause personal injury to people.Therefore,strengthening the research on transformer fault diagnosis technology can find fault symptoms in time and avoid serious accidents.It is of great practical significance to reduce the economic losses caused by power accidents and improve the stability and reliability of power system.Dissolved gas analysis method in oil can cover most transformer fault types and realize real-time monitoring.It is widely used in transformer fault diagnosis.The classical transformer fault diagnosis methods usually diagnose according to the judgment guidelines,and there are some problems,such as missing coding,coding error and so on.In recent years,the research on artificial intelligence technology is very hot,and the application of machine learning algorithm in transformer fault diagnosis is becoming more and more mature.Combined with dissolved gas analysis in oil,it has achieved good diagnosis results.Bayesian networks have won the favor of researchers in the field of fault diagnosis because of its uncertain knowledge expression ability and reasoning ability.In order to avoid the poor optimization effect of Bayesian networks in the process of structure learning,a learning method of improving sparrow search algorithm to optimize the structure of Bayesian networks is proposed.Firstly,10 benchmark functions are selected,and the performance of sparrow search algorithm is compared with other widely used swarm intelligence optimization algorithms such as particle swarm optimization algorithm and gray wolf optimization algorithm.Through the comparison of the optimization results of 10 test functions,it shows that the optimization effect of sparrow search algorithm is better,and using sparrow search algorithm to learn the structure of Bayesian networks has advantages.Then,the improved strategy is introduced into the sparrow search algorithm to further enhance its performance.The maximum support tree is established by introducing the mutual information theory and directional processing to obtain the initial structure of Bayesian networks,that is,the initial population.The tent chaotic map is introduced to initialize the population to make the population evenly distributed;At the same time,a new labor cooperation mechanism is proposed to speed up the convergence speed of sparrow search algorithm;Because the sine and cosine algorithm is periodic in the process of global optimization,it can well balance the local and global search of sparrow search algorithm.In order to prove that the improved algorithm has better optimization performance,10 benchmark functions are tested in combination with the unmodified sparrow search algorithm.By comparing the test results,it can be seen that the improved sparrow search algorithm has better global optimization ability and convergence speed.Therefore,using the improved sparrow search algorithm to learn the structure of Bayesian networks can achieve better results.Taking oil immersed transformer as the research object,the one-to-one correspondence between dissolved gas composition in oil and fault type and the principle of improved codeless ratio are analyzed.The node variables of Bayesian networks fault diagnosis model optimized based on improved sparrow search algorithm are determined,and the fault diagnosis model of oil immersed transformer is established.The fault model is tested by using the fault data of oil immersed transformer.The experimental results show that the fault diagnosis accuracy of the model can reach 97.8%,and its diagnosis effect is better than the transformer fault diagnosis model of Bayesian networks optimized by particle swarm optimization algorithm and the transformer fault diagnosis model of Bayesian networks optimized by gray wolf optimization algorithm.It has feasibility and good application prospect.There are 19 figures,13 tables and 76 references.