Fault Diagnosis Of Power Transformer Based On Improved Extreme Learning Machine

With the continuous development of China’s economy,the reliability requirements of the power system are becoming higher and higher.Power transformers are important transformer equipment in the power system.Their cost is expensive,the operating environment is complex,and the operating time is long,once a failure occurs,the direct and indirect economic losses are huge,so it is of great significance for the safe and stable operation of the power system to promptly and accurately detect and repair the fault of the transformer.The thesis first briefly introduced the gas production mechanism of the transformer,and related the dissolved gas content and gas production rate in the oil to the transformer failure.For transformers that do not exceed the gas content in the oil,there may also have faults.To cope with this situation,the gas production rate of the oil can be used to determine whether it is faulty.Then the basic principle of extreme learning machine was discussed,and the classification method of extreme learning machine was studied in depth.By introducing mixed kernel function to extreme learning machine,a classification method based on mixed kernel extreme learning machine was proposed,its parameters were optimized by artificial fish swarm algorithm with superior optimization effect,and an improved fish school algorithm mixed kernel limit learning machine algorithm(AFSA-MKELM)is proposed.The advantages and disadvantages of this algorithm and single kernel function extreme learning machine,three-ratio method and support vector machine algorithm were compared through simulation experiments.In order to diagnose faults of transformers under different operating conditions,a set of transformer fault diagnosis methods were summarized.Observe the gas content in the transformer oil,if it exceeds the warning value stipulated by international standards and the gas production rate is higher than the absolute gas production rate warning value stipulated by the national standard,it indicates that the transformer has failed,at this time,the AFSA-MKELM algorithm proposed in this paper is used for fault diagnosis.If a transformer has a high gas content for some reason and has exceeded the warning value stipulated by the national standard,but the gas production rate is lower than the absolute value of the national standard stipulated warning value,it cannot indicate that the equipment is faulty,and it may also be caused by oil filling or other reasons.If the generation rate of various gases in a transformer increases rapidly in the short term and exceeds the warning value stipulated by the national standard,but the gas content has not yet exceeded the warning value stipulated by the national standard,it can be judged that there may be a potential fault inside the transformer,which needs further Observed.The feasibility of this method was proved by specific transformer examples.Finally,this thesis built a transformer fault diagnosis system through MATLAB’s GUI(graphical user interface).This system basically covered all the functions required for transformer fault diagnosis.Its operation interface is simple and friendly and users can choose a diagnosis method according to their needs and maintain the database.Finally,the system was tested through an example to prove the application value of the system.