Transformer fault diagnosis method based on cuckoo searching optimized neural network

Abstract

The invention relates to a transformer fault diagnosis method based on a cuckoo searching optimized neural network. On the basis of neural network structure parameters in an artificial intelligence method and combination of a meta-heuristic intelligent method of cuckoo searching, structural parameters of the neural network are optimized by using a cuckoo search method. A stable cuckoo-search-optimization-based neural network structure is obtained by DGA data training; and new data are predicted to solve a classification problem. Therefore, a defect that most of the traditional diagnosis methods are restricted to threshold determination can be overcome and universality is high. Compared with the basic neural network algorithm and other meta-heuristic optimized neural network algorithms, the provided method has advantages of fast convergence speed, low model sensitivity, and high robustness. Moreover, on the basis of the meta intelligent algorithm like cuckoo searching, the neural network structure is optimized, so that the defect that the neural network is restricted to local minimization can be overcome; and the parameter adjusting process has universal regularity.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence and relates to a transformer fault diagnosis method. Background technique [0002] The power transformer is an important equipment in the power system. Any failure during the operation of the transformer may terminate the power supply, resulting in major production accidents and economic losses. Therefore, accurate prediction of transformer faults is very important. [0003] Dissolved gas analysis (DGA) in transformer oil, as an efficient transformer fault diagnosis method, is widely used in the actual operation of power systems. By analyzing data on specific gas content dissolved in transformer oil, patterns in transformer operating conditions can be found. Therefore, researchers have proposed a large number of diagnostic methods based on DGA data, which are mainly divided into two categories: traditional diagnostic methods and intelligent diagnostic methods. [0004] Tradition...

AI Technical Summary

Problems solved by technology

The existing artificial neural network method has the following disadvantages: First, the number of training steps and the correctness of judgment are often not satisfied at the same time, and the training results are often unsatisfactory; second, there is no mature technical means to determine the network structure and adjustment parameters, and the network performance is limited. Good or bad can only rely on manual experience and luck in many cases, and the excellent results obtained cannot be explained, and the experimental results may not be paid in cash, and are sensitive to small changes in parameters

Images