The Research Of Transformer Fault Diagnosis Based On Ant Colony Algorithm

The power transformer is the hub of the power system equipment, its operating state has an important influence on the normal operation of the grid. With the development of the power industry, due to many factors, DGA has been unable to determine very accurately the type of transformer fault. So combining DGA and intelligent approach have become a inevitable trend of transformer fault diagnosis.Combining the DGA with artificial neural networks is the most commonly used diagnostic method at present, artificial neural networks has many advantages, such as adaptive learning、parallel processing、associative memory and nonlinear mapping, those provide a new direction for solving transformer fault problem. The most commonly used is BP neural network combining with DGA by far. But when the training samples with a large number have high accuracy requirements, due to the defects of the BP network’s structure, the shortcomings that BP network has a low speed of convergence and is easy to fall into local minimum reduces the accuracy of fault diagnosis. Therefore, it is necessary for transformer to optimize the BP network.According to the idea that the ant colony with global optimization and heuristic search capability just make up the shortage of BP neural network, this paper puts forward a new theory, which is an optimized BP neural network based on ant colony algorithm used in transformer fault diagnosis. Firstly, it constructs 5-8-5 BP neural network for simulation and fault diagnosis. Using the program written by MATLAB prove that simple BP network can recognize the transformer faults with low accuracy. Secondly, the article describes the basic principle of ant colony algorithm to optimize the BP network, indicating that its core is optimizing the weight parameters of BP network based on ant colony algorithm. Comparing the BP network, ACA-BP proposed in the paper can indentify the transformer fault with high accuracy, it also could effectively avoid getting into local optimum, accelerate network convergence speed, greatly reduce the training time and increase the accuracy of fault identification. Finally, the article summarizes the advantage of ACA-BP and prospects to improve the method further.