Fault diagnosis of transformer based on rough set and genetic algorithm

Focusing on the combination problem of value and ratio research in the dissolved gas-in-oil, a novel reduction method is put forward based on the combination of genetic algorithm and rough set. The reduction method utilizes the capability of searching for global optimum of genetic algorithm and achieves better reduction result compared with classical rough set reduction algorithm. In addition, the break point is divided considering different factors of transformer fault. In transformer fault diagnosis system, it is very important to make unified reduction for value and value signals using rough sets. In this way, the precision of fault identification can be improved. The problems of continuous attribute discretization, attribute reduction and attribute value reduction are translated into the simplification problem of discemibility matrix and genetic algorithm can be employed. Experiment on a certain 220 kV main transformer (diaphragm type) shows that the reduction method using genetic algorithm accelerates the evolutionary process and avoids premature convergence effectively for the system with 16 attributes and 200 records. Compared with other 7 kinds of methods, this method possesses feasibility and effectiveness for the fault diagnosis of complex dissolved gas analysis system with thousands of data.