Fault detection of power transformers using genetic programming method

This paper proposes a novel method for insulation fault detection of power transformer using the genetic programming (GP) method. Fault detection can be seen as a problem of multi-class classification. GP is a way of automatically constructing computer programs using a process analogous to biological evolution. GP methods of problem solving have a great advantage in their power to represent solutions to complex classification problems. The flexibility of representation gives GP the capacity to represent classification problems with means unavailable to other techniques such as neural networks. A binary tree (Bi-tree) structure is presented to transfer an N-class problem into N-1 two-class problems. The proposed method has been tested on the actual records and compared with the conventional methods, fuzzy system method and artificial neural network method. The result shows that GP has advantages over the existing diagnosis methods and provides a new way to solve the problem of fault detection.