Genetic Programming Based Polynomial Networks Model for Insulation Fault Diagnosis of Power Transformers

A Genetic Programming based Polynomial Networks Model (GPPNM) is presented in this paper to promote the diagnostic performance of incipient insulation fault of power transformers. Other than conventional hierarchical architecture to build polynomial networks, the proposed GPPNM constructs it using tree-like structure of Genetic Programming (GP). By means of flexible selection of low-order polynomial functions and feature variables in each node of structure, the polynomial networks is evolving in the global search space by generations to capture the complex and numerical knowledge relationships between dissolved gases and fault types. The proposed model has been applied on the actual fault records and compared with conventional method, artificial neural networks method and self-organizing polynomial networks (SOPN) method. The numeric test testifies that the GPPNM requires less prior knowledge in the process of construction of diagnosis model and has better performance than other methods. Key-Words: Power transformer, Insulation fault diagnosis, Dissolved gas analysis, Polynomial networks, Group Method of Data Handling, Self-organizing polynomial networks, Genetic programming