NOVEL FUSION APPROACHES FOR THE DISSOLVED GAS ANALYSIS OF INSULATING OIL

Dissolved Gas Analysis (DGA) is the most reliable technique to identify the incipient faults in power transformers. There are several DGA techniques in use such as Doernenburg, Rogers, IEC, etc. On the other side there is an increasing tendency to combine data from multiple sources and models to achieve more reliable results than individuals. This investigation proposes two fusion approaches consisting of fusion architectures and respective combination methods to combine DGA techniques and the gas ratios utilized in these techniques. The proposed approaches in this article apply a modified flexible neuro-fuzzy and a gating network as combination methods. Various gas concentration data were used for training and validating the models. Results showed that the proposed approaches have more advantages compared to the conventional DGA techniques. Finally, the importance degree of each gas-ratio to detect each fault was investigated. Keywords– Artificial intelligence, data fusion, neuro-fuzzy systems, neural networks, support vector machines (SVM), dissolved gas analysis (DGA), fault diagnosis, power transformers