Shape-space based negative selection algorithm and its application on power transformer fault diagnosis

Dissolved gas analysis is an effective and important method for power transformer fault diagnosis. The negative selection algorithm has much advantage for some faults which lack a great deal of training sample data in the power transformer faults. It can examine the abnormality of the infinite categories by using fewer detectors, covering with wide space. However, the existing negative selection algorithm also exists some shortages. For these shortages, this paper researches a shape-space based negative selection algorithm, which uses the shape-space model in the mathematics theory, mapped the data of detector, self space and non-self space to the n dimension space, using the calculation of affinity to carry out matching. Experiments show that the method can make use of few fault data (i.e. antigen) to obtain the mature training set, so it is suitable for the small sample fault diagnosis of which the failure data is difficult to get.