PDs in high voltage insulation are local breakdowns of parts of the insulation, which do not result in complete breakdown, but which cause a change of voltage, charge and energy of the test object, seen as a whole. PD measurements have been carried out over the years to assess insulation systems in power apparatus for their integrity and design deficiencies. Digital PD recording, processing and its epitome describing the deterioration of electrical insulation is the recent trend in both industry and testing laboratories. Interpretation of these patterns can lead to evaluation of the cause of PD. Good evaluation of discharges resembles detective work where much patience and use of many small facts is required.

The PD problem is inherently random and also very much influenced by the nature of insulation, amount of ageing, interval of voltageapplication, amplitude of applied voltage etc. In spite of precautions, PD measurements are corrupted by interference and noise. Thus the resulting pattern is quite complex, and also the subsequent patterns (arising due to the same PD source) resemble each the only in a broad sense. Therefore all these factors define a very challenging pattern recognition (PR) task, worth attempting to automate. Therefore, a need arises to look for methods in the domain of PR for automating this process.

In the future, neural nets are expected to solve problems that could not be dealt with by conventional Von Neumann Computers, such as PR or associative memory problems, which call for highly parallel and error tolerant machines. Artificial Neural Nets (ANN) have been used very successfully to address many real world PR problems.

The ability to learn from examples, capability of generalisation, abstraction the case with which it can be adapted to a given problem domain when compared to other Artificial Intelligence Paradigms are the prime reasons for its popularity. Its recent availability in hardware form, as Neural chips has further enhanced the prospects of its real world implementation to practical problems. The biggest problem for the application of NN is the selection of its topology and the connection weights. It has been felt that Gradient-descent learning algorithms i. e., learning algorithms that modify the network parameters following the gradient of an error function, have difficulties in learning the topology (number of nodes and their connections) of the net whose weight distributions they optimize. Hence Genetic algorithms (GA) based ANNs are applied for recognition of PD patterns: Power of GA lies in its ability to direct the search towards relatively prospective regions in the search space.

A user friendly software called PRPDGA (Pattern Recognition of PD based on Genetic Algorithm) have been developed to recognize the PDpatterns. The authors explore the issues related to Genetic evolution of ANN topology and weight distribution to PD problem along with the special traits of GA, PD detection and measurement set up, interfacing system and availability and preparation of PD data. Case studies have been discussed to prove the efficiency of the tool developed; encouraging results have been obtained in case of all power apparatus, in general.