573 / 2022-03-30 12:23:15

Partial Discharge Pattern Recognition of Power Transformer By Distributed Computing Framework

Partial discharge,Distributed,Fault identification

     With the growing size of the electric power industry, various power data also increasing rapidly. Especially when collecting the statistical maps of ultra-high frequency signals of partial discharge, the sampling frequency of the system needs at least several GHz according to the Nyquist sampling theorem. This poses great challenges to data storage in monitoring systems. Difficult to meet the needs of accurate acquisition of equipment status information and network interaction in smart grid. It is urgent to use computing resources efficiently and speed up data transmission. Aiming at the problems of low efficiency and less accuracy of existing partial discharge fault identification algorithms for power transformers, in order to realize fast and accurate partial discharge fault identification of power transformers, a partial discharge fault identification algorithm for power transformers based on distributed computing framework is proposed.The algorithm consists of two parts. In the first part, multiple back propagation neural network pattern recognition classifiers are constructed based on Ray. Then, multiple BP neural network recognition tasks are assigned to each slave node in the Ray cluster using the Ray computing framework to improve the computational efficiency. In the second part, the waveform characteristics and spectrum statistical characteristics of partial discharge UHF signals are extracted, and the pattern recognition is carried out respectively. The recognition results of each identifier are aggregated by using the Ray framework. By introducing trust evaluation mechanism and vector similarity evaluation mechanism, a multi- identifier fusion decision model is constructed, and the final comprehensive decision result is obtained.The pattern recognition rate of the BP neural network used in this algorithm reaches 98.5 %, and the efficiency is also 67 % higher than that of the traditional recognition algorithm. Can be more efficient and accurate recognition of partial discharge UHF model pattern recognition.Based on the partial discharge pattern recognition algorithm under the Ray distributed framework, this algorithm can identify the partial discharge pattern recognition more efficiently and accurately.