Study On Sensor Array Online Monitoring Of Six Kinds Of Transformer Oil Dissolved Gas

Online monitoring the transformer oil dissolved gas is an effective method of analyzing the transformer characteristic gas. It has a positive significance to obtain the insulation condition of the running transformer. Multi gas sensors play an important role in online monitoring transformer multi oil dissolved gas. Because of the cross sensitivity of gas sensor, online monitoring technology based on multi-sensor doesn't improve well. Thus this affects the evaluation of the transformer operating condition. Gas sensor array technology is an effective method to overcome this problem. Sensor array detection of six kinds of transformer oil dissolved gas (H2, CO, CH4, C2H4, C2H2, C2H6) is studied in this thesis. And pattern recognition method is studied step by step. Pattern recognition technology is used to do quantitative analysis after obtaining the multi-dimensional information of the gas composition and concentration by gas sensor array.Characteristic experiments of MQ type gas sensors are carried through based on the requirements of sensitivity, repeatability and response characteristic of online monitoring gas sensor. Research shows that these gas sensors have good performance on sensitivity, repeatability and response characteristic and basically meet the requirements of on line monitoring. But cross sensitivity also exists at the same time.Secondly, BP neural network and principle component analysis are studied. Relevance of sensor array signals is eliminated by adopting principle component analysis. And the conversion results are used as training samples of BP neural network. Pattern recognition consisted of BP neural network and principle component analysis is used in gas sensor array detection. Result shows that this kind of pattern recognition method can solve the cross sensitivity effectively and improve the detecting accuracy of multi-component gas. But traditional BP neural network trains slowly and is trapped into local minimum easily.Thirdly, genetic algorithm is introduced to improve the performance of BP network. And this network is used in the sensor array signals pattern recognition. Results show that genetic algorithm can overcome the shortness of BP neural network effectively, increase the convergence speed and have better recognition accuracy than BP neural network. But it can not ensure that genetic algorithm converges to the global minimum because of the disability of maintaining the group diversity. Finally, immune algorithm is used in the neural network weight adjusting process. Because of the feature that immune algorithm can maintain the diversity of the group, the weights can be adjusted globally. Then BP algorithm is adopted to do local searching on the base of global optimization. A kind of immune neural network is established and used in sensor array signals pattern recognition. Results show that, this kind of neural network overcomes the shortages of both BP neural network and genetic algorithm, and improves recognition accuracy of genetic algorithm neural network.