Research Of PD Pattern Recognition Of Dry-type Transformer In Offshore Platform Based On MobileNet Model

The offshore platform power system is the only power source for offshore oil and gas exploitation,processing and transportation.At present,China's offshore platforms have formed a complex power network,and improving the reliability of offshore multi-platform interconnected power systems is of great significance in ensuring oil and gas production,narrowing the gap between fossil energy supply and demand in China,and promoting sustainable economic development.As the safety level of all kinds of electrical equipment is required for offshore platforms,the transformers used are all large-capacity dry-type transformers.As a manifestation of transformer insulation defects,partial discharge will further lead to the deterioration and damage of insulation materials,thus affecting the safe and stable operation of the transformer and even the whole power grid.However,the offshore platform is far away from the coastline,which leads to the limitation of inconvenient transportation and shortage of space resources Therefore,in order to further improve the stability of the platform and ensure normal production and life,it is more and more important to carry out on-line partial discharge monitoring and fault diagnosis on the transformer in operation.Because different insulation defects in the transformer will lead to different types of partial discharge,which corresponds to different levels of harm,the identification of discharge types can help to understand the internal insulation condition of the transformer and speed up the fault treatment which has a certain practical value.The existing partial discharge pattern recognition classifier mostly adopts the traditional machine learning algorithm suitable for small data samples,but the recognition effect of such algorithm needs to be improved due to the limitation of training mechanism.In this context,the introduction of deep learning algorithm suitable for massive data processing can fully mine the value of detection data and improve the model recognition rate and calculation speed,which can actively promote the development of transformer condition-based maintenance.In this paper,based on the analysis of the cause and process of partial discharge in transformer,four partial discharge defect models are designed to simulate tip discharge,suspension discharge and insulation defect discharge.Build an experimental platform in line with the IEC60270 standard,and use pulse current method and ultra-high frequency method for data acquisition.Because the partial discharge signal is weak and vulnerable to interference,the traditional wavelet threshold denoising method is improved in order to improve the monitoring effect and recognition accuracy.On this basis,the correlation and differences between different types of PD data are studiedCompared with the traditional classifier,the deep learning algorithm has higher accuracy,but it often has some disadvantages,such as complex network structure,large number of parameters,large amount of memory,slow computing speed and so on.In order to solve these problems and make the classifier more suitable for intelligent sensing devices with the trend of miniaturization and low power consumption,a new MobileNet convolution neural network(MCNN)model is proposed to identify discharge patterns.In the construction of the MCNN model,a lightweight attention machine SE module and a nonlinear function h-swish are added to improve the feature expression ability of the model,reduce the number of model parameters and computational complexity,eliminate the potential loss of accuracy in partial discharge pattern recognition,and further improve the recognition rate.The MCNN model was trained and tested with the pre-processed discharge image and a variety of methods are used to visualize the model to verify the effectiveness of the model.The results show that the proposed MCNN can reduce the number of parameters of the network model and improve the calculation speed,and achieve the best performance,on the premise of improving the recognition accuracy,compared with the classification results of partial discharge in various deep neural network models or convolutional network compression models.In order to ensure oil and gas production in China and further improve the reliability of offshore multi-platform interconnected power system,it is necessary to ensure the safe and stable operation of transformers.The realization of integrated on-line monitoring and type diagnosis of partial discharge in transformer can help to fully understand the operation state of transformer and find out its abnormal situation in time,so as to play a certain guiding role in operation,inspection and maintenance.