Research On Key Technologies Of Transformer Condition Monitoring Based On RFID Sensor And Deep Learning

Industrial 4.0 and the concept of energy Internet have increasingly stringent requirements for building a strong smart grid and improving the reliability of power supply.As one of the most important equipments in power grid,transformer plays a connecting role in power transmission,so improving the reliability of transformer operation is essential.Effective condition monitoring of transformers is of great value and significance for transformer operation condition assessment,fast fault diagnosis,precise maintenance and improving the reliability and safety of transformer operation.The existing transformer condition monitoring mainly relies on manual inspection and traditional electrical and chemical measurement methods,which have high cost and poor real-time performance.Therefore,low-cost,effective and real-time condition monitoring method is very important for the development of smart grid.In order to reduce the cost of signal acquisition and transmission,a method of vibration signal acquisition and transmission of transformer based on RFID technology is proposed in this paper.In order to reduce the cost and ensure enough communication distance,photovoltaic cells are used as the energy source of RFID sensors.Considering that photovoltaic cells are greatly affected by weather and cannot acquire energy at night,supercapacitors are used as standby power supply.It realizes the uninterrupted power supply of the RFID sensor and improves the sustainability of its work.Subsequently,due to the poor effect of existing transformer mechanical fault diagnosis methods on extracting early fault features,this paper focuses on the method of extracting transformer early fault features by stacked denoising autoencoder(SDA)technology in deep learning method.The learning rate of each DA in SDA adopts double chain quantum genetic algorithm(DCQGA)for optimization.In addition,the node structure of each hidden layer in SDA has an important impact on the effect of feature extraction,so the hidden layer structure of SDA is also generated by DCQGA algorithm.For the extracted features,the transformer early fault diagnosis model based on support vector machine(SVM)is used to identify all kinds of early faults of transformer windings and cores.Based on the early fault diagnosis of transformer windings and iron cores,the related research on fault prediction is carried out.In view of the current situation that there is little research on transformer mechanical fault prediction system,this paper focuses on a structure of transformer winding and iron core mechanical fault prediction,and proposes a method using Hilbert marginal spectrum of vibration signal as fault prediction feature.According to the Hilbert marginal spectrum obtained,the total harmonic distortion(THD)is calculated as the state index,and the state index of the component is calculated over a period of time.Based on the sample data,the multiple kernel relevance machine(MKRVM)is trained with the sample data,and the related prediction model is established.The fault trend of the components is predicted.The experimental results show that the proposed method has good diagnostic accuracy and prediction accuracy.