Power system requires the system controlled and all aspects can be measured in order to detect problems within the system, and rapid response to ensure security and stability of electrical networks. In our country, high voltage power transmission network has a large configuration of SF6 electrical equipments. And in the twelve five-years, a large number of SF6 electrical equipments will be used in the construction of strong and smart grid too. So event of failure of these devices will be a serious threat to security and stability in the electricity network. Therefore, the detection technique for the state of SF6 electrical equipment is a hot issue people concerned today.The outstanding performances of SF6 electrical equipment depend on its insulating medium SF6. The electrical equipment’s insulating ability degrades when the status of SF6 in the equipment changes,. Lots of researches show that different accidental discharges in the equipments lead to different ramifications. It means the types of SF6’s ramifications can be used to judge the discharge types in the equipments. This diagnostic mode doesn’t have to disintegrate the electrical equipments, and at the same time it can provide sufficient data to make the maintenance strategy.The establishment of electrical equipment’s fault diagnosis and precaution method based on the identification of SF6’s ramifications contains the following contents. First, study on the relationship between the electrical equipments’discharge and the ramifications generated. Second, find out a suitable spectrum process method. Third, study on the reduction of the data dimensions. Fourth, establish the fault diagnosis and precaution model.The relationships between the discharge behavior in the electrical equipments and the ramifications generated are the basic of the fault diagnosis and precaution model. a lot of experiments are needed to find out the relationships. But many predecessors have obtained many achievements in this field, and this article tries to sort them out. As these results are based on many predecessors’ achievements carelessness or omission is unavoidable. So we carried out some experiments with Guang Xi electric power research institute to deal with such omissions and doubt. Finally, we draw a relation table between them. To test the validity of the relationship, a sample spectrum of faulty electrical equipment’s insulating gas is analyzed based on the relationship established. The analysis results match the fact. In this research, infrared spectrum technology is used to analyze the SF6’s ramifications. During the experiments, gas chromatograph technology is used as assistant analysis. So we need a method to deal with the spectrum obtained. This paper proposes a method based on continuous wavelet transform which achieves smoothing, baseline correction and peak finding at the same time. The baseline’s function of original signal is monotone and linear, so after wavelet transform, there is no information of baseline in the wavelet coefficients. What we should consider is the coefficients. Firstly, remove the noise in the coefficients based on Liapunov Exponent. Then, find the ridge mentioned in this paper. The ridge’s position is the peak’s position. The proposed method further simplifies the peak detection procedure. Besides, this paper improves the question of uncertainty threshold in the modulus maximum algorithm based on continuous wavelet transform.Data processing is a crucial step. In the fault diagnosis and precaution strategy, 13 ramifications are considered. There are too many considered datas,so a way how to process these datas in order to reduce the data dimension is needed.. For this goal, both principal component analysis and factor analysis are applied. After comparision, it proved that principal component analysis has a better performance in the data dimension reduction. The principal component analysis reduces the 13 dimensions data into 6 dimensions. And the 6 dimensions data contain 87.58% information of the original data.The reality of fault diagnosis and precaution system which based on identification of SF6’s ramifications is a learning model. In this paper, both radial basis function neural network theory and decision tree theory are applied to build the model. The simulation results show that both the two models have excellent ability of predicting. But penetrate analysis shows that the model based on decision tree has better performance to deal with arc discharge; the model based on RBFN has better performance to deal with spark discharge and corona discharge. In order to increase the predicting accuracy, it proposed to combine the two models. The rate of reject identification (RRI) is proposed in the new model. The simulating result shows that when the RRI is 19.51%, the predicting accuracy is 100%. The sample rejected identification can be diagnosed by experts. |