Research On Capacitive Equipment Insulation On-line Monitoring And Intelligent Fault Diagnosis

With the rapid development of power industry, the structure of power grid continues toexpand, and has become increasingly complicated. However, power system accident causedby electrical equipment fault load to more and more serious loss. Capacitive equipment isimportant power transmission and transformation unit, so the on-line monitoring and faultdiagnosis of their insulation is of great practical for realization of state maintenance. On thebasis of summarizing the electrical equipment insulation on-line monitoring and faultdiagnosis technology at home and abroad, stability and accuracy of monitoring is in-depthresearched and discussed in three aspects of signal frequency extraction, dielectric lossmeasurement, insulation state and fault diagnosis.Based on analysis of spectrum leakage and picket fence effect in DFT method, aself-adaptive complete period sampling algorithm based on DFT is put forward to reduceserious phase error caused by incorrect truncation of sampled signal period. The phaseangle difference of two adjacent signal periods is treated as loop control condition in thismethod. When it is less than pre-established value, the period is found. If the phase angledifference is not less than the pre-established value beyond number of cycles, procedurewill sample again under normal sampling frequency or changed one by hardwareadjustment, and calculate signal frequency again. Sampling frequency is also deduced oncondition that algorithm error has little effect on the dielectric loss angle measurement. Thesimulation results show that this method has better adaptability and flexibility.This paper introduces wavelet denoising method having time-frequency domain aspre-filter of sampling signal against the main factor interfering with calculation accuracy offrequency - Gaussian white noise on the monitoring scene. A comprehensive analysis ofinfluence of denoising factors, such as wavelet functions and level of decomposition, tofrequency calculation is done. On this basis, frequency algorithm based on optimal waveletdenoising under assessment of signal condition is presented, and pattern recognition fromsignal condition to optimal mode of wavelet denoising is realized by BP neural network.This paper also describes adaptive noise cancellation technology in detail. For overcomingthe non-linear-related problem between reference signal and noise, BP neural network isapplied to construct its filter that can learn transmission characteristics of on-site noise byreal-time BP neural network training. Therefore, frequency algorithm based on BP neural network non-linear adaptive filtering is proposed. Numerical simulation shows that twomethods can effectively extract signal frequency, and errors are lower than 0.01Hz. Thispaper also compares the wavelet denoising method and adaptive noise cancellation method,and acquires their applied conditions on the scene.Synthetical relative method including synchronously measuring idea is proposed. Itselects the same phases of all measured equipments connected to same bus, and thencompares the phase angle difference between leakage currents of insulation of theseequipments, called as relative dielectric loss angle. Therefore, it needn't regard voltage onsecondaryside of PT as reference voltage. The method incorporates all of equipments in themeasuring range as an organic whole, and considered the insulation condition of singleequipment from the global view, so it can effectively eliminate similar disturbance andimprove stabilityof the monitoring data, and because of synchronyof signal acquirement ofmultiple equipments, can avoid the inconsistent measurement background caused bysudden transient interference. The corrective measure on symmetrical transform along thelinear part of insulation factory aging curve is also put forward to weaken oscillation ofrelative dielectric loss angle led by the local converse changes of loss angle. Theapplication of the method on the scene is satisfied.Fuzzy diagnosis theory is systematically analyzed in this paper. Regarding relativedielectric loss as feature space of input, diagnostic model based on the reasoning of fuzzyrules for the insulation state is proposed under the guidance of collection of practicalexperience, and its diagnostic results is further used for study of insulation state trend.Diagnostic model based on fuzzy pattern recognition for insulation state trend is built. Theneural network method has also been applied for diagnosis of insulation state and the mainreason of insulation fault. Accordingly, BP and PNN neural network model is established.The model of main insulation fault also can be combined with fuzzy diagnosis to realizecomplete recognition from insulation state to type of fault. The simulation results show thatthese models have better practical value. Integration of fuzzy theory and neural network indiagnosis is further explored.