Research Of The Accuracy Of Oil Chromatography Online Monitoring Data On 220kV Transformer

Online monitoring and fault diagnosis of transformer is the premise and foundation to achieve its state overhaul. The concentration detection of feature gas in the O il is one of the main test project of the transformer and the diagnostic bases, but there will be internal and external module noise inevitably, and the oil chromatogram fault diagnosis method currently Power Authority used is too old,the accuracy and credibility of oil chromatography o nline monitoring results in the practical application are not high. In order to establish a method using data processing means to enhance the accuracy without changing oil chromatography monitoring device hardware. firstly by using the method on wavelet transform simulation to simulate the noise filtering of oil chromatography and the discrimination of its overlapping peaks. After comparing the results obtained a valid set parameters of wavelet noise filtering and proposed a "symmetrical division method" to achieve the analysis of overlapping peaks using lifting wavelet; then use BP neural network to predict the faults of the transformer with using the genetic algorithm to optimize its model parameters and enhance its computing speed.Finally, the verification to testing samples shows the effectiveness of the method,it has certain theoretical significance and practical value.The main findings and research achievements are as follows:1. According to the characteristics of oil chromatogram concentration graph of online monitoring module,using mathematical model Tsaill of stronger generalization to simulate the chromatogram of transfor mer oil and its noise filtering process.with wavelet transform.By comparing peak height error、peak position error、peak area error when choosing different wavelets、decomposition levels、 threshold rules、threshold functions we obtained wavelet noise filtering parameters more suitable for chromatography and achieve a good noise filtering effect.2. Due to the performance degradation for system’s too long operation and the interference of the external environment affect, the chromatogram peaks may overlap.By studying characteristics of overlapping chromatographic peaks, using mathematical models Tsaill to model it.According to the characteristics of lifting wavelet we proposed a "symmetrical division method" to deal with overlapping peaks. The results showed that the range of overlapping peak’s position error after peak separation meets the requirements. And use a problem chromatography from 220 kV AN’Bian transformer substation to test and verify,we find that error between characteristic gas concentration after calculation and the concentration the offline experiment test is small.3. Put forward transformer oil chromatography monitoring data processing methods with the core of the BP neural network algorithm. Put the ratios of the three groups with hydrogen, methane, ethane, acetylene, ethylene as input based on three-ratio method and the transformer fault type as output, use historical data to training and form the model, so as to predict the type of fault within the range of the the training sample, and then compare the results just predicted with the actual fault type after the power test confirmed to determine whether the transformer insulation defect already exists.4. To enhance accuracy and precision of the BP neural network,proposed a optimization method on BP neural network model parameters based on genetic optimization algorithm based, and verify its effectiveness by testing samples.