Research On Transformer Condition Assessment Method And On-line Monitoring Optimization

The transformer is one of the most expensive and important kind of power transmission and transformation equipment, and once failure, enormous asset and outage loss will come up. On-line monitoring is an effective means to ensure the long-term stable operation of transformers, so it is widely used in Chinese power grid. The assessment result of transformer condition can not only help arrange the maintenance activities but also optimize the allocation and operation of on-line monitoring devices. Based on the summary of previous research results and the investigation of on-line monitoring devices application, the following problems still exists:(1) although various condition assessment models have been developed, none of them is used in practice due to the defects and limitations of the models. Existing diagnosis methods were based on off-line DGA data and failed to make full use of on-line data, neglecting the changing trend of data with time. (2) On-line monitoring devices are power equipment themselves and appropriate allocation plans should be scheduled. Besides, once they have been put into practice, the operation and maintenance activities should be scheduled reasonably.Based on above, relevant researches about transformer condition assessment technology and optimization of on-line monitoring have been carried out. Researchse on condition assessment and fault diagnosis technology are carried out firstly, and then an allocation priority assessment model of transformer on-line monitoring devices is set up based on condition assessment results, besides, dynamic monitoring cycle adjustment strategy of transformer chromatography on-line monitoring is established according to the changing trend of DGA. The research results can help find a feasible way of guaranteeing the safe operation of transformers and making full use of on-line monitoring devices and have important theoretical and application value. The specific research works are as follows:(1) A multilayer transformer condition assessment model is built by taking both fuzziness and randomness in uncertainty into consideration. The framework of assessment is divided into three layers, including sub-system assessment, system assessment and overall assessment. In the first layer, sub-system assessment, the relationship between quantitative index relative deterioration degrees and grades is expressed in normal cloud model. The matter-element cloud model provides the association degrees between the quantitative indices and grades. Combining the indices’association degrees and optimal weights, which are got based on moment estimation theory, association degrees of quantitative system with grades can be calculated; association degrees of qualitative system with grades can be obtained based on experts’experiences. D-S evidential reasoning decision-making model is utilized in the second layer and third layer to assess the condition of systems and overall transformer respectively. Feasibility of the model proposed is verified through field test data, providing transformer condition assessment with a new line of thought.(2) A fault data extraction technology based on on-line dissolved gas data is proposed. Firstly, DGA based fault diagnosis flow is summarized and improved based on on-line data, and then proposing an invalid data rejection method, obvious invalid data is rejected by intercomparison and coarse error data is rejected and improved by least square method. Finally, forecasting technology based fault data extraction method is proposed. One set of field data is utilized to testify the method and the results show that the method proposed could improve the classification accuracy in the early stage of the fault and improve diagnosis sensitivity and accuracy.(3) Multiple feature selection and particle swarm optimization-fast relevance vector machine (RVM) based transformer fault diagnosis model is proposed in this paper. By improving the iteration process, computation time of RVM can be reduced remarkably. This paper summarizes the fault feature extraction methods of transformer dissolved gas data and proposes the deterioration degree processing algorithm. Parameters of kernel functions and fault feature extraction methods are optimized by particle swarm optimization method and’ binary tree’multi-layer classifier model is built. The numerical examples testify:FRVM could decrease calculation time a lot than RVM, and the classification performance after optimizing both the feature selection methods and kernel function parameters is better than just optimizing one. Besides, the model proposed could provide satisfactory classification results. Compared with existing methods, such as IEC three ratio and SVM model, the proposed model has higher classification accuracy.(4) An appropriate way to improve the allocation comprehensive benefits of on-line monitoring devices is a new issue in power industry. A priority assessment model of on-line monitoring devices for transformer is proposed first in this paper. The assessment model consists of device level and system level. The device level is divided into property assessment and operation condition assessment. The details of various assessment methods were described in following sections, including device property assessment based on fuzzy analytic hierarchy process(FAHP), operation condition assessment method based on condition assessment technology and system level assessment method based on risk benefits index. An actual grid is utilized to validate the model and the numerical results testify that FAHP can give a comprehensive consideration of several relevant device properties that affect allocation priority and through synthesizing a number of experts’ evaluations we can obtain result of device property assessment. With risk benefits chosen to be evaluation index, we can consider comprehensively about influences on fault rate and repair time under circumstances with or without on-line monitoring devices installation. Then we can estimate the benefits of on-line monitoring from the view of system level. Assessment model with thorough consideration on device property, operation condition and risk benefits can avoid unilateral results because of neglect of other relevant aspects and give a more comprehensive result.(5) Operation cost of transformer chromatography on-line monitoring device and its monitoring cycle is closely related. How to guarantee the monitoring efficiency and economical efficiency by adjusting monitoring cycle is the key issue in this paper. In this context, a dynamic monitoring cycle adjustment strategy for chromatography on-line monitoring device is proposed. Based on the theoretical analysis of the impact of monitoring cycle on transformer service life, phase space reconstruction is carried out on gas content time series data with short time intervals and optimal time delay obtained is considered as relative optimal monitoring cycle. Then self-adaptive gas content forecasting model is built based on gravitation search algorithm and fast relevance vector machine. The early warning method proposed according to forecasting results along with other monitoring information is used to adjust monitoring cycle. The numerical results testify that:the parameters of kernel functions could be well optimized through gravitation search algorithm and higher forecasting precision could be obtained with reconstructed data. Besides, compared with the methods based on gas production rate alert value and gas content alert value, the warning method proposed iny this paper is more suitable for on-line data with short time intervals and low content, and can detect abnormal situation effectively.