蚁群优化支持向量机在变压器故障诊断中的应用

发布时间：2018-04-17 00:41

本文选题：变压器 + 溶解气体分析　；参考：《华北电力大学》2014年硕士论文

【摘要】：油色谱在线监测技术对于发现变压器早期潜伏性故障具有重要的实际价值。基于油中溶解气体比值的传统诊断方法具有简单、实用的特点,但诊断准确率不高。本文在分析传统比值诊断方法以及常用智能诊断方法不足的基础上,将蚁群优化支持向量机应用于变压器故障诊断。主要内容如下：收集了517组变压器事故前油中溶解气体数据,并采用大卫三角法,四比值法以及改良三比值诊断法对其进行诊断,通过对诊断结果的分析,论证了传统诊断方法的不足。由于不同种类气体含量数值差距较大,对数据采用标准化的数据变换方法,并通过对比数据变换前后诊断正确率的变化,分析了数据变换对诊断精度的影响。针对目前色谱仪检测的气体种类以及检测精度的差异,采用相关性分析方法与距离可分性判据对特征参量进行选择,并分析了冗余参量及不相关参量对诊断精度的影响。支持向量机诊断的精度与参数的选取密切相关,鉴于蚁群算法良好的优化性能,提出了蚁群优化支持向量机的方法。为证明该方法的可行性及优越性,将其与广泛使用的遗传优化算法进行理论分析与对比,并对蚁群系统的设计进行了详细地介绍。基于现场数据建立了基于蚁群优化支持向量机的变压器故障诊断模型,重点对比了本文方法与遗传算法优化支持向量机方法的诊断效果。考虑到优化算法的不确定性对诊断结果的影响,采用对测试数据进行多次诊断并比较结果,增加了算法对比结果的说服力。最后,将诊断结果与改良三比值的诊断效果进行对比,充分证明了蚁群优化支持向量机故障诊断方法的优越性。基于现场数据建立了基于支持向量机回归算法和时间序列分析的变压器故障预测模型,重点对比了本文方法与灰色预测方法的诊断效果。实例仿真表明,基于蚁群优化支持向量回归的变压器故障预测模型能很好地应用于变压器油中溶解气体含量的预测,并且预测效果优于灰色预测模型,具有较好的泛化能力。
[Abstract]:On-line monitoring of oil chromatography is of great practical value in detecting early latent faults of transformers.The traditional diagnosis method based on dissolved gas ratio in oil is simple and practical, but the diagnostic accuracy is not high.On the basis of analyzing the deficiency of traditional ratio diagnosis method and common intelligent diagnosis method, this paper applies ant colony optimization support vector machine to transformer fault diagnosis.The main contents are as follows:517 sets of dissolved gas data in transformer oil before accident were collected and diagnosed by using David's triangle method, four-ratio method and modified three-ratio diagnostic method. Through the analysis of the diagnostic results, the shortcomings of the traditional diagnostic methods were demonstrated.Because of the large difference in numerical value of different kinds of gases, a standardized data transformation method is used to analyze the effect of data transformation on diagnostic accuracy by comparing the changes of diagnostic accuracy before and after data transformation.Aiming at the difference of gas types and detection accuracy of chromatograph at present, the correlation analysis method and distance separability criterion are used to select the characteristic parameters, and the effects of redundant and irrelevant parameters on the diagnostic accuracy are analyzed.The accuracy of SVM diagnosis is closely related to the selection of parameters. In view of the good performance of ant colony algorithm, a method of ant colony optimization support vector machine is proposed.In order to prove the feasibility and superiority of this method, it is theoretically analyzed and compared with the widely used genetic optimization algorithm, and the design of ant colony system is introduced in detail.Based on field data, a transformer fault diagnosis model based on ant colony optimization support vector machine is established.Considering the influence of the uncertainty of the optimization algorithm on the diagnosis results, the test data are diagnosed several times and the results are compared, which increases the persuasiveness of the comparison results of the algorithm.Finally, the result of diagnosis is compared with that of the improved three ratio, which fully proves the superiority of the ant colony optimization support vector machine (SVM) method for fault diagnosis.Based on field data, a transformer fault prediction model based on support vector machine regression algorithm and time series analysis is established, and the diagnosis results of this method and grey prediction method are compared.The simulation results show that the transformer fault prediction model based on ant colony optimization support vector regression can be applied to predict dissolved gas content in transformer oil, and the prediction effect is better than that of grey prediction model, and it has better generalization ability.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM407

【参考文献】