【作者】 张燕；

【导师】 方瑞明； 熊军；

【作者基本信息】 华侨大学 ， 工程硕士（专业学位）， 2020， 硕士

【摘要】 如今,随着对电网安全性的要求提高,对电力变压器运行的安全性和稳定性也有了更高的需求。本文针对当前基于油色谱在线监测数据的诊断方法未考虑故障特征气体的时序特性,且普遍存在的阈值确定困难、典型缺陷样本数据匮乏等问题,提出将隐马尔科夫和动态网络标志物模型引入基于油中溶解气体在线监测时序数据的变压器缺陷预警与辨识的方法,为识别变压器的早期缺陷提供理论指导。首先,将隐马尔科夫模型引入变压器的缺陷预警中,利用变压器运行过程中的不同动态特性与待诊断变压器自身油中溶解特征气体浓度变化之间的映射关系,基于待诊断变压器自身的在线监测气体数据进行动态建模,对变压器状态转变的预警信号进行检测,以达到通过变压器自身气体浓度变化对其早期缺陷实时预警的目的。案例分析表明基于隐马尔科夫模型的方法与传统试验室色谱分析手段相比,能更加及时有效的发现变压器早期缺陷。其次,分析了油中溶解气体在线监测数据与变压器内部状态之间的映射关系,通过筛选出当前状态下对变压器状态转换过程影响较大的主要特征气体,建立基于油中溶解气体的变压器当前状态动态网络描述模型,基于临界慢化理论分析了变压器由健康状态向故障状态转换的相变过程中,指出其间存在着一个临界状态;并进一步分析动态网络模型中各特征气体的标准差及气体间相关性的动态变化,提取相变过程中的临界状态点,可对变压器的早期缺陷进行预警和辨识。相比传统阈值比值法,基于动态网络标志物模型的方法可以较为及时地识别变压器内部异常。最后,基于在线监测时序数据,建立待诊断变压器的隐马尔科夫模型和动态网络标志物模型相结合的组合预警模型。先采用隐马尔科夫模型检测变压器可能的“临界点”,再根据“临界点”处筛选的主要特征气体建立相应的动态网络标志物模型,进一步分析动态网络模型的动态变化,验证变压器的运行状态,并对故障进行识别。案例分析结果表明,该组合模型相比单一预警模型,能够较全面地把握变压器运行状态的动态演变过程,提前对未达到注意阈值的过热异常案例进行预警和识别,有助于对变压器的动态发展进行监测分析。本文提出的基于油中溶解气体的变压器缺陷预警与辨识方法,基于待诊断变压器在线监测装置的时序数据,数据获取较为方便,具有一定的推广价值。更多还原

【Abstract】 Nowadays,with the improvement of the safety of power grid,there is a higher demand for the safe and stable operation of power transformers.The current diagnostic method based on on-line monitoring data of oil chromatography fails to consider the temporal characteristics of fault gases.It has difficulty in determining the threshold and lacks typical fault sample data.In this paper,the hidden Markov model and dynamical network marker are introduced into the method of fault warning and identification of transformer based on the on-line monitoring time-series data of dissolved gas in oil is proposed,provide theoretic instruction for early identification of transformer defect.Firstly,the hidden Markov model(HMM)is introduced into transformer defect warning.In this paper,the dynamic modeling is based on the on-line monitoring gas data of the transformer by using the mapping relation between different dynamic characteristics of transformer operation and the variation of dissolved characteristic gas concentration in transformer.The early warning signal of transformer state transition is detected to achieve the purpose of real-time warning of transformer’s early defects through the change of transformer’s gas concentration.The case study shows that the method based on hidden Markov model can detect the early defects of transformer more timely and effectively than the traditional laboratory chromatography.Secondly,the mapping relationship between on-line monitoring system and internal state of transformer was analyzed.The main characteristic gases which have great influence on the transformer state transition process were selected by analyzing the difference between the actual monitoring data and the predicted data.Then,the dynamical network description model of the current state was established.And there is a critical state in the process of transformer transition from healthy state to fault state based on the critical slowing down theory.The dynamic change of the standard deviation and correlation of the characteristic gases in the dynamical network marker was further analyzed to find out the critical point in the phase transition process,and the early defect warning of transformer could be realized.Compared with the traditional threshold and ratio method,the method based on the dynamical network marker can identify the inner abnormality of transformer in time.Finally,based on the on-line monitoring time-series data,the combination model of hidden Markov model and dynamical network marker for transformer is established.First,the hidden Markov model is used to detect the transformer’s possible “critical point”,and then the corresponding dynamical network marker is established according to the main characteristic gas screened at the “critical point”.Then,the dynamic changes of the dynamical network marker are further analyzed to verify the operation state of the transformer and identify the fault.The method was verified by the field cases.The results show that,compared with single early warning model,the combination model can fully grasp the dynamic evolution process of the transformer operation state,and it can make early warning and identification for the cases of abnormal overheating that have not reached the attention threshold.The transformer defect warning and identification method based on dissolved gas in oil in this paper is completely relied on on-line monitoring time-series data of transformer,true and reliable,and have a certain popularization value.更多还原