电力变压器绕组状态在线监测关键问题研究

When vibration analysis method is applied to on-line monitor the winding conditions of power transformer, it is important to carry out the theoretical researches of winding vibration characteristics and to extract the features of vibration signal of power transformer under sudden short-circuit impulse. First, the simulated model of power transformer is established under the environment of electromagnetic transient simulation software EMTP and finite element analysis software ANSYS to obtain the distributions of winding electrodynamic force. Then the winding axial vibration multi-freedom model is constructed with the nonlinear effects of insulated block considered based on the previous calculated results of dynamic electrodynamic force to obtain the motion laws of transformer winding. Meanwhile, the three-dimension finine element model of winding is bulit to study the winding vibration varied patterns with preset typical winding deformation or looseness faults. Then the mathematics morphology and multi-fractral analysis methods are selected in time domain and the local mean decomposition method is applied in frequency domain to analysize the vibration signals of power transformer under the sudden short-circuit impulse experiment. Consequently, the time and frequency domain indexes are defined and calculated to describe the changing tendency of vibration characteristics of transformer winding in the evolutionay process, where the winding are varied from normal conditions to damaged coniditions. Furthemore, the clustering analysis method based on the grid and density is proposed to deal with the large amount of data of voltage, curent and vibration provided by the on-line monitoring system of power transformer. Meanwhile, the improved optimized general regressin neural network is made and trained to predict the vibration signal when the voltage, current and power factor are selected as the inputs of neural network and the vibration signals are selected as the output of neural network. Finally,the layer fuzzy comprehensive evaluation method is presented and applied to evaluate the winding condition of power transfomer,which could provide the foundation for the winding condition monitoring of power transformer based on vibration analysis method.

振动分析法在线监测变压器绕组状态时，突发短路时变压器绕组振动特性的理论研究和振动信号的特征量提取方法是关键。首先使用电磁暂态软件EMTP和有限元软件ANSYS通过建模研究突发短路时的绕组电动力分布，进而分别建立计及动态短路电动力与垫块非线性的变压器绕组轴向振动多自由度模型和可对变压器绕组变形典型故障进行预设的有限元模型，研究突发短路时变压器绕组振动特性规律。然后分别在时域采用数学形态学与多重分形和在时频域采用局域均值分解法对突发短路时的变压器振动信号进行分析，据此定义时频域特征指数描述突发短路时变压器绕组发生变形乃至损坏演变过程中的振动特性。根据变压器在线监测系统提供的海量电压电流和振动监测数据，提出使用网格密度聚类分析法和自回归广义神经网络法对变压器正常运行时振动信号进行监测和预测。最后，应用分层次模糊综合评判法对变压器绕组状态进行评估，为基于振动分析法的变压器绕组状态监测研究奠定基础。

为更加准确的识别变压器绕组状态的变化，确保变压器的安全可靠运行，本研究综合理论建模、信号分析方法及试验研究对突发短路下变压器绕组的振动特性及振动信号的特征量提取方法展开了若干研究。首先建立了基于电磁暂态软件ATP和有限元分析软件ANSYS的变压器电磁场与机械场仿真模型，基于该模型，实现了突发短路下系统短路电流及作用在变压器上的电动力的计算分析，及由短路电动力激励而引起的变压器绕组振动响应的全过程仿真计算。仿真计算结果与试验结果的良好吻合有效验证了计算结果的正确性。此外，突发短路时，由于短路电动力剧增，使得变压器绕组的振动幅值远大于其正常运行时的振动，且振动信号的频谱含量较为丰富。变压器绕组振动的位移峰值随短路电流峰值增大而增大，但增幅逐渐放缓，而变压器绕组振动信号中的谐波含量随电流峰值增大而增幅逐渐增加。同时，突发短路的短路电流值越大，变压器绕组的振动强度越大，同时多次短路冲击会对变压器绕组机械特性的破坏产生累积作用。针对突发短路下变压器振动信号的非平稳和识别特性及短路电动力作用于变压器绕组时的振动响应特性，提出了综合EEMD算法与数学形态学的振动信号波形特征的提取方法，据此实现了突发短路下变压器振动信号的时域分段，进而使用多尺度数学形态学对其进行特征提取，根据振动信号的形态谱波形特征及其谱熵值，提出了基于形态组合滤波与形态谱分析的变压器绕组状态检测方法。所提出的改进LMD算法可更加准确地对突发短路下的振动信号进行分解，进而基于Hilbert谱可实现振动信号频谱特性的准确描述。针对变压器正常运行时振动信号与其运行参数如电压、电流等的关系，所提出的优化广义回归神经网络在具有局部逼近能力及针对非线性模型良好拟合特性的同时，通过设定置信区间的方法对改进广义回归神经网络的输入量进行预处理，有效地抑制了现场干扰信号对测试数据的影响，提高其拟合和预测精度。基于粒子群与人工免疫融合相结合的优化算法具有收敛速度快和局部搜索能力强的特点，可得到较为满意的散布常数的最优值。综合前述研究结果，建立了综合综合变压器稳态运行与突发短路下振动信号的时频域特征的绕组状态评估方法，可对变压器绕组状态进行准确评估。研究结果可为基于振动分析法的变压器绕组状态检测方法提供重要的依据和指导。