【作者】 高明；

【导师】 孙秋野； 魏辉；

【作者基本信息】 东北大学 ， 电气工程（专业学位）， 2020， 硕士

【摘要】 电力变压器为现代电力系统提供电能传输的中转与枢纽,其运行情况直接影响着发电站的电能外送以及终端用户的用电质量,重要性不言而喻。不断提升对电力变压器的故障预测研究水平,有助于及时发现变压器的缺陷与潜在故障点,方便运维检修人员提前安排检修计划以及制定周密的防护措施,保证变压器的稳定运行以及电网安全。现阶段,国内外许多专家学者都积极开展对电力变压器的故障预测研究,但是预测的效果还有一定的提升空间。因此,积极开展对电力变压器故障预测的研究,是一个十分重要且有价值的研究课题。本文主要基于油中溶解气体的含量指标进行变压器故障的预测和诊断,由于溶解气体含量不受外部电磁场因素的干扰,因此该指标对判断变压器故障类型有很强的指导意义,是及时查明变压器缺陷和潜伏性故障的重要方法。本文主要研究工作如下:首先,构建DGA(Dissolved gas analysis,溶解气体分析)组合预测模型。本文通过获取变压器油中溶解气体含量的数据信息,分别使用支持向量机、BP(Back Propagation,径向基函数)神经网络与灰色模型三种单一预测模型,对七种特征气体含量进行预测。使用改进粒子群算法确定三种单一模型的权重,通过计算得到的最优权重,形成组合预测结果。通过分析对比,可以看出组合预测效果明显优于单一的预测方法,能较好地反映特征气体的变化趋势。其次,结合实例进行变压器故障诊断研究。以预测得到的油中溶解气体数据为基础,重建新的故障特征集合,进行主成分分析提取新的特征量,优化核函数的参数选择。建立基于主成分析与参数优化支持向量机(support vector machine,SVM)的变压器故障诊断模型,通过现场实例进行对比分析,证明该模型具备一定的准确性、实用性与有效性。最后,探索变压器故障预测与诊断的工程应用。结合组合预测模型与变压器故障诊断模型,创新构建变压器故障预测诊断模型,并设计了一款可应用至实际项目的变压器故障预测诊断软件。更多还原

【Abstract】 In modern electric power system,the significance of power transformers is obvious.As a transfer station and hub for power transmission,transformers affect the electricity output and power quality directly.Continuously improving the level of research on fault prediction,which will help find out defects and potential failure of power transformers in time,work out maintenance plans and develop protective measures.Then,it will ensure the security and stable operation of power systems.At this stage,both domestic and foreign experts and scholars are carrying out researches on fault prediction and diagnosis of power transformers actively.However,there is still a lot of room for improvement.Therefore,it is still very important and valuable to conduct this kind of studies.In this paper,the research about fault prediction and diagnosis is based on the index of dissolved gas content in transformer oil.As it is free from external electromagnetic interference,the content of dissolved gas will help determine the fault types,and which is an important method to identify transformer defects and latent faults.The research contents are as follows:Firstly,a combined forecasting model based on DGA(dissolved gas analysis)in transformer oil is founded.In this paper,the author obtained the data of the dissolved gas content in transformer oil at first,and then used three single prediction models respectively,which were support vector machines,BP(Back Propagation)neural network and gray model,to get the prediction results based on seven kinds of characteristic gases.By using the improved particle swarm algorithm to determine the weight of the three single models,and with the obtained optimal weights,a combined prediction result will be formed.Through analyzing the final results,it proves that the effect of combined forecasting prediction is obviously better than any single prediction method.The combined prediction model in this paper will definitely reflect the change trend of the characteristic gas better.Secondly,a study with examples on transformer fault diagnosis is conducted in this thesis.Based on the predicted data of dissolved gas,we reconstruct a new set of fault features,distill new characteristic quantities by analysis on principal component,and then optimize the parameters selection of the kernel function.Therefore,a transformer fault diagnosis model based on main component analysis and parameter optimization of SVM(support vector machine)is established.By comparison and analysis of examples,it will verify the accuracy,practicability and effectiveness of this model.Finally,we explore the engineering application on fault prediction and diagnosis of power transformer.Combining the combined model and the transformer fault diagnosis model,a new transformer fault prediction and diagnosis model is formed.Based on the foregoing theories,a transformer fault prediction and diagnosis software for practical projects is constructed.更多还原