面向数字孪生应用的变压器温度场有限元降阶建模方法研究

Research on Finite Element Reduced Order Modeling Method of Transformer Temperature Field for Digital Twin Application

针对油浸式变压器瞬态流固耦合温度场求解存在模型自由度高导致耗时长达小时级的缺陷,构建了变压器数字孪生降阶模型,基于该模型可快速求解瞬态温度场。首先建立基于物联网(internet of things,IoT)的变压器数字孪生实现架构,搭建瞬态流固耦合温度场伽辽金有限元全阶模型。其次,提出将本征正交分解(properorthogonal decomposition,POD)与有限元结合建立瞬态温度场降阶模型,并结合实测数据给出数字孪生建模与降阶计算流程。最后,开展实际变压器温升试验确保全阶模型准确性,并应用不同阶数降阶模型快速计算温度场分布,比较各阶模型的计算误差与时间。结果表明:计算值与实测值的温升误差绝对值满足在1.5℃以内;3阶模型计算结果与全阶模型结果符合POD误差规范要求;降阶模型与全阶模型相比其计算时间由小时级降至秒级。研究结果验证了降阶模型的准确性与时效性,可以在保证数字孪生模型求解精度的同时最大限度提高求解效率。

In order to solve the transient fluid solid coupling temperature field of oil immersed transformer,the digital twin reduced order model of transformer is constructed,which can quickly solve the transient temperature field.Firstly,the digital twin implementation architecture of transformer based on internet of things(IoT)is established,and the full order Galerkin finite element model of transient fluid solid coupling temperature field is built.Secondly,a reduced order model of transient temperature field is established by combining proper orthogonal decomposition(POD)with finite element method,and the digital twin modeling and reduced order calculation flow are given based on the measured data.Finally,the actual transformer temperature rise test is carried out to ensure the accuracy of the full order model;meanwhile,the temperature field distribution is quickly calculated by using different order reduced models,and the calculation error and time of each order model are compared.The results show that the temperature rise error between the calculated value and the measured value is with in 1.5℃.The results of the third-order model and the full order model meet the requirements of POD error specification.Compared with the full order model,the computational time of the reduced order model is reduced from hours-level to seconds-level.The research results verify the accuracy and timeliness of the reduced order model,which can ensure the accuracy of the digital twin model and maximize the efficiency of the solution.