基于近似模型的大型汽轮发电机定子端部绕组动力响应预测研究

Dynamic response prediction of large turbo-generator stator end winding based on approximate model

针对电磁力作用下大型汽轮发电机定子端部绕组动力响应建模及计算复杂问题,提出一种基于数据驱动的支持向量回归动力响应预测方法。通过少量典型样本构建某600 MW汽轮发电机动力响应近似模型,从而代替复杂耗时的有限元模型预测不同结构参数下的动态性能。以端部绕组鼻端位移峰值作为动力响应的关键指标,首先选取绑环刚度、径向支架刚度以及滑销和径向支架之间的固定约束数目为设计变量,通过正交试验设计获取样本,在ABAQUS软件中分别建立试验样本对应的有限元模型,再对其进行计算获得鼻端位移时程曲线;然后通过遗传算法对支持向量机中的参数寻优构建端部绕组动力响应的近似模型,对比结果显示,该近似模型精度优于基于响应面法和克里金插值法构建的预测模型;最后基于近似模型探讨了设计参数对鼻端位移峰值的影响规律。该方法可用于后续的优化设计以及装备数字孪生系统中电气和力学性能的实时求解与计算。

Here,aiming at the complex problem of modeling and calculation for dynamic response of stator end winding of a large turbogenerator under electromagnetic force,a data-driven support vector regression(SVR) dynamic response prediction method was proposed.An approximate model for dynamic response of a certain 600 MW turbogenerator was established using a small number of typical samples to replace complex and time-consuming finite element model for predicting dynamic performance under different structural parameters.Firstly,the peak displacement of nose end of stator end winding was taken as key index of dynamic response,binding ring stiffness,radial support stiffness and fixed constraint number between sliding pins and radial supports were selected as design variables.Samples were obtained through orthogonal test design.In the software ABAQUS,finite element models corresponding to test samples were established,and then they were used to do calculations,and obtain nose end displacement time history curves.Furthermore,genetic algorithm was used to optimize parameters in SVR,and construct an approximate model for dynamic response of stator end winding.The comparison results showed that the accuracy of this approximate model is superior to those of prediction models based on response surface method and Kriging interpolation method.Finally,effects of design parameters on nose end peak displacement were explored based on this approximate model.It was shown that this method can be used in subsequent optimization designs as well as real-time solving and calculating electrical and mechanical properties of equipment digital twin systems.