基于EBE策略的有限元法温度场神经网络计算

Neural Network Simulation in Temperature Field Based upon EBE Strategies

基于Element By Element(EBE)策略 ,提出了一种反馈网络和前向网络共同组成的多层神经网络以实现热传导有限元分析。输入层与中间层的联接权值为单元刚度矩阵与总体刚度矩阵的联系矩阵A ,中间层与输出层的联接权值为A的转置阵 ,中间超单元联接权值为求解问题的单元刚度矩阵。分析和数值仿真结果表明 ,该神经网络稳定收敛于有限元模型解 ,为有限元分析并行计算提供一种新的方法。

Based upon element by element(EBE) iterative strategies,a feedback multi layer neural model is proposed for the heat conduct problem FEM analysis.Linking weight coefficients LWC between input layer and middle layer is the relation matrix A between the element siffness matrix and collectivity stiffness matrix.LWC between the output layer and the middle layer is the transpositive matrix.LWC of the middle layer super cells is the element stiffness matrix about the problem resolved.Numerical simulation shows that the method is consistent with FEM.This method presents a new way for the parallel analysis by FEM.