摘要
针对有限元法计算量大的缺点,将神经网络的并行结构应用于有限元法的运算过程,提出一种求解三维磁场的有限元神经网络模型,实现了有限元法的并行分布式计算。采用共轭梯度法作为神经网络的训练算法对仿真模型进行分析,得到了磁场强度的矢量图、剖面的等磁动势线图。与有限元方法相比,三维有限元神经网络将复杂运算分布到各个神经网络节点并行求解,简化了有限元法的运算过程,是一种精确快速的磁场分析方法。
In order to solve the problem of finite element method （FEM） with computational cost, this paper proposes a novel finite-element neural network （FENN） embedding a finite-element model in neural network parallel architecture. The novel network carried out parallel distributed calculation of FEM for solving 3D magnetic field while conjugate gradient （CG） method is used as the learning algorithm. The vector plot of magnetic field intensity and magnetic potential contour lines of cross section are obtained. The results indicate that, compared with FEM, FENN divides complex calculation into parallel distributed nodes, simplifies the network of FEM, and is well-behaved magnetic field calculation model with rapidness and accuracy.
出处
《电工技术学报》
EI
CSCD
北大核心
2012年第11期125-132,共8页
Transactions of China Electrotechnical Society
基金
国家973计划
河北省自然科学基金(E2008001258)资助项目
