摘要
利用神经网络预测横向磁通感应加热中连续运动金属薄板表面涡流分布和温度分布。所采用的两条神经网络中,一条对涡流场分布进行预测,另一条对温度场分布进行预测。在抽取的检测样本中,预测温度分布的相对误差平均值为1.6—3.2％,以神经网络预测结果作为非线性有限元离散方程组迭代求解的初值,比薄板初始温度作为初值的情况节省55.6～67.6％的迭代次数。
With neural network the eddy current and temperature distribution prediction on the continuously moving thin conducting strips in transverse flux induction heating (TFIH) equipment is presented. There were two neural networks used. One was for the eddy current field and the other for the temperature field. The average relative errors of the temperature prediction for the four groups of tested samples are 1.6~3.2%. Taking the prediction results instead of environment temperature as the initial values for solving non-linear finite element discrete equations, the iteration numbers required can be saved 55.6-67.6%.
出处
《中国电机工程学报》
EI
CSCD
北大核心
2004年第8期119-123,共5页
Proceedings of the CSEE
基金
河北省自然科学基金资助项目(602078)
霍英东教育金会青年教师基金资助项目(71055)~~
