微波陶瓷测试技术的神经网络实现  

Neural Network Realization of the Microwave Ceramics Measurement Technique

作　　者：王国庆[1] 吴顺华[1] 王伟[1] 李媛[1] 孙萍[1] 

机构地区：[1]天津大学电子信息工程学院

出　　处：《天津大学学报（自然科学与工程技术版）》2005年第8期735-739,共5页Journal of Tianjin University：Science and Technology

基　　金：国家高技术研究发展计划(863计划)资助项目(2001AA325110).

摘　　要：针对微波陶瓷的开式腔谐振法算法复杂、计算量大、计算速度慢的问题,提出了一种基于神经网络的替代算法.对完全神经网络模型和部分神经网络模型进行了对比,并讨论了网络拓扑结构对在线训练效果、训练时间和离线计算精度的影响.研究表明,采用1-6-6-2结构即单输入双输出的三层部分神经网络模型,计算速度可比传统算法提高1 280倍以上,介电常数的计算误差不超过0.030/0,品质因数的计算误差不超过0.80/0.与传统算法相比,用神经网络实现的算法具有运算量小、速度快等优点,并保持了足够高的计算精度.An artificial neural network（ANN） based algorithm was proposed to substitute for the traditional calculations used in open resonant cavity method for microwave ceramics measurement, in order to simplify calculations and increase the calculation speed. Comparison was made between an all-ANN model and a partial-ANN model; also the effects of ANN topology on the online training performance, training time, and offline calculation precision were discussed. It was revealed that the 1-6-6-2 structured three-layer partial-ANN model,which had one input and two outputs, could increase the calculation speed by more than 1 280 times with errors for dielectric constant and quality factor not exceeding 0.03% and 0. 8%, respectively. Compared with the traditional algorithm, the method based on ANN is less computation consuming, faster and also of sufficient precision.

关 键 词：人工神经网络 回归分析 开式腔谐振法 介电性能 

分 类 号：TP39[自动化与计算机技术—计算机应用技术] TQ174[自动化与计算机技术—计算机科学与技术]