基于混合学习算法的材料计算数据误差估计  

Error Estimation for Material Simulation Data Based on Hybrid Learning Algorithm

作　　者：王娟[1,2] 杨小渝 王宗国[1] 任杰[1,2] 赵旭山[1] Wang Juan1,2, Yang Xiaoyu , Wang Zongguo1, Ren Jie1,2, Zhao Xushan1(1. Computer Network Information Center, Chinese Academy of Sciences, Beijing 100190, China; 2. University of Chinese Academy of Sciences, Beijing 100049, China)

机构地区：[1]中国科学院计算机网络信息中心,北京100190 [2]中国科学院大学,北京100049

出　　处：《系统仿真学报》2018年第10期3843-3852,3860,共11页Journal of System Simulation

基　　金：国家自然科学基金(61472394;11547177)

摘　　要：鉴于低误差的基于密度泛函理论的材料计算模拟数据在新材料设计与发现中的重要性,提出一种基于BP神经网络和粒子群优化(particleswarmoptimization,PSO)混合学习算法的材料计算数据误差估计建模方法。结合PSO的全局搜索和BP算法的局部搜索优点,将不含速度项的简化PSO算法和BP算法相结合,提出一种PSO和BP混合的学习方法(tsPSO-BP),用于训练材料计算模拟数据误差估计神经网络模型,并以立方晶系二元合金弹性常数计算模拟数据误差估计为应用实例。应用结果表明ts PSO-BP训练后的弹性常数计算模拟误差预测神经网络模型预测的C_(11),C_(12)和C_(44)的计算模拟数据误差的准确率分别达到88.19%,87.83%和88.26%。In order to obtain high quality material simulation data from Density Functional Theory material calculation software package, a modeling method based on BP neural network was proposed to build model estimating the error of material simulation data. A novel hybrid algorithm combining simple particle swarm optimization algorithm that excludes speed item with BP algorithm, also referred to tsPSO-BP, was proposed to optimize the connection weights of the BP neural network. The hybrid learning algorithm not only makes use of strong global searching ability of the PSO, but also strong local searching ability of the BP algorithm. The BP neural network model was trained using tsPSO-BP on the dataset of experimental and calculation data of elastic constants for binary alloys with cubic crystal system, and the results show that the prediction accuracy of the error of C11, C12 and C44 were 88.19%, 87.83% and 88.26%, respectively.

关 键 词：材料计算模拟数据误差估计 神经网络 粒子群优化 BP算法 混合学习方法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.9[自动化与计算机技术—控制科学与工程]