摘要
为了估计岩土材料的力学模型参数,讨论了基于经典BP神经网络的参数反演方法的基本框架和算法。分析了经典BP神经网络所存在的某些缺陷及其改进方法。结合两个基于深度学习神经网络估计锂离子电池电化学模型参数的例子,介绍了基于深度学习神经网络估计模型参数的基本思路。讨论了深度学习神经网络超参数确定方法,分析了提高深度学习神经网络学习效率和泛化能力的某些行之有效的策略。
In order to estimate mechanical model parameters of materials, the basic scheme and algorithm of parameter inversion based on classical BP neural network are discussed. Some drawbacks and improvement methods for classical BP neural network are analyzed. Combined two examples of estimating model parameters of electric-chemical model of Li-ion cells, the basic thinking of esti-mating model parameter procedure based on deep learning neural network is introduced. How to determine hyper parameters of deep learning neural network is discussed. How to improve learning efficiency and generalization ability of deep learning neural network is developed to es-timate model parameters of materials.
出处
《人工智能与机器人研究》
2020年第2期100-109,共10页
Artificial Intelligence and Robotics Research
基金
国家重点基础研究发展计划“973”项目(2015CB057804)
国家自然科学基金资助项目(11572079)。
