基于集成学习的复杂仿真模型验证方法

针对复杂仿真模型验证中海量数据的相似性分析问题,提出了一种基于集成学习的仿真模型验证方法。将仿真时间序列与参考时间序列的相似性分析问题转换为相似性等级分类问题,进而利用神经网络、支持向量机、集成学习等机器学习方法,设计了一种集成分类系统对时间序列的相似性等级进行分类。为了增强基分类器的多样性,提出了基于惩罚因子的多样性筛选准则;通过挑选具有最大差异性的基分类器,构造高性能集成分类系统。最后利用相关数据,对所提出的方法进行应用研究,验证了方法的有效性。

Complex simulation model validation method based on ensemble learning

To resolve the similarity analysis of massive datasets, a complex simulation model validation method based on ensemble learning is proposed. The similarity analysis between simulated time series and observed time series is formulated as a similarity degree classification problem. Machine learning techniques, including back propagation neural network, error correcting output coding support vector machine (ECOC-SVM) and ensemble learning, are utilized to construct an ensemble classification system (ECS). Improve the diversity among base classifiers is essential for building high performance ECS. A screening criterion based on the punish factor is designed to choose base classifiers with maximum diversity. Finally, the proposed model validation method is examined with an application example.