Rethinking validation: Efficient search of the space of parameters for an agent-based model

摘要本篇文章描述了如何将敏感度分析用作模型验证和证实的一个部分。敏感度分析指出了未来数据验证流程应该关注的地方, 以及哪些数据输入可用作模型降阶。我们比较了两种方法, 一种使用系统参数值的变化, 另一种使用优化算法以便更有效地搜寻参数空间。我们对在业务网络范围内探索创新出现的基于主体建模展开了分析, 成功的创新被视为网络范围内知识和财政资源的增加。两种敏感度分析方法无论在时间效率还是在提供的信息类型方面都各不相同。当中, 系统个体敏感度分析协助我们确定哪些输入对结果带来实质性影响并为模型简化提供解决方案; 而优化搜索则为网络资源如何达到更高层次的创新提供见解。遗传算法确定了某些参数值在基于主体建模中就产生不同的结果。

This paper shows how sensitivity analysis can be used as part of model verification and validation Sensitivity analysis provides insights on where future data validation processes should focus and which inputs may be considered for model reduction. We compared two approaches, one using a systematic variation of parameter values, another using an optimised algorithm to make more efficient the search of their space. Analysis was conducted on an agent-based model that explores the emergence of innovation within business networks, where successful innovation is considered an increase in knowledge and financial resources within the network. The two sensitivity analysis approaches differed both on their time efficiency and on the type of information provided. While the systematic individual sensitivity analysis assisted us in identifying inputs with substantial impact upon the results and suggest solutions for model simplification, the optimised search provided insights on the network resources likely to achieve higher levels of innovation. Genetic algorithms found parameter values that produced different results in the agent-based model.