摘要： 在基于效用的行为选择模型基础上对多Agent系统中个性建模问题进行研究。利用人工神经网络能够学习到人类难以理解的目标函数的特点，结合心理学中个性的五因素模型建立Agent个性神经网络，通过不同参数反映个性对效用变化的影响方式，具有更强的个性表征能力。设计梯度下降的学习算法训练Agent相应的个性神经网络。实验验证了该模型刻画Agent个性的有效性。

关键词: 智能体, 行为选择, 神经网络

Abstract: Based on utility-based action selection model, this paper researches on personality modeling in multi-Agent system. That artificial neural network can learn target function which is hard to comprehend draws attention. So an individualized neural network is built according to five-factor model in psychology. Different setting on parameters reflects the way that personality influences utility of actions. This model has stronger ability to describe personality. A gradient descending learning algorithm is proposed to train the individualized neural network. The model is validated on personality exhibition during action selection.

Key words: Agent, action selection, neural network

中图分类号: 

• TP18