摘要
研究Java规则引擎和机器学习在Robocode坦克机器人战斗仿真引擎中的应用。探讨利用Jess规则引擎对Robocode坦克机器人决策规则库部分进行开发与维护,满足仿真环境实时性和策略易扩展性的要求。同时为了提高坦克机器人的在线自适应和自学习能力,结合机器学习方法进行角色训练,利用人工神经预测网络对产生式系统中瞄准规则的火炮瞄准角度值进行优化。结合遗传算法与产生式系统,设计混合的移动策略选择器,并对神经网络预测瞄准和基于遗传算法的移动策略进行了实验,给出了实验结果。
Based on Java Rule Engine and machine learning, a new method to construct Robocode decision-making system was presented. This method developed and maintained the decision-making system by Jess, making the decision-making system more real-time reactive and extendible. At the meantime, this method used machine learning to train tank fighters, enhancing the online adaptive and self-learning ability. To complete the job, a hybrid moving action selector was proposed by an integration of genetic algorithm and production system, and a neural network was chosen to optimize the firing angle of aiming rules. The results in the experiment show the effectiveness of aiming system and moving system.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2006年第z2期912-915,共4页
Journal of System Simulation