Abstract: The current researches of sequential decision-making on ambient intelligence mainly are focused on the problem of Agents interaction decision-making over uncertain context,and solution for perturbation context is not involved.For this problem,the Agent interactive decision-making mechanism was proposed.The entity-spatio-temporal contexts based on the modified context ontology was acquired and calculated,and then the semantic-based metacognitive loop was used to detect and evaluate perturbation context so as to feedback to user-serving Agent.Ultimately,experiments in a smart home environment showed that the proposed method improved the accuracy of decision-making by more than 10% on the basis of several representative machine learning processing methods,while the response time increased by less than 5%,which achieved the expansion in the application field and enhanced the practicability.

Key words: Agent, sequential decision-making, ambient intelligence, perturbation context, spatio-temporal context, metacognitive loop

摘要： 针对目前在环境智能方面的序贯决策研究成果主要集中于不确定环境下的多智能体(Agent)交互决策问题,而未涉及到Agent在异常情景下对于该问题的解决思路,提出一种异常情景中Agent交互决策机制。首先基于改进的情景本体对情景中Agent所观察的实体进行“时—空”状态的获取和计算;其次,结合元认知环结构的语义推理算法对异常情景进行检测和评估,并反馈于Agent,最终做出符合当前情景下用户需求的动作或反应。经过在智能家居环境中的实验验证,在原有几种具有代表性的机器学习处理方法基础上,所提方法在其决策精确性上平均提高10%以上,响应时间则增加5%左右,且实现了在应用领域上的拓展,增强了实用性。

关键词: 智能体, 序贯决策, 环境智能, 异常情景, 情景本体, &ldquo, 时&mdash, 空&rdquo, 状态, 元认知环