Research And Implementation Of The Probabilistic Plan Recognition System Based On Plan Knowledge Graph

Plan recognition is an important research area in artificial intelligence recently. Plan recognition is the process of inferring the goals/plans from observations of an agent's impoverished and fragmented actions. A plan inferred by a plan recognizer not only can add some unobserved actions that not present in the observation stream but were actually executed, but can predict the future actions as well—infer the actions which the agent may adopt future reasonably. Plan recognition is widely used in many fields. It is significant to develop efficient recognition algorithms.Based on plan knowledge graph, this paper presents a new algorithm which provides a more reasonable explanation for the observed actions, that is the probabilistic plan recognition based on plan knowledge graph (KGPPR). In the algorithm, it represents the plan by using the decomposability and abstraction. It analysis the lack of the plan knowledge graph algorithm and redefines the abstraction, specialization, whole and part relationships between events. It gives the method of computing the plan's probabilistic and the algorithm of the solution graph. Then it gives the design of the algorithm.The structure description of KGPPR is simple and it adopts and/or nodes describing the relationship of plans. It adds the information of ordering constraint and probabilistic so as to explain the observed action in reason. When the plan recognition is executed, it is able to infer the agent's actions by using the probabilistic and the ordering information. It can build the solution graph without extracting the least plan when the probabilistic is computed. So it improves the efficiency of plan recognition. It can handle partial observation of domains, partially ordered plans and so on. The implementation of this algorithm will have a very considerable prospect in computer network security and intrusion detection.Based on the algorithm proposed above, we have implemented a new plan recognition system-KGPPR, which can not only recognize the plan and the goal the agent being performed efficiently, but also indicate the actions that the agent will take. The result of the experiment proves that the KGPPR is reasonable and efficient.