Research On Plan Recognition Based On Hierarchical Logical Hidden Markov Model

Intent Recognition is an important kind of cognitive behavior and also an important part of the modeling of combat simulation system.Plan recognition is a kind of complex and important intent recognition problem.Research plan recognition is an important part of human behavior modeling.It has some theoretical research value and is of great significance to improve the credibility of simulation system.Based on Logistic Hidden Markov Model,this thesis presents a model,with related algorithms and algorithm evaluation method to solve the plan recognition problems with complex hierarchical structure and multiple relational data.The main work of the theses is as follows:Firstly,by combining Logical Hidden Markov Model with Hierarchical Markov Model,Hierarchical Logical Hidden Markov Model(Hi LHMM)is proposed to improve the ability of the model to describe complex hierarchical plan;and the forward and backward algorithms,Viterbi algorithm and Baum-Welch algorithm are,respectively,generalized and improved for evaluation,maximum likelihood estimation and parameter estimation.Secondly,in order to reduce the computational complexity and improve the efficiency of recognition algorithm,the method combining with the consistency matching is proposed,where the obtained observation sequence is matched with the plans in the plan library before applying Hi LHMM to reduce the search space of the recognition algorithm.In Environment Driven Hierarchical Logical Hidden Markov Model(ED-Hi LHMM),according to the scenario matching and similarity measure,the scenario activation probability is defined,and the scenario activation probability is taken as the factor that influences the state transition in Hi LHMM,which improved the accuracy of the recognition results.Thirdly,in the experimental part,this thesis analyzes and compares the performance of different algorithms by designing symbolic plans with certain characteristics and domain independence.We mainly compares the performance of different algorithms with different degree of plan timing constraints,plan library size,plan depth.The experimental results show that the Hi LHMM and ED-HiLHMM models can better adapt and solve the plan recognition problems with more complicated plan structure due to the stronger descriptive ability.Then,the thesis uses the Hi LHMM model to recognize the typical tactical mission planning of the UAV.The recognition result is acceptable,which shows that the model is feasible and effective.