Situaiton Assessment Using Probabilistic Graphical Models

Situation assessment is the second-level of the multi-source data fusion system. It extracts and analyzes the current battlefield platform's information which comes from the first-level fusion, and then gives the enemy's situation, which will be passed to the threat assessment system. Situation assessment mainly contains three parts: event detection, target clustering and situation forecast. In this paper, we study the related theory of situation assessment deeply.The process of aggregating the platform is called target clustering, which could provide the information of force level for the military decision-making. Traditional target clustering generally utilizes hierarchical clustering method or K-means algorithm. Owing to battlefield diversity, it is particularly difficult to select a termination condition in the hierarchical clustering method. At the same time, the K-means algorithm falls into a local minimum easily. In order to solve these problems, we propose a target clustering method based on HSC algorithm. This method utilizes rigorous mathematical transformation to substitute the non-differentiable objective function with a differentiable function which has single extreme point. Right now, it can be solved by optimization method perfectly.In this paper, we study the situation forecast based on probabilistic graphical model in depth. This method firstly transforms the probabilistic graphical model to a clique tree, which requires nodes elimination sequence to triangulate the probabilistic graphical model, and secondly utilizes the clique tree propagation algorithm to inference. In the triangulation, the traditional method has to input the sequence of nodes elimination manually. And, this method is inefficiency and obtains a non-optimal nodes elimination sequence usually. To solve these problems, we propose a genetic algorithm to address the sequence. This method has strong global searching capability, and could find the optimal nodes elimination sequence correctly.In addition, we conducted a large number of experiments to compare the above methods with other related methods. Experimental results show that the proposed methods could obtain the results more accurately and efficiently.