Particle Swarm Optimization in Multi-Agent System for the Intelligent Generation of Test Papers

Agent-oriented design is one of the most active areas in the field of deployment of web-based distance education, and test is a popular measurement tool of learners' knowledge in order to verify the learner's level of understanding and select corresponding educational strategy. In this paper, an innovative approach to seamless integration of the particle swarm optimization (PSO) and multi-agent system (MAS) is proposed. In order to generate a test paper automatically, a modified genetic particle swarm optimization (GPSO) is presented, in which the values of parameters will be decreased linearly with the number of iterations for improving the late convergence rate. For the implementation of GPSO based on multi-agent system, a core agents TPAgent(TPA) is provided to undertake the operations of GPSO and will control the evolution operations of each generation of population. To keep communication between different nodes at a minimum cost, fitness evaluation tasks are implemented by the TPAgents at local nodes, only the local minimum fitness and the corresponding best particle are sent to center node so as to get the global best particle in the parallel computing environment. For avoiding the prematurity, the global best particle will be dispatched to remote node randomly. Based on the JADE, a prototype system is setup, and the simulation results show that the proposed approach is feasible and robust.