Focusing on the problems of low prediction accuracy and slow convergence speed of network security situation prediction model, a prediction method based on Improved Particle Swarm Optimization Extreme Learning Machine (IPSO-ELM) algorithm was proposed. Firstly, the inertia weight and learning factor of Particle Swarm Optimization (PSO) algorithm were improved to realize the adaptive adjustment of the two parameters with the increase of iteration times, so that PSO had a large search range and fast speed at the initial stage, strong convergence ability and stability at the later stage. Secondly, aiming at the problem that PSO is easy to fall into the local optimum, a particle stagnation disturbance strategy was proposed to re-guide the particles trapped in the local optimum to the global optimal flying. The Improved Particle Swarm Optimization (IPSO) algorithm obtained in this way ensured the global optimization ability and enhanced the local search ability. Finally, IPSO was combined with Extreme Learning Machine (ELM) to optimize the initial weights and thresholds of ELM. Compared with ELM, the ELM combining with IPSO had the prediction accuracy improved by 44.25%. Experimental results show that, compared with PSO-ELM, IPSO-ELM has the fitting degree of prediction results reached 0.99, and the convergence rate increased by 47.43%. The proposed algorithm is obviously better than the comparison algorithms in the prediction accuracy and convergence speed.

Variable Precision Rough Set (VPRS) can effectively process the noise data, but its portability is not good. Aiming at this problem, an improved vulnerability threat assessment model was proposed by introducing the threshold parameter α. First of all, an assessment decision table was created according to characteristic properties of vulnerability. Then, k-means algorithm was used to discretize the continuous attributes. Next, by adjusting the value of β and α, the attributes were reducted and the probabilistic decision rules were concluded. Finally, the test data was matched with the rule base and the vulnerability assessment results were obtained. The simulation results show that the accuracy of the proposed method is 19.66 percentage points higher than that of VPRS method, and the transplantability is enhanced.