关键词: 吞吐率预测, 接入点, 参数选择, 特征选择, nu-支持向量回归, 并行混合粒子群优化

Abstract:

|Access point is the key device connecting wired and wireless facilities, its performance information is crucial for package routing, bandwidth allocation and management of quality of service parameter. This paper addresses the problem of generating multistepahead throughput prediction for access point. A modeling strategy is introduced based on NuSVR（Nu-Support Vector Regression), and a PHPSO（Parallel Hybrid Particle Swarm Optimization) algorithm is proposed, for the purpose of combinational optimization to prediction model, including feature selection and hyperparameter selection. The evaluation results have shown that NuSVR model can achieve higher accuracy in throughput prediction of multistepahead, and its performance can be remarkably improved by PHPSO algorithm with fast convergence rate.

Key words: throughput prediction, access point, hyperparameter selection, feature selection, nu-support vector regression（Nu-SVR) parallel hybrid particle swarm optimization（PH-PSO）