CAJ | 학술논문

针对带混沌特性的网络流量在线预测，提出一种融合自适应粒子群算法（ APSO）和递推式最小二乘支持向量机回归的流量模型。对流量序列嵌入重构得到多维状态输入矢量，将其作为初始LSSVM的训练样本，其中采用自适应粒子群算法对模型的特征参数、嵌入维数寻优，避免早熟停滞。对于在线预报过程中的吸收样本、删减样本采用核矩阵迭代式求解，动态调整回归机，使得模型具有在线学习能力，由此得APSO-LSSVM在线流量预测模型，并考察网络负荷度与嵌入维数关系。仿真实验表明：该方法能有效预测网络流量，实现较高精度实时流量估计。
침대대혼돈특성적망락류량재선예측，제출일충융합자괄응입자군산법（ APSO）화체추식최소이승지지향량궤회귀적류량모형。대류량서렬감입중구득도다유상태수입시량，장기작위초시LSSVM적훈련양본，기중채용자괄응입자군산법대모형적특정삼수、감입유수심우，피면조숙정체。대우재선예보과정중적흡수양본、산감양본채용핵구진질대식구해，동태조정회귀궤，사득모형구유재선학습능력，유차득APSO-LSSVM재선류량예측모형，병고찰망락부하도여감입유수관계。방진실험표명：해방법능유효예측망락류량，실현교고정도실시류량고계。
For online prediction of network traffic with chaos characteristics , a new network traffic prediction model based on adaptive par-ticle swarm optimisation algorithm (APSO) and recursive least square support vector machine (LSSVM) is proposed.First, the network traf-fic sequence is conducted the embedment and reconstruction to acquire multidimensional states input vectors , and they are used as the initial training sample of LSSVM .The adaptive particle swarm optimisation is used to optimise the characteristic parameters and the embedded di -mensions of the model for avoiding premature stagnation .At last, the iterative expression of kernel matrix is adopted in solution of absorption and deduction of the samples in online prediction process , thus to dynamically adjust the regression machine and have the model possesses on -line learning ability.In this way, the online APSO-LSSVM network traffic prediction model is formed , and then it need to explore the relation-ship between the network load and the dimensions embedded as well .Simulation experiment shows that the APSO-LSSVM model is able to predict network traffic effectively and get higher accuracy in real time traffic estimation .