一种自相似业务量预测的卡尔曼滤波算法

A Kalman Filtering Algorithm for Self-Similar Traffic Prediction

针对网络拥塞控制中不能准确预测自相似业务量的问题,提出了一种噪声在线估计卡尔曼滤波(NOEKF)算法.NOEKF算法不依赖于业务源反馈信息,通过观测节点处当前和过去时刻的业务量来预测下一时刻的业务量,并建立了业务量的卡尔曼滤波状态方程和观测方程,给出了递推形式的状态向量最佳估计形式.考虑到未知状态方程和观测方程噪声的统计特性,采用在线估值法,并引入遗忘因子对噪声的统计特性进行估计.NOEKF算法预测准确、偏差小.仿真结果表明,与经典卡尔曼滤波算法和时间序列预测方法比较,NOEKF算法能够更精确地预测自相似业务量,预测误差可降低60%以上.

A noise on-line estimation Kalman filtering （NOEKF） algorithm is presented to deal with the inaccurate self-similar traffic prediction in network congestion control. The proposed algorithm is independent of the feedback information from traffic sources, and predicts the traffic through observing both the current and previous traffics in a node. Both the state equation and the observation equation are established, and then an optimal recursive formula for the estimation of the state vector is given. By taking the unknown noise statistics of both the state equation and the observation equation into account, an on-line estimation method with forgetting factor is used to estimate the noise statistics. Comparisons with existing algorithms show that the NOEKF algorithm has the advantages of high accuracy and minor prediction error. Simulation results show that the NOEKF algorithm predicts self-similar traffic more accurately than the classical Kalman filtering and time series prediction algorithms do, and that the prediction error is reduced by more than 60%.