基于强化学习的网络流量非线性多步预测方法  

Network Traffic Nonlinear Multi-step Prediction Method Based on Reinforcement learning

作　　者：廉佐政[1] 王海珍[2] 李大辉[2] 滕艳萍[2] 

机构地区：[1]齐齐哈尔大学计算中心,黑龙江齐齐哈尔161006 [2]齐齐哈尔大学计算机与控制工程学院,黑龙江齐齐哈尔161006

出　　处：《计算机仿真》2013年第6期284-287,共4页Computer Simulation

基　　金：黑龙江省自然科学基金项目(F201218);齐齐哈尔大学青年教师科研启动支持计划项目(2011k-M05);黑龙江省高等学校教改工程项目(JG2012010679)

摘　　要：网络流量具有分形特性,用线性方法来预测非线性的网络流量,预测精度不高。为了提高测性能,提出了网络流量的非线性多步预测问题,利用一种结合分形神经网络、强化学习的非线性多步预测方法,用多重分形性质将网络流量序列分解为短相关序列,设计了一种强化学习神经网络(MRLA)流量预测模型,利用强化学习的Q算法训练BP神经网络,预测尺度系数、计算权值,最后构建MRLA网络进行仿真,预测网络流量。实验分析显示,相对MMLP网络,新预测方法具较好的多步预测性能。Because network traffic has the characteristic of fractal, the accuracy is not high adopting the linear method to predict the non - linear network traffic. In order to improve the performance, the paper studied the problem of nonlinear multi - step prediction, and proposed a nonlinear multi - step prediction method which combines fractal neural network with reinforcement learning. A reinforcement learning neural network（MRLA） traffic prediction model was designed on basis of multi - fractal nature, and thenetwork traffic sequence was broken up into range depend- ence. With the model we trained BP neural network by Q - reinforcement learning algorithm predicting model scaling factor, and then calculated the right value. Finally, the simulation experiment was carried out to predict network traf- fic by building MRLA network. The experiment analysis shows that compare with the MMLP network, the method has better multi - step prediction performance.

关 键 词：网络流量 分形特性 非线性多步预测 强化学习 神经网络 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术]