基于BP神经网络的TCP协议改进研究

The improvement of TCP based on BP NN

为解决Internet中的网络拥塞问题,提高网络传输性能,本文提出一种改进的RED算法;RED算法会使得队列长度经常保持在最大限制队列长度附近,从而使网络中数据包的传输延迟较大,网络吞吐量较低。本文运用BP神经网络的预测功能对网络状况进行预测分析,控制路由器上包的平均队列的长度,使得队列能够平滑的渡过拥塞,减少出现TCP断流情况,降低包的传输延迟。用NS2进行仿真实验,结果表明,与RED算法比较,改进后的BPRED算法能有效减小网络数据包的延迟,提高TCP数据流在混合网络中的竞争力,并提高网络的吞吐量。

Network congestion is a serious problem for a hybrid network in Internet. In this paper we have improved the arithmetic of RED. The RED always cause that the length of average queue approximates to the max-limited length on router, and then leads to the larger delay and lower throughputs. Therefore, BP NN, which can forecast the situation of the network for a certain time, is adopted to analyses and control the length of queue on router. And then prevent the TCP stream from congestion so that the queue can transit smoothly and shorten the delay. Simulate in NS2, experimental results show that the delays of the network packets are smaller and the throughputs are higher when use BPRED on router than RED under the same Internet condition. And the TCP data stream is more powerful now.