基于神经网络的改进算法在入侵检测中的应用

Application of Improved Algorithm of Neural Network in Network Intrusion Detection

为解决传统BP算法在网络入侵安全检测中耗时比较长、容易陷入局部最小、均方误差降低率振动剧烈的问题,提出一种BP神经网络的改进算法,通过改变传统中固定学习率或通过某一常数改变学习率,引入动态变化,根据均方误差的变化而改变学习率。最后通过仿真实验,解决了传统算法中收敛速度较慢、均方误差下降时震动剧烈的问题。

In order to solve the problems in the traditional algorithm that they are relatively long time-consuming, easy to fall into local minimum and to lead to severe vibration to reduce the mean square error, we put forward a BP improved algorithm. The improved algorithm of a BP neural network is presented in this paper. To change traditional method which has been fixed learning rate or has changed learning rate by controlling a constant, we introduce a new method that the change is dynamic. And we change learning rate based on the changes of the mean square error. By simulation, we solve the traditional algorithms convergence slower and the method that it leads to severe vibration to reduce the mean square error.