多策略改进RBF神经网络入侵检测方法研究

论文提出了一种多策略改进RBF神经网络入侵检测方法。该方法采用减聚类算法确定隐含层节点数,具有自适应确定隐层节点的能力,避免了调整隐层节点的人为干扰。采用粒子群算法和梯度下降法相结合的方法分别对基函数的中心值、宽度以及隐含层与输出层之间的权值进行全局优化以及局部优化,避免了参数选取的局部性。实验证明,该方法能够有效提高入侵检测系统的检测率,并降低误报率。

Study on the Intrusion Detection Method with Multi-strategy Improved RBF NN

This thesis proposes an intrusion detection method with multi-strategy improved RBF neural networks. The method can be used to determine the number of hidden layer nodes adaptively by using subtractive clustering, avoiding the man-made interference for adjusting the hidden layer nodes. Using PSO algorithm and gradient descent method were applied to optimize position of data centers, widths and weights of RBF neural networks, avoiding the locality in parameter selection. The experiment shows that this method can effectively enhance the intrusion detection system detection rate and reduce the rate of false positives.