基于回声状态网络优化的宽间隔混沌跳频码预测研究

传统宽间隔混沌跳频码预测方法无反馈结构,记忆能力差;且训练过程烦琐,泛化性能差,对预测精度产生不好的影响。为此,提出一种新的基于回声状态网络优化的宽间隔混沌跳频码预测方法。设计回声状态网络,其由输入层、递归层以及输出层三个部分构成。在递归层内部各神经元间引入连接权值稀疏矩阵,使递归层内部存在动态记忆。回声状态网络储备池规模、储备池内部连接权谱半径、储备池稀疏度、储备池输入单元尺度对宽间隔混沌跳频码预测准确性产生不同的影响。通过和声搜索方法对四种储备池参数进行优化,实现回声状态网络的改进。确定优化回声状态网络输入输出数据,建立优化回声状态网络。确定储备池参数,通过训练数据激活储备池,计算回声状态网络输出连接权矩阵,对宽间隔混沌跳频码进行预测。实验结果表明,所提方法预测结果可靠,和其他方法相比有很高的预测精度。

Prediction of wide interval chaotic frequency hopping code based on echo state network optimization

the traditional wide interval chaotic frequency hopping code prediction method has no feedback structure, memory ability is poor, and the training process is cumbersome, and the generalization performance is poor, which has a bad effect on prediction accuracy. To this end, a new prediction method of wide interval chaotic frequency hopping code based on echo state network optimization is proposed. The echo state network is designed, which consists of three parts: the input layer, the recursive layer and the output layer. The connection weight sparse matrix is introduced among the neurons in the recursive layer, so that there is a dynamic memory inside the recursive layer. Echo state network pool size, internal connection pool right spectral radius, pool sparsity, reserve pool input unit scale on the wide interval prediction of chaotic frequency hopping codes have different effects on the accuracy, the harmony search four pool parameter optimization method to achieve improved echo state network. The input and output data of the echo state network are optimized, and the optimal echo state network is established. The parameters of reserve pool are determined, activated pool is activated through training data, and the connection weight matrix of echo state network is calculated, and the wide interval chaotic frequency hopping code is forecasted. The experimental results show that the proposed method is reliable and has high prediction accuracy compared with other methods.