基于遗传神经网络的超混沌时间序列预测

研究了超混沌系统的预测问题。通过分析混沌时间序列，建立具有多个隐节点的3层前馈网络，基于泛化性考虑采用剪枝算法训练，在保证预测精度的基础上消去部分隐节点以降低网络复杂性，利用遗传算法具有的全局寻优能力重新训练网络，利用具有局部寻优能力BP算法再次训练该网络。对Mackey-Glass时滞混沌系统预测实验结果表明，改进算法的泛化性能优于经典BP网络，归一化预测精度提高10倍多，能够较好地解决超混沌系统的预测问题。

Hyperchaotic Time Series Forecast Based on Genetic Neural Networks

This paper studies the forecast of the hyperchaotic system. After analyzing the chaotic time series, a three-layer forward artificial neural network is built up with many notes in the hidden layer. Considering generalization ability of the net, Weight-Elimination(WE) algorithm is adopted to delete some hidden notes for reducing complexity of the net and assure the net forecast precision. Genetic Algorithm(GA) is introduced to train the net over again for its global search ability. And the acquired net is trained again by classical BP arithmetic with its localized search. Experiments on the Mackey-Glass time lag chaos system illustrate that the improved method is better than the classical BP arithmetic, and the normalized forecast precision is enhanced by more than 10 times, so it can resolve the prediction of the super chaotic system.