摘要
针对传统BP神经网络、RBF神经网络及AR模型预测精度不高、结构复杂,提出了相空间重构与Bernstein神经网络组合预测的新方法,并结合PSO算法进行组合预测模型的参数优化。分别以Sprott-J混沌系统和交通流系统为模型,利用自相关法和Cao方法相结合对混沌时间序列进行相空间重构;利用重构时间延迟相量及Bernstein神经网络建立预测模型,并与传统的BP神经网络、RBF神经网络及AR模型进行对比分析。仿真结果表明,相空间重构与Bernstein神经网络组合预测较传统模型结构简单、模拟效果好、预测精度高。
In view of the low prediction accuracy and the complex structure of traditional BP neural network, RBF neural network and AR model, a new prediction method with the combination of phase space reconstruction and Bernstein neural network was proposed, and PSO algorithm was used for parameters optimization of combination forecast model. Taking Sprott-J chaotic system and traffic flow system as models respectively, the combination of autocorrelation and Cao method was used to reconstruct phase space of chaotic time sequence, the refactoring phasor of time delay and Bernstein neural network were used to establish the prediction model, and do comparative analysis with traditional BP neural network, RBF neural network and AR models. The simulation results show that the combination prediction of phase space reconstruction and Bernstein neural network has a simple structure and can get more preferable simulation effect and higher prediction accuracy.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2016年第4期880-889,共10页
Journal of System Simulation
基金
国家自然科学基金(61463047)
