论文部分内容阅读
时间序列预测问题广泛存在于科学研究和工程实践的各个领域之中.传统的时间序列预测方法,多寻求建立静态映射关系来实现预测,从而忽略了时间序列背后研究对象的动态系统本质.本文的研究采用回声状态网络训练算法建立起时间序列的动态预测模型.通过理论分析和数值实验,揭示了回声状态网络训练算法的局限性,并据此提出对训练样本集插值来强化训练的解决方案.在此基础上,本文提出一种包含分叉结构的迭代预测方法,用于训练强化的回声状态网络预测模型.在滑坡位移预测的实际工程案例中,本文方法的优势和有效性得到了验证.
The problem of time series forecasting exists widely in all fields of scientific research and engineering practice.Traditional time series forecasting methods mostly seek to establish the static mapping to realize the forecasting so as to neglect the dynamic system essence of the research object behind the time series.In this paper, The echological state network training algorithm is used to establish the dynamic prediction model of time series.Through the theoretical analysis and numerical experiments, the limitation of the echo state network training algorithm is revealed, and the solution of training sample set interpolation to strengthen the training is proposed. Based on this, an iterative prediction method with bifurcation structure is proposed to train enhanced echo state network prediction model.In the actual case of landslide displacement prediction, the advantages and validity of the proposed method are verified.