论文部分内容阅读
基于回声状态网络(ESN)预测模型,结合小波分析和主元分析,提出一种组合预测方法.首先对含噪非线性时间序列进行小波降噪,并重构时间序列产生训练样本,再将训练样本通过主元分析进行降维处理,降维后的时间序列数据则输入ESN模型进行预测分析.对控制飞机动力输出的动压参数非线性时间序列数据进行了仿真对比实验,结果表明:组合预测方法的5步和单步预测速度累计提高了66.97%,预测的平均平方误差、标准均方根误差和归一化绝对误差也均有较大提高.该方法与传统基于ESN的预测模型相比,能有效地提高预测的效率和精度,是一种有效的非线性时间序列预测方法.
Based on the ESN prediction model and combined with wavelet analysis and principal component analysis, a combined prediction method is proposed.Firstly, wavelet denoising is performed on the noisy nonlinear time series and reconstructed time series to generate training samples, and then training The principal component analysis is used to reduce the dimension of the sample and the dimensionless time series data are input into the ESN model for predictive analysis.The simulation experiments are carried out on the nonlinear time series data of the dynamic pressure parameters controlling aircraft power output.The results show that the combined forecast The 5-step and single-step prediction speed of the method has been increased by 66.97%, and the average squared error, standard root mean square error and normalized absolute error have also been greatly improved. Compared with the traditional ESN-based prediction model , Which can effectively improve the efficiency and accuracy of prediction. It is an effective nonlinear time series prediction method.