偏置剪枝叠式自编码回声状态网络的时序预测
Stacked auto-encoder echo state network with Biased Drop-weight for applications of time series prediction
刘丽丽 1刘玉玺 2王河山2
作者信息
- 1. 陕西师范大学物理学与信息技术学院,陕西西安 710062
- 2. 郑州大学电气工程学院,河南郑州 450001
- 折叠
摘要
针对大多数模型对时间序列预测数据的预测准确率较低,为提升时间序列的预测精度,提出一种基于Biased Drop-weight的偏置剪枝叠式自编码回声状态网络(BD-AE-SGESN)的深度模型.以叠式ESN为多层深度网络框架,提出一种生成式AE算法生成每一层的输入权值,利用BD算法根据输入权重激活值进行剪枝.对比实验结果表明,该模型能够有效提升预测准确率,在3个不同的数据上,相比其它模型有着较小的预测误差和较高的稳定度.
Abstract
For the low prediction accuracy of most models on time series prediction data,to improve the accuracy of multivariate time series prediction,a depth model based on Biased Drop-weight stacked self-coding echo state network(BD-AE-SGESN)was proposed.The proposed stacked ESN was used as a multilayer depth network framework and a generative self-coding algorithm was proposed to generate the input weights for each layer,and the BD algorithm was used to prune the input weights according to their activation values.The proposed model can effectively improve the prediction accuracy with long memory,and it has smaller prediction error and higher stability than other models on three different data.
关键词
多变量时间序列/回声状态网络/预测模型/剪枝/自编码/深度网络/权重优化Key words
multivariable time series prediction/echo state network/prediction model/pruning/auto-encoder/deep network/weight optimization引用本文复制引用
基金项目
国家自然科学基金项目(61603343)
河南省高等学校重点科研基金项目(22A413009)
出版年
2024
NETL
NSTL
万方数据