[1]杨珺,佘佳丽,刘艳珍.基于深度置信网络的时间序列预测[J].深圳大学学报理工版,2019,36(No.6(599-724)):718-724.[doi:10.3724/SP.J.1249.2019.06718]
　YANG Jun,SHE Jiali,and LIU Yanzheng.Time series prediction based on deep confidence networks[J].Journal of Shenzhen University Science and Engineering,2019,36(No.6(599-724)):718-724.[doi:10.3724/SP.J.1249.2019.06718]
点击复制

基于深度置信网络的时间序列预测()
分享到：

《深圳大学学报理工版》[ISSN:1000-2618/CN:44-1401/N]

卷:
第36卷
期数:
2019年No.6(599-724)
页码:
718-724
栏目:
电子与信息科学
出版日期:
2019-11-20

文章信息/Info

Title:
Time series prediction based on deep confidence networks
文章编号:
201906018
作者:
杨珺1; 佘佳丽2; 刘艳珍1
1） 江西农业大学软件院，江西南昌330045 ；2）南昌工学院，江西南昌330099
Author(s):
YANG Jun1;  SHE Jiali2;  and LIU Yanzheng1
1) Jiangxi Agricultural University Software Institute, Nanchang 330015, Jiangxi Province, P.R.China 2) Nanchang Institute of Engineering, Nanchang 330099, Jiangxi Province, P.R.China
关键词:
计算机神经网络; 时间序列预测; 深度神经网络; 深度置信网络; 农机总动力; 预测模型; 股票预测
Keywords:
computer neural networks;  time series prediction;  deep neural network;  deep confidence network;  agricultural machinery total power;  prediction model;  stock forecast
分类号:
TP391
DOI:
10.3724/SP.J.1249.2019.06718
文献标志码:
A
摘要:
针对传统的计算机神经网络存在梯度弥散、局部最小值、非线性时间序列预测的长期预测性能不佳和高维序列数据复杂度高等问题，提出时序深度置信网络模型(timing deep belief network model, T-DBN)．该模型预训练阶段采用改进的贪婪预训练算法，并在预训练过程中使用梯度修正并行回火采样（gradient fixing parallel tempering, GFPT）算法，采用重构误差确定网络深度，反向调整阶段则采用拟牛顿法（BFGS算法），以获得更加准确的预测精度．结合相空间重构理论和BP神经网络，对江西省2016—2020年农业机械总动力进行了预测．针对高非线性的股票数据，提取同花顺软件1990-12-20—2018-03-30时间段内的上证指数特征信息，分别采用T-DBN、DBN和LSTM模型进行股票预测，预测准确率分别为79.3%、77.9%和74.6%，T-DBN模型的预测准确率高于DBN和LSTM模型．
Abstract:
The timing deep belief network model (T-DBN) is proposed to solve the problems of gradient dispersion, local minimum, poor long-term prediction performance of nonlinear time series prediction and high complexity of high-dimensional sequence data in traditional computer neural networks. In the pre-training stage, the improved greedy pre-training algorithm is adopted. In the pre-training process, the gradient fixing parallel tempering (GFPT) algorithm is used to determine the depth of the network by reconstruction error. In the stage of reverse adjustment, quasi-newton method (BFGS algorithm) is adopted to obtain more accurate prediction accuracy. Combined with the theory of phase space reconstruction and BP neural network, the total power of agricultural machinery in jiangxi province from 2016 to 2020 was predicted. For the highly nonlinear stock data, the characteristics of the Shanghai Stock Exchange index in the period of 1990-12-20-2018-03-30 were extracted from tonghuashun software . The T-DBN, DBN and LSTM models were used for stock forecasting. The prediction accuracy was 79.3%, 77.9% and 74.6% respectively, the prediction accuracy of the T-DBN model is higher than that of the DBN and LSTM models.

相似文献/References:

[1]陈玉佳,姜波.基于小波神经网络的加工番茄产量预测模型[J].深圳大学学报理工版,2015,32(No.5(441-550)):546.[doi:10.3724/SP.J.1249.2015.05546]
　Chen Yujia and Jiang Bo.A wavelet neural network model for processing tomato yield forecasting[J].Journal of Shenzhen University Science and Engineering,2015,32(No.6(599-724)):546.[doi:10.3724/SP.J.1249.2015.05546]

备注/Memo

备注/Memo:
Received：2019-04-21；Accepted：2019-05-31 Foundation：National Natural Science Foundation of China (61502213); Science and Technology Project of Jiangxi Provincial Department of Education (GJJ10422) Corresponding author：Associate professor YANG Jun. E-mail: ycjun515@163.com Citation：ANG Jun, SHE Jiali, LIU Yanzheng. Time series prediction based on deep confidence networks［J］. Journal of Shenzhen University Science and Engineering, 2019, 36(6): 718-724.（in Chinese）
更新日期/Last Update: 2019-11-30