摘要
针对流程生产连续性强、时序耦合复杂等特点,传统神经网络不具备长期记忆能力,且在深层次网络训练时易出现训练参数灾难、梯度爆炸等问题,提出基于马尔可夫优化的融合门控循环单元(GRU)与长短期记忆网络(LSTM)的组合预测模型(Mar-G LSTM)。首先在循环神经网络结构中融入门控机制构建深度LSTM神经网络模型,对流程生产时序数据信息进行选择性记忆,学习时序数据序列的信息依赖,进而解决训练过程中的梯度爆炸问题;同时结合马尔可夫链对GRU-LSTM模型的预测结果进行修正优化,在降低模型的复杂度的情况下进一步提高了模型的预测精度。最后,结合某流程生产线的工艺数据进行分析验证,结果表明,Mar-G LSTM算法在预测精度上较随机森林模型、门控循环单元神经网络模型(GRU)、长短期记忆神经网络模型(LSTM)和卷积神经网络与门控循环单元网络组合模型(CNN-GRU)分别提高了37.42%、21.32%、17.91%和12.56%,所提Mar-G LSTM算法可实现流程生产质量的准确预测,为降低工艺参数调控任务的完成时间提供了思路和实现途径。
Aiming at the characteristics of process production with strong continuity and complex temporal coupling,and the problem that traditional neural networks do not have long-term memory capability and are prone to training parameter disasters and gradient explosion during deep network training,a combined prediction model based on incorporates Gated Recurrent Units(GRU)of Markov optimization and Long and Short-Term Memory(LSTM)networks named Mar-G LSTM was proposed.A deep LSTM neural network model was constructed by incorporating the gating mechanism into the recurrent neural network structure to selectively memorise the process production timing data information and learn the information dependence of timing data sequences,thus solving the gradient explosion problem during training.At the same time,the prediction results of the GRU-LSTM model were modified and optimised by combining Markov chain,which further improved the prediction accuracy while reducing the complexity of the model.The prediction accuracy of the model was further improved.The results showed that the Mar-G LSTM algorithm improved the prediction accuracy by 37.42%,21.32%,17.91%and 12.56%compared with the random forest model,the GRU model,the LSTM model and the combined Convolutional Neural Network and GRU network(CNN-GRU)model respectively.The proposed Mar-G LSTM algorithm could achieve accurate prediction of process production quality,which provided an idea and a way to reduce the completion time of process parameter regulation tasks.
作者
阴艳超
苏逸凡
唐军
林文强
蒲昊苒
汪霖宇
YIN Yanchao;SU Yifan;TANG Jun;LIN Wenqiang;PU Haoran;WANG Linyu(Faculty of Mechanical and Electrical Engineering,Kunming University of Science and Technology,Kunming 650500,China;China Tobacco Kunming Industrial Co.Ltd.,Kunming 650024,China)
出处
《计算机集成制造系统》
EI
CSCD
北大核心
2024年第3期942-957,共16页
Computer Integrated Manufacturing Systems
基金
国家重点研发计划资助项目(2023YFB3308401)
国家自然科学基金资助项目(52065033)
云南省重大科技资助项目(202202AG050002)。
