作者：曹伟, 何非

作者单位：南京理工大学机械工程学院工业工程系, 江苏 南京 210094

关键词：温度预测模型;分步预测方法;LSTM循环神经网络;时间序列数据

摘要：

分解炉出口温度是水泥分解工艺的重要指标，温度是否合理对于水泥产品质量有重要意义。为对水泥分解炉出口温度进行预测，结合质量影响因素分析选取的工艺参数，基于LSTM算法建立水泥分解炉出口温度预测模型，模型分为直接预测模型及分步预测模型。在验证集上采用直接预测模型进行预测并与BP神经网络模型进行对比，在实际工况的测试集上将基于状态变量预测的分步预测模型与采用近似值的直接预测模型进行对比，结果表明，分步预测模型针对实际工况有更好的泛化性能，预测误差为0.42 ℃，误差率仅为0.05%。该模型的建立可以为后续分解工艺参数优化及分解炉出口温度控制提供研究基础。

Research on step-by-step prediction method of cement decomposition furnace outlet temperature based on LSTM
CAO Wei, HE Fei
Department of Industrial Engineering, School of Mechanical Engineering, Nanjing University of Science and Technology, Nanjing 210094, China
Abstract: Decomposition furnace outlet temperature is an important indicator of the cement decomposition process, and whether the temperature is reasonable is of great significance to the quality of the cement preduct. In order to predict and control the cement decomposition furnace outlet temperature, combined with the selected process parameters of quality influencing factors analysis, the LSTM algorithm is used to establish a cement decomposition furnace outlet temperature prediction model. This model is divided into a direct prediction model and a step-by-step prediction model. The direct prediction model is used on the validation set to predict and compare with the BP neural network model. The step-by-step prediction model based on state variable prediction is compared with the direct prediction model using approximate values on the test set that simulates actual working conditions. The results show that the step-by-step prediction model has better generalization performance for actual working conditions, the error of prediction is 0.42℃, and the error rate is only 0.05%. The establishment of this model can provide a research basis for the subsequent optimization of decomposition process parameters and the control of the outlet temperature of the decomposition furnace.
Keywords: temperature prediction model;step-by-step prediction method;LSTM recurrent neural network;time series data
2023, 49(5):23-30  收稿日期: 2021-07-04;收到修改稿日期: 2021-09-19
基金项目:
作者简介: 曹伟(1997-),男,江苏泰州市人,硕士研究生,专业方向为工业大数据分析与智能制造
参考文献
[1] 陈全德, 兰明章. 新型干法水泥技术原理与应用讲座[J]. 建材发展导向, 2005(4): 22-28
[2] 李剑锋. 分解炉温度自动控制系统[J]. 水泥技术, 2007(4): 40
[3] 段鑫. 水泥分解炉控制模式的研究[D]. 石家庄: 河北科技大学, 2015.
[4] 考宏涛. 预分解窑系统在稳定运行条件下的用风[J]. 水泥, 2011(2): 17-20
[5] 袁铸钢, 苏哲, 张强. 水泥分解炉出口温度T-S建模[J]. 控制工程, 2016, 23(2): 211-217
[6] YAO Z, LI X, WANG Z. Research and application of neural network PID control in cement industry[C]. 2010 2nd International Conference on Industrial and Information Systems. IEEE, 2010(2): 424-427.
[7] 王祥民, 董学平, 于广宇. 基于动态主元分析和极限学习机的分解炉出口温度预测[J]. 测控技术, 2019, 38(12): 35-39
[8] GRAVES A. Supervised sequence labelling with recurrent neural networks [M]. Berlin: Springer, 2012: 37-45.
[9] SZÉLES B, PARAJKA J, HOGAN P, et al. Stepwise prediction of runoff using proxy data in a small agricultural catchment[J]. Journal of Hydrology and Hydromechanics, 2021, 69(1): 65-75
[10] 程义明, 罗滇生, 何洪英, 等. 分步预测法在省级电网短期负荷预测中的应用[J]. 电力系统及其自动化学报, 2012, 24(4): 54-58
[11] 朱坤财, 徐郑攀, 赵自奇, 等. 基于航迹预测的水面无人艇动态避障方法[J/OL]. 中国测试: 1-6. http://kns.cnki.net/kcms/detail/51.1714.TB.20210708.1426.040.html.
[12] 胡城豪, 胡昌华, 司小胜, 等. 基于MSCNN-LSTM的滚动轴承剩余寿命预测方法[J]. 中国测试, 2020, 46(9): 103-110
[13] 王鑫, 吴际, 刘超, 等. 基于LSTM循环神经网络的故障时间序列预测[J]. 北京航空航天大学学报, 2018, 44(4): 772-784
[14] KINGMA D P, BA J. Adam: A method for stochastic optimization[C]//ICLR 2015, 2015: 1-15.
[15] 武伟宁, 刘小燕, 徐学奎, 等. 水泥熟料质量软测量模型中的时序分析方法[J]. 控制理论与应用, 2018, 35(7): 1029-1036