基于LSTM神经网络的流化床干燥器内生物质颗粒湿度预测

doi:10.11949/0438-1157.20190692

摘要/Abstract

摘要：

生物质作为一种储量丰富、环境友好且易于获取的可再生能源，日渐成为能源研究利用领域的热点。生物质湿度是影响生物质利用效率的关键因素，因此干燥是生物质利用之前的必要步骤。流化床由于其良好的传热传质特性，在干燥过程中得到了广泛的应用。为了实时监测生物质颗粒的干燥过程，利用弧形静电传感器阵列，结合用于时间序列建模的长短期记忆（LSTM）神经网络，实现了流化床干燥器内生物质颗粒湿度的预测。在实验室规模的流化床干燥器上进行了多工况实验获取训练和测试数据，通过模型参数优化确定了LSTM模型。通过与标准循环神经网络（RNN）模型的预测结果的对比表明，LSTM神经网络模型的平均相对误差较小，能够较为准确地预测流化床干燥器内生物质颗粒的湿度。

关键词: LSTM模型, 生物质, 湿度, 神经网络, 静电传感器阵列, 流化床

Abstract:

As a common form of renewable energy with abundant reserves, environmental friendliness and easy access, biomass has aroused general interests in the field of energy research and utilization. Biomass moisture content is a significant factor for the efficient utilization of biomass resources, therefore drying becomes a necessary step before the biomass utilization.Fluidized bed has been widely used in the drying process due to its good heat and mass transfer characteristics. For the online monitoring of the biomass drying process, arc-shaped electrostatic sensor arrays along with a LSTM (long short-term memory) neural network are deployed to establish a neural network model for the prediction of biomass moisture content in the fluidized bed dryer. Experimental tests under different conditions are carried out on a lab-scale fluidized bed dryer to obtain data for model training and testing. Besides, the LSTM neural network model is determined through the optimization of the model parameters. Compared with normal RNN (recurrent neural network) model,the prediction results from the LSTM neural network model have smaller mean relative error, which can predict the biomass moisture content in the fluidized bed dryer with certain accuracy.

Key words: LSTM model, biomass, moisture content, neural network, electrostatic sensor arrays, fluidized bed

中图分类号:

TK 6

方黄峰, 刘瑶瑶, 张文彪. 基于LSTM神经网络的流化床干燥器内生物质颗粒湿度预测[J]. 化工学报, 2020, 71(S1): 307-314.

Huangfeng FANG, Yaoyao LIU, Wenbiao ZHANG. Biomass moisture content prediction in fluidized bed dryer based on LSTM neural network[J]. CIESC Journal, 2020, 71(S1): 307-314.

图/表 12

图1 弧形静电传感器阵列

Fig.1 Arc-shaped electrostatic sensor arrays

图2 实验装置

Fig.2 Experimental setup

表1 实验条件

Table 1 Experimental condition

入口空气温度/℃	入口空气体积流量/(m³/h)
入口空气温度/℃	25	30	35
45	E1	E4	E7
60	E2	E5	E8
75	E3	E6	E9

图3 不同工况下生物质颗粒湿度变化

Fig.3 Biomass moisture content under different experimental conditions

图4 工况E3下初始流化阶段和平稳流化阶段时的传感器A-1、A-2和A-3的静电信号

Fig.4 Electrostatic signal from sensors A-1, A-2 and A-3 at the initial fluidization state and the smooth fluidization state respectively under the experimental condition E3

图5 RNN神经网络结构

Fig.5 Diagram of RNN neural network

图6 LSTM神经网络结构图

Fig.6 Diagram of LSTM neural network

表2 生物质颗粒湿度与输入参数的相关性

Table 2 Correlation between biomass moisture content and input parameters

V_C	R_H	T	DC	AC	N	SKE	KUR	AFF	EN	SF
-0.92	0.19	-0.87	0.59	-0.86	0.42	-0.86	0.79	0.40	0.58	-0.04

图7 隐含层神经元个数、隐含层层数与时间步长的优化

Fig.7 Optimization of number of neurons, number of hidden layers and time step

图8 损失函数变化趋势

Fig.8 Trend of loss function

图9 预测值与参考值对比与相对误差

Fig.9 Comparison and relative errors between predicted and reference values

图10 LSTM与RNN模型预测效果对比

Fig.10 Comparison of LSTM and RNN models

参考文献 33

1	Vitousek P. Beyond global warming-ecology and global change[J]. Ecology, 1994, 75(7): 1862-1876.
2	Stevens C V. Thermochemical processing of biomass: conversion into fuels, chemicals and power[M]//Brown R C， ed. Thermochemical Processing of Biomass. New York: Wiley, 2011: 22-30.
3	阎立峰, 朱清时. 以生物质为原材料的化学化工[J]. 化工学报, 2004, 55(12): 1938-1943.
	Yan L F, Zhu Q S. New chemical industry based on biomass[J]. Journal of Chemical Industry and Engineering (China), 2004, 55(12): 1938-1943.
4	常圣强, 李望良, 张晓宇, 等. 生物质气化发电技术研究进展[J]. 化工学报, 2018, 69(8): 27-39.
	Chang S Q, Li W L, Zhang X Y, et al. Progress in biomass gasification power generation technology[J]. CIESC Journal, 2018, 69(8): 27-39.
5	Garcia R, Pizarro C, Lavin A G, et al. Biomass sources for thermal conversion. Techno-economical overview[J]. Fuel, 2017, 195: 182-189.
6	Dietz T, Rosa E A. Effects of population and affluence on CO₂ emissions[J]. Proceedings of the National Academy of Sciences, 1997, 94(1): 175-179.
7	Mckendry P. Energy production from biomass (Ⅰ): Overview of biomass[J]. Bioresource Technology, 2002, 83(1): 37-46.
8	Donald L. Biomass for Renewable Energy, Fuels, and Chemicals[M]. San Diego: Academic Press, 1998: 53.
9	Verma M, Loha C, Sinha A N, et al. Drying of biomass for utilizing in co-firing with coal and its impact on environment – a review[J]. Renewable and Sustainable Energy Reviews, 2017, 71: 732-741.
10	Da Silva C A M, Butzge J J, Nitz M, et al. Monitoring and control of coating and granulation processes in fluidized beds – a review[J]. Advanced Powder Technology, 2014, 25(1): 195-210.
11	Wang H G, Senior P R, Mann R, et al. Online measurement and control of solids moisture in fluidized bed dryers[J]. Chemical Engineering Science, 2009, 64(12): 2893-2902.
12	Carter A, Briens L. Inline acoustic monitoring to determine fluidized bed performance during pharmaceutical coating[J]. International Journal of Pharmaceutics, 2018, 549: 293-298.
13	Aghbashlo M, Sotudeh-Gharebagh R, Zarghami R, et al. Measurement techniques to monitor and control fluidization quality in fluidized bed dryers: a review[J]. Drying Technology, 2014, 32: 1005-51.
14	Kadlec P, Gabrys B, Strandt S. Data-driven soft sensors in the process industry[J]. Computers and Chemical Engineering, 2009, 33(4): 795-814.
15	Momenzadeh L, Zomorodian A, Mowla D. Experimental and theoretical investigation of shelled corn drying in a microwave-assisted fluidized bed dryer using artificial neural network[J]. Food and Bio-products Processing, 2011, 89(1): 15-21.
16	Chokphoemphun S, Chokphoemphun S. Moisture content prediction of paddy drying in a fluidized-bed drier with a vortex flow generator using an artificial neural network[J]. Applied Thermal Engineering, 2018, 145: 630-636.
17	Zhang W B, Yan Y, Yang Y R, et al. Measurement of flow characteristics in a bubbling fluidized bed using electrostatic sensor arrays[J]. IEEE Transactions on Instrumentation and Measurement, 2016, 65(3): 703-712.
18	Yang Y, Zhang Q, Zi C, et al. Monitoring of particle motions in gas-solid fluidized beds by electrostatic sensors[J]. Powder Technology, 2017, 308: 461-471.
19	Shi Q, Zhang Q, Han G D, et al. Simultaneous measurement of electrostatic charge and its effect on particle motions by electrostatic sensors array in gas-solid fluidized beds[J]. Powder Technology, 2017, 312: 29-37.
20	Qian X C, Shi D P, Yan Y, et al. Effects of moisture content on electrostatic sensing based mass flow measurement of pneumatically conveyed particles[J]. Powder Technology, 2017, 311: 579-588.
21	Zhang W B, Cheng X F, Hu Y H, et al. Measurement of moisture content in a fluidized bed dryer using an electrostatic sensor array[J]. Powder Technology, 2018, 325: 49-57.
22	Zhang W B, Cheng X F, Hu Y H, et al. Online prediction of biomass moisture content in a fluidized bed dryer using electrostatic sensor arrays and the random forest method[J]. Fuel, 2019, 239: 437-445.
23	He C, Bi X T, Grace J R. Simultaneous measurements of particle charge density and bubble properties in gas-solid fluidized beds by dual-tip electrostatic probes[J]. Chemical Engineering Science, 2015, 123: 11-21.
24	He C, Bi X T, Grace J R. Measurement of particle charge-to-mass ratios in a gas–solids fluidized bed by a collision probe[J]. Powder Technology, 2003, 135/136: 181-191.
25	Zachary C L, John B, Charles E. A critical review of recurrent neural networks for sequence learning[J]. Computer Science, 2015, arXiv: 15060019V4.
26	Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735.
27	Bengio Y. Learning long-term dependencies with gradient descent is difficult[J]. IEEE Transactions on Neural Networks, 1994, 5(2): 157-166.
28	Colah. Understanding LSTM Networks[EB/OL].[2015-8-27]. https: //colah.github.io/posts/2015-08
	-Understanding-LSTMs/.[29] Peng L, Liu S, Liu R, et al. Effective long short-term memory with differential evolution algorithm for electricity price prediction[J]. Energy, 2018, 162: 1301-1314.
30	Gers F A, Schmidhuber J, Cummins F. Learning to forget: continual prediction with LSTM[C]//International Conference on Artificial Neural Networks. 2002.
31	窦珊, 张广宇, 熊智华. 基于LSTM时间序列重建的生产装置异常检测[J]. 化工学报, 2019, 70(2): 61-66.
	Dou S, Zhang G Y, Xiong Z H. Anomaly detection of process unit based on LSTM time series reconstruction[J]. CIESC Journal, 2019, 70(2): 61-66.
32	Yao K, Cohn T, Vylomova K, et al. Depth-gated recurrent neural networks[J]. Computer Science, 2015, arXiv: 1508.03790.
33	李泽龙, 杨春节, 刘文辉, 等. 基于LSTM-RNN模型的铁水硅含量预测[J]. 化工学报, 2018, 69(3): 992-997.
	Li Z L, Yang C J, Liu W H, et al. Research on hot metal Si-content prediction based on LSTM-RNN[J]. CIESC Journal, 2018, 69(3): 992-997.

[1]	温凯杰, 郭力, 夏诏杰, 陈建华. 一种耦合CFD与深度学习的气固快速模拟方法[J]. 化工学报, 2023, 74(9): 3775-3785.
[2]	郑佳丽, 李志会, 赵新强, 王延吉. 离子液体催化合成2-氰基呋喃反应动力学研究[J]. 化工学报, 2023, 74(9): 3708-3715.
[3]	诸程瑛, 王振雷. 基于改进深度强化学习的乙烯裂解炉操作优化[J]. 化工学报, 2023, 74(8): 3429-3437.
[4]	闫琳琦, 王振雷. 基于STA-BiLSTM-LightGBM组合模型的多步预测软测量建模[J]. 化工学报, 2023, 74(8): 3407-3418.
[5]	尹刚, 李伊惠, 何飞, 曹文琦, 王民, 颜非亚, 向禹, 卢剑, 罗斌, 卢润廷. 基于KPCA和SVM的铝电解槽漏槽事故预警方法[J]. 化工学报, 2023, 74(8): 3419-3428.
[6]	陈佳起, 赵万玉, 姚睿充, 侯道林, 董社英. 开心果壳基碳点的合成及其对Q235碳钢的缓蚀行为研究[J]. 化工学报, 2023, 74(8): 3446-3456.
[7]	吴文涛, 褚良永, 张玲洁, 谭伟民, 沈丽明, 暴宁钟. 腰果酚生物基自愈合微胶囊的高效制备工艺研究[J]. 化工学报, 2023, 74(7): 3103-3115.
[8]	徐野, 黄文君, 米俊芃, 申川川, 金建祥. 多源信息融合的离心式压缩机喘振诊断方法[J]. 化工学报, 2023, 74(7): 2979-2987.
[9]	董茂林, 陈李栋, 黄六莲, 吴伟兵, 戴红旗, 卞辉洋. 酸性助水溶剂制备木质纳米纤维素及功能应用研究进展[J]. 化工学报, 2023, 74(6): 2281-2295.
[10]	高学金, 姚玉卓, 韩华云, 齐咏生. 基于注意力动态卷积自编码器的发酵过程故障监测[J]. 化工学报, 2023, 74(6): 2503-2521.
[11]	张媛媛, 曲江源, 苏欣欣, 杨静, 张锴. 循环流化床燃煤机组SNCR脱硝过程气液传质和反应特性[J]. 化工学报, 2023, 74(6): 2404-2415.
[12]	杨峥豪, 何臻, 常玉龙, 靳紫恒, 江霞. 生物质快速热解下行式流化床反应器研究进展[J]. 化工学报, 2023, 74(6): 2249-2263.
[13]	陈巨辉, 张谦, 舒崚峰, 李丹, 徐鑫, 刘晓刚, 赵晨希, 曹希峰. 基于DEM方法的旋转流化床纳米颗粒流动特性研究[J]. 化工学报, 2023, 74(6): 2374-2381.
[14]	黄磊, 孔令学, 白进, 李怀柱, 郭振兴, 白宗庆, 李平, 李文. 油页岩添加对准东高钠煤灰熔融行为影响的研究[J]. 化工学报, 2023, 74(5): 2123-2135.
[15]	袁子涵, 王淑彦, 邵宝力, 谢磊, 陈曦, 马一玫. 基于幂律液固曳力模型流化床内湿颗粒流动特性的研究[J]. 化工学报, 2023, 74(5): 2000-2012.