大连海事大学学报 >

2025 , Vol. 51 >Issue 2: 154 - 164

DOI: https://doi.org/10.16411/j.cnki.issn1006-7736.2025.02.017

基于LSTM的含油污水膜分离效果实验与预测方法

展开
  • (大连海事大学 轮机工程学院，辽宁 大连 116026) 
鱼亨洋（1998 — ），男，硕士生，研究方向：船舶含油污水膜分离效果，E-mail：1935864937@qq.com；张彬*(1982—），男，博士，教授，研究方向：船舶污染防治与船舶安全防护。 E-mail：zb_2010@dlmu.edu.cn。

网络出版日期: 2025-04-29

基金资助

国家重点研发计划项目（2023YFB4301700）；工信部高科技船舶项目（2019-360-4）

收起

Experimental and prediction method for membrane separation efficiency of oily waste water based on LSTM

Expand
  • (Marine Engineering College, Dalian Maritime University, Dalian 116026, China)

Online published: 2025-04-29

Fold

摘要

为探究船舶油水分离器中陶瓷膜的分离机理与预测方法，基于自建实验平台，系统考察了含油浓度、膜孔径与跨膜压差对分离效率及通量的影响。结果显示，无论浓度、孔径或压差如何变化，截留率均大于99%；但通量衰减受该三者制约显著，影响顺序为：含油浓度＞跨膜压差＞膜孔径。以孔径1 μm、压差0.10 MPa为例，当浓度由50 mg/L升至200 mg/L时，通量衰减率从2%增至12%；以浓度100 mg/L为例，压差从0.05 MPa升至0.20 MPa，可使初始通量达583 L/(m2·h)，但平均衰减率达8%；在0.10 MPa条件下，孔径由0.5 μm增至2 μm时，高浓度下通量由368 L/(m2·h)降至312 L/(m2·h)，衰减率达16%。为此，提出了结合影响机制的膜分离效果预测网络Flow-LSTM，该模型通过时空特征提取、注意力融合与残差连接，有效缓解了传统LSTM对历史信息筛选不足与损耗问题，并将关键变量排序结果作为输入控制模型。与MLP、RNN、LSTM及GRU基线对比，Flow-LSTM在R2、MSE和MAE指标上均优于基线，较原始LSTM模型，R2提升5%，MSE和MAE分别下降21%和22%。


关键词： 陶瓷膜; 油水分离; 分离效果; 膜通量; 长短期记忆网络（LSTM）

本文引用格式

鱼亨洋, 张彬, 刘小超 . 基于LSTM的含油污水膜分离效果实验与预测方法[J]. 大连海事大学学报, 2025 , 51(2) : 154 -164 . DOI: 10.16411/j.cnki.issn1006-7736.2025.02.017

Abstract

To explore the separation mechanism and prediction method of ceramic membranes in marine oilwater separators, based on a selfbuilt experimental platform, the effects of oil concentration, membrane pore size and transmembrane pressure difference on separation efficiency and flux were systematically investigated. The results show that regardless of how the concentration, pore size or pressure difference changes, the retention rate is all greater than 99%. However, the flux attenuation is significantly restricted by above three factors, and the order of influence is: oil concentration > transmembrane pressure difference > membrane pore size. Taking a pore size of 1 μm and a pressure difference of 0.10 MPa as an example, when the concentration increases from 50 mg/L to 200 mg/L, the flux attenuation rate increases from 2% to 12%. Take 100 mg/L as an example, when the pressure difference increases from 0.05 MPa to 0.20 MPa, the initial flux can reach 583 L/(m2·h), but the average attenuation rate is 8%. Under the condition of 0.10 MPa, when the pore size increases from 0.5 μm to 2 μm, the flux at high concentration decreases from 368 L/(m2·h) to 312 L/(m2·h), and the attenuation rate reaches 16%. Based on this, a membrane separation effect prediction network FlowLSTM combined with the influence mechanism is proposed. This model effectively alleviates the problems of insufficient screening and loss of historical information by traditional LSTM through spatiotemporal feature extraction, attention fusion and residual connection, and takes the ranking results of key variables as the input to control the model. Compared with the baselines of MLP, RNN, LSTM and GRU, FlowLSTM outperforms the baselines in terms of R2, MSE and MAE indicators. Compared with the original LSTM model, R2 increases by 5%, and MSE and MAE decreases by 21% and 22% respectively.


Key words： ceramic membrane; oil-water separation; separation efficiency; membrane flux; long short-term memory network (LSTM)

参考文献

[1]HOSSEINI M K, LIU L, HOSSEINI P K, et al.  Performance evaluation of a pilot-scale membrane  filtration system for oily wastewater treatment:    CFD modeling and scale-up design[J].Journal of  Water Process Engineering, 2023, 52:103570.
[2]LAY H T, WANG R, CHEW J W. Influence of foulant particle shape on membrane fouling in dead-end microfiltration[J].Journal of Membrane Science, 2022,647:120265.
[3]DU X, LIU Y, MA R, et al. Gravity-driven ceramic membrane (GDCM) filtration treating manganese-contaminated surface water: effects of ozone(O3)-  aided pre-coating and membrane pore size[J]. Chemosphere,2021,279:130603.
[4]王帅帅.电场破乳-动态陶瓷膜串联组合含油污水处理研究[D].北京：北京石油化工学院,2023.DOI: 10.27849/d.cnki.gshyj.2023.000101.
WANG S S. Study on treatment technology of only wastewater by combination of electric field demulsification and dynamic ceremic membrane[D].Beijing: Beijing Institute of Petrochemical Technology, 2023. DOI: 10.27849/d.cnki.gshyj.2023.000101. (in Chinese)
[5]王琰璟.膜过滤过程中甘蔗汁风味物质及特征演变规律的研究[D].南宁：广西大学,2024.DOI:10.27034/d.cnki.ggxiu.2024.000404.
WANG Y J. Sugarcane juice flavor substances during membrane filtration and the study of the characteristic evolution law[D]. Nanning:Guangxi University, 2024.DOI:10.27034/d.cnki.ggxiu.2024.000404. (in Chinese)
[6]张汉泉.亲水性陶瓷复合膜的制备及其分离性能研究[D].广州:华南理工大学,2015.
ZHANG H Q. Preparation of hydrophilic ceramic composite membrane and its separation performance research[D]. Guangzhou: South China University of Technology,2015.(in Chinese)
[7]KRAMER F C, SHANG R, RIETVELD L C, et al. Influence of pH, multivalent counter ions, and membrane fouling on phosphate retention during ceramic nanofiltration[J]. Separation and Purification Technology,2019,227:115675.
[8]张芳,吴鹏,何观伟,等.尿素水解法制备高活性 Cu-Ag/SiO2催化剂及其催化草酸二甲酯加氢性能[J].工业催化,2023,31(7):45-49.
ZHANG F, WU P, HE G W, et al. Highly active Cu-Ag/SiO2 catalysts prepared by urea hydrolysis method in dimethyl oxalate hydrogenation[J]. Industrial Catalysis, 2023,31(7):45-49. (in Chinese)
[9]ABDULLAH S, KARMAKAR S, PRADHAN R C, et al. Pressure‐driven crossflow microfiltration coupled with centrifugation for tannin reduction and clarification of cashew apple juice: Modeling of permeate flux decline and optimization of process parameters[J]. Journal of Food Processing and Preservation, 2022, 46(6): e16497.
[10]USMAN J L, SALAMI B A, GBADAMOSI A, et al. Intelligent optimization for modelling superhydrophobic ceramic membrane oil flux and oil-water separation efficiency: Evidence from wastewater treatment and experimental laboratory[J]. Chemosphere, 2023, 331:138726.
[11]IDRIS I, AHMAD Z, OTHMAN M R, et al. Application of artificial neural network to predict water flux from pre-treated palm oil mill effluent using direct contact membrane distillation[J]. Materialstoday: Proceedings, 2022, 63(Supl.1): S411-S417.
[12]PARK S, SHIM J, YOON N, et al. Deep reinforcement learning in an ultrafiltration system: Optimizing operating pressure and chemical cleaning conditions[J]. Chemosphere, 2022, 308(Part2):136364.
[13]刘耀辉.基于BP神经网络模型的海水淡化反渗透膜污染预测研究[D]. 青岛:青岛理工大学,2023.DOI:10.27263/d.cnki.gqudc.2023.000704.
LIU Y H. Prediction of seawater desalination reverse osmosis membrane contamination based on BP neural netword model[D]. Qingdao: Qingdao University of Technology, 2023. DOI:10.27263/d.cnki.gqudc.2023.000704. (in Chinese)
[14]SHIM J, PARK S, CHO K H. Deep learning model for simulating influence of natural organic matter in nanofiltration[J]. Water Research, 2021, 197:117070.
[15]YOGARATHINAM L T, VELSWAMY K, GANGASALAM A, et al. Performance evaluation of whey flux in dead-end and cross-flow modes via convolutional neural networks[J]. Journal of Environmental Management, 2022, 301:113872.
[16]陈艳,牛亚林,彭兴,等.基于深度学习模型的中试纳滤系统膜污染预测研究[J].兰州交通大学学报,2024,43(5):103-112.
CHEN Y, NIU Y L, PENG X, et al. Prediction of membrane fouling in pilot nanofiltration system based on deep learning model[J]. Journal of Lanzhou Jiaotong University, 2024,43(5):103-112. (in Chinese)
[17]武华华,匡海波,宋扬.基于VMD-FFT-LSTM模型的BDI指数预测[J].大连海事大学学报,2019,45(3):9-16.
WU H H, KUANG H B, SONG Y. Prediction of BDI based on VMD-FFT-LSTM model[J]. Journal of Dalian Maritime University,2019,45(3):9-16. (in Chinese)
[18]EREN M Ş A, ARSLANOĞLU H.α-Alumina (α-Al2O3) ceramic microfiltration membranes in industrial wastewater treatment: Production, design, filtration behavior and performance[J].Ceramics International,2025,51(8):10234-10241. DOI:10.1016/j.ceramin-t.2024.12.454.
[19]MOU Y, CHEN W Z, LIU J G. Total Partial Least Square Regression and its application in infrared spectra- quantitative analysis[J].Measurement, 2025,247:116794.

Options
文章导航

/

〈 〉

版权所有 © 《大连海事大学学报》编辑部
电话：0411-84727810　
E-mail：xuebao@dlmu.edu.cn
技术支持：北京玛格泰克科技发展有限公司