Improving the accuracy of mechanistic models for dynamic batch distillation enabled by neural network:An industrial plant case

Neural networks are often viewed as pure‘black box’models,lacking interpretability and extrapolation capabilities of pure mechanistic models.This work proposes a new approach that,with the help of neural networks,improves the conformity of the first-principal model to the actual plant.The final result is still a first-principal model rather than a hybrid model,which maintains the advantage of the high interpretability of first-principal model.This work better simulates industrial batch distillation which separates four components:water,ethylene glycol,diethylene glycol,and triethylene glycol.GRU(gated recurrent neural network)and LSTM(long short-term memory)were used to obtain empirical parameters of mechanistic model that are difficult to measure directly.These were used to improve the empirical processes in mechanistic model,thus correcting unreasonable model assumptions and achieving better predictability for batch distillation.The proposed method was verified using a case study from one industrial plant case,and the results show its advancement in improving model predictions and the potential to extend to other similar systems.

Mechanistic Drifting Forecast Model for A Small Semi-Submersible Drifter Under Tide–Wind–Wave Conditions

Understanding the drifting motion of a small semi-submersible drifter is of vital importance regarding monitoring surface currents and the floating pollutants in coastal regions. This work addresses this issue by establishing a mechanistic drifting forecast model based on kinetic analysis. Taking tide–wind–wave into consideration, the forecast model is validated against in situ drifting experiment in the Radial Sand Ridges. Model results show good performance with respect to the measured drifting features, characterized by migrating back and forth twice a day with daily downwind displacements. Trajectory models are used to evaluate the influence of the individual hydrodynamic forcing. The tidal current is the fundamental dynamic condition in the Radial Sand Ridges and has the greatest impact on the drifting distance. However, it loses its leading position in the field of the daily displacement of the used drifter. The simulations reveal that different hydrodynamic forces dominate the daily displacement of the used drifter at different wind scales. The wave-induced mass transport has the greatest influence on the daily displacement at Beaufort wind scale 5–6; while wind drag contributes mostly at wind scale 2–4.

Validation and Evaluation of a Mechanistic Model of Phasic and Phenological Development in Wheat

Three sets of data from the field experiments with different wheat( Triticum L. ) varieties and sowing dates in China and USA were used to test the performance of the mechanistic model of wheat development. The results showed that the absolute prediction errors for most phasic and phenological stages ranged within 0 - 5 days, and the root mean square errors were generally less than 5 days. The model was of high accuracy and low error especially for emergence, tillering, stamen and pistil initiation, and heading stages, reflecting an enhanced level of mechanism and prediction.

Mechanistic Model for Predicting NO_3-N Uptake by Plants and Its Verification

Some mechanistic models have been proposed to predict the No3- concentrations in the soil solution at root surface and the NO3-N uptake by plants, but all these relatively effective non-steady state models have not yet been verified by any soil culture experiment. In the present study, a mathematical model based on the nutrient transport to the roots, root length and root uptake kinetics as well as taking account of the inter-root competition was used for calculation, and soil culture experiments with rice, wheat and rape plants grown on alkali, neutral and acid soils in rhizoboxes with nylon screen as a isolator were carried out to evaluate the prediction ability of the model through comparing the measured NO3-concentrations at root surface and N uptake with the calculated values. Whether the inter-root competition for nutrients was accounted for in the model was of less importance to the calculated N uptake but could induce significant changes in the relative concentrations of NO3- at root surface. For the three soils and crops, the measured NO3-N uptake agreed well with the calculated one, and the calculated relative concentrations at root surface were approximate to the measured values. But an appropriate rectification for some conditions is necessary when the plant uptake parameter obtained in solution culture experiment is applied to soil culture. In contrast with the present non-steady state model, the predicted relative concentrations, which show an accumulation, by the Phillips' steady-state model were distinct from the measured values which show a depletion, indicating that the present model has a better prediction ability than the steady-state model.

Mechanistic Model versus Artificial Neural Network Model of a Single-Cell PEMFC

Model-based controllers can significantly improve the performance of Proton Exchange Membrane Fuel Cell (PEMFC) systems. However, the complexity of these strategies constraints large scale implementation. In this work, with a view to reduce complexity without affecting performance, two different modeling approaches of a single-cell PEMFC are investigated. A mechanistic model, describing all internal phenomena in a single-cell, and an artificial neural network (ANN) model are tested. To perform this work, databases are measured on a pilot plant. The identification of the two models involves the optimization of the operating conditions in order to build rich databases. The two different models benefits and drawbacks are pointed out using statistical error criteria. Regarding model-based control approach, the computational time of these models is compared during the validation step.

Model-based risk assessment on dynamic control of twin-column continuous capture under feedstock variations

Dynamic control is essential to guarantee the stable performance of continuous chromatography.AutoMAb dynamic control strategy has been developed to ensure a consistent protein load in twincolumn CaptureSMB continuous capture by integrating the UV signal of breakthrough.In this study,the process risk of CaptureSMB continuous capture under AutoMAb control towards the feedstock variations was assessed by a mechanistic model developed by us.The effects of target protein and impurities under the variation range of±10 mAU·min^(-1) on load amount,protein loss,process productivity,and resin capacity utilization were investigated.The results showed that the CaptureSMB process could be successfully controlled by AutoMAb towards increased or slightly decreased concentration of feedstock.However,the load process would be out of control with drastically decreased target protein or impurities,and the decreased impurities would lead to protein loss.It was found that AutoMAb control would cause 44.7%non-operational areas and 18.3%protein loss areas in the variation range of±10 mAU·min^(-1).To improve the stability of the CaptureSMB process,a modified AutoMAb control that would stop the load procedure when the absolute value of the integral area reached the preset value,was proposed to reduce the risk of protein loss and the non-operational area.

On-line Estimation in Fed-batch Fermentation Process Using State Space Model and Unscented Kalman Filter

On-line estimation of unmeasurable biological variables is important in fermentation processes,directly influencing the optimal control performance of the fermentation system as well as the quality and yield of the targeted product.In this study,a novel strategy for state estimation of fed-batch fermentation process is proposed.By combining a simple and reliable mechanistic dynamic model with the sample-based regressive measurement model,a state space model is developed.An improved algorithm,swarm energy conservation particle swarm optimization(SECPSO) ,is presented for the parameter identification in the mechanistic model,and the support vector machines(SVM) method is adopted to establish the nonlinear measurement model.The unscented Kalman filter(UKF) is designed for the state space model to reduce the disturbances of the noises in the fermentation process.The proposed on-line estimation method is demonstrated by the simulation experiments of a penicillin fed-batch fermentation process.

Model Estimates of Nutrient Uptake by Red Spruce Respond to Soil Temperature

A better understanding of the mechanisms that control nutrient acquisition in the context of plant and ecosystem responses to climate change is needed. Mechanistic nutrient uptake models provide a means to investigate some of the impacts of temperature change on soil nutrient supply and root uptake kinetics through the simulation of key soil and plant processes. The NST 3.0 model, in combination with literature values on plant and soil parameters from a red spruce (Picea rubens L.) site in the southern Appalachians, was used to conduct a series of model simulations focused on the combined effects of changes to the maximal rate of nutrient influx at high concentrations (Imax), root growth rate (k), concentration of nutrient occurring in the soil solution (Cli), and the ability of the soil solid phase to buffer changes to the soil solution nutrient concentration (b). Previous research has indicated that these four parameters are responsive to changes in root zone temperature. Simulated uptake of NH4 increased by a factor of up to 2.6 in response to increases in soil temperature of 1°C to 5°C. The model also projected an increase in P uptake coupled with up to an 80% reduction in solution P concentration in response to a 1°C -5°C increase over a 147-d simulation period. These hypothetical changes, if validated, have interesting implications for plant growth and competition and point to a need for additional studies to better define the impacts of soil temperature on soil nutrient supply and root uptake.

质子交换膜燃料电池多物理场建模及模型参数敏感性分析

受限于外部运行工况及内部多物理场参数的耦合作用机制,质子交换膜燃料电池的精确机理建模成为其研发和应用的瓶颈。该文通过探究燃料电池内部“电-热-流”多场耦合关系,构建准一维电堆动态模型,并基于单参数敏感性和多参数敏感性分析,得到不同模型参数的敏感度指标,完成了建模中部分关键参数的权重分析。为验证模型精度及耦合参数变动对输出特性的影响,在搭建Ballard NEXA 1.2 kW燃料电池测试平台的基础上开展变载及参数变动实验。结果表明,所构建的模型能够较为精确的模拟电堆特性,且对称因子、活化面积和压力系数的变动对电堆输出的影响较大。结果可为燃料电池电堆的测试及改型提供理论基础,并揭示电堆故障及老化过程中主要参数的衰退变化趋势。

碳纳米纤维混凝土抗盐-冻性能研究

为研究碳纳米纤维混凝土在盐-冻融循环作用下的抗冻性能变化规律,以3.5%NaCl溶液作为冻融循环介质,对4种不同碳纳米纤维(CNFs)质量掺量(0%、0.1%、0.2%、0.3%)的混凝土试件进行快速冻融循环试验,通过扫描电镜、X射线衍射和氮气吸附等微观测试手段分析纤维改性机理,并建立冻融损伤模型以评价盐-冻融循环条件下碳纳米纤维混凝土的损伤演化规律。结果表明,CNFs可提高混凝土抗盐-冻性能,改善效果与碳纳米纤维掺量呈正相关,质量掺量为0.3%时抗冻等级及耐久性指数较基准组提高100%。CNFs通过桥接裂缝、控制纳米级裂纹、成核、改善孔结构提高混凝土密实度,进而改善混凝土的抗盐-冻性能。基于不同评价指标建立的冻融损伤模型符合威布尔分布且R2均大于0.9,可用于评价和预测盐-冻融循环条件下碳纳米纤维混凝土损伤情况。

机理模型与遥感数据同化的草地生物量估算相关研究

受人为和气候变化的影响,草地退化、沙化的现象愈发严重,对草地的监测和保护亦变得愈发重要。生物量是评价草地生态状况的重要指标之一,草地生物量的估算对草地资源管理、生态修复与保护都有着重要的意义。随着遥感技术的发展与植被机理模型的广泛应用,针对草地生物量的估算方法也不断发展,主要分为遥感反演和植被机理模型估算两种方法。本文系统地阐述了遥感反演中的经验回归方法和辐射传输模型,还有基于植被机理的作物生长模型和草地生长模型。此外,本文还介绍了在农业估产领域已经被广泛应用的数据同化技术,分析了在草地生物量同化估算过程中,遥感数据与机理模型存在的互补优势,提出了目前存在的问题,并指出在简化模型的同时,保证估算的稳定性和准确性并兼顾植被的机理性是未来草地生物量估算的研究重点。

基于物理信息神经网络的生物质气化产物分布预测方法

机器学习方法已经在生物质气化建模中展现出广阔的应用前景。然而,机器学习模型主要依赖于实验数据,并不考虑气化中的反应机理,在数据样本不充分的情况下模型所表现出的实际关联特性与机理规律之间存在严重偏差。为此,提出一种基于物理信息神经网络(PINN)的生物质气化产物分布预测方法,该方法将真实实验数据与先验机理进行无缝衔接,在人工神经网络(ANN)模型中嵌入边界约束和关键参数间的单调性关系,通过自动微分技术进行辅助优化,实现模型的高效训练。结果表明:PINN模型的决定系数大于0.89,均方根误差小于4%,其总体预测精度要优于随机森林(RF)、支持向量机(SVM)和ANN 3种纯拟合机器学习模型;PINN模型能够严格服从边界约束和先验机理单调性关系,表现出更好的可解释性和泛化能力。

面向高温部件故障检测研究的某型燃气轮机模型

燃气轮机的高温部件故障检测,一般是通过涡轮出口的排温热电偶间接地监测。在利用燃气轮机数据分析不同干扰因素对排温的影响时,由于实际数据的限制,往往无法全面地描述不同因素对燃气轮机参数的影响,并且由于燃气轮机成本较高,无法利用破坏性实验分析燃气轮机故障对燃气轮机参数的影响。针对这些问题,一般的解决办法是建立燃气轮机模型,灵活控制各种变量,实现真实燃气轮机无法实现的实验内容和实验效果,并能反映典型部件故障对燃气轮机参数的影响。本文以某型燃气轮机为对象,建立了一个用于高温部件故障检测研究的燃气轮机模型,该模型具有多火焰筒、多涡轮通道结构,考虑了高温燃气在热通道中的掺混及旋转,能够反映典型故障和各种复杂扰动对排气温度分布的影响,并基于建立的模型分析了不同扰动对燃气轮机参数的影响,为后续研究提供理论依据。

华北平原冬小麦-夏玉米农田生态系统土壤自养和异养呼吸模型构建

土壤呼吸是陆地生态系统碳循环的重要过程,准确估算土壤呼吸对估算陆地生态系统碳源汇具有重要意义。通过在华北平原典型农田内开展土壤呼吸及其组分的原位观测实验,构建了适用于华北平原冬小麦-夏玉米轮种制农田生态系统的半经验半机理土壤异养呼吸和土壤自养呼吸模型。结果表明,冬小麦-夏玉米农田土壤异养呼吸模型可表达为土壤温度和土壤水分的函数,其中,土壤温度对土壤异养呼吸的影响适合用Arrhenius方程描述,而土壤水分的影响适合用对称的倒抛物线描述。验证表明,该模型的R^(2)和RMSE分别为0.68和0.52μmol m^(-2)s^(-1)。土壤自养呼吸模型包括维持呼吸和生长呼吸两个模块,其中,维持呼吸表达为土壤温度和叶面积指数的函数,其形式分别为Van′t Hoff指数方程和米氏方程;生长呼吸表达为总初级生产力与维持呼吸之差的线性函数。冬小麦季和夏玉米季土壤自养呼吸模型的结构相同,但是两种作物的模型参数差异较大。验证表明,冬小麦季土壤自养呼吸模型的R2和RMSE分别为0.64和0.50μmol m^(-2)s^(-1),夏玉米季土壤自养呼吸模型的R2和RMSE分别为0.67和0.37μmol m^(-2)s^(-1)。相比于不区分土壤异养呼吸和土壤自养呼吸的土壤总呼吸模型,本研究构建的土壤异养呼吸和土壤自养呼吸模型能够更加准确地模拟土壤呼吸的季节变化和年际变化过程,可为华北平原冬小麦-夏玉米轮种制农田生态系统的土壤呼吸估算提供方法依据。

机理模型与集成学习混合驱动的机器人关节摩擦建模方法

机器人一体化关节广泛应用于医疗、协作机器人等领域,其摩擦特性是影响机器人性能的关键因素.为此,提出了一种机理模型与集成学习混合驱动的机器人关节摩擦建模方法,以提高模型精度.首先,综合考虑转速、负载等关节摩擦特性的影响因素及其周期波动特性,基于先验知识和物理分析分别建立了伺服电机与谐波减速器的参数化机理模型,描述摩擦特性的变化规律.然后,针对机理建模中因线性假设、忽略高阶项等产生的非线性残差,提出了基于eXtreme gradient boosting(XGBoost)的残差补偿模型建模方法,通过采用Boosting集成学习策略,提高残差补偿模型的泛化能力.同时,采用贝叶斯优化方法进行XGBoost模型的超参数寻优,以提高模型精度和训练效率.相比于传统的参数化机理模型,本文所提出的混合驱动模型具有更高精度.与反向传播神经网络、支持向量机、长短时记忆神经网络等多种典型方法的对比实验表明,本文所提出的基于XGBoost的残差补偿模型具有更强的特征提取能力,能够较好地预测强非线性的波动摩擦残差,有效地提高了整体模型的精度.

半潜式油气生产平台闪蒸气压缩机组关键参数实时监测方法

针对半潜式油气生产平台闪蒸气压缩机组关键参数测量难的问题,提出一种基于虚拟计量的关键参数实时监测方法。对闪蒸气压缩机组的工艺原理进行深入分析,建立压缩机排气量、洗涤器排液量、后冷器换热效率指数三个关键参数的机理方程组,结合实际工业需求建立相应的虚拟计量模型,在此基础上开发基于虚拟计量的关键参数在线监测系统。在实际工业数据上的应用测试结果表明,所建立的模型能够有效地描述压缩机排气量、洗涤器排液量和后冷器换热效率的变化趋势,解决了关键参数虚拟计量问题。

Multi-scale observation and cross-scale mechanistic modeling on terrestrial ecosystem carbon cycle

To predict global climate change and to implement the Kyoto Protocol for stabilizing atmospheric greenhouse gases concentrations require quantifying spatio-temporal variations in the terrestrial carbon sink accurately. During the past decade multi-scale ecological experiment and observation networks have been established using various new technologies (e.g. controlled environmental facilities, eddy covariance techniques and quantitative remote sensing), and have obtained a large amount of data about terrestrial ecosystem carbon cycle. However, uncertainties in the magnitude and spatio-temporal variations of the terrestrial carbon sink and in understanding the underlying mechanisms have not been reduced significantly. One of the major reasons is that the observations and experiments were conducted at individual scales independently, but it is the interactions of factors and processes at different scales that determine the dynamics of the terrestrial carbon sink. Since experiments and observations are always conducted at specific scales, to understand cross-scale interactions requires mechanistic analysis that is best to be achieved by mechanistic modeling. However, mechanistic ecosystem models are mainly based on data from single-scale experiments and observations and hence have no capacity to simulate mechanistic cross-scale interconnection and interactions of ecosystem processes. New-generation mechanistic ecosystem models based on new ecological theoretical framework are needed to quantify the mechanisms from micro-level fast eco-physiological responses to macro-level slow acclimation in the pattern and structure in disturbed ecosystems. Multi-scale data-model fusion is a recently emerging approach to assimilate multi-scale observational data into mechanistic, dynamic modeling, in which the structure and parameters of mechanistic models for simulating cross-scale interactions are optimized using multi-scale observational data. The models are validated and evaluated at different spatial and temporal scales and real-time observational data are assimilated continuously into dynamic modeling for predicting and forecasting ecosystem changes realistically. in summary, a breakthrough in terrestrial carbon sink research requires using approaches of multi-scale observations and cross-scale modeling to understand and quantify interconnections and interactions among ecosystem processes at different scales and their controls over ecosystem carbon cycle.

基于合成孔径雷达数据的农作物长势监测研究进展

[目的/意义]农作物长势监测能及时提供农作物的生长状态信息,对于加强中国作物生产管理、确保国家粮食安全具有重要的意义。卫星遥感技术的发展为大面积的作物长势监测提供了契机。然而,在雨热同期的作物生长旺季,光学遥感数据的获取经常受到天气的限制。因此,近年微波雷达遥感技术受到了广泛重视。[进展]梳理了利用合成孔径雷达(Synthetic Aperture Radar,SAR)数据进行农作物长势监测的国内外研究现状,从农作物长势SAR遥感监测指标、农作物长势SAR遥感监测数据和农作物长势SAR遥感监测方法3个方面对基于SAR数据农作物长势监测研究进展与标志性成果进行总结。在分析常用于农作物长势监测的方法及其适用性的基础上,对它们在长势监测中应用情况进行分析。[结论/展望]提出了4个国内外SAR监测农作物长势所存在的问题:1)基于SAR数据的农作物长势监测方法研究整体较少;2)微波散射特征挖掘不够,特别是对极化分解参数的长势监测应用研究还有待深入;3)针对农作物长势监测中的雷达植被指数相对较少,其应用尚未得到充分发挥;4)基于SAR散射强度的农作物长势监测主要采用经验模型,难以推广到不同地区和类型的农作物上。最后,展望未来的研究应聚焦于挖掘微波散射特征、利用SAR极化分解参数、发展和优化雷达植被指数以及深化散射模型来监测农作物长势。

电潜泵气液两相流工况水力增压性能预测模型

电潜泵自20世纪以来广泛应用于海上和非常规油气田,来提高油井的生产效率,是一种革命性的油田生产方法,主要适用于自然压力和能量不足的井。潜水离心泵最初是用于抽取矿井中积水而开发的。阿迈斯·阿鲁图诺夫开发了第一台用于油井生产的电潜泵。此后电潜泵逐渐在石油行业中广为人知并流行起来。然而,电潜泵对流动条件非常敏感,为了提高其耐受性、效率、可靠性以及适应恶劣和复杂的井下流动条件,电潜泵经历了许多重大的技术改进。目前电潜泵通常由电动机、密封和一系列离心泵级组成,非常适合于高产的深井和偏斜井中,也常用于非常规井,如页岩油井。因此,电潜泵的耐气性、粘度影响以及长期开采能力已经引起了重大关注。由于其内部流动规律复杂,两相混输风险高,高粘/气液两相流工况下耐受极限和长效开采能力受限,严重制约了电潜泵在深水和非常规油气田安全高效开采中的运用。本文基于欧拉方程,利用最佳流速概念推导回流、摩擦和泄漏等损失,提出了适用于井下旋转电潜泵复杂工况下增压性能预测的理论模型。针对气液两相流动,该模型基于离心场中两相的力学分析,建立旋转叶轮内部体含气率模型,并根据流型选择曳力系数,计算两相滑移效应,修正混合相密度,进而提升扬程预测的准确性。模型预测和实验数据的对比验证了该模型预测的精度和可靠性均高于文献中常见的计算方法。本文所提出的方法可准确预测油的粘度、含水乳化液和气液流对泵性能的影响。该模型可帮助泵工程师开发新的电潜泵几何形状,同时帮助人工举升工程师改进油井完井举升工艺设计。

Dynamic prediction of building subsidence deformation with data-based mechanistic self-memory model

This paper describes a building subsidence deformation prediction model with the self-memorization principle.According to the non-linear specificity and monotonic growth characteristics of the time series of building subsidence deformation,a data-based mechanistic self-memory model considering randomness and dynamic features of building subsidence deformation is established based on the dynamic data retrieved method and the self-memorization equation.This model first deduces the differential equation of the building subsidence deformation system using the dynamic retrieved method,which treats the monitored time series data as particular solutions of the nonlinear dynamic system.Then,the differential equation is evolved into a difference-integral equation by the self-memory function to establish the self-memory model of dynamic system for predicting nonlinear building subsidence deformation.As the memory coefficients of the proposed model are calculated with historical data,which contain useful information for the prediction and overcome the shortcomings of the average prediction,the model can predict extreme values of a system and provide higher fitting precision and prediction accuracy than deterministic or random statistical prediction methods.The model was applied to subsidence deformation prediction of a building in Xi'an.It was shown that the model is valid and feasible in predicting building subsidence deformation with good accuracy.