Machine learning-assisted design of Al2O3–SiO2 porous ceramics based on few-shot datasets

A machine learning model was proposed to accelerate the preparation of Al₂O₃–SiO₂ porous ceramics (ASPC) based on few-shot datasets. Phosphate tailings and bauxite were used as raw materials to prepare ASPC and obtain the initial dataset. The accuracy and generalization ability of the model are heavily influenced by the dataset partitioning method. 10-fold cross-validation and random sampling were compared to identify the optimal approach for training and testing the model on the limited data. Four commonly used regression algorithms of RF, KNN, SVR and BPNN were selected to predict the mechanical properties of ASPC. Based on the evaluation of model performance, the random forest algorithm was found to be the most suitable for small sample datasets. Using the trained random forest model, the feature importance of ASPC performance was analyzed, and a new ASPC database was established. Experimental schemes with the desired performance were obtained through screening from ASPC database. Finally, the ASPC with the target performance was prepared experimentally. The result demonstrated that employing machine learning techniques on few-shot datasets could effectively tackle the issues of low efficiency and insufficient data samples.     

Ensemble locally weighted partial least squares as a just‐in‐time modeling method

The predictive ability of soft sensors, which estimate values of an objective variable y online, decreases due to process changes in chemical plants. To reduce the decrease of predictive ability, adaptive soft sensors have been developed. We focused on just‐in‐time soft sensors, especially locally weighted partial least squares (LWPLS) regression. Since a set of hyperparameters in an LWPLS model has to be set beforehand and there is only onedataset, a traditional LWPLS model is difficult to accurately predict y‐values in multiple process states. In this study, we propose to combine LWPLS and ensemble learning, and predict y‐values with multiple LWPLS models, whose datasets and sets of hyperparameters are different. The weights of LWPLS models are determined based on Bayes’ theorem, considering their predictive ability. We confirmed that the proposed model has higher predictive accuracy than traditional models through numerical simulation data and two industrial data analyses. © 2015 American Institute of Chemical Engineers AIChE J, 62: 717–725, 2016     

Identification and Online Validation of a pH Neutralization Process Using an Adaptive Network-Based Fuzzy Inference System

In this study, the application of adaptive neuro-fuzzy inference system (ANFIS) architecture to build prediction models that represent the pH neutralization process is proposed. The dataset used to identify the process was obtained experimentally in a bench scale plant. The prediction model attained was validated offline and online and demonstrated as able to precisely predict the one step-ahead value of effluent pH leaving the neutralization reactor. The input variables were the current and one past value of the acid and base flow rates and the current value of the output variable. Variance accounted for (VAF) indices greater than 99% were achieved by the model in experiments in which the disturbances in the acid and basic solutions flow rates were applied separately. For tests with simultaneous disturbances, conditions never seen in the training and suffering from reactor level oscillations, the prediction model VAF index was still approximately 96%. The validations demonstrated the capability of ANFIS to build precise fuzzy models from input–output datasets. R² values achieved were always larger than 0.96.     

Prediction of Anticancer Peptides with High Efficacy and Low Toxicity by Hybrid Model Based on 3D Structure of Peptides

Recently, anticancer peptides (ACPs) have emerged as unique and promising therapeutic agents for cancer treatment compared with antibody and small molecule drugs. In addition to experimental methods of ACPs discovery, it is also necessary to develop accurate machine learning models for ACP prediction. In this study, features were extracted from the three-dimensional (3D) structure of peptides to develop the model, compared to most of the previous computational models, which are based on sequence information. In order to develop ACPs with more potency, more selectivity and less toxicity, the model for predicting ACPs, hemolytic peptides and toxic peptides were established by peptides 3D structure separately. Multiple datasets were collected according to whether the peptide sequence was chemically modified. After feature extraction and screening, diverse algorithms were used to build the model. Twelve models with excellent performance (Acc > 90%) in the ACPs mixed datasets were used to form a hybrid model to predict the candidate ACPs, and then the optimal model of hemolytic peptides (Acc = 73.68%) and toxic peptides (Acc = 85.5%) was used for safety prediction. Novel ACPs were found by using those models, and five peptides were randomly selected to determine their anticancer activity and toxic side effects in vitro experiments.     

Intelligent Modeling and Analysis of Moisture Sorption Isotherms in Milk and Pearl Millet–Based Weaning Food “Fortified Nutrimix”

Soft computing–based intelligent models have been proposed to predict moisture sorption isotherms in milk and pearl millet–based weaning food, “fortified Nutrimix,” at four temperatures, 15, 25, 35, and 45°C over the water activity range 0.11–0.97. Connectionist and adaptive neuro-fuzzy inference system (ANFIS) models were investigated. A back-propagation algorithm with Bayesian regularization/Levenberg-Marquardt optimization mechanisms was employed to develop connectionist models. The ANFIS model was based on the Sugeno-type fuzzy inference system. In addition, several empirical models were explored for fitting the sorption data. The soft computing models, in particular, ANFIS, outperformed the conventional sorption models for predicting isotherms in Nutrimix.     

Application of online support vector regression for soft sensors

Soft sensors have been widely used in chemical plants to estimate process variables that are difficult to measure online. One of the crucial difficulties of soft sensors is that predictive accuracy drops due to changes in state of chemical plants. Characteristics of adaptive soft sensor models such as moving window models, just‐in‐time models and time difference models were previously discussed. The predictive accuracy of any traditional models decreases when sudden changes in processes occur. Therefore, a new soft sensor method based on online support vector regression (SVR) and the time variable was developed for constructing soft sensor models adaptive to rapid changes of relationships among process variables. A nonlinear SVR model with the time variable is updated with the most recent data. The proposed method was applied to simulation data and real industrial data, and achieved higher predictive accuracy than traditional ones even when time‐varying changes in process characteristics happen. © 2013 American Institute of Chemical Engineers AIChE J 60: 600–612, 2014     

IN SITU COMBUSTION IN ENHANCED OIL RECOVERY (EOR): A REVIEW

Enhanced oil recovery (EOR) refers to the technologies developed to increase extraction of crude oil from reservoirs after primary production. In situ combustion (ISC) is one of the methods developed for EOR. This review examines studies done by researchers worldwide to improve our understanding of the mechanism of oil cracking kinetics, which is one of the fundamental mechanisms of in situ combustion. Good agreement between the laboratory and field results has encouraged further research in this field. Extensive research at the laboratory scale to understand the pyrolysis and oxidation behavior of coke formed from medium and light oil and also to propose more realistic models to mimic the true behavior of in situ combustion has been undertaken in recent years. Apart from the classical Arrhenius model, researchers have come up with other models (two-step oxidation model) based on the type of combustion activity observed from their samples, thus modeling the process more accurately. Research work showing optimization of the parameters of ISC and improving the economic viability of the entire process is been one of the main focuses of this article. The review also explains the nature of the various experiments, sheds light on some of the concepts that remain unexplained, and opens the way for fresh thinking in those areas. It also highlights the possibility of developing global solutions for numerical simulation of this EOR process.     

IN SITU COMBUSTION IN ENHANCED OIL RECOVERY (EOR): A REVIEW

Enhanced oil recovery (EOR) refers to the technologies developed to increase extraction of crude oil from reservoirs after primary production. In situ combustion (ISC) is one of the methods developed for EOR. This review examines studies done by researchers worldwide to improve our understanding of the mechanism of oil cracking kinetics, which is one of the fundamental mechanisms of in situ combustion. Good agreement between the laboratory and field results has encouraged further research in this field. Extensive research at the laboratory scale to understand the pyrolysis and oxidation behavior of coke formed from medium and light oil and also to propose more realistic models to mimic the true behavior of in situ combustion has been undertaken in recent years. Apart from the classical Arrhenius model, researchers have come up with other models (two-step oxidation model) based on the type of combustion activity observed from their samples, thus modeling the process more accurately. Research work showing optimization of the parameters of ISC and improving the economic viability of the entire process is been one of the main focuses of this article. The review also explains the nature of the various experiments, sheds light on some of the concepts that remain unexplained, and opens the way for fresh thinking in those areas. It also highlights the possibility of developing global solutions for numerical simulation of this EOR process.     

Grade efficiency for sieve classification processes

Modelling of the screening performance for classification processes is important to obtain a first estimate for a new process in the planning phase. In this work especially the grade efficiency curves of sieve classifications with vibrating screens were examined. A sensitivity study was performed by changing the operating parameters of the sieving machine and the parameters of the charging material (i.e. mass flow, particle size, etc.). The aim was to correlate the input parameters with the grade efficiency curve of the classification process. The main aspect of the presented work is to find an appropriate method to adjust four screening parameters in a way that the measured grade efficiency curve is modelled correctly. Several methods for this adjustment step are reviewed. A sensitivity study using a tumbling screen was performed previously. It is apparent that for that study, different methods and models for the parameter adjustment need to be used. Furthermore it is shown that data reconciliation is necessary, since the mass balance of the particle streams may not be closed correctly. In summary this work is the first step to predict the screening performance of a sieving machine without material‐ and time‐consuming experiments.     

Fuzzy Inference System for Modeling of Zinc Removal Using Micellar-Enhanced Ultrafiltration

Discrete event computer simulation is one of the most widely used modeling tools for production systems. The major objective of this study is to develop a fuzzy rule-based model for the evaluation of micellar-enhanced ultrafiltration (MEUF) performance. In the current work, the foldover fractional factorial designs were applied as a screening experiment to determine all the influential factors affect Zn²⁺ rejection and permeate flux response. According to analysis of the variance (ANOVA) a three-screened significant factors, three-level full factorial designs (3 ^k ) was applied. Finally, a fuzzy model has been developed to predict and calculate the output variables. These mathematical models are found to be reliable predictive tools with an excellent accuracy with AARE ±1.08%, ±3.75%, in comparison with experimental values for permeate flux and rejection, respectively. It was observed that there is acceptable agreement between and fuzzy model results with experimental data.     

Prediction of thermodynamic properties of polyvinylpyrrolidone solutions by artificial neural networks coupled with genetic algorithm

In the present study, the artificial neural networks coupled with the genetic algorithm (ANN–GA) models were used to predict the thermodynamic properties of polyvinylpyrrolidone (PVP) solutions in water and ethanol at various temperatures, mass fractions, and molecular weights of polymer. The genetic algorithm (GA) was used to find the best weights and biases of the network and improve the performance of ANNs. The proposed model was composed of three input variables including the temperature of the solution, the mass fraction, and molecular weight of the polymer. Density, viscosity, and surface tension of PVP solutions with various molecular weights (10,000, 25,000, and 40,000) in water and ethanol have been measured in the temperature range 20–55°C and various mass fractions of polymer. The ANN–GA models were trained by the experimental datasets and the prediction of density, surface tension, and viscosity of PVP solutions was performed using these models. The predicted values were compared with the experimental ones and the mean absolute relative error was less than 0.5% for the density and surface tension and about 3% for the viscosity of solutions.     

Introduction of a generally applicable method to estimate retrieval of active molecules for similarity searching using fingerprints

Fingerprints are bit string representations of molecular structure and properties and are among the most widely used computational tools for similarity searching and database screening. Various fingerprint designs are available and their search performance is in general strongly dependent on the compound classes under study and the chemical characteristics of screening databases. Currently, it is not possible to predict the probability of identifying novel hits through fingerprint searching. However, for practical applications, such estimations would be very useful because one might be able, for example, to prioritize fingerprints and compound selection strategies or decide whether or not a similarity search campaign with subsequent experimental evaluation of candidate compounds would be promising at all. We have developed a method that makes it possible to predict the outcome of similarity search calculations using any type of keyed fingerprint. The methodology incorporates bit frequency distributions of reference molecules and the screening database into an information-theoretic function and determines the principally possible recall of active compounds within selection sets of varying size. We calibrate the function on diverse compound classes and accurately predict compound recovery in retrospective virtual screening trials. Furthermore, we correctly predict fingerprint search performance on two experimental high-throughput screening data sets (HTS). Our findings indicate that given a set of reference molecules, a fingerprint, and a screening database, we can readily estimate how likely it will be to retrieve active compounds, without knowledge about the distribution of potential hits in the database.     

Pollutant monitoring in tail gas of sulfur recovery unit with statistical and soft computing models

In this article, data-driven models are developed for real time monitoring of sulfur dioxide and hydrogen sulfide in the tail gas stream of sulfur recovery unit (SRU). Statistical [partial least square (PLS), ridge regression (RR) and Gaussian process regression (GPR)] and soft computing models are constructed from plant data. The plant data were divided into training and validation sets using Kennard-Stone algorithm. All models are developed from the training data set. PLS model is designed using SIMPLS algorithm. Three different ridge parameter selection techniques are used to design the RR model. GPR model is designed using four hyper parameter selection methods. The soft computing models include fuzzy and neuro-fuzzy models. Prediction accuracy of all models is assessed by simulation with validation dataset. Simulation results show that the GPR model designed with marginal log likelihood maximization method has good prediction accuracy and outperforms the performance of all other models. The developed GPR model is also found to yield better prediction accuracy than some other models of the SRU proposed in the literature.     

基于MIC的支持向量回归及其在化工过程中的应用

在化工生产中,软测量方法可以有效解决某些关键变量由于仪表故障而无法实时获取数据的问题。在建立软测量模型时,变量及回归方法的选取会直接影响模型的准确率。特别是在现代化工中,过程变量众多且变量间存在着冗余且复杂的非线性关系。对此,本文提出了一种基于最大信息系数的支持向量回归算法,利用最大信息系数在非线性相关性度量的优势,选择合适的辅助变量,避免了全部变量作为输入所造成的数据冗余。在此基础上,利用支持向量回归方法建立软测量模型,实现对软测量目标的预测。该方法被应用于存在仪表故障的某催化重整装置进料换热器热端压降的软测量中,结果表明该方法可以有效地实现对压降的软测量,实现了对仪表故障时的数据校正。     

ADAPTIVE FUZZY MODEL BASED PREDICTIVE CONTROL OF AN EXOTHERMIC BATCH CHEMICAL REACTOR*

An adaptive fuzzy model based predictive control (AFMBPC) approach is presented to track the desired temperature trajectories in an exothermic batch chemical reactor. The AFMBPC incorporates an adaptive fuzzy modeling framework into a model based predictive control scheme to derive analytical controller output. This approach has the flexibility to cope with different fuzzy model structures whose choice also lead to improve the controller performance. In this approach, adaptation of fuzzy models using dynamic process information is carried out to build a predictive controller, thus eliminating the determination of a predefined fixed fuzzy model based on various sets of known input-output relations. The performance of the AFMBPC is evaluated by comparing to a fixed fuzzy model based predictive controller (FFMBPC) and a conventional PID controller. The results show the better suitability of AFMBPC for the control of highly nonlinear and time varying batch chemical reactors.     

Improved graph-based multitask learning model with sparse sharing for quantitative structure–property relationship prediction of drug molecules

The quantitative structure–property relationship (QSPR) is a fundamental technique for evaluating and screening potentially valuable molecules in the field of drug discovery. There is an urgent need to speed up pharmaceutical research and development and a huge chemical space to explore, which necessitate effective and precise computer-aided QSPR modeling methods. Previous studies with various deep learning models are limited because they are trained on separate small datasets, known as the small-sample problem. Using transfer learning, this article describes a sparse sharing method that uses advanced graph-based models to construct an efficient and reasonable multitask learning workflow for QSPR prediction. The proposed workflow is systematically and comprehensively tested with four benchmark datasets containing different targets, and several precisely predicted molecular examples are illustrated. The results demonstrate that an obvious improvement in the prediction of molecular properties is achieved, along with the ability to predict multiple properties simultaneously.     

基于JIT-MOSVR的软测量方法及应用

针对传统多模型软测量方法在面对复杂、多变工况时缺少在线更新机制、更新时输出精度降低等问题,提出了一种基于即时学习算法（JIT）的多模型在线软测量方法（MOSVR）。离线阶段首先采用模糊C均值聚类（FCM）对训练数据进行聚类,接着采用SVR建立初始模型集。在线部分以多模型输出作为主要输出,当出现新工况时,通过在线模型更新策略（OSMU）将输出模式切换为JIT,同时多模型集进行在线更新。该方法不仅拥有多模型输出的快速性、精确性,而且在模型更新时通过JIT模式还能保证输出的连续性、稳定性、精确性。最后将该软测量方法进行了数值仿真并运用到乙烷浓度软测量中,验证了该方法的有效性。     

基于FCMISVM的自适应软测量建模算法及其在PX吸附分离过程中的应用

To overcome the problem that soft sensor models cannot be updated with the process changes, a soft sensor modeling algorithm based on hybrid fuzzy c-means (FCM) algorithm and incremental support vector machines (ISVM) is proposed. This hybrid algorithm FCMISVM includes three parts: samples clustering based on FCM algorithm, learning algorithm based on ISVM, and heuristic sample displacement method. In the training process, the training samples are first clustered by the FCM algorithm, and then by training each clustering with the SVM algorithm, a sub-model is built to each clustering. In the predicting process, when an incremental sample that represents new operation information is introduced in the model, the fuzzy membership function of the sample to each clustering is first computed by the FCM algorithm. Then, a corresponding SVM sub-model of the clustering with the largest fuzzy membership function is used to predict and perform incremental learning so the model can be updated on-line. An old sample chosen by heuristic sample displacement method is then discarded from the sub-model to control the size of the working set. The proposed method is applied to predict the p-xylene (PX) purity in the adsorption separation process. Simulation results indicate that the proposed method actually increases the model’s adaptive abilities to various operation conditions and improves its generalization capability.     

甲醇制芳烃K-means-PSO-SVR局部建模及优化

针对甲醇制芳烃（MTA）过程数据样本趋同、维度高、非线性、强耦合、局部差异大的特性,提出了一种K-means-PSO-SVR的局部建模方法,用以解决单一全局模型预测精度低,鲁棒性不强的问题。该方法首先用K-means算法对样本空间的数据进行聚类,实现对样本空间k个区域的划分,再用经过粒子群优化算法（PSO）优化过超参数的支持向量回归算法（SVR）在划分好的样本空间上建立相互独立的局部模型,最终将建立的k个相互独立的局部模型组合起来组成覆盖整个样本空间的集成模型。在不同噪声水平下将K-means-PSO-SVR方法的建模效果与单一全局SVR、BP神经网络和线性回归3种算法的建模效果进行了比较分析,结果表明：K-means-PSO-SVR局部建模方法的性能在所有水平的噪声下都明显优于其他3种建模方法,并且该方法对噪声具有很强的鲁棒性。以建立的数据模型为基础优化了两段式固定床甲醇制芳烃的关键工艺参数,并用5次独立重复实验验证了优化结果的可靠性,得出当一段温度为446.2℃、二段温度为467.3℃、甲醇体积空速为0.4h^-1、压力为0.64MPa时反应产物中苯、甲苯和二甲苯（BTX）的总收率最高,最高收率为44.30%。