Industrial Debutanizer Column Research Articles

A knowledge-refined hybrid graph model for quality prediction of industrial processes

The complexity of industrial processes has spurred the application of soft sensor techniques for predicting key quality variables based on easy-measurable process variables. Currently, data-driven soft sensors based on Artificial Intelligence techniques have become the mainstream. However, these soft sensing models deeply rely on the quality of training data, where the domain knowledge is often ignored. Meanwhile, a significant amount of labeled data is not fully utilized. To address these issues, this paper proposes a supervised framework based on a knowledge-refined hybrid graph network, which contributes to the artificial intelligence application of nonlinear dynamic soft sensors. The problems of applying traditional artificial intelligence models in soft sensor have been addressed by reconstructing the input module of graph neural networks with knowledge-guided approaches. Both spatial and temporal correlations of process data are captured and the hybrid network significantly improves the reliability and interpretability of the soft sensing model. By incorporating labeled data into the model, the representation of quality information is also enhanced. Finally, the proposed framework was applied to an industrial debutanizer column, and the experimental results fully demonstrate the effectiveness and superiority of the method.

Soft sensor modeling based on self-organizing fuzzy neural network with clustering, merging, and splitting scheme

Abstract As a pivotal role in the control, optimization, and monitoring of contemporary industrial processes, soft sensors are frequently employed in the prediction of key quality variables. To achieve accurate prediction of key quality variables in industrial processes, a soft sensor modeling method based on the self-organizing fuzzy neural network with the clustering, merging, and splitting scheme (SOFNN-CMS) is proposed. First, the supervised fuzzy C-means clustering algorithm is proposed to identify the appropriate initial center and width of the fuzzy neural network, obtaining appropriate initial fuzzy rules. Then, a neuron merging and splitting strategy is designed to adjust the structure of the fuzzy neural network, by merging and splitting the hidden neurons according to the distance of clusters, increasing the adaptability of the fuzzy neural network. Besides, to accelerate the convergence of estimation errors, an improved Levenberg Marquardt algorithm is utilized to update neural network parameters in the training phase, realizing the soft sensor modeling of key quality variables. The effectiveness of the proposed SOFNN-CMS neural network is demonstrated on two benchmark problems and an industrial debutanizer column. Finally, the experiments showcase that the proposed SOFNN-CMS neural network can obtain better soft sensor modeling performance with a compact structure.

A denoising and multiscale residual deep network for soft sensor modeling of industrial processes

Deep learning plays an important role in soft sensors of industrial processes for the timely measurement of key quality variables. However, since sensors are often operated under noisy and nonstationary industrial conditions, the collected industrial process data exhibit extreme complexity, which severely restricts the learning capacity and measurement accuracy of deep learning methods. In this paper, a novel denoising and multiscale residual deep network (DMRDN) is proposed for soft sensor modeling. Firstly, a stacked denoising autoencoder with level-aware attention is developed to denoise the process data, in which denoised features on different levels are learned and fused. Secondly, the denoised features are fed into multiscale residual convolutional neural network with scale-aware attention, which is designed to capture and fuse deep dynamic features from different scales. Finally, experiments were conducted on an industrial debutanizer column. The experimental results demonstrate that the proposed DMRDN greatly strengthens the learning ability and achieves better prediction performance compared with other methods.

Soft Sensor Design Using Multi-State Dependent Parameter Methodology Based on Generalized Random Walk Method

The main problem of developing data-driven soft sensors is the existence of contamination (i.e., outliers) and missing values in the industrial real-time data. In this paper, a new soft sensor modeling method has been extended using a generalized random walk model (GRW) in order to access a robust estimation of parameters in the presence of missing data and outliers. The method termed as generalized random walk-multi-state-dependent parameter (GRW-MSDP) was established based on MSDP models. The model parameters are estimated in multivariable state space by employing the Kalman filter (KF) and fixed-interval smoothing (FIS) algorithms. The Kalman filter has been applied to identify the best state estimation values and reduce the effect of outliers by assigning low weight to them. Although in the optimization of KF hyper-parameters the missing values are not taken into account, the FIS algorithm implements a predictor-corrector type estimator to handle the missing values. The prediction step of FIS can be used for interpolation directly without parameterization. The main privilege of the GRW-MSDP method is the not necessity of data pre-processing for fitting the best models. A simulation case and an industrial debutanizer column are utilized to illustrate the effectiveness and advantages of the proposed method. Results indicate that the non-linearity of the process can be addressed under this modeling method using fewer input variables and the change of the process is also well-tracked when missing values exist in the time series data. In addition, the GRW-MSDP method obtains significant improvements in the smoothing of parameters.

Intelligent Control of an Industrial Debutanizer Column

AbstractA debutanizer column located at a refinery in Malaysia produces LPG as its top product and light naphtha as its bottom product. Control of the product compositions of the column is very important to maintain the desired purities. The current control design of the debutanizer column is based on the classical proportional integral derivative approach, which has been found to be less effective in controlling the variations in the column as the process is characterized by strong nonlinear and uncertain dynamics. Here, intelligent control based on an adaptive neuro‐fuzzy inference system (ANFIS) is investigated and compared against an artificial neural network, for the special case of limited training data of the industrial debutanizer column. The ANFIS‐based controller outperforms the other control configurations for both set point tracking and disturbance rejection cases and has better generalization performance even when trained with limited data samples.

Quality Variable Prediction for Nonlinear Dynamic Industrial Processes Based on Temporal Convolutional Networks

Soft sensors have been extensively developed to estimate the difficult-to-measure quality variables for real-time process monitoring and control. Process nonlinearities and dynamics are two main challenges for accurate soft sensor modeling. To cope with these problems, two temporal convolutional network (TCN)-based soft sensor models are developed in this paper. With a hierarchy of temporal convolution kernels and large receptive fields, TCN is able to describe the nonlinearities and long dynamic dependence of process variables. Thus, a TCN model is first designed for nonlinear dynamic soft sensing. However, the original TCN model neglects the auto-correlations of the quality variable, as well as the cross-correlations between the quality variable and process variables. Hence, an autoregressive TCN (AR-TCN) is further constructed by integrating the lagged quality data in the past neighboring samples for quality prediction of the current sample. Thus, AR-TCN is able to capture the auto-correlations and cross-correlations between the quality variables and process variables, which are more beneficial for quality prediction. The effectiveness of two proposed models is validated on an industrial debutanizer column and an industrial hydrocracking process.

Prediction of industrial debutanizer column compositions using data-driven ANFIS- and ANN-based approaches

The work in this paper is based on an industrial debutanizer column in a petroleum refinery located in Malaysia, which produces LPG (liquefied petroleum gas) as the top stream and light naphtha as the bottom stream. The controlled outputs, which are the critical product quality to be measured, are the concentrations of the top and the bottom streams. It is required to maintain the product quality for profitability, and recent trends in soft-computing have seen a plethora of methods developed for this purpose. However, due to practical constraints experienced by the plant that limit extensive plant step tests, data available for the development of the soft sensor for this industrial debutanizer column is scarce. Hence, in this paper, a neuro-fuzzy technique (ANFIS) is investigated and compared against the widely used ANN model for the special case of limited training data samples of the industrial debutanizer column. It is observed that in comparison to ANN, ANFIS has better generalization performance and can work well with limited training data samples. From the results, it was observed that the prediction performance of the ANFIS model generally improved the RMSE by approximately 10 times in comparison to that of ANN.

Double-Level Locally Weighted Extreme Learning Machine for Soft Sensor Modeling of Complex Nonlinear Industrial Processes

In order to provide the stable and reliable prediction of key quality variables for industrial processes with complicated nonlinear characteristic, this paper proposes a Double-level Locally Weighted Extreme Learning Machine (DLWELM) based soft sensor modeling method. Firstly, a just-in-time learning based extreme learning machine framework (JITELM) is designed for adaptive nonlinear soft sensor modeling. Then, one double-level similarity measure methodology is presented with considering both the variable importance and the sample distance. At the first level, the mutual information between different input variables and output variable is computed for evaluating the importance of the input variables, which results in the variable weight coefficients. Furthermore, the variable-weighted sample distances are obtained at the second level, which are utilized to build the similarity measure for the relevant samples searching. Lastly, with the relevant samples as the modeling dataset, the locally weighted ELM (LWELM) model is developed, which assigns different weights to the training samples according to their double-level similarity values. Three cases including one numerical system, the industrial debutanizer column plant and the industrial polypropylene production process, are used to test the methods and the results demonstrate that the proposed DLWELM method has higher prediction precision compared to the basic ELM methods.

Soft sensor model for dynamic processes based on multichannel convolutional neural network

Soft sensors have been extensively used to predict the difficult-to-measure key quality variables. The robust soft sensors should be able to sufficiently extract the local dynamic and nonlinear features of process data for accurate prediction. Convolutional neural network (CNN) has shown powerful performance in local feature representation that is suitable for soft sensor modeling. However, the process variables that have a distant topological structure usually cannot be covered within the same convolution kernel when applying CNN to process data, which results in the fact that local correlations of those distant process variables are not captured. Therefore, a new multichannel CNN (MCNN) is proposed for various local dynamic feature representation. As a key step, a multichannel 3-D tensor is augmented for each sample as the input to the MCNN model. For the 3-D tensor, each channel has specific local correlations of certain variables, while the variables have different neighborhood relationships for different channels, which refer the various local correlations of different combination variables. Combining with the time axis of each channel, the various local dynamic correlations of different variable combinations can be learnt using MCNN regardless of their distance. The feasibility and effectiveness of MCNN-based soft sensor are demonstrated on the industrial debutanizer column and hydrocracking process.

Deep learning for quality prediction of nonlinear dynamic processes with variable attention‐based long short‐term memory network

AbstractIndustrial processes are often characterized with high nonlinearities and dynamics. For soft sensor modelling, it is important to model the nonlinear and dynamic relationship between input and output data. Thus, long short‐term memory (LSTM) networks are suitable for quality prediction of soft sensor modelling. However, they do not consider the relevance of different input variables with the quality variable. To address this issue, a variable attention‐based long short‐term memory (VA‐LSTM) network is proposed for soft sensing in this paper. In VA‐LSTM, variable attention is designed to identify important input variables according to their relevance with quality prediction. After that, different attention weights are calculated and assigned to further obtain a weighted input sample at each time step. Finally, the LSTM network is exploited to capture the long‐term dependencies of the weighted input time series to predict the quality variable. The performance of the proposed modelling method is validated on an industrial debutanizer column and a hydrocracking process.

Optimization of time‐variable‐parameter model for data‐based soft sensor of industrial debutanizer

SummaryThe enhancement of modern process control methods has caused the popularity of soft sensors in online quality prediction. It is significant to consider the reduction of model complexity, the performance increment, and decrement of input variables in soft sensor design, simultaneously. The aim of this paper is designing and applying a new data‐based soft sensor with minimum input variables for the enhancement of product quality estimation. Time‐varying‐parameter model by employing the Kalman filter and fixed interval smoothing algorithms has been developed to determine the dynamic transfer function and parameters setting based on time. A novel hybrid method with a dynamic autoregressive exogenous variable model and genetic algorithm has been presented for both state identification and parameter prediction. The combinatorial optimization problem has constructed based on a selection of input variables and an evaluation of Akaike information criterion as a fitness function. An industrial debutanizer column has been used for soft sensor performance validation. The result has indicated that the final soft sensor model in comparison to other presented soft sensing methods for this case has less complexity, fewer input variables, more robust and higher predictive performance. Due to fewer input variables, rapid convergence, and low complexity of this model, it can be efficient in industrial processes control, time‐saving, and improvement of quality prediction.

Hierarchical Quality-Relevant Feature Representation for Soft Sensor Modeling: A Novel Deep Learning Strategy

Deep learning is a recently developed feature representation technique for data with complicated structures, which has great potential for soft sensing of industrial processes. However, most deep networks mainly focus on hierarchical feature learning for the raw observed input data. For soft sensor applications, it is important to reduce irrelevant information and extract quality-relevant features from the raw input data for quality prediction. To deal with this problem, a novel deep learning network is proposed for quality-relevant feature representation in this article, which is based on stacked quality-driven autoencoder (SQAE). First, a quality-driven autoencoder (QAE) is designed by exploiting the quality data to guide feature extraction with the constraint that the potential features should largely reconstruct the input layer data and the quality data at the output layer. In this way, quality-relevant features can be captured by QAE. Then, by stacking multiple QAEs to construct the deep SQAE network, SQAE can gradually reduce irrelevant features and learn hierarchical quality-relevant features. Finally, the high-level quality-relevant features can be directly applied for soft sensing of the quality variables. The effectiveness and flexibility of the proposed deep learning model are validated on an industrial debutanizer column process.

Deep quality-related feature extraction for soft sensing modeling: A deep learning approach with hybrid VW-SAE

Soft sensors have been extensively used to predict difficult-to-measure quality variables for effective modeling, control and optimization of industrial processes. To construct accurate soft sensors, it is significant to carry out feature extraction for massive high-dimensional process data. Recently, deep learning has been introduced for feature representation in process data modeling. However, most of them cannot capture deep quality-related features for output prediction. In this paper, a hybrid variable-wise weighted stacked autoencoder (HVW-SAE) is developed to learn quality-related features for soft sensor modeling. By measuring the linear Pearson and nonlinear Spearman correlations for variables at the input layer with the quality variable at each encoder, a corresponding weighted reconstruction objective function is designed to successively pretrain the deep networks. With the constraint of preferential reconstruction for more quality-related variables, it can ensure that the learned features contain more information for quality prediction. Finally, the effectiveness of the proposed HVW-SAE based soft sensor method is validated on an industrial debutanizer column process.

Nonlinear Dynamic Soft Sensor Modeling With Supervised Long Short-Term Memory Network

Soft sensor has been extensively utilized in industrial processes for prediction of key quality variables. To build an accurate virtual sensor model, it is very significant to model the dynamic and nonlinear behaviors of process sequential data properly. Recently, a long short-term memory (LSTM) network has shown great modeling ability on various time series, in which basic LSTM units can handle data nonlinearities and dynamics with a dynamic latent variable structure. However, the hidden variables in the basic LSTM unit mainly focus on describing the dynamics of input variables, which lack representation for the quality data. In this paper, a supervised LSTM (SLSTM) network is proposed to learn quality-relevant hidden dynamics for soft sensor application, which is composed of basic SLSTM unit at each sampling instant. In the basic SLSTM unit, the quality and input variables are simultaneously utilized to learn the dynamic hidden states, which are more relevant and useful for quality prediction. The effectiveness of the proposed SLSTM network is demonstrated on a penicillin fermentation process and an industrial debutanizer column.

Probabilistic density-based regression model for soft sensing of nonlinear industrial processes

Process nonlinearity is a challenging issue for soft sensor modeling of industrial plants. Traditional nonlinear soft sensing methods are not achieved through the probabilistic manner, which only give single point estimation for output variables but do not provide the prediction uncertainty. To meet the probabilistic soft sensor requirement, a novel density-based regression method, which is called weighted Gaussian regression (WGR), is proposed in this paper. By taking the weights of training samples into consideration, a local weighted Gaussian model (WGM) is first built to model the joint density P(x, y) of input and output variables around the query sample. Then, the output variables can be estimated by taking the conditional distribution P(y|x). The new method can successfully approximate the nonlinear relationship between output and input variables. Moreover, WGR can provide more detailed information of uncertainty for the prediction. The effectiveness and flexibility of WGR are validated through a numerical example and an industrial debutanizer column process.

Data-driven soft sensor approach for online quality prediction using state dependent parameter models

The goal of this paper is to design and implementation of a new data-driven soft sensor that uses state dependent parameter (SDP) models to improve product quality monitoring. The SDP model parameters assumed to be function of the system states, which are estimated by data-based modeling philosophy and SDP method. Soft sensing performance of the proposed method is validated on a simulated continuous stirred tank reactor and an industrial debutanizer column. A comparative study of different soft sensing methods for online monitoring of debutanizer column is also carried out. The results show that the process non-linearity can also be addressed under this modeling method and the change of the process is also well tracked when missing data exist in the observed data. The results indicate that the new model is much more robust and reliable with less model parameters, which make it useful for industrial applications. In addition, the performance indexes show the superiority of the proposed model over other conventional soft sensing methods.

Adaptive soft sensor based on time difference Gaussian process regression with local time-delay reconstruction

Apart from strong nonlinearity and time-varying behaviors in industrial processes, the hidden time-delay information, which is unfortunately overlooked in most existing modeling methods, should also be taken into account in soft sensor modeling. In view of this, a novel soft sensor, referred to as local time-delay reconstruction based moving window time difference Gaussian process regression (LTR-MWTDGPR), is proposed in this paper. To deal with the time-delay, a fuzzy curve analysis based local time-delay parameter extraction procedure is performed along with a strategy of a moving window, which simultaneously captures the process time-varying feature. Then the local window training dataset and new query sample are reconstructed according to the time-delay parameter set at the next sampling instant. Afterwards, the time difference Gaussian process regression is employed to handle the drifting feature of local reconstructed dataset. The effectiveness and accuracy of the proposed LTR-MWTDGPR approach in predicting quality variables are verified through a real sulfur recovery unit and an industrial debutanizer column.

Supervised neighborhood preserving embedding for feature extraction and its application for soft sensor modeling

AbstractNeighborhood preserving embedding (NPE) is a useful tool for learning the manifold of high‐dimensional data. As a linear approximation of nonlinear locally linear embedding, NPE can be applied to dimensionality reduction by neighborhood preserving. However, the original NPE algorithm is an unsupervised method, which extracts features without any reference to the output information. In this paper, a supervised NPE framework is proposed for output‐related feature extraction in soft sensor applications. In the supervised NPE framework, the output information is utilized to guide the procedures for constructing the adjacent graph and calculating the weight matrix, with which the intrinsic structure of the data can be better described. For performance evaluation of the proposed method, experiments on a numerical example and an industrial debutanizer column process are carried out. The results show the effectiveness of the proposed framework.

Online soft sensor design using local partial least squares models with adaptive process state partition

We propose a soft sensing method using local partial least squares models with adaptive process state partition, referring to as the LPLS-APSP, which is capable of effectively handling time-varying characteristics and nonlinearities of processes, the two major adverse effects of common industrial processes that cause low-performance of soft sensors. In our proposed approach, statistical hypothesis testing is employed to adaptively partition the process state into the unique local model regions each consisting of certain number of consecutive-time data samples, and partial least squares is adopted to construct local models. Advantages of this adaptive strategy are that the number of local models does not need to be pre-defined and the local model set can be augmented online without retraining from scratch. Moreover, to improve the prediction accuracy, a novel online model adaptation criterion is proposed, which not only takes the current process dynamics into account, but also enables mining the information contained in the neighborhood of the query sample. The guidelines for tuning the model parameters are also presented. The LPLS-APSP scheme is applied to develop the dynamic soft sensors for a simulated continuous stirred tank reactor and a real industrial debutanizer column, and the results obtained demonstrate the effectiveness of this proposed approach, in comparison to several existing state-of-the-art methods, for online soft sensor design.

Performance improvement and efficiency enhancement of a debutanizer column (a case study in South Pars gas field)

Due to the growing demands for processes such as gas to liquid and steam-methane reforming, in which precious chemicals especially methanol and gasoline are produced, performance enhancement of the debutanizer columns has become the center of attention. Debutanizer column, that is one of the main units of refineries and petrochemical complexes, has the responsibility of separating C4 cuts and lighter components from natural gas liquids, typically gasoline. In this paper, an industrial debutanizer column was simulated applying a steady state flow sheet simulator in order to investigate possible sources of low-efficiency separation problem. In this regards, the reflux ratio and the feed tray location were changed theoretically to achieve higher performance of the column. The results clearly revealed that the 19th tray can be selected as the tray with the best separation performance. Moreover, the optimum reflux ratio was determined and it was proved that better separation can be achieved by applying both modifications simultaneously. In addition, the feedstock liquefaction and the optimum number of liquid furnaces were studied. The results demonstrated that applying three furnaces is more beneficial than one, two or four furnaces and feedstock liquefaction has a slight positive effect on the quality of the separation. In order to investigate whether the application of the optimum reflux ratio and three furnaces can practically improve the separation performance, the aforementioned conditions were subjected to the industrial debutanizer column. The results confirm the improvement of the separation process leading to substantial savings in energy and cost.