Dexian Huang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Dexian Huang has a total of 71 co-authors. Here are the top co-authors by number of shared publications:
- Chao Shang
Department of Automation, Tsinghua University, Beijing, China
12 shared publications - Fan Yang
Department of Automation, Tsinghua University, Beijing 100084, China
11 shared publications Yongheng Jiang
Beijing National Research Center for Information Science and Technology (BNRist), Beijing 100084, People’s Republic of China
10 shared publicationsXiaoyong Gao
Institute for Ocean Engineering, China University of Petroleum, Beijing 102249, China
5 shared publicationsChao Shang
Beijing National Research Center for Information Science and Technology (BNRist), Beijing 100084, People’s Republic of China
4 shared publications- Tao Chen
School of Chemistry and Chemical Engineering, University of Surrey, Guildford GU2 7XH, UK
4 shared publications
Latest Publications
- 24 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-12
A Novel Accuracy-Constrained Scheme for Efficient Trend Extraction of Industrial Time-Series Data
Data trend extraction provides a useful means to qualitatively capturing the underlying variations of time series data. A class of algorithms are built upon segmentation and piecewise polynomial fitting, where the model complexity is primarily controlled by the number of data segments. However, it is not trivial to specify when tackling datasets of different sizes, and expensive computations are required in current global optimization algorithms. To address these issues, we propose a novel data trend extraction and segmentation method based on accuracy-constrained polynomial fitting. Two normalized indices are coined to define constraints on the accuracy of piecewise polynomial fitting, which allows for an interpretable and clear tuning guideline to regulate model complexities of segmentation when facing data trajectories of different lengths. By exploiting the structure of the constrained fitting problem, a breadth-first search algorithm is established, with two branch pruning strategies designed to remarkably improve the solution efficiency. In particular, we prove that the proposed solution algorithm has a desirable O(n2) complexity that does not grow with the number of segments and is much lower than that of generic global optimization algorithms. Comprehensive case studies show that compared with conventional methods, our approach enjoys better empirical performance, easier tuning of parameters, and lower computational cost.
- 24 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-12
A Novel Accuracy-Constrained Scheme for Efficient Trend Extraction of Industrial Time-Series Data
Data trend extraction provides a useful means to qualitatively capturing the underlying variations of time series data. A class of algorithms are built upon segmentation and piecewise polynomial fitting, where the model complexity is primarily controlled by the number of data segments. However, it is not trivial to specify when tackling datasets of different sizes, and expensive computations are required in current global optimization algorithms. To address these issues, we propose a novel data trend extraction and segmentation method based on accuracy-constrained polynomial fitting. Two normalized indices are coined to define constraints on the accuracy of piecewise polynomial fitting, which allows for an interpretable and clear tuning guideline to regulate model complexities of segmentation when facing data trajectories of different lengths. By exploiting the structure of the constrained fitting problem, a breadth-first search algorithm is established, with two branch pruning strategies designed to remarkably improve the solution efficiency. In particular, we prove that the proposed solution algorithm has a desirable O(n2) complexity that does not grow with the number of segments and is much lower than that of generic global optimization algorithms. Comprehensive case studies show that compared with conventional methods, our approach enjoys better empirical performance, easier tuning of parameters, and lower computational cost.
- 19 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Control Systems Technology
- Vol. 33 (5) , 1921-1928
Causality-Informed Data-Driven Predictive Control
As a useful and efficient alternative to generic model-based control scheme, data-driven predictive control (DDPC) is subject to bias-variance tradeoff and is known to not perform desirably in face of uncertainty. Through the connection between direct data-driven control and subspace predictive control (SPC), we gain insight into the reason being the lack of causality as a main cause for their high variance of implicit prediction. In this brief, we derive a new causality-informed formulation of DDPC and its regularized form that balances between control cost minimization and implicit identification of a causal multistep predictor. Since the proposed causality-informed formulations only call for block-triangularization of a submatrix in the generic noncausal DDPC based on LQ factorization, our causality-informed formulation of DDPC enjoys computational efficiency. Its efficacy is investigated through numerical examples and application to model-free control of a simulated industrial heating furnace. Empirical results corroborate that the proposed method yields obvious performance improvement over existing formulations in handling stochastic noise and process nonlinearity.
- 19 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Control Systems Technology
- Vol. 33 (5) , 1921-1928
Causality-Informed Data-Driven Predictive Control
As a useful and efficient alternative to generic model-based control scheme, data-driven predictive control (DDPC) is subject to bias-variance tradeoff and is known to not perform desirably in face of uncertainty. Through the connection between direct data-driven control and subspace predictive control (SPC), we gain insight into the reason being the lack of causality as a main cause for their high variance of implicit prediction. In this brief, we derive a new causality-informed formulation of DDPC and its regularized form that balances between control cost minimization and implicit identification of a causal multistep predictor. Since the proposed causality-informed formulations only call for block-triangularization of a submatrix in the generic noncausal DDPC based on LQ factorization, our causality-informed formulation of DDPC enjoys computational efficiency. Its efficacy is investigated through numerical examples and application to model-free control of a simulated industrial heating furnace. Empirical results corroborate that the proposed method yields obvious performance improvement over existing formulations in handling stochastic noise and process nonlinearity.
- 19 December 2024
- Published by American Chemical Society (ACS) in Industrial & Engineering Chemistry Research
- Vol. 64 (1) , 564-574
Risk-Averse PID Tuning Based on Scenario Programming and Parallel Bayesian Optimization
- 19 December 2024
- Published by American Chemical Society (ACS) in Industrial & Engineering Chemistry Research
- Vol. 64 (1) , 564-574
Risk-Averse PID Tuning Based on Scenario Programming and Parallel Bayesian Optimization
- 15 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 22 (15455955) , 5614-5626
Dynamic-Inner White Component Analysis: A New Latent Variable Model for Dynamic Data Analytics and Process Monitoring
Recently, dynamic latent variable (DLV) models have been prevalent in dynamic data modeling and process monitoring. They maximize the covariance or canonical correlation between latent components and one-step ahead prediction thereof; however, auto-correlations may still be present in residuals, resulting in unmodeled dynamics and hence compromising the efficacy of generic variability-based monitoring indices. In this work, a novel dynamic-inner white component analysis (DiWCA) model is put forth. Its salient feature lies in that residuals of DLVs are enforced to be as “white” as possible to suppress unmodeled dynamics such that process dynamics can be adequately described. The whiteness index is used to quantify the closeness of finite-length innovation time series to white noise, based on which a bi-level program is formulated to extract DLVs. Thanks to the enforced whiteness, two new monitoring statistics can be constructed in DiWCA to effectively detect whether static variations become “colore” and thus yield interpretable information that complements generic variability-based indices. Finally, case studies on the Tennessee Eastman process and a real-world vacuum furnace system show that DiWCA is capable of adequately capturing process dynamics from data and providing deep insight into process operating status via two whiteness-based monitoring statistics. Note to Practitioners —Interpretability is a crucial aspect of statistic design in process monitoring schemes. Generic test statistics in DLV models are mostly constructed based upon variability of residuals, which are only a partial indicator of the existence of temporal dynamics. In the proposed DiWCA model, residuals are endowed with desirable noise-like properties so that new whiteness-based monitoring statistics can be constructed to testify whether the residuals are “colored”. This offers meaningful information about the existence of temporal dynamics in an interpretable way that well complements generic variability-based statistics. Furthermore, the integration of whiteness-based and variability-based monitoring statistics in DiWCA can provide a panoramic view of process operating status. Moreover, on realistic industrial data DiWCA shows better generalization capability than existing DLV models, indicating that process dynamics can be faithfully captured through the maximization of whiteness.
- 15 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 22 (15455955) , 5614-5626
Dynamic-Inner White Component Analysis: A New Latent Variable Model for Dynamic Data Analytics and Process Monitoring
Recently, dynamic latent variable (DLV) models have been prevalent in dynamic data modeling and process monitoring. They maximize the covariance or canonical correlation between latent components and one-step ahead prediction thereof; however, auto-correlations may still be present in residuals, resulting in unmodeled dynamics and hence compromising the efficacy of generic variability-based monitoring indices. In this work, a novel dynamic-inner white component analysis (DiWCA) model is put forth. Its salient feature lies in that residuals of DLVs are enforced to be as “white” as possible to suppress unmodeled dynamics such that process dynamics can be adequately described. The whiteness index is used to quantify the closeness of finite-length innovation time series to white noise, based on which a bi-level program is formulated to extract DLVs. Thanks to the enforced whiteness, two new monitoring statistics can be constructed in DiWCA to effectively detect whether static variations become “colore” and thus yield interpretable information that complements generic variability-based indices. Finally, case studies on the Tennessee Eastman process and a real-world vacuum furnace system show that DiWCA is capable of adequately capturing process dynamics from data and providing deep insight into process operating status via two whiteness-based monitoring statistics. Note to Practitioners —Interpretability is a crucial aspect of statistic design in process monitoring schemes. Generic test statistics in DLV models are mostly constructed based upon variability of residuals, which are only a partial indicator of the existence of temporal dynamics. In the proposed DiWCA model, residuals are endowed with desirable noise-like properties so that new whiteness-based monitoring statistics can be constructed to testify whether the residuals are “colored”. This offers meaningful information about the existence of temporal dynamics in an interpretable way that well complements generic variability-based statistics. Furthermore, the integration of whiteness-based and variability-based monitoring statistics in DiWCA can provide a panoramic view of process operating status. Moreover, on realistic industrial data DiWCA shows better generalization capability than existing DLV models, indicating that process dynamics can be faithfully captured through the maximization of whiteness.
- Open Access
Data-Driven Output Prediction and Control of Stochastic Systems: An Innovation-Based Approach
Recent years have witnessed a booming interest in data-driven control of dynamical systems. However, the implicit data-driven output predictors are vulnerable to uncertainty such as process disturbance and measurement noise, causing unreliable predictions and unexpected control actions. In this brief, we put forward a new data-driven approach to output prediction of stochastic linear time-invariant (LTI) systems. By utilizing the innovation form, the uncertainty in stochastic LTI systems is recast as innovations that can be readily estimated from input-output data without knowing system matrices. In this way, by applying the fundamental lemma to the innovation form, we propose a new innovation-based data-driven output predictor (OP) of stochastic LTI systems, which bypasses the need for identifying state-space matrices explicitly and building a state estimator. The boundedness of the second moment of prediction errors in closed-loop is established under mild conditions. The proposed data-driven OP can be integrated into optimal control design for better performance. Numerical simulations are carried out to demonstrate the outperformance of the proposed innovation-based methods in output prediction and control design over existing formulations.
Open AccessData-Driven Output Prediction and Control of Stochastic Systems: An Innovation-Based Approach
Recent years have witnessed a booming interest in data-driven control of dynamical systems. However, the implicit data-driven output predictors are vulnerable to uncertainty such as process disturbance and measurement noise, causing unreliable predictions and unexpected control actions. In this brief, we put forward a new data-driven approach to output prediction of stochastic linear time-invariant (LTI) systems. By utilizing the innovation form, the uncertainty in stochastic LTI systems is recast as innovations that can be readily estimated from input-output data without knowing system matrices. In this way, by applying the fundamental lemma to the innovation form, we propose a new innovation-based data-driven output predictor (OP) of stochastic LTI systems, which bypasses the need for identifying state-space matrices explicitly and building a state estimator. The boundedness of the second moment of prediction errors in closed-loop is established under mild conditions. The proposed data-driven OP can be integrated into optimal control design for better performance. Numerical simulations are carried out to demonstrate the outperformance of the proposed innovation-based methods in output prediction and control design over existing formulations.
- 7 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Control Systems Letters
- Vol. 7, 3639-3644
Data-Driven Predictive Control Using Closed-Loop Data: An Instrumental Variable Approach
Current data-driven predictive control (DDPC) methods heavily rely on data collected in open-loop operation with elaborate design of inputs. However, due to safety or economic concerns, systems may have to be under feedback control, where only closed-loop data are available. In this context, it remains challenging to implement DDPC using closed-loop data. In this paper, we propose a new DDPC method using closed-loop data by means of instrumental variables (IVs). We point out that the original DDPC fails to represent all admissible trajectories due to feedback control. By drawing from closed-loop subspace identification, the use of two forms of IVs is suggested to address this issue and the correlation between inputs and noise. Furthermore, a new DDPC formulation with a novel IV-inspired regularizer is proposed, where a balance between control cost minimization and weighted least-squares data fitting can be made for improvement of control performance. Numerical examples and application to a simulated industrial furnace showcase the improved performance of the proposed DDPC based on closed-loop data.
- 7 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Control Systems Letters
- Vol. 7, 3639-3644
Data-Driven Predictive Control Using Closed-Loop Data: An Instrumental Variable Approach
Current data-driven predictive control (DDPC) methods heavily rely on data collected in open-loop operation with elaborate design of inputs. However, due to safety or economic concerns, systems may have to be under feedback control, where only closed-loop data are available. In this context, it remains challenging to implement DDPC using closed-loop data. In this paper, we propose a new DDPC method using closed-loop data by means of instrumental variables (IVs). We point out that the original DDPC fails to represent all admissible trajectories due to feedback control. By drawing from closed-loop subspace identification, the use of two forms of IVs is suggested to address this issue and the correlation between inputs and noise. Furthermore, a new DDPC formulation with a novel IV-inspired regularizer is proposed, where a balance between control cost minimization and weighted least-squares data fitting can be made for improvement of control performance. Numerical examples and application to a simulated industrial furnace showcase the improved performance of the proposed DDPC based on closed-loop data.
- Open Access
Data-Driven Predictive Control Using Closed-Loop Data: An Instrumental Variable Approach
Current data-driven predictive control (DDPC) methods heavily rely on data collected in open-loop operation with elaborate design of inputs. However, due to safety or economic concerns, systems may have to be under feedback control, where only closed-loop data are available. In this context, it remains challenging to implement DDPC using closed-loop data. In this paper, we propose a new DDPC method using closed-loop data by means of instrumental variables (IVs). By drawing from closed-loop subspace identification, the use of two forms of IVs is suggested to address the closed-loop issues caused by feedback control and the correlation between inputs and noise. Furthermore, a new DDPC formulation with a novel IV-inspired regularizer is proposed, where a balance between control cost minimization and weighted least-squares data fitting can be made for improvement of control performance. Numerical examples and application to a simulated industrial furnace showcase the improved performance of the proposed DDPC based on closed-loop data.
Open AccessData-Driven Predictive Control Using Closed-Loop Data: An Instrumental Variable Approach
Current data-driven predictive control (DDPC) methods heavily rely on data collected in open-loop operation with elaborate design of inputs. However, due to safety or economic concerns, systems may have to be under feedback control, where only closed-loop data are available. In this context, it remains challenging to implement DDPC using closed-loop data. In this paper, we propose a new DDPC method using closed-loop data by means of instrumental variables (IVs). By drawing from closed-loop subspace identification, the use of two forms of IVs is suggested to address the closed-loop issues caused by feedback control and the correlation between inputs and noise. Furthermore, a new DDPC formulation with a novel IV-inspired regularizer is proposed, where a balance between control cost minimization and weighted least-squares data fitting can be made for improvement of control performance. Numerical examples and application to a simulated industrial furnace showcase the improved performance of the proposed DDPC based on closed-loop data.
- 15 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Control Systems Technology
- Vol. 31 (4) , 1787-1802
Efficient Relay Autotuner of Industrial Controllers via Rank-Constrained Identification of Low-Order Time-Delay Models
Relay feedback tuning has been a state-of-the-art autotuning technique in process control, where frequency-domain analysis is generically carried out using limit cycle data for model identification and controller tuning. However, due to noise and disturbance, nonnegligible approximation errors exist in frequency information distilled from data, resulting in model mismatch and a huge obstacle for automating the tuning procedure. In this article, we propose an efficient time-domain identification approach toward relay feedback autotuning. A new rank-constrained nonparametric formulation of model identification is proposed, which is in correspondence to low-order parametric transfer functions. Based on this, an efficient initialization strategy for nonparametric models is put forward using kernel-based regularization, which helps to circumvent local optima and eventually deliver better model estimates than conventional gradient-based approaches. Moreover, nonzero initial conditions can be tackled by means of regularization heuristics. Numerical examples and experiments on a real-world system showcase that the proposed identification approach not only improves accuracy but also effectively circumvents unfaithful models contradicting prior knowledge, thereby substantially benefiting the automated tuning of industrial controllers.
- 15 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Control Systems Technology
- Vol. 31 (4) , 1787-1802
Efficient Relay Autotuner of Industrial Controllers via Rank-Constrained Identification of Low-Order Time-Delay Models
Relay feedback tuning has been a state-of-the-art autotuning technique in process control, where frequency-domain analysis is generically carried out using limit cycle data for model identification and controller tuning. However, due to noise and disturbance, nonnegligible approximation errors exist in frequency information distilled from data, resulting in model mismatch and a huge obstacle for automating the tuning procedure. In this article, we propose an efficient time-domain identification approach toward relay feedback autotuning. A new rank-constrained nonparametric formulation of model identification is proposed, which is in correspondence to low-order parametric transfer functions. Based on this, an efficient initialization strategy for nonparametric models is put forward using kernel-based regularization, which helps to circumvent local optima and eventually deliver better model estimates than conventional gradient-based approaches. Moreover, nonzero initial conditions can be tackled by means of regularization heuristics. Numerical examples and experiments on a real-world system showcase that the proposed identification approach not only improves accuracy but also effectively circumvents unfaithful models contradicting prior knowledge, thereby substantially benefiting the automated tuning of industrial controllers.
- 7 November 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automatic Control
- Vol. 68 (9) , 5333-5348
From Generalized Gauss Bounds to Distributionally Robust Fault Detection With Unimodality Information
The need for exact distributions in probabilistic fault detection design is hardly fulfilled. The recent moment-based distributionally robust fault detection (DRFD) design secures robustness against inexact distributions but suffers from over-pessimism. To address this issue, we develop a new DRFD design scheme by using unimodality, a ubiquitous property of real-life distributions. To evaluate worst-case false alarm rates, a new generalized Gauss bound is first attained, which is less conservative than known Chebyshev bounds that underpin moment-based DRFD. This also yields analytical solutions to DRFD design problems, which are suboptimal but provably less conservative than known ones disregarding unimodality. A tightened Gauss bound is further derived by assuming bounded uncertainty, based on which convex programming approximation of DRFD problems is developed. Results on physical system data elucidate that the proposed DRFD design can reduce conservatism of moment-based ones by using unimodality information, and attaining a better robustness-sensitivity trade-off than prevalent data-centric design with moderate sample sizes.
- 7 November 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automatic Control
- Vol. 68 (9) , 5333-5348
From Generalized Gauss Bounds to Distributionally Robust Fault Detection With Unimodality Information
The need for exact distributions in probabilistic fault detection design is hardly fulfilled. The recent moment-based distributionally robust fault detection (DRFD) design secures robustness against inexact distributions but suffers from over-pessimism. To address this issue, we develop a new DRFD design scheme by using unimodality, a ubiquitous property of real-life distributions. To evaluate worst-case false alarm rates, a new generalized Gauss bound is first attained, which is less conservative than known Chebyshev bounds that underpin moment-based DRFD. This also yields analytical solutions to DRFD design problems, which are suboptimal but provably less conservative than known ones disregarding unimodality. A tightened Gauss bound is further derived by assuming bounded uncertainty, based on which convex programming approximation of DRFD problems is developed. Results on physical system data elucidate that the proposed DRFD design can reduce conservatism of moment-based ones by using unimodality information, and attaining a better robustness-sensitivity trade-off than prevalent data-centric design with moderate sample sizes.