arXiv subjects

John Bagterp Jørgensen

Publications and source records attributed to John Bagterp Jørgensen.

48 records · Page 3Linked to original sources

arXiv2022

State Estimation Methods for Continuous-Discrete Nonlinear Systems involving Stochastic Differential Equations

In this work, we present methods for state estimation in continuous-discrete nonlinear systems involving stochastic differential equations. We present the extended Kalman filter, the unscented Kalman filter, the ensemble Kalman filter, and a particle filter. We implement the state estimation methods in Matlab. We evaluate the performance of the methods on a simulation of the modified four-tank system. We implement the state estimation methods for non-stiff systems, i.e., using an explicit numerical integration scheme. The implementation of the extended Kalman filter utilises the Joseph stabilising form for numerical stability. We evaluate the accuracy of the state estimation methods in terms of the mean absolute percentage error over the simulation horizon. We show that each method successfully estimates the states and unmeasured disturbances of the simulated modified four-tank system. Finally, we present conclusions.

math.OC↗

arXiv2022

Large-scale Virtual Clinical Trials of Closed-loop Treatments for People with Type 1 Diabetes

We propose a virtual clinical trial for assessing the safety and efficacy of closed-loop diabetes treatments prior to an actual clinical trial. Such virtual trials enable rapid and risk-free pretrial testing of algorithms, and they can be used to compare different treatment variations for large and diverse populations. The participants are represented by multiple mathematical models, consisting of stochastic differential equations, and we use Monte Carlo closed-loop simulations to compute detailed statistics of the closed-loop treatments. We implement the virtual clinical trial using high-performance software and hardware, and we present an example trial with two mathematical models of one~million participants over 52~weeks (i.e., two~million simulations), which can be completed in 2~h 9~min.

math.OC↗

arXiv2022

A whole-body multi-scale mathematical model for dynamic simulation of the metabolism in man

We propose a whole-body model of the metabolism in man as well as a generalized approach for modeling metabolic networks. Using this approach, we are able to write a large metabolic network in a systematic and compact way. We demonstrate the approach using a whole-body model of the metabolism of the three macronutrients, carbohydrates, proteins and lipids. The model contains 7 organs, 16 metabolites and 31 enzymatic reactions. All reaction rates are described by Michaelis-Menten kinetics with an addition of a hormonal regulator based on the two hormones insulin and glucagon. We incorporate ingestion of food in order to simulate metabolite concentrations during the feed-fast cycle. The model can simulate several days due to the inclusion of storage forms (glycogen, muscle protein and lipid droplets), that can be depleted if food is not ingested regularly. A physiological model incorporating complex cellular metabolism and whole-body mass dynamics can be used in virtual clinical trials. Such trials can be used to improve the development of medicine, treatment strategies such as control algorithms, and increase the likelihood of a successful clinical trial.

q-bio.QM↗

arXiv2022

High-performance Uncertainty Quantification in Large-scale Virtual Clinical Trials of Closed-loop Diabetes Treatment

In this paper, we propose a virtual clinical trial for assessing the performance and identifying risks in closed-loop diabetes treatments. Virtual clinical trials enable fast and risk-free tests of many treatment variations for large populations of fictive patients (represented by mathematical models). We use closed-loop Monte Carlo simulation, implemented in high-performance software and hardware, to quantify the uncertainty in treatment performance as well as to compare the performance in different scenarios or of different closed-loop treatments. Our software can be used for testing a wide variety of control strategies ranging from heuristical approaches to nonlinear model predictive control. We present an example of a virtual clinical trial with one million patients over 52 weeks, and we use high-performance software and hardware to conduct the virtual trial in 1 h and 22 min.

math.OC↗

arXiv2022

Nonlinear Model Predictive Control and System Identification for a Dual-hormone Artificial Pancreas

In this work, we present a switching nonlinear model predictive control (NMPC) algorithm for a dual-hormone artificial pancreas (AP), and we use maximum likelihood estimation (MLE) to identify model parameters. A dual-hormone AP consists of a continuous glucose monitor (CGM), a control algorithm, an insulin pump, and a glucagon pump. The AP is designed with a heuristic to switch between insulin and glucagon as well as state-dependent constraints. We extend an existing glucoregulatory model with glucagon and exercise for simulation, and we use a simpler model for control. We test the AP (NMPC and MLE) using in silico numerical simulations on 50 virtual people with type 1 diabetes. The system is identified for each virtual person based on data generated with the simulation model. The simulations show a mean of 89.3% time in range (3.9-10 mmol/L) and no hypoglycemic events.

math.OC↗

arXiv2022

A one-size-fits-all artificial pancreas for people with type 1 diabetes based on physiological insight and feedback control

We propose a model-free artificial pancreas (AP) for people with type 1 diabetes. The algorithmic parameters are tuned to a virtual population of 1,000,000 individuals, and the AP repeatedly estimates the basal and bolus insulin requirements necessary for maintaining normal blood glucose levels. Therefore, the AP can be used without healthcare personnel or engineers customizing the algorithm to each user. The estimates are based on bodyweight, measurements from a continuous glucose monitor (CGM), and estimates of the meal carbohydrate contents. In a virtual clinical trial with all 1,000,000 individuals (i.e., a Monte Carlo closed-loop simulation), the AP achieves a mean time in range of more than 87% and almost 89% of the participants satisfy several glycemic targets.

math.OC↗

arXiv2019

Reduced Order Modeling for Nonlinear PDE-constrained Optimization using Neural Networks

Nonlinear model predictive control (NMPC) often requires real-time solution to optimization problems. However, in cases where the mathematical model is of high dimension in the solution space, e.g. for solution of partial differential equations (PDEs), black-box optimizers are rarely sufficient to get the required online computational speed. In such cases one must resort to customized solvers. This paper present a new solver for nonlinear time-dependent PDE-constrained optimization problems. It is composed of a sequential quadratic programming (SQP) scheme to solve the PDE-constrained problem in an offline phase, a proper orthogonal decomposition (POD) approach to identify a lower dimensional solution space, and a neural network (NN) for fast online evaluations. The proposed method is showcased on a regularized least-square optimal control problem for the viscous Burgers' equation. It is concluded that significant online speed-up is achieved, compared to conventional methods using SQP and finite elements, at a cost of a prolonged offline phase and reduced accuracy.

math.NA↗

arXiv2019

Filtering and Prediction of the Blood Glucose Concentration using an Android Smart Phone and a Continuous Glucose Monitor

In this paper we numerically assess the performance of Java linear algebra libraries for the implementation of nonlinear filters in an Android smart phone (Samsung A5 2017). We implemented a linear Kalman filter (KF), an extended Kalman filter (EKF), and an unscented Kalman filter (UKF). These filters are used for state and parameter estimation, as well as fault detection and meal detection in an artificial pancreas. We present the state estimation technologies used for glucose estimation based on a continuous glucose monitor (CGM). We compared three linear algebra libraries: The Efficient Java Matrix Library (EJML), JAMA and Apache Common Math. Overall, EJML provides the best performance for linear algebra operations. We demonstrate the implementation and performance of filtering (KF, EKF and UKF) using real CGM data.

math.OC↗

arXiv2018

Dynamic Optimization of Thermodynamically Rigorous Models of Multiphase Flow in Porous Subsurface Oil Reservoirs

In this paper, we consider dynamic optimization of thermal and isothermal oil recovery processes which involve multicomponent three-phase flow in porous media. We present thermodynamically rigorous models of these processes based on 1) conservation of mass and energy, and 2) phase equilibrium. The conservation equations are partial differential equations. The phase equilibrium problems that are relevant to thermal and isothermal models are called the UV and the VT flash, and they are based on the second law of thermodynamics. We formulate these phase equilibrium problems as optimization problems and the phase equilibrium conditions as the corresponding first order optimality conditions. We demonstrate that the thermal and isothermal flow models are in a semi-explicit differential-algebraic form, and we solve the dynamic optimization problems with a previously developed gradient-based algorithm implemented in C/C++. We present numerical examples of optimized thermal and isothermal oil recovery strategies and discuss the computational performance of the dynamic optimization algorithm in these examples.

math.OC↗

arXiv2018

A Family of ESDIRK Integration Methods

In this paper we derive and analyze the properties of explicit singly diagonal implicit Runge-Kutta (ESDIRK) integration methods. We discuss the principles for construction of Runge-Kutta methods with embedded methods of different order for error estimation and continuous extensions for discrete event location. These principles are used to derive a family of ESDIRK integration methods with error estimators and continuous-extensions. The orders of the advancing method (and error estimator) are 1(2), 2(3) and 3(4), respectively. These methods are suitable for obtaining low to medium accuracy solutions of systems of ordinary differential equations as well as index-1 differential algebraic equations. The continuous extensions facilitates solution of hybrid systems with discrete-events. Other ESDIRK methods due to Kværnø are equipped with continuous-extensions as well to make them applicable to hybrid systems with discrete events.

math.NA↗

arXiv2018

Risk minimization in life-cycle oil production optimization

The geology of oil reservoirs is largely unknown. Consequently, the reservoir models used for production optimization are subject to significant uncertainty. To minimize the associated risk, the oil literature has mainly used ensemble-based methods to optimize sample estimated risk measures of net present value (NPV). However, the success in reducing risk critically depends on the choice of risk measure. As a systematic approach to risk mitigation in production optimization, this paper characterizes proper risk measures by the axioms of coherence and aversion. As an example of a proper measure, we consider conditional value-at-risk, $\text{CVaR}_α$, at different risk levels, $α$. The potential of $\text{CVaR}_α$ to minimize profit loss is demonstrated by a simulated case study. The case study compares $\text{CVaR}_α$ to real-world best practices, represented by reactive control. It shows that for any risk level, $α$, we may find an optimized strategy that provides lower risk than reactive control. However, despite overall lower risk, we see that all optimized strategies still yield some unacceptable low profit realizations relative to reactive control. To remedy this, we introduce a risk mitigation method based on the NPV offset distribution. Unlike existing methods of the oil literature, the offset risk mitigation approach minimizes the risk relative to a reference strategy representing common real-life practices, e.g. reactive control. In the simulated case study, we minimize the worst case profit offset to reduce the risk of realizations that do worse than the reactive strategy. The results suggest that it may be more relevant to consider the NPV offset distribution than the NPV distribution when minimizing risk in production optimization.

math.OC↗

arXiv2017

A fast and memory-efficient spectral Galerkin scheme for distributed elliptic optimal control problems

Many scientific and engineering challenges can be formulated as optimization problems which are constrained by partial differential equations (PDEs). These include inverse problems, control problems, and design problems. As a major challenge, the associated optimization procedures are inherently large-scale. To ensure computational tractability, the design of efficient and robust iterative methods becomes imperative. To meet this challenge, this paper introduces a fast and memory-efficient preconditioned iterative scheme for a class of distributed optimal control problems governed by convection-diffusion-reaction (CDR) equations. As an alternative to low-order discretizations and Schur-complement block preconditioners, the scheme combines a high-order spectral Galerkin method with an efficient preconditioner tailored specifically for the CDR application. The preconditioner is matrix-free and can be applied within linear complexity where the proportionality constant is small. Numerical results demonstrate that the preconditioner is ideal in the sense that appropriate Krylov subspace methods converge within a low number of iterations, independently of the problem size and the Tikhonov regularization parameter.

math.OC↗