Real-time Model Predictive Control Research Articles

Autonomous Needle Navigation in Subretinal Injections via iOCT.

Subretinal injection is an effective method for direct delivery of therapeutic agents to treat prevalent subretinal diseases. Among the challenges for surgeons are physiological hand tremor, difficulty resolving single-micron scale depth perception, and lack of tactile feedback. The recent introduction of intraoperative Optical Coherence Tomography (iOCT) enables precise depth information during subretinal surgery. However, even when relying on iOCT, achieving the required micron-scale precision remains a significant surgical challenge. This work presents a robot-assisted workflow for high-precision autonomous needle navigation for subretinal injection. The workflow includes online registration between robot and iOCT coordinates; tool-tip localization in iOCT coordinates using a Convolutional Neural Network (CNN); and tool-tip planning and tracking system using real-time Model Predictive Control (MPC). The proposed workflow is validated using a silicone eye phantom and ex vivo porcine eyes. The experimental results demonstrate that the mean error to reach the user-defined target and the mean procedure duration are within an acceptable precision range. The proposed workflow achieves a 100% success rate for subretinal injection, while maintaining scleral forces at the scleral insertion point below 15mN throughout the navigation procedures.

Towards a Digital Twin Framework in Additive Manufacturing: Machine Learning and Bayesian Optimization for Time Series Process Optimization

Laser directed-energy deposition (DED) offers notable advantages in additive manufacturing (AM) for producing intricate geometries and facilitating material functional grading. However, inherent challenges such as material property inconsistencies and part variability persist, predominantly due to its layer-wise fabrication approach. Critical to these challenges is heat accumulation during DED, influencing the resultant material microstructure and properties. Although closed-loop control methods for managing heat accumulation and temperature regulation are prevalent in DED literature, few approaches integrate real-time monitoring, physics-based modeling, and control simultaneously in a cohesive framework. To address this, we present a digital twin (DT) framework for real-time model predictive control of process parameters of the DED for achieving a specific process design objective. To enable its implementation, we detail the development of a surrogate model utilizing Long Short-Term Memory (LSTM)-based machine learning which uses Bayesian Inference to predict temperatures across various spatial locations of the DED-built part. This model offers real-time predictions of future temperature states. In addition, we introduce a Bayesian Optimization (BO) method for Time Series Process Optimization (BOTSPO). Its foundational principles align with traditional BO, and its novelty lies in our unique time series process profile generator with a reduced dimensional representation. BOTSPO is used for dynamic process optimization in which we deploy BOTSPO to determine the optimal laser power profile, aiming to achieve desired mechanical properties in a DED build. The identified profile establishes a process trajectory that online process optimizations aim to match or exceed in performance. This paper elucidates components of the digital twin framework, advocating its prospective consolidation into a comprehensive digital twin system for AM.

An energy management strategy for fuel cell hybrid electric vehicle based on a real-time model predictive control and pontryagin’s maximum principle

ABSTRACT In order to maintain the battery SOC, the fuel cell power will fluctuate dramatically, as well as frequent start-stop, which will greatly increase the life attenuation of the fuel cell and reduce the durability. An optimization-based energy management strategy with a real-time model predictive control and pontryagin’s maximum principle for FCHEV is proposed in this paper, both the fuel economy and the fuel cell durability are considered in the optimization. A novel model predictive control is studied to achieve energy distribution. After the calculation of predicted speed sequence through back propagation neural network, pontryagin’s maximum principle is introduced to solve the optimal control problem in each prediction horizon and obtain the ideal control strategy. In addition, the fuel cell degradation model is introduced in the modeling process, the minimum power point of the fuel cell system is designed to improve the fuel economy and durability of the fuel cell. Compared with the rule-based strategy, the proposed MPC strategy has better performance to reduce the total equivalent hydrogen consumption, which can save up to 8.44% in the test case of the mid-size fuel cell passenger car while maintaining the stability of the battery’s state of charge.

Real-time Learning-based Nonlinear Model Predictive Control of a virtual motorcycle employing grey-box modeling through Gaussian processes

Virtual prototyping tools simulations are nowadays largely adopted methods in the automotive industry. However, to achieve effective simulative tests for two-wheeled vehicles, a virtual rider, i.e. a controller for virtual motorcycles, is typically needed due to the inherent instability of the system. Different control strategies have been employed to deal with this control task, and promising results have been obtained using Nonlinear Model Predictive Control (NMPC). Yet, performance of NMPC highly relies on the plant characterization, that can be highly complex to obtain analytically for 2-wheel vehicles. To improve it, learning dynamics approaches can be used within a Learning-based Nonlinear Model Predictive Control (LbNMPC) framework, exploiting the combination of data-driven techniques and NMPC strategy. In this manuscript, we present a tailored real-time capable LbNMPC for a virtual motorcycle that relies on a continuous grey-box model based on Gaussian Processes (GPs), method that has proven to be effective for 4-wheel vehicle applications. To cope with the real-time requirement, a feature selection procedure and sparse GP approximations have been adopted. A comparison with a physics-based model NMPC implementation taken from the literature has been carried out, analyzing the impact of different sparse GP representations. Remarkable improvements in both model accuracy and tracking performance have been obtained, and the generalization properties of the grey-box model have been assessed.

A Terminal State Feasibility Governor for Real-Time Nonlinear Model Predictive Control Over Arbitrary Horizons

A Terminal State Feasibility Governor for Real-Time Nonlinear Model Predictive Control Over Arbitrary Horizons

Real-Time Model Predictive Control Framework for a Point Absorber Wave Energy Converter With Excitation Force Estimation and Prediction

Real-Time Model Predictive Control Framework for a Point Absorber Wave Energy Converter With Excitation Force Estimation and Prediction

Active Battery Cell Balancing by Real-Time Model Predictive Control for Extending Electric Vehicle Driving Range

Active Battery Cell Balancing by Real-Time Model Predictive Control for Extending Electric Vehicle Driving Range

Modeling and aggregated control of large-scale 5G base stations and backup energy storage systems towards secondary frequency support

A significant number of 5G base stations (gNBs) and their backup energy storage systems (BESSs) are redundantly configured, possessing surplus capacity during non-peak traffic hours. Moreover, traffic load profiles exhibit spatial variations across different areas. Proper scheduling of surplus capacity from gNBs and BESSs in different areas can provide sustainable frequency support for the power system without compromising the operation of 5G network. In this paper, a comprehensive strategy is proposed to safely incorporate gNBs and their BESSs (called “gNB systems”) into the secondary frequency control procedure. Initially, an aggregated model is developed using a state space method to capture the state of a cluster of heterogeneous gNB systems (gNBs-cluster). Subsequently, a utility function is defined to evaluate the quality of 5G network operation in a normalized manner. Based on this utility function, an aggregated control method is proposed, including real-time available power estimation and model predictive control (MPC) for the gNBs-cluster, which ensures the 5G network operation within the security constraint during demand response processes. Simulations, utilizing actual device data, demonstrate the effectiveness of the proposed method in improving power system frequency performance while guaranteeing the safety and reliability of the 5G network.

Real-time model predictive control of urban drainage system in coastal areas

In the context of global climate change and urbanization, model predictive control (MPC) has been increasingly applied in the field of real-time control of urban drainage systems to cope with frequent urban flooding events caused by extreme rainstorms. However, the computation time of the predictive model calculation greatly hinders further development of MPC in this field. In this study, we propose two surrogate models, namely the water tank-water balance model 1 (TWBM1) and the water tank-water balance model 2 (TWBM2), to replace the storm water management model (SWMM) as the prediction model for MPC, thus improving MPC computational efficiency. Both surrogate models can obtain the total outflow process at the downstream outfall of the study area and the corresponding overflow process. In comparison with TWBM1, TWBM2 requires 26 % fewer calibration parameters, and its structure is simpler. These results have been verified in the real-time optimal scheduling of a rainwater drainage system in the Doumen area of Fuzhou City, China. The results show that TWBM1 and TWBM2 have shorter computation time compared to SWMM, which improves the efficiency of MPC computation. Additionally, in comparison with the existing scheduling results, MPC has better performance in terms of economy, security and computational efficiency.

Physically consistent deep learning-based day-ahead energy dispatching and thermal comfort control for grid-interactive communities

Grid-interactive communities are an emerging solution for optimal energy dispatching, improving grid stability, and enhancing the usage of on-site renewable energy by leveraging community flexibilities. This work introduces a novel Physically Consistent Deep Learning (PCDL)-based optimization framework for day-ahead energy dispatching and thermal comfort control within grid-interactive communities. Specifically, the PCDL model is developed to capture the thermal dynamics within the community while ensuring strict adherence to physics consistency. This consistency is defined with the aim of generating physically viable solutions in response to modified system inputs. Subsequently, the PCDL model is utilized for load estimation and indoor climate prediction within a proposed scheduling and control strategy: (1) an optimal energy dispatching plan is calculated based on the day-ahead grid market; (2) a real-time model predictive control (MPC) method is applied to minimize the utilization error and indoor comfort constraint violations caused by following the scheduled energy dispatching plan. A simulation case of a residential hall community at Cornell University with on-site renewable energy resources is implemented to demonstrate the effectiveness of the proposed approach. Comparative simulations, which include the Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Gaussian Process (GP) regression models, confirm the performance advantage of capturing community thermal dynamic and control-oriented generalization ability when using the PCDL model. These results leads to notable improvements in indoor thermal comfort control, ranging between 65.4 and 68.7%, 63.6–79.3%, and 60.4–62.2% for each comparative model, respectively. Moreover, by harnessing community demand flexibility, we achieve energy cost savings between 29.5 and 39.7% compared to the baseline controller.

Intelligent state estimation for fault tolerant integrated frequent RTO and adaptive nonlinear MPC

Combinations of real-time optimization (RTO) and model predictive control (MPC) have been widely employed in the process industry for tracking the economic optimum in the face of drifting disturbances and parameters. Online update of model parameters is a critical step in the implementation of RTO. In this work, an intelligent state and parameter estimation approach is developed by combining a fault diagnosis approach with a simultaneous state and parameter estimator. Since faults are more likely to develop as slow drifts, a nonlinear generalized likelihood ratio (GLR) approach available in the literature is modified by considering ramp models for the progression of faults with time. When a fault is isolated by the fault diagnosis and identification (FDI) component, the magnitude of the isolated fault is refined using a moving window state and parameter estimator until the fault magnitude continues to change. The estimation of fault magnitudes is carried out only when required and triggered by the fault identification scheme. Thus, the subset of parameters/faults that are being estimated online can change with time. The intelligent state and parameter estimator is further combined with an online optimizing control scheme consisting of integrated frequent RTO and adaptive MPC. The integrated optimizing control scheme has embedded intelligence to auto-correct models used for estimation, control, and optimization and decide whether the detected changes require the invocation of RTO. The proposed approach employs a single model to carry out four different tasks: process monitoring, state and parameter estimation, nonlinear predictive control, and real-time optimization. This eliminates difficulties that can arise due to model mismatch between different components of the online optimizing control scheme. The efficacy of the proposed scheme is investigated using the benchmark Williams Otto reactor and a continuously operated fermenter. The economic optimum operating point of these systems is sensitive to mean shifts in unmeasured disturbances or system parameters. The proposed ramp model based approach successfully isolates the parameter/ unmeasured disturbance/ sensor bias/ actuator bias that has undergone slow drift and tracks the shifting economic optimum without significant delays. Thus, the proposed integrated approach has the ability to handle normal and abnormal operating envelopes of the system.

Data-Driven Linear Koopman Embedding for Networked Systems: Model-Predictive Grid Control

This article presents a data-learned linear Koopman embedding of nonlinear networked dynamics and uses it to enable real-time model predictive emergency voltage control in a power grid. The approach involves a novel data-driven “basis-dictionary free” lifting of the system dynamics into a higher dimensional linear space over which a model predictive control is exercised, making it both scalable and rapid for practical real-time implementation. A <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Koopman-inspired deep neural network</i> (KDNN) encoder–decoder architecture for the linear embedding of the underlying networked dynamics under distributive control is presented, in which the end-to-end components of the KDNN, comprising of a triple of transforms is learned from the system trajectory data in one go: a neural network (NN)-based lifting to a higher dimension, a linear dynamics within that higher dimension, and an NN-based projection to the original space. This data-learned approach relieves the burden of the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ad hoc</i> selection of the nonlinear basis functions (e.g., radial or polynomial) used in conventional approaches for lifting to a higher dimensional linear space. We validate the efficacy and robustness of the approach via application to the standard IEEE 39-bus system.

A Novel Autonomous Landing Method for Flying–Walking Power Line Inspection Robots Based on Prior Structure Data

Hybrid inspection robots have been attracting increasing interest in recent years, and are suitable for inspecting long-distance overhead power transmission lines (OPTLs), combining the advantages of flying robots (e.g., UAVs) and climbing robots (e.g., multiple-arm robots). Due to the complex work conditions (e.g., power line slopes, complex backgrounds, wind interference), landing on OPTL is one of the most difficult challenges faced by hybrid inspection robots. To address this problem, this study proposes a novel autonomous landing method for a developed flying–walking power line inspection robot (FPLIR) based on prior structure data. The proposed method includes three main steps: (1) A color image of the target power line is segmented using a real-time semantic segmentation network, fusing the depth image to estimate the position of the power line. (2) The safe landing area (SLA) is determined using prior structure data, applying the trajectory planning method with geometric constraints to generate the dynamic landing trajectory. (3) The landing trajectory is tracked using real-time model predictive control (MPC), controlling FPLIR to land on the OPTL. The feasibility of the proposed method was verified in the ROS Gazebo environment. The RMSE values of the position along three axes were 0.1205,0.0976 and 0.0953, respectively, while the RMSE values of the velocity along these axes were 0.0426, 0.0345 and 0.0781. Additionally, experiments in a real environment using FPLIR were performed to verify the validity of the proposed method. The experimental results showed that the errors of position and velocity for the FPLIR landing on the lines were 6.18×10−2 m and 2.16×10−2 m/s. The simulation results as well as the experimental findings both satisfy the practical requirements. The proposed method provides a foundation for the intelligent inspection of OPTL in the future.

Online model-based predictive control with smart thermostats: application to an experimental house in Québec

This paper tests the impact of model resolution and structure on the performance of Model Predictive Control (MPC) implementation in an unoccupied research house in Québec equipped with smart thermostats. Two low-order models and a high-order multi-zone model were calibrated with measured data, with the structure of the multi-zone model being generated automatically during the calibration procedure. The three models were used to apply real-time MPC to an experimental house in Québec using the established dynamic tariffs for morning and evening peaks. MPC with any of the three models successfully preheated the house before the demand-response events, outperforming the reference reactive controller, reducing cost and thermal discomfort. The high-order multi-zone model performed the best, reducing average cost of electricity by 55% and high-price energy consumption by 71%, compared to the low-order models, which achieved cost reductions of 40% and 44% and energy consumption reductions of 48% and 54% respectively.

Multi-Objective Adaptive Optimization Model Predictive Control: Decreasing Carbon Emissions from a Zinc Oxide Rotary Kiln

The zinc oxide rotary kiln, as an essential piece of equipment in the zinc smelting industrial process, is presenting new challenges in process control. China’s strategy of achieving a carbon peak and carbon neutrality is putting new demands on the industry, including green production and the use of fewer resources; thus, traditional stability control is no longer suitable for multi-objective control tasks. Although researchers have revealed the principle of the rotary kiln and set up computational fluid dynamics (CFD) simulation models to study its dynamics, these models cannot be directly applied to process control due to their high computational complexity. To address these issues, this paper proposes a multi-objective adaptive optimization model predictive control (MAO-MPC) method based on sparse identification. More specifically, with a large amount of data collected from a CFD model, a sparse regression problem is first formulated and solved to obtain a reduction model. Then, a two-layered control framework including real-time optimization (RTO) and model predictive control (MPC) is designed. In the RTO layer, an optimization problem with the goal of achieving optimal operation performance and the lowest possible resource consumption is set up. By solving the optimization problem in real time, a suitable setting value is sent to the MPC layer to ensure that the zinc oxide rotary kiln always functions in an optimal state. Our experiments show the strength and reliability of the proposed method, which reduces the usage of coal while maintaining high profits.

Simplified finite volume-based dynamic modeling, experimental validation, and data-driven simulation of a fire-tube hot-water boiler

This paper proposes and implements a novel approach to simulate the dynamic behavior of hot-water fire-tube boilers, in which physical phenomena-based and data-driven modeling methodologies have both been employed. The first model, which includes a reduced one-dimensional finite volume method to simulate the flue-gas side’s behavior, can be employed for accurate sizing of the unit considering end-users’ dynamic consumption profile. In the data-driven model instead, machine learning algorithms are used to estimate hot water’s supply temperature, which makes it a suitable tool for real-time prediction and model predictive control. Utilizing the latter model in combination with an advanced management system allows reducing the plant’s energy consumption and enhancing its controllability. Employing the measurements performed in an Italian industrial firm, the developed model is validated and is demonstrated to have a limited thermal efficiency estimation bias of 1.2%. Furthermore, the data-driven model achieves a mean absolute relative difference error lower than 6%, demonstrating its acceptable accuracy for various prediction horizons.

Multi-objective real-time energy management for series–parallel hybrid electric vehicles considering battery life

The real-time control of energy management strategy (EMS) is becoming increasingly challenging as the complexity of the model and control strategy increases. To address this issue while ensuring the accuracy of the EMS, a multi-objective real-time EMS based on model predictive control (MPC) that considers battery life is proposed in this paper. Firstly, to balance the accuracy and efficiency of the prediction module, a kernel extreme learning machine based on the whale optimization algorithm is proposed as a short-term speed prediction model. Secondly, an adaptive state of charge (SOC) trajectory planning method is established to plan MPC reference trajectory. Next, to optimize fuel efficiency, electrical energy consumption, and battery aging in real-time, a multi-objective real-time MPC (MOR-MPC) algorithm is proposed. Finally, the effectiveness, real-time performance, and robustness of the proposed strategy are verified. Simulation results demonstrate that the total cost of the strategy is reduced by 6.15% compared to the equivalent consumption minimization strategy (ECMS), with 98.17% of dynamic programming (DP) performance achieved. Real-time performance is improved by 97.89% compared to DP-MPC. Hardware-in-the-loop (HIL) testing is also carried out to evaluate the proposed strategy.

Model predictive real-time architecture for secondary voltage control of microgrids

The integration of new elements and technologies to the power grid, driven by the expansion of electric microgrids, brings various challenges requiring a suitable control architecture for systems to operate ensuring economic and reliable access to electricity. This paper presents the development of a time-varying constraint real-time model predictive secondary voltage control (MPVC) strategy for microgrids built up in a multi-class Python environment with InfluxDB and SQLExpress databases storing the controller and microgrid data, and its experimental implementation with Modbus communication to physical devices at the Universidad Pontificia Bolivariana (UPB) campus microgrid. The microgrid characteristics are described and the experimental setup is presented. The mathematical model of the resource used as case study is obtained through experimental identification, and it is integrated to the MPVC scheme. Polytopic invariant sets are used as terminal sets in the MPVC scheme to ensure stability of the solution, and time-varying constraints of reactive power availability are considered. Validation studies of the proposed control system in computational simulation are presented prior to real application in the UPB campus microgrid. Controller performance results are described for three different implementation scenarios of low, medium and high variability of external disturbances, presenting adequate voltage regulation around the reference value and considerable dependence on the meteorological forecast quality. The model predictive voltage controller is finally benchmarked against a commercial microgrid controller to evaluate its performance, standing out as a more comprehensive and tailor-made solution.

Optimizing heating operation via GA- and ANN-based model predictive control: Concept for a real nearly-zero energy building

Using a simulation- and optimization-based framework that leverages artificial neural networks for the model predictive control of space heating operation, machine learning techniques are explored to predict the energy performance of a real nearly zero energy building located in Benevento (Southern Italy, Mediterranean climate). The framework's goal is to minimize heating energy costs and thermal discomfort by providing optimal values of setpoint temperatures on a day-ahead planning horizon based on weather forecasts. To achieve this, a Pareto multi-objective approach is applied, which considers thermal comfort via the adaptive theory of ASHRAE 55, assessing a comfort penalty function. The optimization problem is solved using a genetic algorithm in the MATLAB® environment. Objective functions are evaluated using the coupling between MATLAB® and EnergyPlus, which is used as building performance simulation tool. Multi-criteria decision-making is performed to select an optimal solution from the Pareto front while setting a limit to the comfort penalty. To test the framework, EnergyPlus weather data is used to simulate weather forecasts for typical days of the winter season. Two different economic scenarios are simulated in order to be able to analyze both conditions and given the variability of the electricity price. For the different days tested, the best results in terms of savings achieved through the proposed solution are on Feb. 28, corresponding to the coldest day and the highest energy cost, with daily savings of 26% compared to a reference control with a fixed setpoint of 21 °C, while maintaining similar comfort performance. The results suggest that optimized control is a crucial aspect of sustainable building design, and the proposed framework can be virtually implemented or integrated into automation systems for real-time model predictive control.

Real-Time Neural MPC: Deep Learning Model Predictive Control for Quadrotors and Agile Robotic Platforms

Model Predictive Control (MPC) has become a popular framework in embedded control for high-performance autonomous systems. However, to achieve good control performance using MPC, an accurate dynamics model is key. To maintain real-time operation, the dynamics models used on embedded systems have been limited to simple first-principle models, which substantially limits their representative power. In contrast to such simple models, machine learning approaches, specifically neural networks, have been shown to accurately model even complex dynamic effects, but their large computational complexity hindered combination with fast real-time iteration loops. With this work, we present Real-time Neural MPC, a framework to efficiently integrate large, complex neural network architectures as dynamics models within a model-predictive control pipeline. Our experiments, performed in simulation and the real world onboard a highly agile quadrotor platform, demonstrate the capabilities of the described system to run learned models with, previously infeasible, large modeling capacity using gradient-based online optimization MPC. Compared to prior implementations of neural networks in online optimization MPC we can leverage models of over 4000 times larger parametric capacity in a 50Hz real-time window on an embedded platform. Further, we show the feasibility of our framework on real-world problems by reducing the positional tracking error by up to 82% when compared to state-of-the-art MPC approaches without neural network dynamics.