Differentiable predictive control: Deep learning alternative to explicit model predictive control for unknown nonlinear systems
Not Available
SEARCH · Engineering Papers
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Not Available
A strategy that calculates an explicit state feedback policy to regulate constrained uncertain discrete‐time uncertain linear systems is presented. We consider uncertain processes, affected by box‐bounded multiplicative uncertainty as well as bounded additive uncertainty with linear state and inputs constraints. The proposed method includes (i) the calculation of a terminal set constraint and (ii) the robust reformulation of state constraints in the prediction horizon. These features allow the derivation of the desired policy by solving a single multiparametric quadratic programming problem that guarantees feasible operation in the presence of uncertainty. Additionally, we employ variable and constraint elimination approaches to enhance the computational performance of the strategy. We demonstrate the steps and benefits of these developments with a numerical example and a chemical engineering case study.
We present a differentiable predictive control (DPC) methodology for learning constrained control laws for unknown nonlinear systems. DPC poses an approximate solution to multiparametric programming problems emerging from explicit nonlinear model predictive control (MPC). Contrary to approximate MPC, DPC does not require supervision by an expert controller. Instead, a system dynamics model is learned from a small dataset of recorded observations of the perturbed system's dynamics and the control law is optimized offline by interaction with the learned system model. The DPC method is based on two sequential steps, i) system identification using a constrained neural state-space model, and ii) optimization of an explicit control law parametrized by another neural network in closed-loop simulation with the identified neural state-space model. The combination of a differentiable closed-loop system and penalty methods for constraint handling of system outputs and inputs allows us to optimize the control law's parameters directly by backpropagating economic MPC loss through the learned system model. By incorporating domain knowledge and leveraging established techniques from optimal control, our method leverages deep neural networks as nonlinear function approximators for system identification and control while avoiding concomitant costs of intractably large datasets, and computationally expensive over-parametrized models. The scalability, data efficiency, and constrained optimal control capability of the proposed DPC method are demonstrated in simulation using a multi-zone building emulator.
The integration of thermal storage and heat pumps in district heating networks (DHNs) can significantly enhance operational flexibility and energy efficiency; however, the practical deployment of advanced control strategies is often hindered by forecasting requirements and computational complexity. This study presents a comparative analysis of thermal storage control strategies in an ultra-low-temperature fourth-generation DHN, focusing on the development of a simplified rule-based control (RBC) explicitly informed by Model Predictive Control (MPC) behavior. The proposed methodology systematically analyzes the charging and discharging decisions of an MPC-controlled system under ideal forecasting conditions and extracts recurrent control patterns as a function of key system variables, including outdoor temperature, thermal demand, and electricity price. These patterns are translated into a set of structured time- and condition-based rules, resulting in an MPC-informed RBC that embeds predictive insights while preserving implementation simplicity and operational transparency. The approach is validated on a realistic mixed-use urban district in Denver, Colorado, USA, equipped with a centralized air-source heat pump, distributed water-to-water heat pumps, and a central thermal storage unit. Results show that the tuned RBC attains approximately 96% of ideal MPC economic performance (-27% of costs), preserves values of technical and environmental indicators (reduction only of 2-3%), and substantially reduces complexity. Sensitivity analyses further demonstrate the robustness of the RBC under varying operational conditions (i.e., ambient temperature, electricity price). Overall, the study demonstrates that MPC-informed rule-based control represents an effective trade-off between control performance and real-world applicability, enabling the integration of additional system components while maintaining simplicity, robustness, and ease of implementation.
The CIRCLES project aims to reduce instabilities in traffic flow, which are naturally occurring phenomena due to human driving behavior. These "phantom jams" or "stop-and-go waves,"are a significant source of wasted energy. Toward this goal, the CIRCLES project designed a control system referred to as the MegaController by the CIRCLES team, that could be deployed in real traffic. Our field experiment leveraged a heterogeneous fleet of 100 longitudinally-controlled vehicles as Lagrangian traffic actuators, each of which ran a controller with the architecture described in this paper. The MegaController is a hierarchical control architecture, which consists of two main layers. The upper layer is called Speed Planner, and is a centralized optimal control algorithm. It assigns speed targets to the vehicles, conveyed through the LTE cellular network. The lower layer is a control layer, running on each vehicle. It performs local actuation by overriding the stock adaptive cruise controller, using the stock on-board sensors. The Speed Planner ingests live data feeds provided by third parties, as well as data from our own control vehicles, and uses both to perform the speed assignment. The architecture of the speed planner allows for modular use of standard control techniques, such as optimal control, model predictive control, kernel methods and others, including Deep RL, model predictive control and explicit controllers. Depending on the vehicle architecture, all onboard sensing data can be accessed by the local controllers, or only some. Control inputs vary across different automakers, with inputs ranging from torque or acceleration requests for some cars, and electronic selection of ACC set points in others. The proposed architecture allows for the combination of all possible settings proposed above. Most configurations were tested throughout the ramp up to the MegaVandertest.
We present a scalable unsupervised learning-based method for obtaining explicit control policies for model predictive control problems for stochastic linear systems with additive uncertainties subject to nonlinear chance constraints. We call the proposed method stochastic parametric differentiable predictive control (SP-DPC), which extends the recently proposed deterministic DPC policy optimization algorithm. We formulate the SP-DPC as a deterministic approximation to the stochastic parametric constrained optimal control problem via independent sampling of the problem's parameters and uncertainties. This formulation allows us to directly compute the policy gradients via automatic differentiation of the problem's value function, evaluated over sampled parameters and uncertainties. In particular, the computed expectation of the problem's value function is backpropagated through the finite-time closed-loop system rollouts parametrized by a known nominal system dynamics model and neural control policy. We also provide theoretical probabilistic guarantees on closed-loop stability and chance constraints satisfaction for systems controlled by learned neural policies. We demonstrate the computational efficiency and scalability of the proposed policy optimization algorithm in three numerical examples, including systems with a large number of states or subject to nonlinear constraints.
This article addresses the problem of wake steering control for wind farms that explicitly consider the tradeoff between farm-level power generation and yaw duty cycle under variable and uncertain wind conditions. A novel stochastic model predictive control (MPC) algorithm is presented, which utilizes a stochastic model of the freestream wind field components in a receding horizon framework to compute optimal yaw set points that maximize the expected value of the farm power while constraining the yaw actuation. Different configurations of the algorithm are evaluated using a steady-state wind farm simulator. The proposed stochastic MPC algorithm can plan control actions over a future prediction horizon based on probabilistic estimates of the incoming wind magnitude and direction.
There are significant economic, environmental, energy, and other societal costs incurred by the road transportation sector. With the advent and penetration of connected and autonomous vehicles there are vast opportunities to optimize the control of individual vehicles for reducing energy consumption and increasing traffic flow. Model predictive control is a useful tool to achieve such goals, while accommodating ego-centric objectives typical of heterogeneous traffic and explicitly enforcing collision and other constraints. In this paper, we describe a multi-agent distributed maneuver planning and lane selection model predictive controller that includes an information sharing and coordination scheme. The energy saving potential of the proposed coordination scheme is then evaluated via large scale microscopic traffic simulations considering different penetration levels of connected and automated vehicles.
Although groundwater discharge is a critical stream temperature control process, it is not explicitly represented in many stream temperature models, an omission that may reduce predictive accuracy, hinder management of aquatic habitat, and decrease user confidence. We assessed the performance of a previously-described process-guided deep learning model of stream temperature in the Delaware River Basin (USA). We found lower accuracy (root mean square error [RMSE] of 1.71 versus 1.35°C) and stronger seasonal bias (absolute mean monthly bias of 1.06 vs. 0.68°C) for reaches primarily influenced by deep groundwater as compared to atmospheric conditions. We then tested four approaches for improving groundwater process representation: (a) a custom loss function leveraging the unique patterns of air and water temperature coupling characteristic of different temperature drivers, (b) inclusion of additional groundwater-relevant catchment attributes, (c) incorporation of additional process model outputs, and (d) a composite model. The custom loss function and the additional attributes significantly improved the predictive accuracy in groundwater-dominated reaches (RMSE of 1.37 and 1.26°C) and reduced the seasonal bias (absolute mean monthly bias of 0.44 and 0.48°C), but neither approach could identify holdout groundwater reaches. Variable importance analysis indicates the custom loss function nudges the model to use the existing inputs more efficiently, whereas with the added features the model relies on a broader suite of inputs. This analysis is a substantial step toward more accurately representing groundwater discharge processes in stream temperature models and will improve predictive accuracy and inform habitat management.
The residential building sector accounts for approximately 37% of total U.S. electricity consumption. Within the residential building sector, heating and cooling is the main target for peak load shifting/reduction since it is the largest contributor to peak demand. In fact, the flexibility of residential HVAC loads can provide continuous variation of demand to provide grid services by varying their demand over a baseline. The performance of HVAC load control to provide grid services relies heavily on the accuracy of indoor air temperature or cooling/heating demand predictions and therefore the quality of building model. Besides forward models, popular building models are data-driven models which can be broken down into two categories: gray-box model, e.g. Resistance-Capacitance (RC) model and black-box model. RC model, also called lumped capacitance or network model, which is constituted with electrical analogue pattern with resistance (R) and capacitance (C). In general, RC models require considerable computation burden and long periods of data to train limited number of model coefficients. Black-box models have gained increasing interest due to their capability in analyzing large-scale data and flexibility in practical applications. But, the data-mining based (machine learning algorithms/techniques based) models tend to have invisible model structures which poses a problem when trying to use the model for optimal control or model predictive control of the HVAC system. Hence, there is a continuing need for efficient online system identification techniques, which can provide explicit parameters for the model. Traditional regression models fit well for this specific purpose. This paper presents an innovative way to predict average indoor temperature in separate floors of typical detached residential house. A rolling horizon linear regression model, which includes online adaptive correction component, is proposed to predict the temperature difference between downstairs and upstairs. A RC model is used to predict the overall mean indoor air temperature. Since the adaptive algorithm needs to be implemented online, a less computation-demanding polynomial fitting algorithm is adopted. This kind of fitting problem can be cast as linear regression problem with multiple variables, parameters of which can be efficiently obtained by well-known gradient descent method.The validation is conducted by comparing the predicted results with the results from data-mining based models as well as measured data from a real typical detached two-floor house. The results show that the developed method has satisfactory performance in predicting the building indoor temperature in 1st and 2nd floors.
Abstract Microstructurally small fatigue‐crack growth in polycrystalline materials is highly three‐dimensional due to sensitivity to local microstructural features (e.g., grains). One requirement for modeling microstructurally sensitive crack propagation is establishing the criteria that govern crack evolution, including crack deflection. Here, a high‐fidelity finite‐element modeling framework is used to assess the performance and validity of various crack‐growth criteria, including slip‐based metrics (e.g., fatigue‐indicator parameters), as potential criteria for predicting three‐dimensional crack paths in polycrystalline materials. The modeling framework represents cracks as geometrically explicit discontinuities and involves voxel‐based remeshing, mesh‐gradation control, and a crystal‐plasticity constitutive model. The predictions are compared to experimental measurements of WE43 magnesium samples subject to fatigue loading, for which three‐dimensional grain structures and fatigue‐crack surfaces were measured post‐mortem using near‐field high‐energy x‐ray diffraction microscopy and x‐ray computed tomography. Findings from this work are expected to improve the predictive capabilities of simulations involving microstructurally small fatigue‐crack growth in polycrystalline materials.
Rapid advances in soil biology are increasingly reflected in mathematical models. From molecular to global scales, models contribute to fundamental understanding and prediction of critical soil processes, such as carbon sequestration. Recent models explicitly incorporate microbial control over soil carbon and nutrient cycling, an approach that offers greater biological realism but raises new challenges in model parameterization and stability. Although they have proliferated at community and ecosystem scales, microbial-explicit models have yet to be incorporated into most Earth system models. Applying soil models across scales requires integration with data to support parameterization and validation using approaches such as Bayesian data assimilation. In conclusion, to catalyze future progress, we recommend breaking down barriers between modeling and empirical disciplines while broadening access to the computational infrastructure that supports soil biological modeling.
Recent advances in connected vehicle technologies enable vehicles and signal controllers to cooperate and improve the traffic management at intersections. This paper explores the opportunity for cooperative and integrated vehicle and intersection control for energy efficiency (CIVIC-E 2 ) to contribute to a more sustainable transportation system. We propose a two-level approach that jointly optimizes the traffic signal timing and vehicles' approach speed, with the objective being to minimize total energy consumption for all vehicles passing through an isolated intersection. More specifically, at the intersection level, a dynamic programming algorithm is designed to find the optimal signal timing by explicitly considering the arrival time and energy profile of each vehicle. At the vehicle level, a model predictive control strategy is adopted to ensure that vehicles pass through the intersection in a timely fashion. Our simulation study has shown that the proposed CIVIC-E 2 system can significantly improve intersection performance under various traffic conditions. Furthermore, compared with conventional fixed-time and actuated signal control strategies, the proposed algorithm can reduce energy consumption and queue length by up to 31% and 95%, respectively.
The dynamics-aware economic dispatch (DED) problem embeds low-level generator dynamics and operational constraints to enable near real-time scheduling of generation units in a power network. DED produces a more dynamic supervisory control policy than traditional economic dispatch (T-ED) that reduces overall generation costs. However, the incorporation of differential equations that govern the system dynamics makes DED an optimization problem that is computationally prohibitive to solve. In this work, we present a new data-driven approach based on differentiable programming to efficiently obtain offline parametric solutions to the underlying DED problem. In particular, we employ the recently proposed differentiable predictive control (DPC) for offline learning of explicit neural control policies based on identified Koopman operator (KO) model of the system dynamics. We demonstrate the high solution quality and five orders of magnitude computational-time savings of the DPC method over the original optimization-based DED approach on a 9-bus test power grid network.
Bayesian inference of microbial soil respiration models is often based on the assumptions that the residuals are independent (i.e., no temporal or spatial correlation), identically distributed (i.e., Gaussian noise), and have constant variance (i.e., homoscedastic). In the presence of model discrepancy, as no model is perfect, this study shows that these assumptions are generally invalid in soil respiration modeling such that residuals have high temporal correlation, an increasing variance with increasing magnitude of CO 2 efflux, and non-Gaussian distribution. Relaxing these three assumptions stepwise results in eight data models. Data models are the basis of formulating likelihood functions of Bayesian inference. This study presents a systematic and comprehensive investigation of the impacts of data model selection on Bayesian inference and predictive performance. We use three mechanistic soil respiration models with different levels of model fidelity (i.e., model discrepancy) with respect to the number of carbon pools and the explicit representations of soil moisture controls on carbon degradation; therefore, we have different levels of model complexity with respect to the number of model parameters. The study shows that data models have substantial impacts on Bayesian inference and predictive performance of the soil respiration models such that the following points are true: (i) the level of complexity of the best model is generally justified by the cross-validation results for different data models; (ii) not accounting for heteroscedasticity and autocorrelation might not necessarily result in biased parameter estimates or predictions, but will definitely underestimate uncertainty; (iii) using a non-Gaussian data model improves the parameter estimates and the predictive performance; and (iv) accounting for autocorrelation only or joint inversion of correlation and heteroscedasticity can be problematic and requires special treatment. Although the conclusions of this study are empirical, the analysis may provide insights for selecting appropriate data models for soil respiration modeling.
Conventional physics-based modeling is a time-consuming bottleneck in control design for complex nonlinear systems like autonomous underwater vehicles (AUVs). In contrast, purely data-driven models, require a large number of observations and lack operational guarantees for safety-critical systems. Data-driven models leveraging available partially characterized dynamics have potential to provide reliable systems models in a typical data-limited scenario for high value complex systems, thereby avoiding months of expensive expert modeling time. In this work we explore this middle-ground between expert-modeled and pure data-driven modeling. We present control-oriented parametric models with varying levels of domain-awareness that exploit known system structure and prior physics knowledge to create constrained deep neural dynamical system models. We employ universal differential equations to construct data-driven blackbox and graybox representations of the AUV dynamics. In addition, we explore a hybrid formulation that explicitly models the residual error related to imperfect graybox models. We compare the prediction performance of the learned models for different distributions of initial conditions and control inputs to assess their suitability for control.
High-energy diffraction microscopy (HEDM) constitutes a suite of combined X-ray characterization methods, which hold the unique advantage of illuminating the microstructure and micromechanical state of a material during concurrent in situ mechanical deformation. The data generated from HEDM experiments provides a heretofore unrealized opportunity to validate meso-scale modeling techniques, such as crystal plasticity finite element modeling (CPFEM), by explicitly testing the accuracy of these models at the length scales where the models predict their response. Combining HEDM methods with in situ loading under known and controlled boundary conditions represents a significant challenge, inspiring the recent development of a new high-precision rotation and axial motion system for simultaneously rotating and axially loading a sample. In this paper, we describe the initial HEDM dataset collected using this hardware on an alpha-titanium alloy (Ti-7Al) under in situ tensile deformation at the Advanced Photon Source, Argonne National Laboratory. We present both near-field HEDM data that maps out the grain morphology and intragranular crystallographic orientations and far-field HEDM data that provides the grain centroid, grain average crystallographic orientation, and grain average elastic strain tensor for each grain. Finally, we provide a finite element mesh that can be utilized to simulate deformation in the volume of this Ti-7Al specimen.