A5047724723- Advanced Control Systems Optimization
- Adaptive Dynamic Programming Control
- Building Energy and Comfort Optimization
- Smart Grid Energy Management
- Fault Detection and Control Systems
- Model Reduction and Neural Networks
- Reinforcement Learning in Robotics
- Control Systems and Identification
- Adaptive Control of Nonlinear Systems
- Power System Optimization and Stability
- Smart Grid Security and Resilience
- Microgrid Control and Optimization
- Frequency Control in Power Systems
- Real-time simulation and control systems
- Neural Networks and Applications
- Energy Efficiency and Management
- Modeling and Simulation Systems
- Wind and Air Flow Studies
- Fuel Cells and Related Materials
- Extremum Seeking Control Systems
- Integrated Energy Systems Optimization
- Oil and Gas Production Techniques
- Energy Load and Power Forecasting
- Adversarial Robustness in Machine Learning
- Gaussian Processes and Bayesian Inference
Pacific Northwest National Laboratory
2017-2024
GE Vernova (United States)
2023
Government of the United States of America
2022
Battelle
2020-2021
Hartford Financial Services (United States)
2011-2015
The University of Texas at Arlington
2007-2012
Raytheon Technologies (Finland)
2012
United Technologies Research Center
2012
Robotics Research (United States)
2008-2010
Gheorghe Asachi Technical University of Iași
2004-2005
Living organisms learn by acting on their environment, observing the resulting reward stimulus, and adjusting actions accordingly to improve reward. This action-based or reinforcement learning can capture notions of optimal behavior occurring in natural systems. We describe mathematical formulations for a practical implementation method known as adaptive dynamic programming. These give us insight into design controllers man-made engineered systems that both exhibit behavior.
This article describes the use of principles reinforcement learning to design feedback controllers for discrete- and continuous-time dynamical systems that combine features adaptive control optimal control. Adaptive [1], [2] [3] represent different philosophies designing controllers. Optimal are normally designed ine by solving Hamilton JacobiBellman (HJB) equations, example, Riccati equation, using complete knowledge system dynamics. Determining policies nonlinear requires offline solution...
It has been proven that advanced building control, like model predictive control (MPC), can notably reduce the energy use and mitigate greenhouse gas emissions. However, despite intensive research efforts, practical applications are still in early stages. There is a growing need for multidisciplinary education on methods built environment to be accessible broad range of researchers practitioners with different engineering backgrounds. This paper provides unified framework technology focus...
Development of new building HVAC control algorithms has grown due to needs for energy efficiency and operational flexibility. However, case studies demonstrating are largely individualized, making algorithm performance difficult compare directly. In addition, the effort expertise required implement in real or simulated buildings limits rapid prototyping potential. Therefore, this paper presents Building Optimization Testing Framework (BOPTEST) associated software simulation-based...
In this paper, we introduce an online algorithm that uses integral reinforcement knowledge for learning the continuous-time optimal control solution nonlinear systems with infinite horizon costs and partial of system dynamics. This is a data-based approach to Hamilton–Jacobi–Bellman equation, it does not require explicit on system's drift A novel adaptive given based policy iteration implemented using actor/critic structure having two approximator structures. Both actor critic approximation...
We present a physics-constrained deep learning method to develop control-oriented models of building thermal dynamics. The proposed uses systematic encoding physics-based prior knowledge into structured recurrent neural architecture. Specifically, our incorporates structural from traditional modeling the architecture network model. Further, we also use penalty methods provide inequality constraints, thereby bounding predictions within physically realistic and safe operating ranges. observe...
We present differentiable predictive control (DPC) as a deep learning-based alternative to the explicit model (MPC) for unknown nonlinear systems. In DPC framework, neural state-space is learned from time-series measurements of system dynamics. The policy then optimized via stochastic gradient descent approach by differentiating MPC loss function through closed-loop dynamics model. proposed method learns model-based policies with state and input constraints, while supporting time-varying...
Physics-informed machine learning (PIML) is a set of methods and tools that systematically integrate (ML) algorithms with physical constraints abstract mathematical models developed in scientific engineering domains. As opposed to purely data-driven methods, PIML can be trained from additional information obtained by enforcing laws such as energy mass conservation. More broadly, include properties conditions stability, convexity, or invariance. The basic premise the integration ML physics...
We present differentiable predictive control (DPC), a method for offline learning of constrained neural policies nonlinear dynamical systems with performance guarantees. show that the sensitivities parametric optimal problem can be used to obtain direct policy gradients. Specifically, we employ automatic differentiation (AD) efficiently compute model (MPC) objective function and constraints penalties. To guarantee safety upon deployment, derive probabilistic guarantees on closed-loop...
Binary Quadratic Programs (BQPs) are a class of NP-hard problems that arise in wide range applications, including finance, machine learning, and logistics. These challenging to solve due the combinatorial search space nonlinearity. In fact, this optimization is so that, many instances, standard algorithms struggle identify feasible solutions within reasonable time. Primal heuristic have been developed quickly BQPs. paper, we propose Cover-Relax-Search, an efficient primal for This approach...
This article presents the framework and results of implementing optimization-based control algorithm for building HVAC systems demonstrates its benefits through reduced energy consumption as well improved thermal comfort along with lessons learned. In particular, a practically effective computationally efficient model predictive is proposed to optimize usage while maintaining in multi-zone medium-sized commercial building. has two themes. Driven by challenge fully evaluating benefit...
In this paper we develop a new online adaptive control scheme, for partially unknown nonlinear systems, which converges to the optimal state feedback solution affine in inputs systems. The derivation of algorithm is presented continuous-time framework. will be obtained direct fashion, without system identification. an approach policy iterations based on critic structure find approximate feedback, infinite-horizon, problem.
In this paper we present two adaptive algorithms which offer solution to the continuous-time optimal control problem for nonlinear, affine in inputs, time-invariant systems. Both were developed based on generalized policy iteration technique and involve adaptation of neural network structures namely actor, providing signal, critic, performing evaluation performance. Despite similarities, differ manner takes place, required knowledge system dynamics, formulation persistence excitation...