A5033406975- Model Reduction and Neural Networks
- Neural Networks and Applications
- Smart Grid Security and Resilience
- Advanced Control Systems Optimization
- Fault Detection and Control Systems
- Control Systems and Identification
- Advanced Multi-Objective Optimization Algorithms
- Reservoir Engineering and Simulation Methods
- Air Traffic Management and Optimization
- Real-time simulation and control systems
- Gaussian Processes and Bayesian Inference
- Target Tracking and Data Fusion in Sensor Networks
- Transportation Planning and Optimization
- Building Energy and Comfort Optimization
- Space Exploration and Technology
- Smart Grid Energy Management
- Adversarial Robustness in Machine Learning
- Space exploration and regulation
- Oil and Gas Production Techniques
- Space Science and Extraterrestrial Life
- Traffic and Road Safety
- Infrastructure Resilience and Vulnerability Analysis
- Embedded Systems Design Techniques
- Software-Defined Networks and 5G
- Autonomous Vehicle Technology and Safety
Pacific Northwest National Laboratory
2020-2024
University of Washington
2012-2020
Seattle University
2015
We present a differentiable predictive control (DPC) methodology for learning constrained laws unknown nonlinear systems. DPC poses an approximate solution to multiparametric programming problems emerging from explicit model (MPC). Contrary MPC, does not require supervision by expert controller. Instead, system dynamics is learned the observed system's dynamics, and neural law optimized offline leveraging closed-loop model. The combination of penalty methods constraint handling outputs...
This paper explores the benefits and challenges of using solar energy to power unmanned aerial vehicles (UAVs) for surveillance purposes. The task persistent requires constant supply input is particular significance in a number applications such as weather monitoring, wildfire control, pollution or contamination detection, target search other long endurance missions. Here, we consider monitoring geographical area events varying priorities design optimal trajectories sufficient continuous...
Neural network modules conditioned by known priors can be effectively trained and combined to represent systems with nonlinear dynamics. This work explores a novel formulation for data-efficient learning of deep control-oriented dynamical models embedding local model structure constraints. The proposed method consists neural blocks that input, state, output dynamics constraints placed on the weights system variables. For handling partially observable systems, we utilize state observer...
In this paper, we provide sufficient conditions for dissipativity and local asymptotic stability of discrete-time dynamical systems parametrized by deep neural networks. We leverage the representation networks as pointwise affine maps, thus exposing their linear operators making them accessible to classical system analytic design methods. This allows us "crack open black box" system's behavior evaluating dissipativity, estimating stationary points state-space partitioning. relate norms these...
Critical infrastructure networks, such as power and transportation are often modeled Cyber–Physical systems (CPSs). With ever increasing complexity of these systems, there is a need for newer more relevant metrics design tools that will co-optimize the physical system components control policies to guarantee resilience against cyber natural threats. To this end, we develop simulation-based co-design computational framework concurrently determine parameters CPS meet prespecified resilience,...
Sequential approaches to system and control design produce sub-optimal solutions due unidirectional coupling between the variables, i.e., prescribes approach but not vice versa. A critical challenge co-design is complex heterogeneous variables: time-independent variables time-varying parameters. Traditional optimization-based methods are unsuitable for large-scale engineering systems where plant accurate control-oriented, white-box models may be available. These challenges have led...
We apply Bayesian Linear Regression to estimate the response rate of drivers variable message signs at Seattle-Tacoma International Airport, or SEA. Our approach uses vehicle speed and flow data measured entrances arrival departure-ways airport terminal, sign data. Depending on time day, we that between 5.5 9.1% divert from departures arrivals when reads “departures full, use arrivals”, conversely, 1.9 4.2% departures. Though lack counterfactual (i.e., what would have happened had...
Industrial control systems are subject to cyber attacks that produce physical consequences. These can be both hard detect and protracted. Here, we focus on deception-based sensor bias made against a hierarchical system where the attacker attempts stealthy. We develop data-driven, optimization-based model use Koopman operator represent dynamics in domain-aware computationally efficient manner. Using this model, compute several different high-fidelity commercial building emulator compare...
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 large number of observations and lack operational guarantees safety-critical systems. Data-driven leveraging available partially characterized dynamics have potential to provide reliable typical data-limited scenario high value systems, thereby avoiding months expensive expert time. this...
Cyber-Physical Systems (CPSs) provide opportunities for cyber attacks to have physical impacts. Advanced Persistent Threats (APTs) are a subclass of threats that act stealthily avoid detection and enable long-term attacks. Here, we build on our past work in APT modelling combine deception-based sensor bias direct actuator manipulations against hierarchical control system. That used the Koopman operator develop data-driven, domain-aware, optimization-based attacker model. Using an expansion...
This paper presents a simple data-driven approach to improve ground target tracking by an unmanned aerial vehicle (UAV) for certain classes of trajectories from learned local linear models. The UAV is assumed be small fixed-wing aircraft equipped with gimbaled camera visual sensing. We attempt build controller measurement data building augmented Linear Quadratic Regulator (LQR) system approximated operator that indirectly captures the properties system. evaluate relative performance...
In this paper, we provide sufficient conditions for dissipativity and local asymptotic stability of discrete-time dynamical systems parametrized by deep neural networks. We leverage the representation networks as pointwise affine maps, thus exposing their linear operators making them accessible to classical system analytic design methods. This allows us "crack open black box" system's behavior evaluating dissipativity, estimating stationary points state-space partitioning. relate norms these...
A simulation-based optimization framework is developed to con-currently design the system and control parameters meet de-sired performance operational resiliency objectives. Leveraging information from both data models of varying fideli-ties, a rigorous probabilistic approach employed for co-design experimentation. Significant economic benefits resilience im-provements are demonstrated using compared existing sequential designs cyber-physical systems.
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, though convenient and quick to obtain, require large number of observations lack operational guarantees safety-critical systems. Data-driven models leveraging available partially characterized dynamics have potential provide reliable typical data-limited scenario high value systems, thereby...