Truong X. Nghiem

Research Interests

My research brings together techniques from Control Theory and Computer Science to control large and complex cyber-physical systems (CPS), with applications in smart and energy-efficient building controls, automotive controls, and embedded control systems. I place great emphasis on model-based approaches, analytical and formal methods, and computational techniques in my research. My work has focused on three broad themes:

  • Control/Scheduling for Peak Demand Management in Energy Systems: I have developed a new class of scalable and lightweight control/scheduling algorithms to reduce the peak demand in dynamical energy systems, with applications in building systems and electric vehicles.

  • Smart Building Controls: I have studied cost-efficient modeling and advanced controls for smart and energy- efficient buildings, and developed a successful tool for integrated modeling, simulation and controls of buildings.

  • Control, Computation, and Verification of Cyber-Physical Systems: I have developed a method to bridge the gap between a high-level control model and its implementation, and a practical testing framework for falsification of temporal properties of nonlinear hybrid systems. I have also focused on the co-design of control and anytime approximate computation for real-time control systems.

For my publications related to each research area, please visit this page.

Control/Scheduling for Peak Demand Management in Energy Systems

Many practical engineering systems consist of a demand side and a supply side. Well-known examples can be found in energy systems, for instance: electric grids, HVAC (Heating, Cooling and Air Conditioning) systems, and electric vehicles. However this characteristic is also present in non-energy-related systems such as traffic systems. In these systems, peaks in the demand are usually bad because they can cause undesirable effects on the supply, either technically or economically. For example, peaks in the electricity demand result in oversized power plants as well as the construction of back-up plants; otherwise they can cause grid instability and power outages. Because of this reason, large electricity customers are often charged at very high rates for their peak demand to discourage them from using electricity during critical time. In battery-powered electric vehicles, peaks in the power drawn from the battery cause the battery to heat up quickly, eventually decrease its lifetime. Peaks in traffic demand often cause traffic jams.

Many approaches, in various applications, have been proposed to reduce peak demand (known as peak demand management). A popular technical approach to tackle this problem is to utilize optimization (dynamic programming, model predictive control) to calculate optimal schedules for the operation of the systems. However, this approach usually requires an accurate model of the system (which might be very difficult to obtain), accurate forecasts of the demand and disturbances (again, might be very difficult to obtain), and high computational capability (to solve large and complex optimization programs). Another approach to the peak demand reduction problem is to introduce some flexibility into the systems, typically via some direct or indirect form of storage, and exploit this flexibility to reduce and smooth out the demand. For example, thermal energy storage can be added to HVAC systems, or the room temperature can be fluctuated within a comfort range, or super-capacitors are used with batteries. In this line of research, I study the control and scheduling of these dynamical systems for peak demand reduction via direct/indirect storage. The goals are to develop control and scheduling architectures and algorithms that

  1. effectively reduce the peak demand for the supply side, while maintaining certain operational or safety constraints of the system; and

  2. does not require highly accurate system model and forecasts; and

  3. does not require high computational capability (particularly for systems with fast dynamics).

Smart Building Controls

In this area, I have studied modeling and advanced control methods for smart and energy-efficient buildings, and have investigated cost-efficient building model training. I have also developed a successful open-source tool for integrated modeling, simulation and controls of buildings.

Building Modeling and Advanced Controls

Buildings consume significant energy, about 41% of primary energy consumption in the United States in 2010. Indoor environments are among the most influential factors in the productivity, safety, and physical and mental health of many humans. Consequently, smart and energy-efficient building controls have recently been at the forefront of research and developments in the U.S. Buildings are large, complex systems with many interacting dynamical subsystems: mechanical components (e.g., valves, ducts, coils) are tightly coupled with electrical components (e.g., fans, pumps) and control components (e.g., thermostats) to form a complete building system. Buildings are also affected by large and highly uncertain disturbances (e.g., weather, occupants). Furthermore, certain classes of buildings are subject to stringent operational or safety requirements. Therefore, buildings are “messy” plants that are difficult to model accurately and control efficiently. My work in this direction has focused on these two challenging problems, namely building modeling and advanced controls.

MLE+: Tool for Integrated Modeling, Simulation and Controls of Buildings

A challenge that many early researchers in advanced building controls, myself included, faced was the unavailability of a good comprehensive tool for both building simulation and control design and analysis. While there exist high-fidelity building simulators, such as the DoE’s EnergyPlus, and high-quality control toolboxes, such as Matlab and Simulink, they do not work well together. Several attempts had been made to enable co-simulations between EnergyPlus and other software, however control researchers often found them difficult to use and unsatisfactory. Having realized this missing link, I developed a toolbox called MLE+ to facilitate advanced building controls research in Matlab/Simulink with EnergyPlus. I have used MLE+ extensively in my own research, and also released it to the public2. MLE+ was selected to be featured on the DoE’s EnergyPlus website3. It won the 2012 ACM Building Systems Symposium (BuildSys) Award for best demonstration. More importantly, it has been used by many researchers worldwide in various types of buildings research projects, including at least two doctoral theses. We have collaborated with the National Renewable Energy Laboratory (NREL) to develop campus-wide buildings modeling and simulation of the NREL campus in Golden, CO using MLE+. My current efforts involve extending the features of MLE+ and integrating MLE+ into the TerraSwarm project.

Cyber-Physical Systems (CPS)

Cyber-Physical Systems (CPS) integrate physical processes with computation and networks. They can be found in almost every complex system in practice, from as small as biological systems and embedded medical devices, to as large as airplanes, space shuttles, and the electric grid. I have started pursuing a research interest in CPS, particularly real-time and embedded control systems and hybrid systems. With the proliferation of embedded computing components and software components in modern control systems, there have been new challenges such as automatic synthesis and verification of control software, and secure and high-confidence control systems. Following are several past and on-going research directions in this research thrust.

Optimal Scheduling of Computation in Time-triggered Control System Implementations

Bridging the gap between high-level modeling or programming abstractions and implementation platforms is one of the key challenges for embedded software research. While high-level modeling provides a useful abstraction for controller design and analysis, the discrepancy between its idealized semantics and its implementation can result in poor control performance, even instability of the closed-loop system. My research developed a method, which combines control and computation techniques, to quantify rigorously the effect of computation on control performance and stability of a digital linear control system on a time-triggered architecture. This helps bridge the gap between model-based design and platform-based implementation. Furthermore, it allows us to compare and partially order different implementations with various scheduling or timing characteristics. I extended this result and studied the optimal scheduling problem to minimize the error of the implementation of embedded control systems. I also proposed a Genetic Algorithm for the problem, which was shown by experimental results to be very promising when compared to random search and exhaustive bounded search.

Falsification of Temporal Properties of Hybrid Systems

As many operational and safety specifications of control systems can be expressed in temporal logics, the problem of verifying temporal properties of a given control system is of great importance. Verifying temporal properties of a nonlinear hybrid systems is particularly challenging due to the complexity of its dynamics. The undecidability of this verification problem over such complex continuous systems mandates the use of lightweight formal methods that usually involve testing. In collaboration with the NEC Laboratories America, I developed a testing framework for falsification of temporal properties of nonlinear hybrid systems. The advantages of our approach are not only that we can falsify nonlinear hybrid systems, but also that we can provide robustness guarantees even to systems that have been proven correct. This work also resulted in my patent application filed by NEC.

Co-design of Control and Anytime Approximate Computation

My current research in this area investigates the co-design of control and approximate computation for real-time control systems. All control computations are approximate, take time, and consume energy. These include not only the control algorithms but also the sensing, the state estimation, the actuation and the communication. Most algorithms we currently use in real-time systems are run-to-completion and provide either an answer upon completion or no answer if interrupted by a deadline before completion. Anytime algorithms, on the other hand, have an increasing utility with the length of execution time (and energy consumption) and can provide an answer with varying quality depending on the deadline. For example, in a computer vision based sensing system, the image quality and the parameters of the image processing algorithm could be adapted to meet the deadline, with an associated cost on the accuracy of the result. I am developing control design techniques and algorithms that can exploit this flexibility of anytime algorithms and adapt the deadline or the mode of computation to achieve an optimal balance between control performance and control cost while ensuring robust safety of the system. The ultimate goal is to co-design control and computation for anytime and approximate computation, with applications in computer vision based autonomous vehicles. The research is funded by the $5.9 million DoT University Transportation Center for safe and effective future vehicle architectures.