This thesis presents progress in overcoming two challenges in achieving a reliable electric power system: frequency regulation and secure state estimation against cyber attacks.

Frequency regulation is a type of ancillary service used to control the grid frequency around its nominal value. Recently, the higher penetration of renewable energy sources has increased the demand for frequency regulation reserves. This thesis explores the feasibility of using commercial buildings for this application. Commercial buildings are a tremendous untapped source because of their large consumption and thermal inertia, as well as being able to adjust their electricity consumption continuously. However, large disturbances such as occupancy and the complexity of the heating, ventilation and air conditioning system of commercial buildings make it challenging to: 1) identify a model that accurately describes its temperature evolution and is amenable to control, 2) design a robust frequency regulation controller. This thesis tackles both challenges. First, it proposes a physics-based and a data-driven method to identify a building model that captures internal gains such as occupancy. Both methods are used to identify models of the same testbed, and a quantitative

comparison of the resulting models is made including open-loop prediction accuracy and closed-loop control performance. It is concluded that a data-driven model may be suitable for temperature critical applications such as frequency regulation. Second, this thesis improves on existing frequency regulation control schemes and proposes a bilevel controller that is suitable for buildings subject to larger uncertainties, where accurate models are unavailable. Finally, eld experiments in accordance with the Pennsylvania, New Jersey and Maryland Interconnection's regulation market rules are conducted on an occupied building during both daytime and nighttime, which demonstrate the suitability of the data-driven building model and the performance of the proposed frequency regulation controller.

Advanced instruments such as phasor measurement units that communicate over wireless networks to achieve better efficiency of the power system are becoming increasingly pervasive, especially under the smart grid initiatives. However, these communication networks are vulnerable to cyber attacks that can be erratic and difficult to predict. To add to the challenge, the dynamics of the power system cannot be

approximated by a linear model when it's under severe disturbances. This thesis develops a secure state estimation method for linear dynamical systems under sensor attack, and then extends it to two classes of nonlinear systems and applies it to the nonlinear power system. Both estimation methods assume that the attack signal can be arbitrary and unbounded, and the set of attacked sensors can change over time. More specifically, we use feedback linearization to transform the nonlinear system into an equivalent linear system. We then formulate the secure estimation problem into a classical error correction problem, from which we propose an l1-optimization

based estimator that is computationally efficient. In addition, we derive the maximum number of sensor attacks that can be corrected with our estimator and propose to use pole placement techniques to design a feedback controller such that the resulting secure estimator can guarantee accurate estimation. Finally, to improve the estimator's practical performance, we propose to combine our secure estimator with

a Kalman Filter (KF), where the KF serves to lter out both occasional estimation attacks by the secure estimator as well as noisy measurements.