My research group studies sensing, communication, learning, and control issues in large-scale networked and cyber-physical systems. The goal of this research is to enable greater autonomy, higher reliability, adaptability, and efficiency in these systems.
Buildings are large-scale distributed systems housing hundreds of sensors, actuators, and processors. Yet they are largely unaware of the presence, preferences, and actions of the occupants. For example, the HVAC system typically runs on a static heating/cooling schedule that does not take occupancy into account, wasting a lot of energy in conditioning empty or partially-occupied spaces. We investigate how time-series data generated by heterogenous sensors can be fused to estimate the number of occupants and their thermal satisfaction, infer their activities, and localize them inside the building. Since sensor data streams are intermittent and noisy in real-world settings, we develop a multimodal fusion model that makes opportunistic use of the available sensor data, captures complementary information across different modalities, and is robust to sensor faults, network issues, and changes in the environment (see Centaur). Additionally, we study how this approximate knowledge of occupancy at the level of individual rooms can be incorporated in control loops.
IoT and mobile devices are equipped with a myriad of sensors to monitor the environment and/or understand user behavior. The sheer amount of data collected by these sensors poses serious threats to our privacy. For example, building occupancy can be inferred from data collected by smart thermostats, people's health and wellness can be inferred from conversations recorded by smart home personal assistant devices, and their age and gender can be determined by a fitness tracking application running on their smartphone. To prevent these unwanted inferences and minimize distortion of useful data, we train machine learning models (e.g., variational autoencoder with discriminator networks, guided diffusion models) on sensor data streams to obscure private information contained in the sensor data. These techniques enable data sharing with third party applications without compromising the user's privacy. See PrivacyThroughDiffusion, ObscureNet, Blinder, and our privacy-preserving, latent space transformation technique.
Distributed energy resources are traditionally operated by simple, reactive controllers, which are suboptimal. We explore how learning-based control of these resources can improve their performance, increase the revenue generated, and minimize detrimental impacts on the power grid. We develop adaptive and distributed control algorithms that run on edge devices and compare their performance with various model-based and model-free control algorithms.
Data-enabled and learning-based control techniques are gradually being integrated into the operation of cyber-physical systems, from electric power systems and buildings to autonomous cars and intelligent transport systems. The operation of these systems is governed by the laws of physics and is subject to various constraints. We explore how physics and operational constraints can be incorporated in the design of these control and optimization techniques to ensure safe, reliable, and robust operation of these systems.
Energy systems are becoming increasingly complex with dynamics that span multiple timescales. Designing/validating new optimization and control algorithms for the smart grid requires simulation tools that accurately model the dynamics of the physical system and its interactions with various other domains, such as communication networks, building systems, transport systems, and electricity markets. We use concepts from operating systems and software engineering to design a modular, efficient, and secure co-simulation platform. Our smart grid co-simulation platform is called Maestro. We use this simulation platform to develop control and state estimation algorithms in addition to exploring cyber attacks (e.g., false data injection).
COBS is the simulation platform we developed for smart building. It is specifically designed to facilitate interactions between Reinforcement Learning (RL) agents and multiple building systems.
Power distribution networks have a sparse topology which is unknown to system operators. For cost reasons, distribution networks are presently monitored only at a small number of points beyond the substation. We leverage this sparsity along with the available sensor measurements and pseudo-measurements to (a) identify the network topology and model parameters, (b) estimate the state, (c) identify and pinpoint events to a small part of the network, and (d) monitor harmonic generation and propagation.
Similarly, in the building domain, we use time-series generated by smart thermostats to identify the thermal model of a home. Knowing this model helps design controllers that minimize energy use and discomfort.
We designed a multi-resolution search algorithm for time series data analytics. Leveraging statistical summaries of data offered by BTrDB for a desired temporal resolution, this algorithm is capable of efficiently identifying rare, critical events across a wide range of temporal scales (micro-seconds to years) in vast amounts of telemetry data. This algorithm has been used to detect tap operations, voltage sags, and switching events as a part of the ARPA-E micro-synchrophasors for distribution systems project.
We designed an efficient algorithm for solving the inverse Power Flow (iPF) problem in a three phase distribution network equipped with phasor measurement units. This algorithm is capable of inferring network topology, i.e., the admittance matrix, and identifying switching operations from voltage and current phasor measurements in cases with and without hidden states.
Variable-power distributed energy resources, such as solar photovoltaic and storage systems, and high-power elastic loads, such as electric vehicle chargers, are being installed at a phenomenal rate in power distribution systems. Such active end-nodes can affect the reliable operation of the grid if they are not controlled properly. The joint control of active end-nodes in quasi real-time based on fast timescale measurements could enable operators to meet their efficiency and fairness requirements, and enhance service reliability by preventing network overloads, reverse flows, and voltage deviations beyond operating limits.
Uncoordinated charging of EVs may overload distribution transformers and branches. Assuming that EV chargers obey control signals sent from the grid, and support variable-rate charging, we propose a distributed control algorithm which adapts the charging rate of EVs to the available capacity of the distribution network in real-time. This algorithm is stable and yields a proportionally fair charging rate allocation. This algorithm is similar to the congestion control algorithms used in packet-switched networks.
We investigate the problem of identifying energy consumption profiles for residential consumers so that they can be used to produce accurate, personalized and thus effective feedback. Since residential energy consumption is driven by a wide range of factors such as routine, occupancy, activity and weather, we pursue a modelling approach to formulate energy profiles that capture these factors. Categorizing consumers based on these profiles can provide valuable insight into their relative energy consumption and forms the foundation for comparative feedback applications.
We developed an application for Microsoft HomeOS which detects abnormal household electricity consumption patterns and subsequently informs homeowners by sending an email or a text message. This application communicates with a Current Cost Envi device to monitor the aggregate power consumption of a home.
We used teletraffic theory to size power distribution networks just as it is used to size telecommunication access networks. Specifically, we proved the equivalence between a model of a distribution branch comprised of a transformer and storage that we want to size for a given underflow probability ε, and a queuing model that we want to size for a given overflow probability ε. Based on this equivalence, we showed how existing teletraffic analysis can be applied to jointly size transformers and storage.
We developed a testbed to obtain fine-grained measurements of active power consumption of homes. This testbed is comprised of 30 measurement nodes; each node has an Envi device to measure the instantaneous power consumption and a netbook to collect and send the measurements to our server. The dataset can be downloaded from this link.
We developed a testbed to monitor temperature, humidity, sound, light, and air pressure at a fast timescale (every six seconds). We deployed 24 wired Weatherduck sensors in graduate students' offices and labs in Davis Centre at the University of Waterloo. These measurements are used to infer room occupancy and reduce energy consumption of AC and electric heater when the room is unoccupied.
I am grateful for the support of the following funding agencies and corporate sponsors: