Duncan Callaway, Professor

Closed (1) Modular microgrid hardware development

Closed. This professor is continuing with Spring 2021 apprentices on this project; no new apprentices needed for Fall 2021.

We're engaged in a few projects that theorize how one could build and control an ensemble of small solar-plus-storage electricity generators. The technology would be useful for a variety of microgrid applications, ranging from electrification projects in parts of the world without electricity access to communities in California facing shutoffs during wildfire season. We want to build a testbed that can support proof of concept experiments, in particular implementing different control algorithms.

The undergraduate researcher will be responsible for (1) sourcing and testing solar panels, batteries and inverters, (2) networking those devices together in a microgrid lab environment, (3) implementing one or more existing control algorithms on the network and (4) exploring the long-run opportunities for real-world implementation of the concepts explored.


Day-to-day supervisor for this project: Jonathan Lee, Graduate Student

Qualifications: We're looking for students that are excited about the application, eager to learn and good at self study. Students should have some knowledge of electricity (one or more of EE137A, 137B, 113 desirable but not essential) and knowledge of programming (Python and / or Matlab most likely to be used).

Weekly Hours: 6-8 hrs

Off-Campus Research Site: This is a remote project that can be performed at home.

Related website: http://emac.berkeley.edu

Closed (2) Wildfire and power grid risk modeling

Applications for fall 2021 are now closed for this project.

We're part of a multi-institution team (Berkeley, UCSB, UCSD, Lawrence Berkeley Lab, Lawrence Livermore Lab) working to build models that describe phenomena at the intersection of wildfire and power grid risk. The team has economists, power system engineers, climate modelers, and we're data scientists of one type or another. Targeted outcomes of the project include depowering rules and detailed assessments of the costs and benefits of public safety power shutoffs.

Qualifications: The undergraduate researcher will be responsible for (1) building python-based notebooks that ingest massive amounts of time series and geospatial data and outputting dataframes that can be passed into machine learning libraries and (2) putting together and vetting exploratory predictive models.

Weekly Hours: 6-8 hrs

Off-Campus Research Site: This is a remote project that can be performed at home.

Related website: http://emac.berkeley.edu

Closed (3) Agent-based transactive energy in autonomous microgrids

Applications for fall 2021 are now closed for this project.

Traditionally, energy dispatch in microgrid settings is done in a centralized fashion, in which a smart operator actively manages generation resources to match electricity supply and demand. However, new decentralized and distributed schemes are becoming increasingly attractive as microgrids grow in size and centralized dispatch methods become harder to use and adapt to new resources and consumers joining the microgrid.

Currently, we are engaged in a project to develop an agent-based transactive energy setup, on which each agent in the microgrid, including generation providers and consumers, must supply cost curves each time step, similar to a pool-based market. In the current implementation, a privacy-preserving algorithm aggregates all curves, to quickly find the operating point and assign the energy breakdown and rewards to each agent. However, this method works in a single time-step approach. We want to extend the algorithm to consider multiple time-steps in order to include storage agents, while exploring how these types of agents will participate in this market setting.


The undergraduate researcher will be responsible for (1) review of storage schemes in pool-based market setups, (2) assistance with algorithm development and implementation and (3) development of computational experiments for testing algorithm performance.


Day-to-day supervisor for this project: Rodrigo Henriquez, Graduate Student

Qualifications: We’re looking for students with knowledge of the Python programming language, who are excited about the specific application and eager to learn more of it. It is desirable (but not required) that students have some knowledge of power systems (EE137A, EE137B) and microeconomics.

Weekly Hours: 6-8 hrs

Off-Campus Research Site: This is a remote project that can be performed at home.

Related website: http://emac.berkeley.edu

Closed (4) Machine Learning Surrogates for Scientific Computing

Applications for fall 2021 are now closed for this project.

In many engineering disciplines extensive computer simulations are carried out to ensure the the systems in question can satisfactory operate across all operating conditions for varying system parameters. These simulations typically require high fidelity models of the system and, as a result, can be time consuming to carry out. Within this project, the URAP student will explore the use of Scientific Machine Learning (SciML), specifically the application of surrogates, to accelerate these computer simulations.

SciML is a growing area of research that aims to blend traditional scientific computing approaches and recent machine learning advancements. In the context of this particular project, SciML will be used to develop surrogate models that accurately approximate the behavior of full-order physical based systems. These surrogates are trained using data-driven approaches. The original system is sufficiently excited to construct a rich training set, such that the resultant surrogates can well approximate the original system outside of the training set. The particular use case of interest within this project is accelerating the simulation of power systems with high penetration of wind and/or solar energy resources to ensure that we can simulate these systems in near-real time. The addition of wind and/or solar energy sources is resulting in these simulations taking significantly longer, under current approaches, due their fast response rates and spatial diversity.


The URAP will be writing Julia code to implement some of the most recent approaches in the field of surrogate modeling. This will include surrogates that seek to learn both the solution of the system as well as the dynamical behavior of a system. Some of the research questions that will be explored include sampling requirements of the original system to ensure sufficient accuracy of the surrogate, how to train systems whose timescale of dynamics span orders of magnitude (this impacts the allowable learning rate) and how different network architectures are more amenable to different use cases.


Day-to-day supervisor for this project: Ciaran Roberts

Qualifications: Undergraduate research apprentices need to have significant programming experience and some experience in training neural networks. The work will be carried out in the Julia programming environment. Experience in Julia is a plus, but not required.

Weekly Hours: 6-8 hrs

Off-Campus Research Site: This is a remote project that can be performed at home.

Related website: http://emac.berkeley.edu