Miguel Heleno, Berkeley Lab
YouTube Stream: https://youtube.com/live/vRNO-7v7_VQ
Join group to receive calendar invite: https://groups.google.com/a/modelingtalks.org/g/talks
Abstract:
Electric power distribution grid infrastructure accounts for the largest share of electric utility capital expenditure, directly impacting consumer electric bills. These high infrastructure costs are driven by electrification of other sectors, such as electric vehicles (EVs) and heat pumps, and by the growing penetration of distributed energy resources, such as solar PV. They are also driven by the need to prepare grid infrastructure for an increasing number of extreme events, whose nature is uncertain and whose impacts can lead to major power outages. This talk will cover core modeling techniques to plan distribution grid capacity infrastructure, including the representation of grid technical constraints in economic problems, as well as methods to deal with uncertainty in the context of reliability and resilience planning decisions.
Bio:
Miguel Heleno is a Staff Scientist at Lawrence Berkeley National Laboratory, where he manages power systems research within the DOE Office of Electricity programs. He is a Senior Member of the IEEE and an Associate Editor of the IEEE Transactions on Energy Markets, Policy and Regulation. He holds an MSc in Electrical Engineering and Computer Science and a PhD in Sustainable Energy Systems from the University of Porto through the MIT Portugal Program. Miguel has more than 15 years of experience in energy systems research and innovation, in Europe and the United States, focusing on grid optimization and planning, power systems economics, and energy policy.
Summary:
Focus: optimization for the power grid, focusing on the distribution portion
Power distribution grid
accounts for 32% of US utility capex / $50b in 2025
riven by increasing electrification
Distribution grid planning
New power lines
New circuit ties
New substations
Capacitor banks
Voltage regulators
New controls/automation
Utility-owned distribution resources (e.g. large-scale batteries)
Problem: optimal/least cost combination of resources that makes the grid feasible, safe and reliable
This is a mixed integer - nonlinear problem
Integer: options for investments (install one component/line or another)
Nonlinear: power flow equations
Demand varies across space and time, hourly resolution
Use-case: increased solar PV adoption by consumers
Planning for the distribution of possible solar PV adoption curves according to simple models
Utilities focus on the average scenario and try to be robust to the min and max scenarios
This is too coarse:
PV generation varies throughout the day and year
The locations where PVs are deployed by consumers matters a lot (e.g. all in the same city)
Applying Bayesian Optimization to analyze Critical Scenarios
Goal: find set of unique scenarios that are non-dominated by other scenarios to capture the diversity of the scenarios the grid may face
Divide grid into groups
Define stress functions that define the worst-case voltage/power flow problems that may occur there
Define a surrogate Gaussian Process model that estimates stresses for unevaluated scenarios
Bayesian Optimization used to find scenarios that lead to unique patterns of grid stress
Approach quickly finds the set of all non-dominated grid scenarios
Use-case: preparing the grid for unexpected heat waves
Today: Utilities predict weather-driven peak demand from historical data
Long-term analysis: peak load
Short-term planning for emergency operation
During extreme weather temperature-load relationship is highly non-linear and we have little data on its exact shape
Hard to predict people’s exact behavior and response to events
E.g. as temperature rises people start turning on AC but once all the ACs are on, additional warming causes no additional demand
Approach: use a physics-based model to predict demand under specific scenarios
Using CityBES https://citybes.lbl.gov/ to model cities and their infrastructure
Modeling consumer behavior within the context of the city using EcoBee thermostat data
Calibrated using 2021 meter data from PG&E
Observation: temperature to grid demand curve is parabolic / curve down
At extreme temperatures utilities engage in emergency operation
Utilities ask for flexible power users to reduce demand and they respond; not fully, but it does significantly and measurably reduce load during extremely hot events
Planning for resilience
Reliability: mitigation of outages caused by events (expected value of interruptions)
Resilients: controlling the risk posed by extreme events (high impact / low probability)
Must co-optimize against both reliability (expected value of cost) and resilience (value at risk) metrics, using a constant to balance their weight
Used to design grid investments for ComEd in Chicago to design under risk aversion
Risk-neutral plan (focused on reliability: small 3 line interconnection but the CVAR cost is $2.6 for an extreme events)
Risk averse plan (focused on resilience: 5 new line interconnections, CVAR is $1m)