US ELECTRICITY
FUTURES

Redesigning the market operator console for setting electricity futures prices

CHALLENGE

Our team at Futuredraft was hired by GE Grid Solutions to redesign the electricity market console, which is used to set futures prices for the US electricity market.

Futures prices are generated by a proprietary algorithm, based on a model of the grid, which the market operators update each business morning to reflect the current status of principal grid elements, the generators, outtages and transmission lines that govern the availability of electricity at any given point, or node.

Sometimes the algorithm generates prices that fall outside of the minimum or maximum thresholds established by the Federal Energy Regulatory Commission, which governs market operations. In such cases, the market operators need to adjust the grid model to smooth out the price spikes and dips.¹

Once the model is updated, the pricing algorithm then generates next-day market prices for each of 100,000+ nodes, for every hour of the day. These prices are then published, and electricity traders can then buy the futures, cross-referencing the published prices against their own predictions for the market.

The time to complete this process is called the "market clearing time", and currently takes around 4 hours.

The purpose of the redesign is to improve the efficiency of the market operations process enough to reduce the market clearing time from 4 hours to ½ hr, thereby improving the efficiency of the futures market.

[In a perfect world, this whole process will be automated, but there are so many disparate systems under distributed control that it's going to take a long time for that to happen. Until then, the goal is to maximize the efficiency of the human operators.]

The futures market in the US is managed according to region by the 9 independent operators that bear collective responsibliity for managing the efficient flow of power to homes, businesses & government facilities across the US. The futures market acts as a kind of shock absorber for the grid, modulating the flow of power when it is scarce, and encouraging consumption when electricity is freely available.

Our solution, once implemented, is intended to make that shock absorber work better. In doing so, it will reduce the incidence of blackouts and and brownouts in grid regions across the US, which currently cost between $18B & $33B/yr.

I was one of six senior designers on the project, and led the contextual inquiry/user research together with a UX designer from GE at US grid command and control centers in Norristown, Pennsylvania (PJM) and Indianapolis, Indiana (MISO). I joined the project a week after the initial discovery workshop, and stayed on through the successful conclusion of the engagement.

DISCOVERY

Futuredraft's engagement began with a two day workshop in Seattle, at GE’s Grid Solutions offices. The engineers who designed the original algorithm sketched out the conceptual models and data flow that underpinned the market clearing process. Three designers from our team, Jorge Arango, Chris Baum and Matt Nish-Lapidus, conducted the workshop and created a synthesis deck that became my point of departure.

Additionally, we had a weekly call with the GE team that provided additional insight. The engineer who designed the original version of the console went over the current system screen by screen, an exercise that proved invaluable in getting a handle on the clearing process.

Today it takes market operators on average of 4 hours to set market pricing for next day electricity futures in the US. This delay causes market inefficiencies that lead to brownouts and blackouts during peak periods, and result in unnecessary surges in pricing during those times.

Next day market pricing is set in each of the nine US electricity markets using a complex algorithm that looks at a model of the grid and sets future prices in a way that mirrors as closely as possible the demand vs. availability of power at each of 100,000+ nodes around the country.

The model takes as input any outages, changes in generator output, or alterations in the transmission status of any of the high power lines that comprise it. Armed with this information, it generates prices for each node for each hour of the day.

DEFINING THE PROBLEM

Currently, the process of adjusting the model is performed using a series of spreadsheets and macros, with elaborate protocols for workflow and order of operations.

In order to improve the workflow, we needed to not only appreciate the environment and context, but develop a functional understanding of how the market operators configured the system, as well as their mental models and tasking.

Initially a few things were clear:

1. The market operators were drowning in an ocean of data with very little in the way of visual hierarchy to help them see which cells of the spreadsheets were of interest.
2. The few visualization tools they did have were poorly implemented, inconvenient to access, and hence not particularly useful.
3. The console itself was patterned after the data but was not well suited to the work process of the operators. As a result, routine tasks were onerous.
4. Shortcomings in the console had been bridged using a series of home-grown applets that became new sources of frustration as their code bases began to age and their original authors moved on.

DESIGNING A SOLUTION

To address these shortcomings, we began sketching the process, and the principal screens, with an eye towards developing a visual hierarchy. We also sought to provide them with tools for creating their own views of the data that were appropriate to the kinds of analysis they were performing:

Shortly thereafter, we were afforded the opportunity to visit PJM headquarters, one of the grid and market operations center for the northeastern US. This was a coup, because up until now, we had not had direct access to the end customer or their users. The second day we spent with the engineers going through the clearing process from beginning to end. Even with all the prep, it was initially quite disorienting. We took a ton of verbatim notes, which we dissected with the engineers as their schedules permitted.

Our efforts were rewarded, though, with a much clearer picture of the users, their mental models, workflow, pains & opportunities.

Upon our return, I shared with my colleagues what we had learned. We divided up the console into sections. Roughly speaking, these were Initialization, Operation, and Execution. I took on Operation, which involved the thorny task of identifying anomalies and adjusting the model to resolve them. This was the most time consuming part of the clearing process, and after having witnessed some of the pain points first hand, I was beginning to formulate a hypothesis on how to solve it.

The idea was to create an initial visualization of the grid model that would allow the operator to quickly identify each anomaly, and then create a system of structured data views that were dynamically created from the dataset in order to lead the operator quickly to the cell or cells that needed to be changed in order to mitigate the anomaly.

We took each set of wires and normalized it against the other two to ensure we were delivering a coherent experience, hashing out inconsistencies as they arose. We then presented to the GE team and got their feedback.

Next we developed some preliminary visual designs and circulated them for review.

Then it was on to visit another market operator, MISO in Indianapolis, which maintains the power grid in many of the midwestern states. There we did some more contextual inquiry, and we shared our latest designs.

Intermediate Wireframes (above)

DELIVERY

We took the rest of the feedback and input from this visit and revised the designs again. Finally we created Invision prototypes of all three flows and delivered them in our final presentation to the client. Here are the visuals for the override flow, for which I was responsible.

As of this writing (9/7/2019), the design and associated proposal from GE is currently in the final evaluation phase by the FERC after competing successfully against proposals from 15 competing firms incl. Siemens, Accenture & Cray.

¹ These adjustments are then converted into field orders, which are dispatched to maintainance personnel to conform the grid to the model.