Advancing Long-Term Forecasting for Tomorrow’s Grid

Moving forward, both system-level and distribution-level planners need to improve their forecasts by incorporating new methodologies that account for the disruptive changes to load. More variables mean forecasting is now more complicated, and the industry must respond to the complexity by understanding every hour of the year.

8,760 hours: The future of long-term modeling and forecasting

The vision for long-term load forecasting builds from the bottom up, using parcel-level load data that can be aggregated up to the feeder or substation level for distribution planners, all the way up to the system level for ISO/RTO forecasts. While the industry as a whole is not there yet, innovative planners and load forecasters across North America are building toward that future.

As forecasting evolves, “load-to-weather” modeling will remain relevant. These models depend on data to find the relationship between consumption behavior and temperature and humidity. However, there is relatively little data on these relationships during events like extreme heat and cold, and models often perform poorly in extreme scenarios. ISOs, RTOs, and research organizations are working to improve load-to-weather models with state-of-the-art climate modeling to account for the effect of increases in extreme weather.

Planners are also working to gain a better understanding of the emerging external load variables that are disturbing traditional models, such as DER and EV adoption and new large loads like data centers. In many instances, the scale of these changes is without historic precedent, and many planners lack the data to effectively model these new variables.

At the cutting edge — forecasts that account for climate changes and changing customer use patterns — planners are moving toward what is called 8,760 forecasting.

There are 8,760 hours in a year, and planners want to model the possibilities for every hour. This allows them to understand the impact of all the variables in the mix. For example, record heat on a mid-summer afternoon might not yield a demand peak on a system with high solar and battery storage adoption. But if demand response programs drain battery storage capacity by the time the sun goes down, when a large number of EV owners plug in to charge, that could be a recipe for a late demand peak. 8,760 forecasting solves those complex scenarios by modeling every variable over every hour of the year—both external variables like customer-owned DERs and EVs but also emerging utility-controlled ones like demand response and flexible load management resources.

ISO and RTO planners are building toward that level of granularity. System-level planners will be the first to develop the 8,760-hour-based models that enable 8,760-model forecasts. Distribution-level planners need to understand and perform some 8,760 modeling, but likely won’t have the resources to achieve 8,760 forecasting in the near term. Eventually, however, everyone will have to get there to achieve the optimal bottom-up load forecasting vision.

Focus areas to prepare long-term forecasting for an era of complexity

Changing long-term load forecasting practices isn’t easy. However, it is possible to model emerging climate, consumer behavior, technology, and market changes. EPE Consulting has worked with planners at every level and recommends focusing on three areas to drive the necessary change:

1. Change processes

In the past, silos between system-level and distribution-level planners, which led to the use of different data and assumptions in separate models, didn’t harm planning outcomes. That’s not true anymore. In one instance, a utility performed a bottom-up study to forecast solar adoption in its territory. The ISO responsible for planning in the utility’s territory did the same without any of the same bottom-up data. The forecasts were hundreds of MW apart in their prediction for solar adoption, leading to a dispute that was settled by FERC in favor of the utility. The answer in these instances is to increase coordination, share data, and create a single common source of truth.

Another important process update is to incorporate the use of more scenarios in forecasting. More load-impacting variables means more complexity. Planners must evaluate more scenarios and improve their multi-scenario analysis to understand the range of risks of different outlooks that could develop. Relying on a single, deterministic future can create a false sense of certainty, obscure early warning signs of emerging rapid changes, and reduce decision-makers understanding of the tradeoffs and risks they are facing across the planning horizon.

2. Upgrade technology

The increased complexity of load forecasting has driven planners from simple spreadsheets to more complicated software tools. These tools provide access to common data sets for use by many and help planners run complex scenario analysis.

Planning organizations must leverage technology to improve their data collection and their use of that data to manage rising complexity.

3. Start collaborating

To fully break down silos and improve forecast quality, broad collaboration is a must.

When cross-departmental or cross-organizational groups are formed to discuss needs, individuals often discover their data requirements and modeling capability needs aren’t unique to their use case. What can be used to improve a distribution-level forecast can also be used to design better customer programs and rates, for instance. A single investment can serve many outcomes.

By developing strategies to evolve processes, technology, and collaboration, planners at every level can develop the modeling and forecasting capabilities required to better identify, understand, and manage the uncertainty associated with the complexities they face.