Tag Archives: load forecasting

Advanced power system modeling need not mean more complex modeling

A recent article by E3 and Form Energy in Utility Dive calls for more granular temporal modeling of the electric power system to better capture the constraints of a fully-renewable portfolio and the requirements for supporting technologies such as storage. The authors have identified the correct problem–most current models use a “typical week” of loads that are an average of historic conditions and probabilistic representations of unit availability. This approach fails to capture the “tail” conditions where renewables and currently available storage are likely to be sufficient.

But the answer is not a full blown hour by hour model of the entire year with many permutations of the many possibilities. These system production simulation models already take too long to run a single scenario due to the complexity of this giant “transmission machine.” Adding the required uncertainty will cause these models to run “in real time” as some modelers describe it.

Instead a separate analysis should first identify the conditions under which renewables + current technology storage are unlikely to meet demand sufficiently. These include drought that limits hydropower, extreme weather, and extended weather that limits renewable production. Then these conditions can input into the current models to assess how the system responds.

The two important fixes which has always been problem in these models are to energy-limited resources and unit commitment algorithms. Both of these are complex problems, and these models have not done well in scheduling seasonal hydropower pondage storage and in deciding which units to commit to meet a high demand several days ahead. (And these problems are also why relying solely on hourly bulk power pricing doesn’t give an accurate measure of the true market value of a resource.) But focusing on these two problems is much easier than trying to incorporating the full range of uncertainty for all 8,760 hours for at least a decade into the future.

We should not confuse precision with accuracy. The current models can be quite precise on specific metrics such as unit efficiency as different load points, but they can be inaccurate because they don’t capture the effect of load and fuel price variations. We should not be trying to achieve spurious precision through more complete granular modeling–we should be focusing on accuracy in the narrow situations that matter.

Not grasping the concept: PG&E misses the peak load shift

Utility peak shifted by solar graph

PG&E in its 2020 ERRA Forecast Proceeding testimony wrote “however, BTM DG [behind the meter distributed generation] has a limited impact to the annual system peak as customer-owned solar photovoltaic (PV) generation is minimal during the peak hour of 7 p.m.” Uh, how does PG&E know that customer-owned solar doesn’t contribute to reducing the system peak if PG&E does not meter that generation?

PG&E actually has it wrong. Customer-owned solar has in fact reduced the former pre-solar peak that used to occur between 2 and 4 p.m. The metered load that PG&E can see, which is customer usage minus solar output (BTM DG), has shifted its apparent peak from 4 p.m. to 7 p.m.–3 hours. The graphic above illustrates how this shift has occurred. (PG&E produced a similar chart of its 2016 loads in its TOU rate rulemaking.) So BTM DG has had a profound impact on the annual system peak.