How AI/ML can optimize renewable unit commitment and dispatch

Unit commitment and economic dispatch are not new problems to solve, yet commonly used modeling techniques and solutions are not able to keep up with the complexity introduced by the influx of renewable generation. Currently the level of uncertainty caused by the forecasting inaccuracy of renewable assets leads to curtailment and overuse of fossil fuels. Power generators and utilities must maximize use of renewables without sacrificing asset reliability today while preparing for a future of significantly more renewables. To meet both these needs, artificial intelligence (AI) and stochastic modeling are needed to account for the increasing complexity and optimize dispatch.

To learn more, Rachel Farr, senior director of Product Management at GE Digital, answers questions on how GE Digital is working with vertically integrated utilities to use AI and machine learning (ML) for reliable and clean energy production.

We see that many utilities have set targets to reach net zero emissions by 2050. To achieve this goal, they are adding substantial renewable generation resources to their fleets every year. We expect the CAGR (compound annual growth rate) for wind and solar generation to be on the order of 10-15% for at least the next 10 years.

As these utilities add renewable generation, they are also retiring older conventional, dispatchable assets like coal. Furthermore, distributed energy resources – rooftop solar as an example – are growing as well. Weather models are much better than they used to be, but the effects of small changes in cloud cover or wind speed/direction significantly impact the dynamics of renewable generation.

All this together creates much more uncertainty day-to-day, hour-to-hour, minute-to-minute than has ever been encountered in the past. If we don’t improve upon the technology used to optimally dispatch the non-renewable resources, we will risk reliability, overcompensate backstopping of renewable generation with more carbon-producing resources, require building excess renewable resources to be certain to meet demand, or all the above.

Yes, this was a very interesting study. ISO New England (ISO-NE) is not a utility but instead an independent system operator, but the study did a great job of highlighting challenges faced by ISOs and utilities alike seeking to decarbonize. As of the study, ISO-NE had 350 generating units or 32.7 GWs of generating capacity with goals across various states of reducing CO2 and meeting net zero by 2050.

The study concluded to reach 56% renewables use, the reserve margin – i.e., how many extra resources are needed to keep the system reliable in times of stress – may need to increase by an order of magnitude by 2040 (i.e., from 15% to 300%). A lack of diversity in the future resource mix may necessitate the construction of many more new resources.

That is a huge challenge to manage; furthermore, the transition won’t happen overnight. Utilities need to find the balance to perform while transforming, and that is where advancements in software and analytics can help to accelerate the value of renewables while extending the value in dispatchable thermal fleet.

Specifically, regarding power generation unit commitment and dispatch, we are doing a number of really exciting things to advance the analytics and software tools to enable utilities to maximize the benefit of renewable energy without sacrificing reliability. We have spent the last several years investing in AI/ML-enabled digital twins for forecasting generation for all resource types – wind, solar and thermal, including heat rate.

As far as I know, we are the only software provider than enables this technology for thermal plants in addition to renewable assets, making sure utilities not only have foresight into uncertainty regarding renewable generation, but also always know the most efficient thermal resources to use to backstop renewables. Furthermore, we are in the process of bringing Fleet Orchestration to market with AI/ML-enabled probabilistic recommendations for unit commitment planning that will facilitate fast, reliable, economic response to changes in renewable generation.

Utility-scale renewable generation and distributed energy resources are making the problem of solving for lowest-cost, reliable unit commitment and dispatch much more challenging due to all the added uncertainty. The legacy tools in this space cannot handle the dynamics or complexity-faced without putting significant margins on operating reserves, which means utilities spend more on fuel or purchased power and are not able to minimize their carbon footprint. If effectively used, AI/ML can be a game changer in helping manage and provide visibility to the uncertainty and risk so it can be more efficiently managed.

Yes, we did a study of a ~6 MW fleet in North America and estimated spinning reserve could be reduced by nearly 60% in certain instances with access to these technologies. That could be worth upwards of $10 million in annual fuel savings, not even accounting for the reduced carbon emissions. Utilities are investing significant capital and Operation & Maintenance resources in renewable projects to help them achieve their sustainability goals. If they do not have access to these digital technologies, they may have to operate their dispatchable thermal fleet in a more conservative way to manage risk, thus rendering some of their clean energy ineffective.

Like with many other software or digital solutions, the value is highly reliant on the organizational alignment. In this space specifically, what was once a day-ahead necessity to produce a unit commitment plan for the next day is changing into a continuum from day-ahead to intra-day to real-time dispatch. This requires not just new technology to facilitate the best decision-making but also a willingness by the organization to bring together what might have been siloed workstreams in the past between generation and operations.