Taking Unit Commitment to the Next Level: How Fleet Orchestration Helps Maximize Renewable Energy, Reliably

As countries, corporations, and organizations across industries set net-zero emissions goals to help tackle the challenges posed by our changing climate,  massively influence the overall carbon footprint of a region. In fact, the electric power sector emits about 30% of carbon dioxide in the U.S. However, by generating and distributing electricity to consumers and businesses, electric utilities are uniquely positioned to help lead the transition to a more sustainable future.

From grid modernization and energy storage to encouraging and supporting energy efficiency initiatives for consumers and businesses, electric utilities can take numerous critical actions to contribute to the net-zero emissions goal. One of the first, most critical steps electric utilities can take to transition to a low-carbon economy is investing in renewable energy sources such as solar and wind.

The challenge, however, for utilities is balancing this green transition with the industry-wide priority of keeping the power on. Renewable energy sources depend on weather conditions, which fluctuate throughout the day and as seasons change. Consequently, many utility companies struggle to effectively predict electricity supply and determine which power generation units to dispatch and when — especially in the face of constant or increased electricity demand.

This Q&A with Daniel Hynum, senior product manager at GE Digital, delves into why utilities with mixed-generation fleets must incorporate stochastic modeling and machine learning-driven performance predictions into their workflows to reduce their carbon footprint while maintaining reliability.

Unit commitment was historically a relatively simple process. The power sector dealt with dispatchable power sources — like coal originally, then gas turbines and nuclear — and there was minimal variation, empowering utilities with high confidence in how much power they could generate.

Fast forward to the 2010s when renewables came on the scene. Suddenly, the generation profile changed, with the amount of planned energy generation no longer matching the amount of energy actually produced. The unit commitment challenge became too difficult to solve by hand.

Enter modeling. The modeling approach uses historical performance information to account for the inherent variability of renewable energy generation. By enabling utilities to better quantify this uncertainty, modeling drives bolstered unit commitment planning and more risk-aware decision-making.

The marketplace has been fairly stagnant, honestly, which is where we saw an opportunity to step in and help. At GE, we’ve harnessed advances on the data science, machine learning and AI side of things, gleaned learnings from those solutions and applied that knowledge to this underserved space.

Accurately understanding variation over time is critical, especially within the constraints of runtime capability.

The Monte Carlo is like a brute force method, giving the model the complete distribution of possibilities. Whenever dealing with large fleets, utilities can't run that data in a reasonable amount of time because the number of permutations is through the roof, probably more than there are stars in the universe.

Stochastic models incorporate weather forecasts, historical production data, and other relevant variables to generate probabilistic renewable energy generation predictions that help set appropriate reserve margins and make more reliable unit commitment decisions. Rather than relying on data that isn’t accurate enough for the time frame needed, now decision-makers can use probabilistic information to run a narrow, focused band of scenarios.

Other models — deterministic, mixed-integer linear programming, mixed-integer quadratic programming, and Monte Carlo simulation — traditionally would give a single-point answer that could be dramatically disconnected from the actual probabilities throughout the day. So, while experienced operators have been doing unit commitment planning for years, even their expertise is lacking as more renewables are brought online.

Enter the need for technology that provides more accurate data. Without precise information to base dispatch on, operators face too much uncertainty, forcing utilities to spend more on fuel to avoid risk.

While other models may better suit simpler scenarios or specific aspects of power system optimization, stochastic modeling allows for the inclusion of uncertainty in renewable energy generation, improving forecasting accuracy, risk assessment, market participation, and bidding strategies.

Our Fleet Orchestration solution creates AI digital twins of each facility inside the fleet and then aggregates those together when feeding that information into the unit commitment process.

Because our tools analyze each plant, the product helps provide teams with more up-to-date understanding of how that plant is currently — and has been — operating. This is opposed to static curves or an outdated document from the original equipment manufacturer stating how that plant should operate. By leveraging AI, we're capturing degradation or even improvements occurring at each plant and reflecting that information in the output capability. These AI-informed enhancements add a higher level of fidelity and accuracy to the unit commitment process — and a single source of truth across the organization.

Fleet Orchestration employs neural net digital twins at each plant, stochastic modeling, and a rapid, multistage probabilistic unit commitment engine for everything from overall system demand forecasting to price forecasting. This unique approach to unit commitment allows for more efficient, sustainable, and cost-effective integration of renewable energy. Delivering 15-minute, day- and week-ahead recommendations, our Fleet Orchestration solution helps provide visibility into the ideal approach for navigating uncertainty to match generation and demand.

We are providing performance predictions to 17GW of power generation plants worldwide today. The majority of that generation is thermal, but the renewable fleet is growing. We have partnerships with several thermal and renewable plants featuring an installed base around the performance prediction side, providing daily probabilistic information for power generation and heat rate.

One of the partnerships is Tampa Electric. The energy company used our performance predictions in combination with our solar APM to help achieve 2025 emissions reduction goals. You can watch the success story here: Tampa Electric: Swiftly Evolving Renewable Needs.

Another partnership includes a cogeneration plant, with results thus far showing that Fleet Orchestration optimizes the use of assets and has reduced operational expenditure by 4% per year.

Luckily, no. We built Fleet Orchestration with user experience as a key CTQ (critical to quality) requirement, conducting several feedback sessions with early users to simplify the user experience or help them determine how our AI-powered guidance makes a difference.

We plan on interacting with a few specific teams the most. First, the planning team is responsible for the day-to-day operations of power plants and for creating unit commitment plans on consistent bases. Then, some participate in energy trading, buying or selling power according to that unit commitment plan. The last group is the grid operators— those with their finger on the button, making a call to start or stop a plant and deviate from that unit commitment plan generated by those upstream teams.

The energy industry — grid operations, in particular — has traditionally relied on established practices and tools. Introducing new solutions like Fleet Orchestration initially faces resistance from professionals accustomed to conventional methods who may be hesitant to embrace unfamiliar technologies. Many decision-makers naturally wonder what makes us think our solution can do this better than they can with either their in-house model or current model. Consequently, building industry wide trust is a priority.

Our multistep process to building trust begins with the planning and marketing teams before seeking grid operator approval. Engaging with industry experts, conducting validation studies, and providing hands-on user training and support help build trust and confidence in our probabilistic unit commitment engine’s capabilities. 

We believe Fleet Orchestration is another important tool in the toolbox — one that helps to augment decision-makers and empower those responsible for the day-to-day with the following to confirm their current practice or adjust course based on near-real-time information:

• Operations of power plants
• Creating unit commitment plans
• Trading energy according to those plans
• Operating the grid

There are a couple of elements of differentiation. Starting at the plant level, having probabilistic predictions for every plant within the fleet and feeding those forecasts into the unit commitment process is a core differentiation. To make the most reliable and economic decisions, utilities need the most up-to-date information possible. If data is even a day old, it’s already outdated and unreliable, so Fleet Orchestration has the ability to analyze thousands of scenarios in seconds, updating recommendations as frequently as every 15 minutes.

Taking the insights to the next level, Fleet Orchestration centralizes all of this probabilistic information across all power plants within the fleet and regarding overall system demand — incorporating uncertainty and variability — to equip decision-makers with more accurate renewable energy forecasts. By utilizing Fleet Orchestration, the software helps utilities gain the ability to effectively predict and handle fluctuations in supply and demand. This in turn helps to reduce the level of reserves needed and costly power purchases while enhancing overall system efficiency and minimizing the carbon footprint.