Can Artificial Intelligence Help Close the Gap between Day-Ahead, Hour-Ahead, and Real-Time Energy Forecasting?

Mitigating the worst effects of climate change requires large-scale adoption of renewable energy and storage technologies across every sector of society. However, current electricity grid infrastructure is ill-equipped to rapidly integrate distributed energy resources (DERs) like solar panels, wind turbines, electric vehicles (EVs), and battery systems.

Just to highlight one green technology, the International Energy Agency (IEA) notes that global solar PV capacity grew 23% from 2019 to 2020 – despite lockdown restrictions. And in just the US alone, the Solar Energy Industries Association (SEIA) reports that the nation’s PV industry has enjoyed an average annual growth rate of 42% over the last decade. When you add in falling PV panel prices, new green incentives, and rising utility rates, it’s clear solar adoption will only accelerate.

And this is just one renewable energy technology of several.

Although this is the right direction for society as a whole, our transition to a truly sustainable future remains in jeopardy unless grid operators can find a way to onboard, manage, and distribute an exponential influx of green energy and storage potential.

Why Are Renewables So Difficult to Manage?

Grid operators are responsible for managing the electricity network, forecasting future demand, and generating enough supply to cover real-time energy needs. Even before the rise of DERs, managing all this data was a constant balancing act that required continuously predicting demand and producing energy at the lowest possible cost (or highest potential profit).

In theory, adding more distributed energy resources should make the entire grid greener since renewable technologies can offset the fossil fuel many energy providers use. Integrating DERs should also make the grid more reliable since power generation, storage, and transmission all become less centralized.

However, transitioning to renewables at scale poses several problems. Already discussed is the sheer number of new solar installations coming online. But batteries, EVs, charging stations, wind turbines, and microgrids are all on similar growth trajectories. And this results in more data than any grid operator can potentially manage.

The great irony, of course, is that a lot of data is missing from the equation. That’s because many distributed energy resources like rooftop solar or electric vehicles are privately owned, making them behind-the-meter and outside the direct control of utility operators.

There is one additional layer of complexity.

Renewables like solar and wind are intermittent, which makes accurate forecasting incredibly difficult. Seasonal variations and changing weather conditions both have an obvious impact on power generation. But even a passing cloud can dramatically decrease real-time output for a mega solar PV farm.

With so many variables to manage, conventional methods for forecasting and capacity planning are insufficient against this deluge of information. Grid operators are increasingly forced to update their day-ahead, hour-ahead, and real-time operating procedures. This often involves migrating to a distributed energy resource management system (DERMS) technology that can better manage this information. But that still relies on human decision-makers who must collect, analyze, and act on many terabytes of streaming real-time data.

Leveraging the Power of Machines and Automation

There is now a growing push to use artificial intelligence (AI) to overcome many of the grid management challenges outlined above. When coupled with adaptive machine learning, AI is capable of:

Analyzing historical load data, weather information, and other network parameters

Generating highly accurate energy demand, generation, and price forecasts using continuous learning leading to improved accuracy over time

Incorporating new data to further refine its predictions and become increasingly more accurate

Optimizing grid conditions by collecting real-time data and sending real-time instructions to distributed energy resources, microgrids, and utility operators

Note that this isn’t just theoretical. Artificial intelligence is already being used in the field alongside (and sometimes instead of) traditional utility operators.

For example, Veritone’s AI-powered Forecaster uses historic and real-time weather, power demand, and DER device data to generate incredibly accurate predictions about the future supply, demand, and price of electricity. It recently chose California as a test market given that:

It’s the most populous state, with nearly 40 million residents¹ dependent on the larger power grid
California is the undisputed leader in solar PV, EVs², and battery storage³ (and an impressive 6th in installed wind capacity⁴).
The California Independent System Operator (CAISO) is responsible for managing 80% of all the energy in the state

When put to the test in November 2021, Veritone’s day-ahead market (DAM) energy Forecaster outperformed CAISO’s forecast by a staggering 31%.

In addition to millions in direct savings to utility customers, this level of forecasting accuracy also helps to make the grid more resilient since energy is generated, sent, stored, or sold – all in real-time and under optimal conditions. Equally important, accurate forecasting also has a tremendous impact on operational planning, with grid operators often struggling to close the gap between the forecasted vs. actual supply, demand, and price of energy.

A 31% improvement not only offers measurable benefits in the form of economic savings and improved efficiency, but it also delivers environmental dividends as well. By becoming more resilient, the grid can more easily integrate distributed renewable energy and storage technologies. This is particularly true when using AI to both predict and manage this integration.

Better still, these predictive capabilities will only become more accurate over time as Veritone’s AI-driven Forecaster continues to collect new data and adjust its forecasts.

Conclusion

The current challenges energy providers face are not going away anytime soon. In fact, they will only become more pronounced as falling technology prices and rising utility rates move the world away from fossil fuels. With more moving parts and data, managing the grid will become even more difficult – resulting in transmission losses, congestion, surges, shortages, equipment degradation, and even blackouts. And of course, consumers also pay a price in the form of less reliable delivery and worsening air quality.

And therein lies the paradox. Rapidly greening the grid has the opposite effect of making energy delivery more expensive and less green.

However, AI and machine learning have the potential to help resolve this paradox – allowing us to rapidly decarbonize while ensuring affordable and reliable power whenever and wherever it is needed at that moment in time.

References: