Powering forward: How AI is reshaping the energy sector
By Paulette van Bossum
AI is emerging as a key driver in the fast-changing world of energy. It supports grid management, energy efficiency and renewable integration. AI is helping drive the energy transition, and improves the ability to control, enhance and predict complex energy systems.
To understand the current state and future potential of AI in energy, we spoke with four leading experts working at the intersection of these fields:
Yuliya Shapovalova, Assistant Professor at Radboud University Nijmegen
Arno van den Berg, Co-Founder at DataSpeak.nl (formerly CTO of Vandebron)
Adil Bohoudi, Founder & Managing Director at Conclusion Intelligence
Alex Imbrea, Engineering Manager at El Niño
From commercial applications to academic research, their insights reveal how AI is helping drive the energy transition and creating new opportunities across the sector.
What You'll Learn:
Why the Netherlands faces unique grid challenges that make AI solutions particularly urgent
How AI is accelerating grid reliability assessments from hours to seconds
Real-world applications delivering financial benefits through smart energy scheduling
The hidden energy costs of AI itself and practical strategies to minimize them
Future directions, such as hyper-personalized energy services and foundation models
The evolution of AI in energy: from simple models to smart grids
The application of AI in energy has evolved significantly over the past two decades. Adil Bohoudi of Conclusion Intelligence reflects on this journey: "I started around 2005 in data and AI in the energy sector. This was the time that the energy market was liberalized, and the start of several commercial companies."
At that time, the focus was mostly on basic asset management. One of the first applications was predictive maintenance, as Bohoudi explains: "One of my first projects was to try to create a 'digital' twin to do progressive maintenance. In 2005-2006, they [the energy companies] didn't have any clue about what they owned or their physical condition."
The use of AI accelerated with the shift toward renewable energy, as Arno van den Berg explains: "It was quite straightforward until green energy came to play, because that brought us new challenges with more volatile production volumes of energy that were very weather dependent, and that's where the data question came up more and more."
Today, AI has evolved from these early applications to sophisticated systems managing complex, renewable-heavy grids.
This journey from basic asset management to real-time grid optimization represents one of the most successful applications of AI's use in critical infrastructure.
The urgent need for AI in energy systems
The energy industry is undergoing a rapid transition, with renewables playing increasingly important roles. However, this shift introduces significant challenges, particularly for power grid management.
Yuliya Shapovalova of Radboud University explains that "for the energy domain, especially currently in the Netherlands, one of the biggest problems is that the energy grid is full and there are essentially physical limitations in terms of being able to connect more consumers to the grid, or expand capacity for existing businesses."
This grid congestion creates an urgent need for better tools to manage existing capacity. Traditional approaches face significant limitations, particularly in computational efficiency.
According to Shapovalova, "many existing, state-of-the-art, classical algorithms are very computationally expensive. And AI approaches can tackle quite a bit of this computational burden. Traditionally, this type of computation takes hours, but with AI, after the model is trained, we can make predictions in seconds."
This dramatic speed improvement enables faster decision-making about grid capacity, which is crucial when the grid is already at its limits.
The inherent volatility of renewable energy sources introduces even more complexity. As Shapovalova notes, "renewable energy generally makes the grid less stable, because renewable energy is not a constant source of energy. So today is a sunny day, tomorrow is a cloudy day, and how much solar energy we get depends on that... And that's where AI forecasting can play a major role."
Key applications of AI in energy
Our experts pointed out several key ways AI is benefiting the energy sector:
Grid management and reliability
AI enhances how power grids are managed, particularly in assessing grid reliability when new customers want to connect.
Shapovalova's research directly addresses this: "If you want to connect a new customer, you want to evaluate whether the grid will still be reliable once a new customer is connected." The dramatic speed-up in calculations that AI enables (from hours to seconds) can accelerate decision-making processes that are critical for grid expansion.
Her research also explores using AI to assist grid operators. She explains: "When you manage a grid, you can take certain actions to navigate challenges. Typically, there are grid operators which make these kinds of decisions... for example, rerouting some electricity through a different path in the grid." Through imitation learning, AI agents can propose actions based on decisions made by human operators in the past.
This collaboration between AI and human operators is crucial, as Shapovalova explains: "These are very key decisions, you don't want to make a wrong one, because the consequences can be bad, in the sense that a whole neighbourhood can be cut off from electricity, or some key operations cannot happen."
Forecasting and prediction
Forecasting emerges as perhaps the most fundamental application of AI in energy systems. According to Shapovalova, "for grids, short-term forecasting is especially important, say predicting how much energy will be consumed in the next 24 hours, so you can manage grid congestion and redistribute loads."
Van den Berg reinforces this: "Forecasting is a huge capability for an energy supplier. There's a need in the energy market to be very precise about what you produce and what you consume." For energy suppliers, accurate forecasting allows them to effectively balance production and consumption.
Alex Imbrea of El Niño describes implementing AI-based forecasting for a solar equipment manufacturer: "For Victron Energy, we use machine learning to forecast solar yield based on solar irradiance data obtained from satellite images. We also forecast consumption patterns for individual devices and entire installations."
Van den Berg emphasises the critical importance of this capability: "AI really helps in building accurate forecasting models to ensure the system stays balanced. You need that to be in control of your business proposition."
Energy optimization and smart scheduling
Beyond forecasting, AI enables smarter scheduling of energy usage, storage and trading.
AI-powered scheduling delivers concrete benefits. According to Imbrea, the financial benefit is the most immediate. “Consumers get a lower electricity bill because of a smarter schedule and they sometimes make use of the solar energy more efficiently. You don't want to use solar and sell it to the grid when everybody does and the prices are quite low. So we will pour it into the battery and sell it at night or some time that is more expensive."
Smart scheduling also improves hardware utilization: "By having the schedule, there is better use of the hardware that they [consumers] have. You already bought the battery, so you want to make use of it."
Imbrea highlights how AI-driven insights influence everyday decisions: "If we show a forecast that says that tomorrow you're going to have a lot of sun, some people might decide it's time to heat up the swimming pool or something that requires a lot of energy. And they better use the energy that we predict is going to come tomorrow from the sun, rather than using the grid."
Intelligent user interfaces and experience
AI is being deployed more frequently to enhance how consumers interact with energy systems.
Imbrea describes an AI co-pilot they're developing "for users to generate charts and widgets. So you can, say, plot the temperatures and the voltages of my system for this week, and it creates a visualisation of those values."
Van den Berg notes that AI is transforming customer interactions: "We were always promised that these chatbots could handle our customer inquiries and help us solve our problems, but in the end, the only thing you try to do is go to a chatbot to figure out how to get to a human. But now, we're finally delivering on those problems by enabling really good digital interaction and having more customized and personalized experiences. Maybe also more ‘human’, because of the natural conversational progress genAI has made."
These improved interfaces help democratize access to energy insights. As Imbrea explains: "For the AI LLM projects, I would say it improves the understanding of what the system does, and it makes it easier to get insights from your system. This lowers the entry barrier for non-expert users to learn more about their systems, but also for expert users to save time and do more in their working days."
Van den Berg highlights the sophistication of modern AI assistants: "You're able to make the customer interaction way more dynamic and linked to way more different data points than before. You're able to make that experience magical."
Freeing people for high-value work
One significant benefit of AI in energy is how it frees up human experts to focus on more complex and valuable tasks, rather than replacing them.
As Shapovalova explains: "I think at the moment we see a lot of improvement in decision-making and support for those working in the energy sector. It’s about saving time, reducing repetitive tasks, and automating the work that may be boring or unpleasant for employees. AI helps build a more efficient decision-making process."
She makes it clear that this is not about replacing jobs: "There are still many challenges in the energy transition sector that need solving. What we are doing is freeing up employees' time so they can focus on tasks that might be less urgent but still important."
The result is a shift in how human expertise is used: "It's not a complete substitute, as in now these people have nothing to do. It's more that they can concentrate on work that is not time-critical but more interesting and meaningful."
This underscores how AI can support and enhance human capabilities, allowing people to focus on higher-value tasks while AI handles routine work.
The Dutch context
The Netherlands presents an interesting case study for AI in energy, with both advantages and challenges.
Shapovalova observes that "Dutch electricity grids are especially overloaded compared to the other countries at the moment. We hear a lot about this in the news and that's the experience that we have working with energy companies."
This pressure creates a unique urgency: "Building the grid takes years. So if you want to build a section of the grid or expand it, that can take 15 years. But the problems at the moment are quite urgent, so we cannot wait 15 years."
However, there's also a silver lining: "The Dutch energy sector, at least when we work with Alliander and TenneT, is very much ahead in terms of using AI and being very open to using machine learning and AI. And they've collected, over the years, quite a bit of data, and that definitely creates quite a lot of opportunities."
This creates a particularly fertile ground for AI research. "When companies have collected good quality data, it gives us researchers in AI the opportunity to make use of it."
Imbrea adds: "In the Netherlands, people are actually passionate about the field, and the adoption of dynamic price contracts is also quite high. That makes it very attractive for people to use these kinds of AI-powered solutions."
Van den Berg points to distinct features of the Dutch market: "The Netherlands has a couple of typical things in the market. One of them is, we are number one in solar production at households, and we also have a lot of electric cars." This brings both challenges and opportunities for AI.
He elaborates on the local context: "I think in the Netherlands there has traditionally been more of a need to fix the data problem. So the investment in the energy world into understanding these energy streams, and building the right products around that, is quite strong, I think."
Technical implementation
Our experts provided insights into how AI for energy applications is actually implemented.
Machine learning models for forecasting
Imbrea details their approach to forecasting: "We started using classic machine learning models, like linear regressions and random forest models that map solar irradiance data we get from third parties, into the solar yield of a specific site."
However, basic models weren't sufficient: "We realized there are things like the seasonality aspect, so we started using STL decompositions." Their current solution combines "an ensemble of models that already exist, like the classical ones, and our own mathematical models."
This highlights the iterative nature of implementing AI in real-world energy systems. Starting with simpler models and progressively refining them based on domain-specific insights is a common pattern we heard from our experts.
Graph neural networks for grid analysis
For understanding complex power grids, specialized AI architectures are necessary. Shapovalova explains that "the whole structure of an electricity grid can be represented as a graph,where substations are nodes of the graph, and then things like electricity cables are the connecting links in the graph."
She continues: "Each node and each link between nodes has certain features, such as maximum capacity of a cable, or how much current load there is in that cable."
To process this type of data, "graph neural networks are essentially a type of a deep learning architecture which can deal with this type of data."
This approach is particularly well-suited to grid applications because it naturally represents the network structure of power systems, allowing AI models to better understand the complex interdependencies within the grid.
Imitation learning for grid operations
Shapovalova describes an innovative approach to assist grid operators: "With imitation learning, the idea is that you collect a lot of data from the real-world agents, so the actual humans who operate the grid, and you train your agent by imitation to replicate their type of action."
The goal isn't to replace human operators but to augment them: "When AI has seen a lot of diverse actions, it can propose these actions, so then an operator can more efficiently spend their time on evaluating which decision is the best."
This approach preserves human expertise while amplifying it through AI, creating a powerful human-machine collaboration that leverages the strengths of both.
LLMs and generative AI for user interfaces
Imbrea outlines their vision for generative AI: "The long-term vision is to have some kind of AI co-pilot in the systems that offers to answer all kinds of questions about the system, about performance, the configuration of devices - basically what users do manually currently in the app."
Implementation requires careful consideration: "We use open source LLM models as well as models from the big providers. With a lot of care and guardrails in place, because in the energy sector, it's important that the right decisions are made, and the quality of the output is high."
Challenges
When it comes to AI in energy, our experts identified some significant challenges and considerations that must be addressed:
Ethical and safety considerations
For critical infrastructure like power grids, Shapovalova emphasizes: "When it comes to decision-making agents, it's important that a human oversees the process, especially for critical decisions. You can't fully delegate that to AI, at least not yet."
This ties into broader questions of accountability: "It's also related, of course, to laws. We see it with autonomous driving too. One of the big limitations is figuring out who is responsible for the decision-making process if something goes wrong."
She also highlights fairness concerns: "If you're using forecasting systems at, say, the household level, you need to ensure AI isn't being used in an unfair way. For example, with dynamic pricing strategies, some consumers shouldn't end up at an unfair disadvantage because of it."
Regulatory and privacy challenges
Van den Berg highlights the regulatory hurdles: "With large language models, we’ve ended up in a cowboy kind of situation. There’s a big fight over compute, and then also the question of where the compute is."
He continues: "Capacity is limited in Europe, and getting access to a bit of that constrained capacity is not easy, especially if you're a smaller company."
European regulations also present complications: "In terms of regulation, it’s another topic. You can be GDPR-compliant and have all your data, but then you still have to ask which companies you're working with. Are they European, or are they American?"
Imbrea explains how they handle privacy: "We never send user or energy data to the models. What we do is ask the model to generate a query, which we then run on our own systems. The data stays with us. The LLM just helps decide what we should be querying from the database."
Data volume and complexity
According to Imbrea, "One aspect is the volume of the data. These systems report real-time measurements every second, every minute, and we store them. Doing something that applies to all systems at once is challenging because of the volume of data."
This challenge intensifies when trying to apply AI across multiple systems or large geographic areas. The sheer scale of data collection in modern energy systems requires sophisticated data management approaches.
LLM limitations
LLMs face specific challenges in energy, as Imbrea explains: "LLMs aren't really tailored or fine-tuned for the energy market. They're trained on general internet knowledge - and most of what's online isn't about energy systems, power, or the technical details that matter in this field."
This highlights the need for domain-specific training or fine-tuning to make LLMs genuinely useful for specialized energy applications.
Van den Berg adds another dimension to this challenge from the point of view of an energy provider: "The accuracy matters. You’re attaching your brand to it, so you need to be sure it’s not hallucinating, making things up, or getting it wrong.”
The energy cost of AI
AI has huge potential to improve energy systems, but there’s a clear paradox: AI itself consumes a lot of energy, which can undercut its benefits.
Bohoudi puts it plainly: "The hidden cost of AI, how to reduce it? There are a lot of studies on this. And of course, there's the question of how transparent big players like AWS and Azure really are about the true energy costs."
He offers an example: "My wife asked me, 'can you give us a workshop at our school about how to generate nice images with AI?' I told her I can do it, but first we go to Albert Heijn and buy a pallet of water bottles. Why? Because generating just one image uses a lot of water and as much energy as charging your phone from zero to 100 percent."
This reveals the often-overlooked footprint of simple AI tasks. Bohoudi says the first step is: "Awareness. That’s one important thing."
He follows with a practical point: "Use efficient architecture. Do you really need that heavy algorithm for your case? Maybe a simple decision tree is enough."
Bohoudi cautions against over-engineering: "It’s costly, not just in time and money, but in energy too. Think about energy-aware optimization and measuring. There are many monitoring tools. Are you measuring your carbon footprint?"
He recommends tools like Carbon Tracker: "Tools like that can show how much carbon your models are producing. Even if your method isn't perfect, just having a measurement is already a big first step."
He also suggests edge computing: "Running models locally with edge technology helps a lot."
When asked about AI’s overall energy impact, Shapovalova gives a balanced view: "It depends on the application. In the energy sector, what we do is usually very beneficial. Traditional methods can take hours, so on a larger scale, AI can actually save energy."
She adds a note of caution: "The real issue is using AI mindlessly, for every little task. That can burn a lot of energy just to generate simple answers. In some domains it makes sense, in others it might use more energy than current solutions."
Future directions and opportunities
Looking ahead, our experts identified several exciting directions for AI in energy:
Foundation models for energy
Shapovalova envisions more comprehensive models, such as the: "Development of potential foundation models. We’ve seen this in weather forecasting, where datasets from different domains and atmospheric levels are combined into a single AI model. That model can then handle multiple tasks, like forecasting precipitation, solar irradiance, and cloud coverage at the same time."
She adds that AI has applications beyond grid management: "AI can also play a role in areas like material development, which indirectly could contribute to the energy sector as well. That’s one of the big applications right now - material science. It helps develop more efficient materials (potentially also for the electricity grids)."
Online automated machine learning
Imbrea highlights a more adaptive approach: "I'm currently working on an online, automated machine learning platform. It’s different from classic batch learning, where models are trained once on a static dataset."
The key benefit is adaptability: "With online automated learning, the model updates over time, which helps it stay accurate for longer (hopefully forever)."
He notes the approach is still not widely adopted: "It's underused in the industry. I think many IoT problems could benefit from it. The maths and algorithms aren't new, but adoption has been quite slow."
Hyper-personalization in energy consumption
Bohoudi points to a growing trend: "One of the big movements is hyper-personalization. You see it everywhere, and I think energy consumption will be hyper-personalized too. That means we’ll know exactly when you're at home and what kind of energy you need."
Van den Berg describes how this could transform customer experience: "If you can innovate your customer interaction, it will feel more personal. The goal isn’t to be more personal, but if you have a more natural, high-quality conversation with your customer, then it automatically becomes more personal."
Multi-stakeholder collaboration
Shapovalova highlights a promising shift in collaboration: "We’ve just established a new flex lab with Alliander, in collaboration with several universities. Both big and small companies are involved. I think this kind of open-source collaboration across the energy sector has real potential, and I hope we’ll see more of it. That’s where really fruitful developments can happen."
She expands on the approach: "The idea is to specifically involve small to medium-sized companies, so they can benefit from the technology as well. It’s not just for the big players with massive datasets. It’s really meant for everyone in the sector."
Advice for AI professionals
Our experts offered some advice for AI professionals potentially looking to work in the energy sector:
Shapovalova emphasizes the positive impact of the field: "It's a very exciting field, because the data is available, and the technology is there. You can definitely fine-tune existing solutions to your problem. You can also find new problems that aren’t solved yet. And I think it's an area where it's quite easy to grow as an AI researcher."
She adds: "One of the nice parts is that you can also feel good about the value you bring to society and the community, because you're working with state-of-the-art technology that does good - whether for your country, for the European region, or for the world."
At the same time, she cautions against overcomplicating solutions: "One of the important parts is to not forget that sometimes simple things also work. Sometimes we forget there are simpler solutions in machine learning and AI that don’t involve large models, but still work well."
Van den Berg highlights the momentum behind the energy sector: "The energy transition is still one of our top priorities as a community, as a country. There's definitely a lot of work to do, but the work is also changing. Traditional data science, data engineering, and software engineering are evolving rapidly."
Imbrea advises that the barrier to entry is lower than many might think: "All the core concepts of AI are the same. The main challenge would be understanding the domain knowledge. It really helps if you have an inclination toward physics or related fields before, but it's also very easy to learn, even if you didn't."
He sees strong opportunities ahead: "At least in the Netherlands, everybody wants a smarter home, smarter grids. I think there are both policy-driven initiatives from the government and interest from individuals to build a smarter energy system."
Bohoudi expands the conversation to responsible development: "I hope we can grow this idea of green AI - so that when we talk about ethical AI, it means being good for both people and the planet."
Conclusion: powering the future with AI
AI is becoming a key part of how we build the energy systems of the future. It helps manage power grids, forecast energy needs, and make better use of renewables.
As energy systems grow more complex, AI makes it easier to respond quickly, use resources wisely, and keep everything running more smoothly.
The experts we spoke with showed that AI is already making a real difference. It speeds up decisions that used to take hours, helps people save money through smart energy planning, and supports better forecasting.
At the same time, challenges remain. These include handling huge amounts of data, making sure AI is accurate and fair, dealing with privacy rules, and keeping energy use from AI itself under control.
The Netherlands is a great example of both the urgency and the opportunity in the sector. The grid is under pressure, but there's strong momentum, a real willingness to apply AI and try new things, and a solid base of data and research. That makes the Netherlands an ideal testing ground for AI-driven energy innovation.
For people working in AI, this is a field where you can apply your skills to real-world problems that matter. From improved battery tech to smarter home energy use, the work is full of purpose and potential.
The future of energy will be smarter, more flexible and more sustainable. And AI, used with purpose and responsibility, will help get us there.
