Hyperight

AI in the energy sector: How neural networks can improve plant operations

energy sector
Photo by Pixabay on Pexels

Although AI is in the early stages of implementation in the energy sector, promising AI projects are being developed in plants that are already showing significant results. One such energy company that put trust in AI is Uniper, a global energy supply company based in DĂĽsseldorf, Germany. 

Uniper is doing some awesome breakthrough work with AI and neural networks that deserves to be divulged. For that reason, Tobias Mathur, Head of AI Operations at Uniper, will reveal their findings at the Data Innovation Summit in August.

Hyperight: Hi Tobias, we are thrilled to have you join us as a speaker at the 5th edition of the Data Innovation Summit. Before going to more in-depth questions, please reveal to us more about yourself and the work that you do at Uniper.

Tobias Mathur - Head of AI Operations at Uniper

Tobias Mathur: Hi Ivana, it is my pleasure. Thanks for having me and for the brilliant organisation during such unusual times. I would call myself a machine learning enthusiast, and I’m excited about innovative products. Actually, I got in closer contact with both of it during my time at Kellogg. And now, here at Uniper, I’m heading a team of experts who are just doing that: developing new products using machine learning. It’s an exciting task.

Hyperight: Your Data Innovation Summit presentation will revolve around Utilizing AI technologies in day to day plant operations. Could you tell us more about where you deploy AI, and what benefits you have experienced by integrating AI into power plant operations?

Tobias Mathur: We indeed use machine learning in the heart of a power plant: we operate the combustion process. Traditionally, especially in waste2energy plants, many operative decisions are made by human operators in the control room of a power plant. In order to support them and improve operations, we trained neural networks with the data from the best human operator and then let it then run 24/7.

As an effect, our projects show a significant increase in efficiency. Additionally, we see fewer flue gases and less CO. Clear commercial upsides for the plant owners.

AI in the energy sector: How neural networks can improve plant operations
Photo by Uniper

Hyperight: The energy industry has recognised the potential of AI to transform and improve energy systems. What are the most significant applications of AI in the energy industry and the value they provide?

Tobias Mathur: AI is already used for quite some time is in plant monitoring, process analysis, risk management, but also in energy trading. With Enerlytics, Uniper has developed its own tool for condition monitoring, and our traders use AI-supported algorithms to trade. So there are some important areas where different forms of AI are already being used. But in operations, this is quite new, especially because of the high challenges in the processes and also for safety reasons. I nevertheless believe that machine learning in operations will be the next big development.

AI in the energy sector: How neural networks can improve plant operations
Photo by Uniper

Hyperight: What is AI’s role in producing renewable and sustainable, but at the same time reliable energy?

Tobias Mathur: AI can definitely help to make plant operations more sustainable and reliable. In our pilots, we already see reductions in flue gases and CO emissions. With increasing emission reduction requirements for many plants around the world, AI could help to match those requirements.

Hyperight: More particularly, you will present how a Neural Network is being used to support human operators to optimise combustion processes, specifically in a Waste2Energy plant. How can neural networks help tackle the challenges a power plant faces such as flexibility, emissions, and efficiency?

Tobias Mathur: For Waste2Energy, higher efficiency is achieved by increasing waste throughput and increasing energy generation. At the same time, downtimes need to be avoided, and the costs for consumables should be reduced. Since the AI optimises the process constantly – it makes decisions every 30 seconds – and by understanding the full complexity of all processes in the combustion – the AI analyses 24 measurements within an 80 min time frame – the AI can make much better decisions. Therefore a usually quite volatile process is smoothened, and the efficiency is increased.

That power of AI is also seen with CO reduction. CO is one of the challenges in waste incineration. The big problem is here that CO develops with a delay – when the sensors start detecting it, it’s already too late to act against it. This dead time between the cause and measuring it is a huge problem for automation. Our neural network, in contrast, understands the patterns of the process, predicts CO developments, directly works against the development and thereby reduces those emissions. In the picture below, you can see that the AI changes the process (green, red) 10 minutes before the CO (purple) shows up.

CO reduction
Photo by Uniper

In sum, the Neural Network can deal with those complex processes much better, simply because it uses more data.

Hyperight: According to your experience, what are the biggest hurdles that stand in the way of energy companies to reach the full benefit of AI adoption?

Tobias Mathur: AI adoption is a process. We are just starting to develop the right tools that really work. Plus we must ensure smooth operations since downtimes will always be more costly than potential improvements. So the first hurdle is to develop the right tools. Furthermore, we also need to challenge ourselves, try new things and always try to find better solutions.

Secondly, an important wise hurdle is safety requirements – power plants are critical infrastructure and highly sensitive. So we wisely set high hurdles in the form of safety standards that each AI solution must comply with.

Thirdly you must keep in mind that we talk about plants that exist for a couple of years. They are not Tesla-Plants built for AI. Actually, no operational plant today is designed for autonomous driving. Some plants are 20 or 30 years old. We need to find solutions to integrate AI and Machine Learning into that existing setup.

AI in the energy sector: How neural networks can improve plant operations
Photo by Uniper

Hyperight: What is the maturity with the widespread adoption of AI in the energy sector? What are your outlooks on the future potential AI, especially neural networks, can bring to the power sector?

Tobias Mathur: AI and Machine Learning is still a quite new thing. Both in Plant Management and Plant Operations, AI solutions are still few and new. So in the wider universe of a power plant, there are plenty of options to use AI. Developing those solutions will certainly lead to more energy generated with the same plants and reduced emissions. AI will help to optimise the processes within the physical limits of a plant.


Add comment