Artificial Intelligence (AI) in the Energy Industry

Artificial Intelligence offers numerous opportunities for utilities to achieve significant efficiency improvements in all primary and secondary processes, from anomaly detection to time series prediction. The editorial team of the "ew" magazine spoke with Andre Wilsdorf from Arvato Systems about the requirements for a successful start with AI and current use cases.

Mr. Wilsdorf, the buzzwords Artificial Intelligence, Big Data, Machine Learning, and Advanced Analytics are currently much discussed in the energy industry. What exactly is behind this trend?

Wilsdorf: These terms are now the subject of much debate. In our consulting practice, however, we are usually confronted with very different ideas of what Artificial Intelligence means. Some of the opinions are also very emotional. But Artificial Intelligence is not a new technology as the theoretical origin was already laid 50 to 70 years ago. At that time, however, there was no technological feasibility to put the available algorithms into practice. These only developed in the last 20 years, which were also supplemented by the topics around Big Data, which functions as a decisive basis for such applications. All this leads to the boost we are experiencing now in the field of Artificial Intelligence. Due to its long history, marked by many summers and winters, I am convinced that AI applications have a huge potential and that in the future, it will be impossible to imagine many areas without them.

A distinction is made between weak and strong AI. What are the differences, and which form is mainly used today?

Wilsdorf: Here, I would like to mention the AI application Alpha-Go as an example. This application has already been able to beat professional "go-players" several times. That is a great achievement, but it is also only a weak AI, as the AI only masters this one topic. That means: AI applications that can distinguish a dog from a cat can only do this, while the chess-playing robot is only good at playing chess. Transferred to the energy industry: We are currently working on a standard solution to determine the probability of changes within the management of private and commercial customer sales. That is only a weak AI, as the solution is limited to one area of application. For telephone customer contact or the preparation of offers, on the other hand, more extensive solutions are required. We divide the following four solution competencies in the field of Artificial intelligence, which are text, image, video as well as data and loT. A combination of different solution competencies tends to develop into a strong AI. To sum up: strong AI goes in the direction of imitating humans by not only reacting but also by acting, being flexible, and being able to handle more than one task.

So at the moment, more so-called weak solutions are used. How do you see the further development towards strong AI?

Wilsdorf: In China and the USA, there is a lot of research into strong AI. But Germany is also quite well positioned in regarding that topic. One example is the European Human Brain Project, in which the Technical University of Munich is involved, among others. Here, tools are to be developed to understand the functions of the human brain then to apply this knowledge into the field of computer technology. However, I assume that several years of research work is still needed. Thus, in the field of energy economics, we will surely only be dealing with methods of weak AI for the foreseeable future.

How long has Arvato Systems been working with AI, and are you using it in the energy industry?

Wilsdorf: Arvato Systems has been working on Artificial Intelligence for about seven years now, but the initial focus was on commerce and media topics, which shifted to the utility sector only about two years ago. However, the conditions in the energy industry are very different: Some companies want to start or have already realized their first projects, while others have never heard of AI. In the meantime, we have developed several use cases specifically for the energy industry. Still, nevertheless, we are often only active in a consulting capacity to introduce customers to the topic in general at the moment. Here I would like to see more courage within the energy industry because I am convinced that many energy suppliers could benefit from AI applications.

What specific applications do you see in the energy industry?

Wilsdorf: As already mentioned, we divide our applications into four solution competencies. Data-driven applications based on pattern recognition are particularly interesting for energy suppliers. These companies have extensive data in their billing and CRM systems that can serve as the basis for such applications. Examples are the consumption load forecast for network operators or energy suppliers from consumption data or, in sales, the estimation of payment or cancellation behavior. Also interesting are applications in complaint management, where for example, complaint texts are automatically classified. In this way, internal processing can be optimized, and better feedback can be achieved for the customer. Also, the video recording of overhead lines or other power systems with drones could be an exciting application for the automation of maintenance work by detecting anomalies.

What requirements must companies fulfill to benefit from AI?

Wilsdorf: We use an "AI implementation house" for this where the areas of the organization, people, and technology are mentioned as crucial success factors. New knowledge and skills which have to be acquired as well as redefined structures and roles within the company, are organizational prerequisites. Above all, analytical mathematical know-how for the evaluation and interpretation of data is essential. In the area of people, corporate culture is a critical success factor as well. The culture of "failing to succeed" is crucial. AI projects cannot be guaranteed to succeed. An AI can not make a prediction of a client terminating their contract with a one hundred percent accuracy, for example. Such statements are always subject to a certain degree of uncertainty. This is another thing you have to learn to deal with. So for an unexplored use case, you won't be able to act according to the motto "buy software and get started," but the pilot and an agile approach play a significant role. It is also recommended to test the first small AI use cases in a " laboratory" at an early stage. Then only the provision of the required information in good data quality from the IT infrastructure is necessary. As an IT service provider, we offer our customers such a laboratory environment.

Data and data quality are often the biggest challenges in AI projects. What is your assessment of the situation in the energy industry?

Wilsdorf: For AI applications, the quantity and quality of data are equally crucial. In terms of data quantity, I assume that it depends on the use case and the goals to be achieved. Ideally, there is the necessary amount of data in appropriate quality so that the desired results can be achieved by using AI. Data quality has always been important, for example, for possible migration projects or campaigns. However, the requirements and the corresponding effort for AI projects are quite high. However, this is worthwhile, as the company also benefits from it in other areas.

How do companies get started with AI applications?

Wilsdorf: First of all, companies need to create a general understanding of AI and create a use case or better a use case portfolio based on current needs. If, for example, the primary wish is to know four weeks in advance that a customer wants to cancel, the corresponding use case has to be designed around that wish. Initially, it is irrelevant whether AI models are required for this or whether classical methods would also be sufficient. We support our customers with the offer of an "AI-Business Day." In a workshop, we present the basics of AI, we take prerequisites of the fields of application within the energy industry into account, present use cases, and work with the customer on the individual sales opportunities. Subsequently, we check in which form, quantity, and quality the necessary data is available to the customer. Finally, it is essential to have the courage to initiate and try out such a project. Usually, this is associated with little effort, so that often concrete results are available within only a few days. In the first step, no proprietary software or extensive know-how for data analysis is required in-house. Suitable partners such as Arvato Systems are available here. With an agile style, including regular reviews every three to four days, the first projects can be set up quickly and with low risk. If successful, these projects can be developed into a scalable product.

How does Arvato Systems support companies in this process?

Wilsdorf: We have a very broad base here and work with various providers - from Amazon to Google to Microsoft. That enables us to support our customers in all areas of the value chain - from the initial idea to the actual implementation. We provide all resources for this and have a central pool of data analysts as well as a Strategy Consulting division. An introduction to the topic of Artificial Intelligence is, for example, done with the help of our AI-Business-Day mentioned above. However, it is important to hold such events not only with the IT department but with all affected departments. That is how you create a broader acceptance for such innovative projects.

What concrete projects have you already implemented in the energy sector?

Wilsdorf: We are currently working on a network load forecast project. Here, the consumption data within the network area of a network operator is used for load forecasting, including autoregression analysis. An excellent forecast can be made from the patterns of historical consumption data. We have implemented a system for controlling the combined heat and power plant (CHP) for another customer. The system provides recommendations on how the operator should react to specific heat, pressure, or vibration conditions at the CHP. We are currently working on a standardized churn management system for predicting change probabilities in private and commercial customer environments. The product should be usable as a standard for every energy supplier without an implementation project or IT adaptation. Based on this, we will build a recommendation logic a la Netflix that proposes customer-specific supplementary energy services - beyond pure electricity or gas sales.

What were the most significant challenges?

Wilsdorf: In churn management, challenges in the context of data protection, according to the GDPR, are, of course, significant. The "culture of failure" that has already been mentioned has yet to be created and lived out in many cases. In terms of the database, however, the energy industry is already well-positioned. Up to now, a fairly good database has always been available - in part already structured in a data warehouse, so that the data could be easily imported into the necessary Software-as-a-Service solutions. So, according to our experience to date in the energy industry, the technical basis does not represent a hurdle. The start of Artificial Intelligence is, therefore, possible with only one small step.

For question about AI in the energy industry, please contact Andre Wilsdorf.