This article explores how generative artificial intelligence (GenAI) can enhance our work and the power grid’s efficiency.
CHALLENGES AND TRANSFORMATION IN THE POWER INDUSTRY
The power grid faces numerous challenges, from aging infrastructure and increasing demand to retiring professionals and integrating new technologies like virtualization. We are moving away from large, centralized power plants towards more renewables such as solar and wind. All these factors combined make power grid reliability a complex and evolving challenge:
- Aging infrastructure. In some countries, the grid is outdated, prone to failures, and requires significant upgrades and maintenance investments.
- Increasing demand. The challenges stemming from decarbonization efforts, such as reducing the use of SF6 and the rise in renewables, electronics, electric vehicles, heat pumps, and air conditioning, are straining the power grid significantly.
- Retiring professionals and knowledge gaps. Many experienced professionals are retiring, leading to a loss of expertise and the challenge of training new workers.
Digital tools are no longer optional but are essential for solving these challenges. Can GenAI be the solution to some of these problems? How can GenAI foster the creation of conversational models that can access vast, complex data sets and help close the knowledge gap?
ARTIFICIAL INTELLIGENCE: WHAT, WHY, AND HOW
To clarify some basic principles, let’s start with two definitions:
- Artificial refers to something created by humans, for example, computer software.
- Intelligence is the ability to learn, understand, and apply knowledge to solve problems, adapt to new situations, and comprehend complex concepts. It involves cognitive functions, such as reasoning and memory, and emotional abilities, like empathy.
Artificial intelligence (AI) combines these two ideas. AI is a branch of computer science focused on creating solutions that can perform tasks that typically require human intelligence. These tasks range from recognizing speech, making decisions, and translating languages to playing games or driving cars.
AI has evolved significantly from early rule-based systems like IBM’s 1997 Deep Blue, a specialized chess software that analyzed millions of moves per second to defeat world champion Garry Kasparov. It advanced through machine learning in the 2000s with
systems like Amazon’s recommendations and progressed to deep learning in the 2010s, exemplified by iOS Face ID.
AI in the Power Grid
AI holds tremendous potential in power grids. Although not yet widely used, it is starting to gain attention. AI can analyze massive amounts of data to predict and detect failures. For example, machine learning helps predict asset failures and optimize grid operations, aiding decision-making. However, it is important to emphasize that humans always make the final decisions.
GenAI, popularized by ChatGPT in 2022, can create conversational models that access vast data sources and provide human-readable answers to complex questions. These models can make critical knowledge easily accessible to new engineers, bridging the knowledge gap and enabling quicker problem-solving and more efficient grid management.
GenAI is advancing very quickly. This rapid change creates significant gaps between what is possible with new technologies, what we can understand, and, most importantly, how we can integrate them into our daily work.
At OMICRON, company culture will help us close these gaps with continuous learning, sharing knowledge, and exploring innovative ideas. Failure isn’t something to avoid; small, manageable failures are the best way to learn and grow, both individually and as a company.
RETRIEVAL-AUGMENTED GENERATION
Large language models (LLMs) are a specific type of GenAI focused on understanding and generating human-like text. LLMs like GPT-4 are incredibly powerful but have significant limitations, particularly from a training perspective. Training these models is extremely expensive due to the massive computational resources and energy required.
Additionally, LLMs need vast amounts of diverse and high-quality data, making data collection and curation challenging. Once trained, updating these models with new information is also complex and time-consuming, often requiring costly retraining or fine-tuning.
Luckily, retrieval-augmented generation (RAG) offers a solution to these limitations. RAG needs to be used with models that have already been trained, such as GPT-4 (OpenAI), Gemini (Google), LLaMA (Meta), or Claude (Anthropic). It combines their power with a real-time retrieval system that accesses relevant information (Figure 1).
This approach reduces the need for constant retraining, making it more cost-effective. Furthermore, RAG can give models access to specific knowledge stored in internal documents and collaboration platforms. This ensures accurate and up-to-date responses
to internal queries. RAG also has an important advantage: The information it retrieves is not used to train the models, which is crucial for maintaining data confidentiality.
Another advantage of RAG is its independence from the specific LLM it uses. This independence makes it easy to switch between different versions or LLMs as needed, ensuring the system can adapt quickly to new advancements or specific requirements without significant reconfiguration.
CHALLENGES OF IMPLEMENTING AI
Implementing AI in any field has several challenges. We had to consider the following factors before implementing AI:
- Data preparation. AI solutions can only be as good as the data they use. Ensuring clean, reliable, and well-organized data helps AI systems learn effectively, achieving more accurate results.
- Integration with existing systems. Integrating AI with existing systems is challenging due to data silos and outdated legacy material. In some cases, these systems store data in disconnected and inconsistent formats. Established processes tied to these legacy systems often lack the flexibility to support AI, making it difficult for AI to access and process data efficiently. Departments will need to collaborate and adapt established processes to ensure seamless data flow and maximize the potential of AI tools.
- Acceptance. Resistance to change, lack of understanding, and lack of trust in AI’s capabilities make accepting AI solutions challenging. Overcoming this reluctance requires training, clear communication, and demonstration of AI’s value. Encouraging a culture that embraces innovation with ongoing support helps smooth the transition and enhance the overall effectiveness of AI initiatives.
CASPER: OUR EXPERIMENTS WITH GenAI AND LLMS
Casper is our first GenAI-powered assistant. It was developed in 2023 as part of an innovation project aimed at helping colleagues find and draft answers related to OMICRON products. It can retrieve information contained in files and other knowledge resources. Along with the answer, it references related documents, allowing the user to obtain more information that helps validate it (Figure 2).
Casper is intended for internal use only. Currently, there are no plans to turn it into an OMICRON product. Experimenting with a GenAI-powered bot enhances our innovative capabilities and lets us explore new ways to improve our solutions.
During 2024 and 2025, we will collect feedback about the technology and its potential in our industry. This feedback will help us identify new opportunities for improving our solutions.
IS CASPER A CUSTOM VERSION OF CHATGPT OR COPILOT 365?
Casper, ChatGPT, and Copilot have different purposes and access to other data. Our bot understands the relationships between our products, such as software and accessories compatible with main devices, which differentiate Casper from ChatGPT. Casper also considers documents that may be less relevant, such as those related to phased-out products or obsolete accessories.
Copilot 365 is an AI-powered assistant integrated into Microsoft 365 applications. It is designed to enhance productivity by providing contextual suggestions and automating tasks. It can access data, such as emails, meeting notes, and documents, during retrieval. Copilot offers personalized insights and support directly within Microsoft Office apps like Word, Excel, and Outlook. This makes it a personal assistant integrated with Microsoft Office and tailored to our needs.
On the other hand, Casper can use data about OMICRON products. It is meant to be a virtual intern, able to quickly answer OMICRON product-related questions, draft responses, grant access, and provide references to related documents or knowledge sources.
CONCLUSION
While still in the experimental stage, GenAI shows great promise in enhancing the value we deliver to customers. We are exploring this technology to boost efficiency and improve internal knowledge management.
If you want to know which parts of this article were AI-generated and which parts we wrote ourselves, email juan-carlos.sanchez@omicronenergy.com or florian.fink@omicronenergy.com.
LISTEN TO THE PODCAST
In this episode, OMICRON digital transformation experts Florian Fink and Juan-Carlos Sánchez discuss how implementing digital tools, such as artificial intelligence (AI), helps optimize power grid reliability. They describe the advantages AI presents to the power industry in removing knowledge barriers and improving the efficiency and accuracy of daily tasks, such as analyzing data to predict
and detect failures or to optimize energy distribution.
Scan the QR Code or visit: omicron.energy/episode80
Florian Fink has been part of OMICRON’s product management team since 2013. He focuses on protection testing and industrial and distribution applications and has been driving the topics of digital transformation and generative AI since 2022. He worked as a project engineer for Cegelec, Germany, from 2009 to 2012 and as a planning engineer for InfraServ Knapsack, Germany, from 2012 to 2013. Fink received his diploma in electrical power engineering from the University of Applied Sciences in Cologne in 2009.
Juan Carlos Sánchez Calle has been part of OMICRON electronics since 2013. He has more than eight years of experience in software engineering and seven years in product management. As the driver of digital transformation, Calle focuses on leveraging Generative AI and advanced technologies to enhance decision-making and innovation. Passionate about agility and lifelong learning, he actively fosters collaboration to drive digital evolution. Calle holds an MSc in artificial intelligence and big data from Universitat Oberta de Catalunya.