What Utility Teams Need to Know About AI
Tony McGrail is Solutions Director of Asset Management and Monitoring Technology at Doble Engineering. In this column, he drills down into the capabilities of AI and clears up some common misunderstandings
What types of applications is AI best suited for, and what should teams avoid when using the technology?
Artificial intelligence (AI) and machine learning (ML) tools provide great benefits for utility teams, especially when it comes to analysing and evaluating large sets of power systems data. But implementation of AI should be focused on specific applications or well-described problems with which teams have sufficient data. On the asset side, AI tools can be used to test power flows, power quality, send alerts for transformers, and interpret signals to identify types of faults within an asset. AI can also help predict when transformers will fail. These systems can also plan maintenance and manage the replacement plans of an aging asset.
Utility teams need to ensure that the data the AI uses is clean, well-organised and complete. Current AI and ML tools can easily be misled by data that’s inconsistent with their training. Because of this, teams should also avoid expecting AI to make reliable predictions or solve general problems if they haven’t taken the time to ensure that the technology is based on solid data. It’s also important to not blindly trust the results and take the time to review and investigate the outputs to ensure they’re accurate.
AI is an incredibly valuable tool, but it isn’t perfect. The goal is the ability to use the original training data to approximate the unanalysed, ‘testing’ data. However, the qualities and characteristics of the unanalysed data may change over time, and we must know when a change is large enough to render the AI 'poor'. This can be addressed by using an 'out of distribution' tool within the AI to ensure whether the new data even looks like the training data before attempting to classify.
What is the biggest misunderstanding or challenges you’re seeing today when it comes to AI implementation?
A common mistake is expecting AI to do too much. These high expectations for AI often result in disappointment. In other words, utility teams are mistakenly applying the technology in a broad sense when it should support engineers in addressing very specific problems, such as evaluating dissolved gas analysis results or interpreting power factor results.
It’s important to keep AI implementations focused and avoid using the same tools to solve more than one problem. AI can work well when interpolating within a dataset, but not as much when extrapolating the data to create new insights. Even one slight change can require a new set of training and rules. Because of the lack of data uniformity in the power industry, utility teams need to carefully select where and when they apply AI.
For AI to work properly, it also needs well-ordered and normalized data. In fact, it’s estimated that 95% of AI’s possible benefits can be achieved through data clean-up and standard statistical methods. In a perfect world, every transformer would have detailed and complete information on maintenance history and fault data. We’d also have clear standards for data and apply well understood analytic tools to display health, probability of failure, remaining life, etc. However, the industry commonly faces incomplete and ambiguous data. That’s why it’s important to start in a controlled manner where the data’s limitations are recognized, and the application is well understood—one which can be checked and verified before moving into areas with more unknowns.
What approach should utility teams take when utilising AI technology?
Utility teams should take a targeted and hybrid approach toward implementing AI that involves a foundation of reliable data and tapping human expertise. For example, AI can be applied to data sources from DGA, but experts need to work with the technology to properly set standards and guidelines and interpret the most important and critical data points.
Essentially, apply AI tools where they are strong, such as analysing data to identify standard or normal cases, and allow the subject matter experts to concentrate on the data that is unclear or requires real attention.
Teams should also work closely with the AI so that the inputs and outputs are always accessible. Other approaches that encourage complete automation and keep algorithms a mystery can cause major issues down the line. These types of “black box” methods to AI and ML can cause confusion around why certain decisions are being made. These methods also create added difficulties for teams when mistakes pop up. Utility team members should never blindly assume that the AI model is right and instead should closely communicate with subject matter experts to ensure models are on track and reflect the real world. This way, decision-making based on the algorithms is as accurate as possible.
What skills gaps are you seeing in the space? How real and urgent are they?
The industry is facing two critical skills gaps, both of which are essential for the long-term strength of the power grid. As the industry takes on more responsibility in supporting climate goals through adopting renewables, demand and consumption of power increases, and implementing more decentralised system, AI will be a valuable tool to maintain grid reliability. But utilities need to be prepared to properly execute it. While the power industry has made strides in increasing investments in these technologies, recent surveys show that workers lack the proper skills when it comes to digital literacy and AI.
Understanding how these technologies function and can support utilities is important to ensuring grid reliability in the future. There’s also an urgent need to upskill incoming workers on the more manual and functional aspects of the power grid. As we begin to rely more on advanced technology within the industry, we’re also seeing much of the workforce age out. Preserving this wealth of technical expertise and knowledge on the ins and outs of the power grid is vital to properly address the industry’s pressing issues.
Addressing both the digital and technical skill gaps will be crucial to ensuring that utility teams can properly implement AI where it’s needed most urgently in the power grid.
What can teams do to bridge the digital skills gap and where specifically do we need to upskill?
It’s not enough anymore to know how to test an asset. Engineers need to interpret all the available data and apply the insights to solve real problems and deliver value.
Invest in practical skills training in digital competencies, such as data science, interpretation and analytics, and foster a cooperative culture to empower workers to understand context and add value beyond tech implementation. Encourage close relationships between inexperienced staff and experienced engineers to help the rising workforce hone technical skills and transfer knowledge on traditional power engineering that can get lost as employees retire.
Mentorship programs and specialised training webinars are also effective and powerful ways to improve specific skills gaps. These mediums can also help workers strengthen soft skills such as flexibility, communication and critical reasoning, which are increasingly needed in the industry today.
When it comes to digital transformation success, why are humans just as important as adopting the right technology?
Human expertise and analysis will always be vital to the power industry and for achieving digital transformation. A successful transformation initiative requires workers to be ready to embrace change and exhibit a willingness to learn and collaborate. While AI is not perfect, there's a lot of potential and opportunity for its application. The industry needs workers committed to working with AI systems and ensuring that utilities get the most value out of the systems today, and over time.
Engineers who understand the how, when and why for AI implementation and seasoned subject matter experts set the foundation for an effective digital adoption initiative. AI can be extremely powerful but without a clear strategy and engineers with strong background knowledge and understanding of how to implement it, organisations will never achieve the goals they set out to accomplish.