Machine learning can help electric utilities right now (and for the next decade)

With the media’s daily bombardment of new information on artificial intelligence’s advancing capabilities, even the most level-headed professionals can be excused for wondering how long before the machines rise up and take over?

Engineers and decision-makers in the utility industry may not share the alarmist’s view on artificial intelligence. Instead, many are seeking to learn how to get ahead of the AI curve and use its powers for business-good.

But how many energy professionals truly understand not just the different types of AI, but how those types progressively build off one another, starting with a predictive analytics foundation?

Let’s define and explore AI’s ancestral components and subsets with an eye toward identifying which can best help utilities analyze and defend their grid infrastructures now and in the future.

Predictive Analytics

Before diving head-first into AI, it is important to understand predictive analytics, a core foundational element upon which AI builds. Predictive analytics involves ingesting historical data and making forecasts and predictions about outcomes and events in the future. It is a key building block of AI that relies on statistical modeling to predict the future by identifying patterns in the historical data. 

Common modeling techniques include Regression and Decision Trees which have been used for decades. It’s only recently, however, that advancements in software have enabled even entry-level data analytics professionals to take advantage of these techniques.

Examples of predictive analytics for utilities include: 

• Asset Health/Predictive Maintenance – Using asset and other applicable data sources  to identify assets to upgrade or replace prior to failure, enabling better project prioritization and increased reliability over time
• Outage Prediction – Using outage, weather, and other applicable data sources to identify areas of service territory most susceptible to outages, enabling better project prioritization and increased reliability over time

Artificial Intelligence (AI)

Now that the AI foundation is built, we can start to cover more advanced concepts. Broadly speaking, Artificial Intelligence is a field of computer science with the goal of developing systems or machines that exhibit intelligent behavior. While AI is sometimes used with a narrow definition describing only systems that seek to emulate human intelligence, it is more commonly understood to include a wide array of approaches and techniques including predictive analytics, machine learning, deep learning, natural language processing, generative AI and more. AI can perceive their environment, make decisions based on their environment, learn from sets of data, and interact with other systems, including humans.

Machine Learning

Machine learning (ML) is how a computer system develops its intelligence.  ML is a subset of artificial intelligence focusing on models and algorithms that learn from the data (like predictive analytics) but can make these predictions and decisions without being explicitly programmed. ML utilizes a portion of the data to “train” a model. A portion of the training data is “held out” to validate that the model is likely to successfully generalize the data unseen during training. Then, in production, the data under consideration is run through the resulting model and the ensuing predictions or output are used by the end-user or by downstream business logic. Using statistical and computational techniques, along with additional data, models trained by ML can improve and become more accurate over time.

Examples of machine learning for utilities include: 

• Object Identification – Detect assets on a utility pole using various types of imagery, for inspection and/or auditing purposes
• Pole Viability – Determine if a pole suitable to attach equipment, and what degree of make-ready engineering is required 

Deep Learning 

A subset and a more advanced type of machine learning, deep learning is inspired by the functions and structures of the human brain. Interconnected layers of artificial neurons process and transform the data. It is a specific approach to ML and is used to learn and represent complex patterns in data. Deep Learning is particularly performant when applied to non-tabular data, such as images or natural language. Although the technique is often poorly understood, that has not prevented deep learning from being the favorite technique to use in image recognition and natural language processing.

Examples of deep learning for utilities include:

• Advanced Object Identification & Anomaly Detection – Detect assets and anomalies signaling possibly damaged or defective equipment on a utility pole using various types of imagery, for inspection and/or auditing purposes
• Text and Sentiment Analysis – Analyze  free-form text fields from any database (customer service, internal survey etc.) to identify trends and topics, so that actionable insights can be gained

Generative AI

Generative AI refers to the ability of AI systems to generate new content or data that resembles and captures the characteristics of the original input data. Generative AI models can create realistic images, text, audio, and even videos that are nearly indistinguishable from real human-generated content. This approach is often used in areas such as creative arts, content generation, and simulation. 

An example of generative AI for utilities includes: 

• Streamline Compliance & Governance: Create a large language model (LLM) with internal policy documents (engineering standards) and industry policy (NESC) to assist with self-service compliance in pole loading

Making Business Sense of Data Analytics

Before diving head-first into AI’s deep and as-of-yet uncharted waters, it’s important for utilities to understand the evolution of analytics as just described in a business context. Simply put, how can data analytics (not just AI!) help your organization achieve its goals? 

Establishing an effective analytics culture within your organization is the first step.

Find analytical pioneers in your organization who were the “data scientists before there were data scientists” to establish governance. These are the people who have already mastered predictive analytics techniques and have established the modern habit of defending their positions with numbers rather than anecdotes or declarations of superior insight from personal experience. 

Make these data proponents the leaders of your organization’s data team and give them grounds to establish governance on how data is to be used at your company. 

What Data Analysis Tools Can Best Help Utilities RIGHT NOW?

Electric utilities are tasked with (above all) maintaining the grid’s reliability and safety both in the present and in the future. Utilities often find themselves in the process of analyzing the grid’s infrastructure today and predicting how it might fare tomorrow.

It would seem then that predictive analytics as we have just defined them–models that ingest historical data and make forecasts and predictions about outcomes and events in the future–would be the key tool for forward-thinking utilities seeking to shore up their grids of the future.

But if we examine our definition of machine learning–models and algorithms that learn from the data but can make these predictions and decisions without being explicitly programmed–we find the term also defines a set of tools whose goal is making predictions while adding a layer of automation to help achieve insights at scale.

It’s that layer of automation and the ability to achieve insights at scale that make machine learning solutions (not just AI!) a utility’s best friend in 2023 and beyond. 

The Case for Machine Learning and Human-Driven Automation

That US utilities have their work cut for them when it comes to preparing their grids for the future while facing a dearth of human resources to complete the immense workload has been well-documented.  AI’s nascence and seemingly limitless (for better or worse) potential has also been well-documented. 

AI, as our definition proclaims, seeks to develop systems or behaviors that exhibit intelligence. If there is a component of AI (or more to the point generative AI) that is causing electric utility professionals to give pause, it’s the notion of relinquishing control of a critical system to a machine. 

With machine learning, however, no such relinquishment takes place. A human is always in control and always involved in the process of deciding whether to act or not act on the model’s output.  But that human is not doing all of the work. The identification and output are done by the computer, but the work–i.e. the thinking–is done by humans. 

Whereas the automation component is what differentiates machine learning from predictive analytics, the human element of machine learning is what differentiates it from AI and should also assuage any fears associated with generative AI rising above human control. 

For these key reasons, machine learning is what utilities will be using to drive innovation now and for the next decade. 

The grid was conceived and designed by humans using machinery to make advancements at scale. Now and for the next decade, machine learning can help it stay that way.

Matt Fernandez–a sentient human being and NOT a proxy for generative AI –is a Project Manager and Certified Scrum Master for ikeGPS. He has eight years of experience with a large public utility, performing various roles from managing IT projects to capital planning.  He holds a bachelor’s degree in Information Systems Management from Duquesne University and an MBA from The University of Akron.