7 Ground-Breaking Machine Learning Applications for Utilities

Technology, Dev Tales

Digital Transformation is an ongoing process for utilities today. However, to be successful they must focus on technologies that deliver the services customers want. Machine Learning offers enormous potential for utilities to discover more about their customers and for solving the common issues utilities face every day.

Digital Transformation and Machine Learning

Today, it is undisputed that Digital Transformation is essential for utilities. However, organizations often find the results of their Digital Transformation efforts disappointing.

One major reason for this underperformance is that efforts are focused on new technologies rather than on customer needs.

Machine Learning and Data Analytics offer a solution for utilities. This powerful combination can deliver insights that give utilities a better understanding of their customers. These insights can then support focused, customer-centric Digital Transformation.

Machine Learning is already helping to solve many of the issues that utilities face on an on-going basis. Next generation Machine Learning models promise to solve many more of these common utility problems and generate new business value.

In this blog, we’ll examine where leading-edge models are already being used to understand customers and solve common energy utility problems. We’ll also take a closer look at areas where great strides are being made in the development of next-generation Machine Learning models.

Firstly, we’ll dig a bit deeper into why Machine Learning and Data Analytics are vital components in utilities’ Digital Transformation.

Why Digital Transformations Fail

A recent survey by Deloitte found that 95% of energy executives believe that ‘Digital transformation is a top strategic priority’. However, only 5% of companies report achieving or even exceeding expectations for their digital transformation.

If you think this sounds a bit discouraging, you are not alone. Many hours have been spent analyzing the statistics and much has been written about this lack of success.

So, let’s look at two very common reasons for Digital Transformation efforts falling short of expectations. They are actually flip sides of the same coin.

1. Investing without Understanding Customer Need

      One common reason for the failure of Digital Transformation efforts is an attempt to move too quickly through transformations. Many organizations, and utilities are no exception, attempt to rapidly expand their digital capabilities, anticipating demand that doesn’t exist, or doesn’t yet exist. They adopt a “if you build it, they will come” approach without understanding their customers’ needs.

      2. A Slow Adoption Rate of the Technologies that Deliver Customer Insights

          Machine Learning and Data Analytics are currently hot topics in the sector. Utilities are keen to explore the value these technologies can bring to the large volumes of data they now have access to by delivering insights into customers’ behavior.

          However, the adoption rate has been slow as organizations are either still considering how they should use these technologies or are in the very initial stages of implementing solutions. There are multiple reasons for this, from technical difficulties such as issues with data quality to a lack of technical know-how within the organization.

          The answer?

          Proper Machine Learning and Data Analytics. This powerful combination allows utilities to understand customers’ needs and the actions required to meet these needs.

          To gain an edge in a competitive market, utilities need to act. But they must act in a focused and customer centric way.

          Firstly, we’ll look at some of the ground-breaking ways Machine Learning is already being used by some utilities to achieve this.

          Ground-breaking Machine Learning Applications

          1. Identifying Suspicious Metering Points

          There are many possible reasons for suspicious metering points including malfunctioning end points, installation problems, wrongly configured SoR data or even the customer being the victim of fraud.

          This is an everyday problem for utilities, but the consequences are not trivial, either for utilities or customers. The customer notices their energy bills skyrocketing and may switch provider without the utility ever knowing the reason for having lost a customer.

          The Machine Learning Solution

          Suspicious metering points such as unstable or unexpectedly large consumption, gives an early warning into unusual behavior.

          However, detecting these metering points is not an easy task.

          Detecting suspicious values in the amount of meter data is challenging, especially because electric consumption heavily depends on the context. For instance, a customer’s energy consumption profile may change during vacation or weekends compared to normal working days”.

          Insights from data can clearly help solve this situation. Machine Learning and data analysis facilitate customer clustering based on consumption patterns. If a utility receives automatic detection of suspicious consumption data, it can be proactive in helping the customer find the cause of the unexpectedly high bills.

          2. Assure Incoming Grid Data Integrity

          As utilities becoming increasingly reliant on data, data quality has become a major concern. Poor quality data affects the accuracy and effectiveness of electricity data mining and energy big data analytics.

          An example of such an issue is power outages. When an outage hits an area, it affects a particular distribution station or substation. However, some metering points assumed to be connected to the station may still be sending values whereas others not connected to that station have stopped sending data.

          This means that information related to the metering point connection is incorrect and could potentially lead to some miscalculations and prediction errors. The utility needs to know about these issues so the problem can be fixed.

          The Machine Learning Solution

          Machine Learning models are already providing constant background verification and validation of incoming data. If there are errors or inconsistencies the utility is notified leading to more accurate results from its data.

          3. Grid Edge Insights

          Utility customers vary enormously in their consumption habits. Some have predictable energy usage patterns with conventional working and commuting times whereas others may have sporadic or extreme patterns, requiring large amounts of power at specific times of the day.

          In fact, these more unusual patterns have increased in areas which have experienced a Coronavirus lockdown and may not return to ‘normal’ even after restrictions have been lifted.

          The rapid increase in innovation at the Grid Edge adds another layer of complexity. Electric vehicle charging, solar panels and a growing number of behind-the-meter appliances will influence the way users produce or consume energy. Understanding how different lifestyles and technologies affect electricity use allows utilities to analyze where loads are located at the Grid Edge based on specific data patterns and enables utilities to tailor energy efficiency solutions.

          The Machine Learning Solution

          Machine Learning models are already being applied to predict the likelihood of certain Grid Edge devices or applications being used and their location based on energy usage and consumption patterns. This is enabling utilities to improve their service delivery.

          4. General Real-Time Statistics

          Real-time reporting and visualization enable organizations to make better decisions, operational efficiencies and improve customer service and profitability.

          It is not surprising then that reporting and visualization of the general state of the system is one of the features most commonly requested by management.

          The Machine Learning Solution

          Machine Learning models are already being used to capture real-time data. The type of data used varies depending on the issues a utility is facing.

          One common use case is applying the model to data that shows the percentage of active metering points sending valid data in real-time. However, the issues and data where the Machine Learning model is adopted can be altered or added to over time as the utility’s priorities change.

          Next Generation Machine Learning Models

          The following set of real-world use cases are common to utilities the world over. Using Machine Learning models to solve these issues will improve customer service and lower operational costs. Great strides are already being made in the research and development of these next-generation models.

          5. Predicting Outages

          Power outages are costly for companies and disruptive for consumers. An ability to predict where outages are likely to occur and then prevent them would be a massive step change for utilities.

          Next-Generation Machine Learning Solution

          Research has already begun on developing models to predict outages so they can be prevented, or the time customers are left without power can be reduced. In the U.S., The Department of Energy has experimented with Machine Learning to detect weaknesses in the grid and proactively repair them before any outages occur. In addition, data can be used to locate vegetation that could cause issues and allocate crews to address those issues.

          Weather events are a massive concern for utilities and are often responsible for outages. If utilities can analyze weather data, asset and metering point behavior patterns, they are better placed to find risk time windows and grid vulnerability areas.

          Predicting outages due to weather events requires a complex Machine Learning model because it involves many different factors such as meteorological data, power consumption information and so on. However, work is already well underway on finding a solution for this universal issue.

          6. Energy Theft

          Electricity theft and other “non-technical losses” (NTL) total a staggering $96 billion per year globally.

          NTL is often a hidden cost and has enormous financial consequences for utilities, customers and even governments. They lead to higher prices for customers, safety issues and can threaten the financial sustainability of electric utilities.

          One of the main causes of NTL in distribution networks is electricity theft. This causes significant harm to power grids, affecting power supply quality and reducing operating profits.

          Next Generation Machine Learning Solution

          It is already possible to use Machine Learning to provide real-time data on NTL. However, understanding where future theft is likely to happen would be an invaluable tool. This requires a complex Machine Learning model, but research is already being carried out to develop such a model. Utilities could then put measures in place to make vulnerable areas more secure (link to White Paper).

          7. Consumption Predictions

          One of the new challenges utilities face as a result of the coronavirus pandemic includes their understanding of energy consumption and as a result has highlighted vulnerabilities within utility ecosystems.

          Long term electricity consumption forecasting forms the basis for energy and grid investment planning. Poor forecasting, or a total lack of forecasting, has a negative impact on utilities, resulting in wasted resources, higher operational costs, possible energy outages and financial losses.

          Increasing the accuracy of electricity consumption modeling and a better understand these changes on grid loads helps utilities to avoid making costly mistakes and make informed smarter grid investment decisions.

          Next Generation Machine Learning Solution

          There has already been a considerable amount of research into Machine Learning models for consumption prediction based on different factors. One method is to calculate prediction performance. Another route is to predict energy demand. While both models are a work-in-progress, a viable model will be an invaluable tool for power utilities.

          Some of these Machine Learning use-cases sound like a glimpse into a future world. But there are already some products on the market to help utilities to tackle these scenarios. One of these is Utilihive Datalake from Greenbird Integration Technologies.

          Utilihive is already helping utilities to identify suspicious metering points, to assure incoming data integrity, delivering Grid Edge insights and real-time statistics and visualization. What’s more, developers at Greenbird are currently working on solutions to provide the next generation of Machine Learning models for utilities.

          Machine Learning already offers utilities the tools to understand their customers better. Next generation solutions have the potential to solve the complex issues utilities are tackling every day.


          Utilihive empowers utilities to manage their data flow faster and smoother than traditional system integration models while accelerating utilities’ journey towards the energy revolution.

          For more information on the Machine Learning solutions Utilihive offers and the next-generation solutions our developers are currently working on, download our Whitepaper here.

          Greenbird offers out-of-the-box system integration for utilities. We are a true DevOps company, delivering unique time-to-market and reliability. We were named a Gartner ‘Cool Vendor’ in 2018 because of our domain specific and flexible integration capabilities, crucial for creating easy-to-consume integrated solutions.