AI to power smarter and more resilient energy systems

Artificial Intelligence (AI) is transforming the energy sector. From detecting faults in infrastructure before they happen, to improving the way power flows across the grid, AI is becoming an essential tool for power utilities transitioning to smarter and more sustainable energy system operations.

Unlocking efficiency gains for power utilities

AI and machine learning can already unlock significant operational benefits for power utilities. For example:

• Detecting non-technical losses, such as electricity theft, using anomaly detection models.
• Supporting faster connection assessments by analysing connection request patterns and automating some of the assessment processes.
• Optimising maintenance schedules based on asset condition and usage, not just calendar dates.

These applications of AI have the potential to save our clients time, reduce costs, and increase reliability across their networks. Ricardo is already supporting organisations, including UK DNOs, to do just this. 

AI in action

One of the biggest opportunities lies in how we maintain our energy infrastructure. Extreme weather, particularly heatwaves, can cause a spike in faults across power systems. But what if we could predict where and when those failures are likely to happen?

In a recent project for the Department for Energy Security and Net Zero (DESNZ), Ricardo developed a machine learning model to do just that. By analysing historical fault and weather data, an AI algorithm was trained to spot patterns between temperature spikes and asset failures.

Our approach

We used the National Fault and Interruption Scheme (NaFIRS) dataset, which records faults reported by Distribution Network Operators (DNOs) from 2018 to 2022, alongside historical UK temperature data. We established the relationship between temperature and fault rates using supervised machine learning (specifically, Extreme Gradient Boosting or XGBoost), looking at a range of factors including maximum daily temperature, minimum daily temperature, and daily rainfall.  

The results

This resulted in a predictive model that forecasts fault risk during heatwaves with reasonable accuracy – giving network operators valuable time to prepare and respond. This is illustrated in the figure below, showing fault prediction results at low voltage.

 
Key lessons learned:

We learned important lessons along the way to improve the accuracy and reliability of such predictive models going forward:

• The accuracy of the predictive models increases not just with the size of the dataset, but also and more significantly with its locational granularity and precision. Specifically, the closer and the more central the temperature sensor to the fault area, the easier it was to establish a relationship between fault occurrence and ambient temperature. Predictive AI models such as this one would benefit from tapping onto more localised temperature measurements, including at substations, for example.
• Although we initially relied on one single dataset (REMIT, i.e. self-reported faults data from generation, transmission and consumption units in the GB wholesale market) we quickly established this was not sufficient. Expanding to more datasets including those privately recorded by distribution network operators was lengthy but instrumental in achieving better performance for the predictive AI model. Collaboration and partnerships between power utilities to pool historical dataset and thereby increase the accuracy of predictive models used by all would undoubtedly bring mutual benefits.

Benefits for power utilities

Overall, such an approach enables smarter maintenance, improves system resilience and ultimately helps to keep the lights on during some of the most challenging conditions caused by climate change.

The future of AI

AI is set to play an even more significant role going forward. Emerging techniques such as deep reinforcement learning, federated learning, and digital twins have significant potential to transform not just the digital tools used by power utilities, but their operating model altogether.

• Deep reinforcement learning (DRL) is a branch of artificial intelligence where algorithms learn to make decisions by interacting with an environment and receiving feedback in the form of rewards or penalties, using deep neural networks to handle complex input data. DRL can be applied to optimize real-time grid operations, such as balancing electricity supply and demand or integrating renewable energy sources efficiently. 
• Federated learning is a machine learning approach that enables model training across multiple decentralised devices or servers while keeping data localised, thus preserving privacy and security. Federated learning could allow power companies to collaboratively develop accurate forecasting models or detect faults using data from multiple locations without sharing sensitive information. 
• Digital twins are virtual models of physical systems or assets that are continuously updated with real-time data, allowing for simulation, analysis, and optimisation. Digital twins enable utilities to simulate and predict the behaviour of the grid under various conditions including in real-time conditions, test new approaches and technologies including virtual power plants, and proactively address maintenance needs or system failures, thereby improving reliability and resilience.

Why Ricardo?

Our strength lies in combining deep technical expertise with a systems-thinking approach to sustainable energy solutions. We not only understand how AI tools can benefit the power system, but how they can help to solve our clients’ challenges in the wider  transition. We bring together specialists in data science, energy systems, digital infrastructure and regulation, and  tailor our expertise towards supporting our  clients’ goals.

As AI becomes more embedded in the way we manage electricity, Ricardo can support your organisation in building power systems that are smart, resilient and ready to incorporate transformational change.

Learn about our solutions for the clean power transition >