AI is an enabler of rapid transformation in the energy sector. Utilities are moving away from simply supplying energy to an energy-as-a-service model, where the customer experience is what distinguishes power providers.
Hover over the infographic in order to share it
Decarbonisation and decentralisation play a paramount role in the long-term transformation of the electricity system. Moving towards 100% decarbonisation requires the integration of distributed renewable electricity generation, mainly photovoltaic sources, with consumers and, soon, with storage devices. In a decentralized energy system, energy flows happen bi-directionally and the boundaries between producers and consumers fade. As result, the grid becomes more complex to manage and operate because system operators must deal with situations in which the peripherical production is limited or exceeds consumption. Renewable resources such as solar and wind are variable by their nature and can increase volatility in the energy production: therefore, future carbon-free electric grids must rely on other forms of controllable generation or consumption to match the supply of energy to demand.
A necessary step towards a decarbonised system is grid digitalisation where the main goal is to guarantee a reliable and seamless operation leveraging on data coming from the field. Utilities understand this and are increasingly investing to improve data transmission channels (5G, fiber), in the internet of things solutions/edge computing and in smart meters and sensors. All these technologies are becoming cheaper with time, and they are enabling new operational capabilities for the power grids of the future.
At the same time, the widespread roll-out of smart meters will allow for the application of AI-enabled demand response and flexibility services that will facilitate the greening of the grid by better matching customer usage with intermittent renewable output. Effective monitoring and customer behaviour forecasting are essential to better match up supply and demand with flexibility tools, for example with smart home and smart electric vehicle charging services.
One solution to control this volatility is presented by virtual power plants (VPP), systems that aggregate heterogeneous distributed energy sources (DER) such as stationary batteries, solar panels, electric vehicles and flexible loads to modulate aggregated power exchanged with the grid and deliver energy services needed to maintain grid stability. A VPP can efficiently provide flexibility, thus allowing the system to better react to fluctuations. AI and advanced analytics tools play a key role to manage such complexity; they enable the automation of the processes, the maximisation of financial returns from participation to energy markets and lead to an overall increased stability of the grid.
Organisations are recognising the value of AI: 84% of C-suite executives believe they must leverage AI to achieve their growth objectives, according to Accenture, and 80% of utilities executives believe they risk going out of business in five years if they don't scale AI. Automation not only drives consistency and scalability to business processes, it also drives growth. But rather than searching for ways to implement AI projects, utilities should identify pain points they have and conduct analysis to determine which digital tools best fit the problem. Accenture found that 87% of executives acknowledge they know how to pilot, but struggle to scale AI across the business.
An AI strategy should be seen as a journey, which should start with walking before running. A solid foundation is required to craft a different approach and build new skills.
Many utilities find that predictive maintenance and asset optimisation is a good place to start testing AI solutions, as efficiency gains offer fast returns on investment. These solutions save time and money while improving quality and safety.
For example, ABB's Asset Health application allows models and current operational data to predict asset failure and allow for condition-based maintenance schedules. Dutch company OneWatt’s technology employs smart acoustic sensors that use sound data and machine learning to predict motor failures.
Offshore wind turbines are difficult to maintain due to their remote location, and so present a prime opportunity for AI-driven solutions. Machine learning and AI is used in the Horizon 2020 Romeo project to optimise the operation and cost of offshore wind turbines through predictive decision-making. Portuguese data analytics platform provider Jungle AI forecasts wind plant failures before they happen which reduces downtime and increases availability of turbines. French start-up Sterblue provides an autonomous drone inspection solution for wind turbines with an AI back end to analyse the inspection data. Spanish utility Iberdrola uses AI to plan maintenance, monitor electricity usage and optimise distribution. Italian utility Enel uses AI in the whole value chain, on both structured and unstructured data, from predictive maintenance for the grip optimisation, production forecast, flexibility and optimisation for grid decarbonisation, sentiment analysis for brand reputation, customer value and image recognition for asset management and anomaly detection.
Possibly the most extensive example is in China, where State Grid is using AI to manage equipment, including fault detection, control, and diagnosis in facilities. China State Grid uses power grid demand forecasting, identifying type and severity of grid breakdowns, computer vision to identify defects in transmission lines, and predictive maintenance of electricity equipment. Texas-based utility ONCOR uses AI and geospatial technologies to help predict where vegetative growth is most likely to interfere with power lines - which can cause blackouts and wildfires - allowing managers to better plan for preventative maintenance.
One of the biggest challenges for grid management as the share of renewable generation grows is the variable nature of its output based on weather conditions. Many solutions are emerging to address this challenge, possibly the best-known being Google DeepMind’s wind energy output forecaster which it claims boosted its ‘value’ by about 20%. Options for solar forecasting are also being tested. OpenClimate Fix is using artificial intelligence to predict how clouds affect solar output using a ‘nowcasting’ model working with National Grid ESO and the Met Office, and using satellite data with cloud imagery and production statistics from around 600 PV systems from PVoutput.org. Deep learning algorithms are also able to learn by trial and error. For example, Norway’s Agder Energi developed an algorithm to optimise water usage in hydropower plants.
AI-driven data analysis can also help determine the optimal location for low-carbon technologies by predicting the impact on network capacity, as demonstrated by the UK’s Western Power Distribution and Electralink who are seeking to improve congestion management. Danish transmission system operator Energinet engaged IBM to design a ‘virtual operator’, which could estimate risks of operational limit violations based on simulation data.
The Enel group widely deploys chatbot and virtual assistants in the retail sector. Both in Italy and in Spain, Enel is developing customer solutions with an AI-enabled voice assistant technology to advise them on more efficient energy consumption and optimising costs or for consumption reporting and client assistant. In Italy, Enel’s customer management can be made more efficient thanks to the virtual assistant Elena available on a 24/7 basis on different channels: Telegram, Enel Energia web page, Facebook Messenger, phone channel and WhatsApp. Using a complex systems of AI algorithms, Elena can guarantee for every customer an easy, fast and complete user experience providing support on the different Enel Energia services. For instance, Elena can give information about energy readings or the status of supply activation, the different methods for paying a bill or where you can find the nearest service point.
In the US, CubeLogic’s Know Your Customer application uses data from social media and the internet to give insight into market trends and activity.
Smart devices such as Amazon Alexa, Google Home and Google Nest, often described as part of a ‘smart home’, allow customers to set their thermostats and other control systems to automatically monitor their energy consumption based on desired comfort levels. Appliances can be switched off when power is expensive, or cars and other batteries can be charged when power is cheap. Energy suppliers can use AI to forecast likely customer behaviour and plan accordingly.
A division of Ireland’s largest utility Electricity Supply Board ESB Networks is rolling out a smart metering programme using AI to support the installation process. The installer takes photos before and after the home installation. Computer Vision is then used to ensure the job has been completed and the appropriate standards have been applied.
Iberdrola STAR project in Spain installed over 10 million smart meters and automated secondary substations and now Iberdrola is running developments such as fault and fraud detection, and an Automatic Replenishment Algorithm (ARA), resulting in an Increase of reliability (> 99.995%). Iberdrola has also become 1.8 times faster in network fault recovery due to its AI-based ARA tool, automatically detecting and isolating the smallest area of the network to speed up recovery, and reduced operating costs by 12%.
AI and digital twin technologies can also create ‘smart cities’, where public lighting, mobility and planning can be coordinated remotely. Enel’s Urban Futurability project in São Paulo, Brazil is a 3D digital model that matches the local electricity infrastructure via thousands of sensors installed on the actual grid, each communicating information on grid status in real-time to both the distributor and local stakeholders. This digital twin can be used to create greater awareness of energy use and efficiency.
One of the most important ways AI will drive the energy transition is in optimising distributed resources and battery storage with customer demand, including electric vehicles. As the electricity grid morphs from a one-day transmission of electricity from large centralised power plants to hundreds of thousands of small generators and smart devices with bidirectional flows, system management requires a superhuman effort.
Many virtual power plant and internet of things projects are already in operation, where flexibility resources are aggregated and can provide services to the grid. Some were supported by the EU’s Horizon 2020 programme, for example a smart system of renewable energy storage based on INtegrated EVs and bAtteries to empower mobile, Distributed and centralised Energy storage in the distribution grid’ (INVADE) - a cloud-based flexibility management system incorporating machine learning and other advanced analytical techniques. The project explores how big data, machine learning and analytics can be integrated into existing smart home solutions.
Electric vehicles present an opportunity for flexibility where vehicle-to-grid technology can be used to optimise charging, and AI is being deployed to enhance this. A team from Stanford University, the Massachusetts Institute of Technology and the Toyota Research Institute published findings from battery testing aimed at cutting electric vehicle charging times down to 10 minutes using AI without increasing the degradation of lithium-ion batteries.
Habitat Energy is using cutting edge AI to optimise grid scale batteries taking into account the chemistry, the degradation factors, the replacement cost and how the individual batteries can be optimised every five mins across multiple traded markets.
The benefits go beyond the power sector: Utonomy is helping to create a smart gas grid by using AI along with hardware to optimise pressure in gas grids in order to reduce leakage but also to allow greater access of biomethane and hydrogen.
And AI can even be used to address a problem it creates itself, that big computer servers burn a lot of energy: Google’s Deepmind says it has used reinforcement learning to reduce energy use in its data centres by 15%.
Iberdrola is utilising its own predictive maintenance system ASPA (Advanced System of Predictive Analysis), for its onshore wind fleet, since 2017. Using digital twins for the wind turbines Iberdrola can identify signs of turbine failures before they occur, reducing corrective maintenance time and, thereby, enhances the reliability of the turbine’s energy generation. Thanks to this improvements Iberdrola is saving 50,000 man-hours/year, -10% major correctives and -20% parts supply.