This is an extract from a recent report “Energy Security and AI” by the Parliamentary Office of Science and Technology, United Kingdom. This report investigates the significant role that artificial intelligence (AI) and Machine Learning (ML) are set to play in the UK’s transition towards its Net Zero targets 

The UK Government has committed to decarbonising the national electricity grid by 2030 under the Department of Energy and Net Zero Mission Control initiative, pulling the previous target forward by five years, which largely requires increasing electricity generation from renewable sources. The use of AI has the potential to reduce strain on the network and speed-up renewable connections as the UK works to meet its Net Zero targets, while also reducing costs from the operator to the consumer. Greenbyte AB, a renewable energy management platform, estimates that the world’s wind turbines generate more than 400 billion points of data annually, data which could be used to better understand turbine performance. There is potential for AI and ML technologies to leverage this type of big data to improve the efficiency of energy generation, storage and use. However, there are uncertainties around the use of AI tools, with some stakeholders expressing concerns over privacy, security, transparency, and ethical use.

What is AI in the context of energy security?

AI and ML: AI does not currently have a universally agreed definition. The UK Government’s 2023 policy paper defines AI as “products or services” that are “adaptable” and “autonomous”. Industry stakeholders indicate that predictive AI will provide key benefits to the energy system. In all cases, access to data is required for training the AI model

Energy security: AI is already being developed and used in the UK grid, predominantly by the national transmission network and energy system operators for forecasting and maintenance. Wider implementation of AI and ML in the energy system could optimise energy planning, generation and use. Ofgem suggests that benefits will apply across the value chain, from operators through to consumers. Academic stakeholders expect improved efficiency to allow lower cost energy for consumers and increased capacity to balance the grid to reduce or delay the need for large infrastructure investments. However, industry stakeholders note that investment in physical grid infrastructure is critical to ensuring long-term energy security of the UK. AI could help to increase the number of renewable generators connected to the network, which could accelerate the transition towards a decarbonised grid and Net Zero targets.

Applications of AI and ML in the energy sector

Technical barriers
Explainability: High performance AI models are generally complex, so decision-making processes can be complex. This is often referred to as ‘black-box AI’, where it can be difficult to understand and explain how an AI model decided on a certain outcome. The electricity system is classified as a Critical National Infrastructure, so the ability to trace the chain of decision-making is of particular importance. The lack of explainability around the decisions taken by AI has impacted stakeholder confidence and slowed uptake within the energy sector. Advances in explainable AI are important to support deployment of applications as the energy system transitions to decarbonisation.

Culture: The IEA suggests that the energy industry is traditionally resistant to rapid change and has a relatively conservative management culture. Stakeholders suggest that while risk averse planning is done with the right intentions, this mindset acts as a blocker to innovation. Smart meter data is currently mainly limited to energy suppliers and Distribution Network Operators (DNOs). Many of the datasets are siloed due to commercial sensitivity or regional inconsistencies in DNO data privacy policy making it unclear what data can be shared. However, aggregated smart meter data is now classified as open energy system data with DNOs required to publish aggregated consumption data.

Data access: The quality of an AI model is dependent on the quality and volume of data on which it is trained. Smart meter data is considered as personally identifiable under GDPR and access is governed by Data Access and Privacy Framework (DAPF). Many of the datasets are siloed due to commercial sensitivity or regional inconsistencies in DNO data privacy policy making it unclear what data can be shared. Stakeholders identified access to data, especially granular data, as a limiting factor for innovators in this area. In July 2024 Ofgem released a consultation on the governance of data sharing infrastructure, with the intention of driving greater availability and standardisation of data.

Jobs and skills: The UK is facing a skills gap in personnel with energy sector knowledge and digital literacy. A survey of 604 organisations across 16 countries, conducted by IBM and Ponemon Institute, reported that the cyber skills gap has grown by 26.2% between 2023 and 2024. Industry stakeholders point to the need to make positions within energy more competitive to attract the necessary AI talent, and workforce upskilling initiatives. According to a survey conducted by Microsoft, Masdar and the Abu Dhabi National Oil Company (ADNOC), 78% of business leaders consider talent and training a challenge in the adoption of AI.

Regulation: The IEA states that “government policies and regulations will play an important role in the deployment of digital technologies”, recommending that regulators consider removing older regulations and introduce new statutes. Industry stakeholders suggest that the established benefit-measuring metrics that drive decision making, funding and bring together industry and government may not fully capture the benefits of AI and developing technologies. Changes to policy and regulation will likely be critical in ensuring the digital transformation of energy is fully realised and to accelerating decarbonisation.

Infrastructure barriers

Communication infrastructure: Communication is central to the real-time operation of power systems as envisioned with smart grids. Distributed edge devices need contact with each other, the main server and low latency. However, older power system infrastructure is not thought, by some stakeholders, to be equipped to handle the demands of emerging technologies. A 2022 report from the Energy Digitalisation Taskforce noted the need to improve interoperability and recommended updating standards to ensure smart devices are able to communicate with one another. Upfront costs required to upgrade software and hardware necessary for AI integration can be prohibitive. They suggest co-planning and co-locating infrastructure is required for a productive outcome.

Computing power: AI is reliant on substantial computing power to train, tune and deploy models. Large-scale models, in particular, require on average 100 times more computing power than other contemporary AI models. Intergovernmental stakeholders such as the Organisation for Economic Co-operation and Development (OECD) suggest that sufficient computing power is an important component in expanding integration of AI in the energy system. Others suggest it will be necessary to address connection bottlenecks and environmental concerns surrounding the scalability of data centres that provide these computing services

Challenges for energy system AI use

Privacy and security risks: Academic stakeholders have raised concerns about data privacy and ownership. According to research consortium EnergyREV, privacy was the most common concern raised by energy sector stakeholders. Energy companies have access to large volumes of personal data, in some cases every half-hour. This information could be used to determine socio-economic and demographic profiles, or insight into daily routines of a household. There are cyber security concerns regarding system operations. Legacy infrastructure designed before cyber security was a concern may expose vulnerabilities, and the globalised nature of asset supply chains makes it difficult to ensure products are procured from trusted sources.

Fairness and accessibility risks: Ensuring that the use of AI does not exacerbate existing inequalities was identified by intergovernmental stakeholders as a potential concern. Sustainable Energy for All states that in “the AI sector, where the workforce is heavily male dominated…only 12% of positions requiring over 10 years of experience [are] held by women”. The lack of diversity in AI development teams could lead to technologies that reflect gender biases. AI models are shaped by the historical data used in their training. There is a risk that underlying biases present in initial datasets could lead to unfair outcomes for protected characteristics. If AI applications are designed to obtain the optimal outcome economically, there are concerns that this could lead to unethical and biased decision-making.

Technical and operational risks: With the ‘black box’ nature of some AI applications, and outsourcing of AI development to private companies, there is a risk that system operators would be using programmes they do not, and in instances cannot, fully understand, making human intervention difficult. Academic stakeholders note the need for explainable and contestable AI, where dynamic human-machine interaction is used to explain and revise the decision-making process, to ensure that AI powered decisions align with intended goals. According to a 2023 World Economic Forum report, 23% of jobs globally may be disrupted by AI in the next 5 years. There are concerns that automation and advancements in AI could reduce the number of staff required in traditional roles within the energy sector, such as field technicians, maintenance staff and data analysts.

Potential mitigation approaches: Privacy-preserving technologies ‘Privacy preserving/enhancing’ technologies can be used to maintain the security of sensitive personal information. For example, Federated Learning has data kept locally, and the model is trained locally, with only parameters shared to central servers. Homomorphic Encryption allows operations to be performed on encrypted data without the need for decryption, ensuring the underlying data cannot be accessed. These are just two examples of a range of possible techniques. However, these technologies are complex and may therefore use more energy.

Enhanced cyber security and digital literacy: The EU AI Act sets out how to implement robust cyber security measures and investing in resilience to protect AI systems. The US National Institute of Standards and Technology (NIST) released an updated framework for dealing with cyber security threats in 2024, while the National Cyber Security Centre (NCSC) produced guidelines for secure AI system development in 2023. The IEA suggests that digital energy security should be built around:

• Resilience – the ability to withstand shocks and adapt quickly. 

• Security by design – where security objectives are a core part of the design process. 

• Cyber hygiene – with precautious access right allocation and training in digital literacy for staff. 

Stakeholders state that adherence to frameworks such as these, and implementation of clear regulation, can help to reduce the cyber security risks associated with AI in the energy system.

Balanced model training and validation: Fair datasets hold the key to unbiased and equitable outcomes, so it is important that the data used to train models is as representative as possible. Development of AI applications that are explainable and interpretable will contribute to improved accuracy, reliability and robustness, to limit potential biases and fairness concerns in models.

Financial support: To realise the full potential of AI optimisation in the energy system, academic and industry stakeholders suggest clear regulations and incentives are needed to attract investment. Investment from government and research funders in sector specific research for energy data and AI could create an ethics-conscious, pro-innovation culture. Careful planning and significant investment in AI technologies could overcome the challenges of AI.

Standardised processes and ethical oversight: Clear protocols, regular re-evaluations and regulatory oversight could help move approaches beyond legal compliance towards ethical best practice. Stakeholders suggest that there is an opportunity for the UK to establish itself as a global leader by setting ‘gold-standard’ best practice guidelines and ethical frameworks. Stakeholders also note the importance of forward planning and the need to approach risks in a less siloed way. For instance, regulation of infrastructure surrounding AI such as green data centre hosting requirements or sustainability reporting schemes currently being explored and implemented by the European Commission.

Impacts on energy systems and market approaches: As the grid becomes more distributed, with increased generation capacity and greater rates of digitisation, AI could also help automate bidirectional markets.AI could manage complex interactions between local markets or microgrids, and rapidly react to granular market signals to facilitate dynamic real-time peer-to-peer trading. Expansion of mechanisms such as demand side response, coupled with small scale renewable generation, could transform end users from passive consumers to active participants in the energy system. The IEA has suggested that distributed generation developments could transform the way that electricity supply functions. Industry stakeholders suggest that while AI is a tool for optimising transition of the energy system, it is not the driver of these changes. Some stakeholders also suggest that there will be very little impact in the next decade, as the current energy sector is not designed for rapid adaptive changes.

Access the brief here