This article was published by the International Energy Agency on Nov. 2, 2023.
Managing the grids of the future will require more powerful analytical tools, with a critical role for AI
Power systems are becoming vastly more complex as demand for electricity grows and decarbonization efforts ramp up. In the past, grids directed energy from centralized power stations. Now, power systems increasingly need to support multi-directional flows of electricity between distributed generators, the grid and users. The rising number of grid-connected devices, from electric vehicle (EV) charging stations to residential solar installations, makes flows less predictable. Meanwhile, links are deepening between the power system and the transportation, industry, building and industrial sectors. The result is a vastly greater need for information exchange – and more powerful tools to plan and operate power systems as they keep evolving.
This need arrives just as the capabilities of artificial intelligence (AI) applications are rapidly progressing. As machine learning models have become more advanced, the computational power required to develop them has doubled every five to six months since 2010. AI models can now reliably provide language or image recognition, transform audio sounds into analyzable data, power chatbots and automate simple tasks. AI mimics aspects of human intelligence by analyzing data and inputs – generating outputs more quickly and at greater volume than a human operator could. Some AI algorithms are even able to self-programme and modify their own code.
It is therefore unsurprising that the energy sector is taking early steps to harness the power of AI to boost efficiency and accelerate innovation. The technology is uniquely placed to support the simultaneous growth of smart grids and the massive quantities of data they generate. Smart meters produce and send several thousand times more data points to utilities than their analogue predecessors. New devices for monitoring grid power flows funnel more than an order of magnitude more data to operators than the technologies they are replacing. And the global fleet of wind turbines is estimated to produce more than 400 billion data points per year.
This volume is a key reason energy firms see AI as an increasingly critical resource. A recent estimate suggests that AI already serves more than 50 different uses in the energy system, and that the market for the technology in the sector could be worth up to USD 13 billion.
AI and machine learning can unlock flexibility by forecasting supply and demand
One of the most common uses for AI by the energy sector has been to improve predictions of supply and demand. Developing a greater understanding of both when renewable power is available and when it’s needed is crucial for next-generation power systems. Yet this can be complicated for renewable technologies, since the sun doesn’t always shine, and the wind doesn’t always blow.
That’s where machine learning can play a role. It can help match variable supply with rising and falling demand – maximizing the financial value of renewable energy and allowing it to be integrated more easily into the grid.
Wind power output, for example, can be forecast using weather models and information on the location of turbines. However, deviations in wind flow can lead to output levels that are higher or lower than expected, pushing up operational costs. To address this, Google and its AI subsidiary DeepMind developed a neural network in 2019 to increase the accuracy of forecasts for its 700 MW renewable fleet. Based on historical data, the network developed a model to predict future output up to 36 hours in advance with much greater accuracy than was previously possible.
This greater visibility allows Google to sell its power in advance, rather than in real time. The company has stated that this, along with other AI-facilitated efficiencies, has increased the financial value of its wind power by 20 per cent. Higher prices also improve the business case for wind power and can drive further investment in renewables. Notably, Google’s proprietary software is now being piloted by a major energy company.
Additionally, with a more accurate picture of peaks in output, companies like Google are able to shift the timing of peak consumption, such as during heavy computing loads, to coincide with them. Doing so avoids the need to buy additional power from the market. This capacity, if expanded more widely, could have a significant impact on the promotion of load shifting and peak shaving – especially if combined with better demand forecasts. For example, Swiss manufacturer ABB has developed an AI-enabled energy demand forecasting application that allows commercial building managers to avoid peak charges and benefit from time-of-use tariffs.
AI can also prevent grid failures, increasing reliability and security
Another key AI application is predictive maintenance, where the performance of energy assets is continuously monitored and analysed to identify potential faults ahead of time. Maintenance typically happens on a regular schedule; poles on a transmission line, for example, might be examined once within a pre-defined period and repairs carried out as needed. This one-size-fits-all approach can lead to inefficiencies if maintenance happens too early or, more problematically, too late.
To address this, a range of utilities are developing AI-enabled schemes to help monitor physical assets and use past data on performance and outages to predict when intervention is required. Utility company E.ON, for instance, has developed a machine learning algorithm to predict when medium voltage cables in the grid need to replaced, using data from a range of sources to identify patterns in electricity generation and flag any inconsistencies. E.ON’s research suggests that predictive maintenance could reduce outages in the grid by up to 30 per cent compared with a conventional approach.
Similarly, in 2019 Italy-based utility Enel began installing sensors on power lines to monitor vibration levels. Machine learning algorithms allowed Enel to identify potential problems from the resulting data and discern what caused them. As a result, Enel has been able to reduce the number of power outages on these cables by 15 per cent. Meanwhile, Estonian technology startup Hepta Airborne uses a machine learning platform with drone footage of transmission lines to identify defects, and State Grid Corporation of China uses AI extensively to carry out actions such as analysing data from smart meters to identify problems with customers’ equipment.
Potential uses for AI across power systems are likely to soar in the years to come. In addition to better forecasting of energy supply and demand and predictive maintenance of physical assets, applications could include:
- Managing and controlling grids, using an array of data from sensors, smart meters and other internet-of-things devices to observe and control the flow of power in the network, particularly at the distribution level.
- Facilitating demand response, using a range of processes such as forecasting electricity prices, scheduling and controlling response loads, and setting dynamic pricing.
- Providing improved or expanded consumer services, using AI or machine learning processes in apps and online chatbots to better customers’ billing experiences, for instance. Firms such as Octopus Energy and Oracle Utilities are already exploring this.
The technology will enable digitalization, but addressing risk is also essential
Without AI, system operators and utilities will only be able to make effective use of a fraction of the new data sources and processes offered by emerging digital technologies, and they will miss out on a significant proportion of the benefits on offer. However, risks associated with AI must also be considered and addressed before the technology is scaled across the sector. These include, but are not limited to, threats to cybersecurity and privacy, the influence of biases or errors in data, and miscorrelations due to insufficient training, data or coding mistakes.
The availability of workers with the right skills is a significant challenge for any sector looking to tap AI’s potential. Across the global workforce, AI and machine learning specialists are the profession experiencing the fastest growth in demand, creating a recruitment bottleneck. In June 2022, there were only 22 000 AI specialists globally across all industries, and 61 per cent of large firms surveyed in the United Kingdom and United States reported lacking staff with sufficient AI experience. The energy industry will need to compete to recruit the best data scientists and programmers, while firms looking to retain staff that understand the sector should consider uptraining and reskilling parts of their existing workforce. Digital training courses, supported by governments with input from the private sector, will be vital to these efforts. However, the availability and quality of such courses is not yet consistent across the largest global economies.
AI also uses more energy than other forms of computing – a crucial consideration as the world seeks to build a more efficient energy system. Training a single model uses more electricity than 100 US homes consume in an entire year. In 2022, Google reported that machine learning accounted for about 15 per cent of its total energy use over the prior three years. However, data is not systematically collected on AI’s energy use and wider environmental impacts, and there is a need for greater transparency and tracking – especially as models grow. The most efficient computing infrastructure and AI algorithms should be prioritized to prevent it from offsetting efficiency gains.
Furthermore, increased use of automated and self-learning software raises questions about who is responsible for the outputs or outcomes of these systems. Operators frequently purchase AI technology or a related service from IT companies and startups. This can result in decision making on electricity balancing or investments, for example, based on models they do not understand or control, leading to questions about accountability for public spending, energy prices or outages.
In an effort to address some of these issues, the OECD AI Principles – adopted in 2019 by OECD member governments and many non-member governments – provide guidance on pursuing a human-centric approach to trustworthy AI. Clearer national, regional and international frameworks may also be necessary, given that the energy sector underpins the global economy and is crucial to meeting climate goals. The European Union’s AI Act, first proposed in 2021 and currently under negotiation by EU institutions and member states, aims to develop better conditions for the technology’s development and use while guaranteeing robust protections for the environment, among other goals.
For AI to be an effective ally towards efficient, decarbonised and resilient power systems, governments will also need to develop mechanisms for data sharing and governance. A coordinated global approach can enable internationally applicable and replicable solutions, transfer learnings globally, and expedite the energy transition while reducing its costs.