The effects of climate change are becoming ever more apparent across the world, with storms, droughts, fires and flooding increasing in both frequency and severity. At the same time, a global energy transition is in progress, with societies moving away from the fossil fuels that drive climate change towards low-carbon energy systems. Artificial intelligence (AI) and machine learning (ML) have the potential to speed up this transition and help halt the rise in global temperatures by making both better predictions and better decisions about how technologies are used. AI is being applied in areas as diverse as transportation, buildings and governmental policy. In this blog, Alexander Kell, research assistant at the Sustainable Gas Institute (SGI), explores how five sectors may be transformed through the use of intelligent analytics and big data. Note that this is a non-exhaustive list, and the possibilities span far further than those which have been highlighted here.
The electricity system will play a central role in future energy systems. Not only does the electricity system produce a large amount of data, which can be utilised by AI algorithms, for example, from smart meters installed in the home, but electricity can be used to decarbonise other sectors. To achieve this, a transition from stable but carbon-emitting energy sources, such as coal and gas, to variable sources, such as solar and wind, is required. This demands a paradigm shift in the way the electricity system is managed.
One way in which AI could reduce the dependence on coal and gas plants, is to better predict wind speeds, solar irradiance and electricity demand. This way, the increasing amounts of variable sources can be proactively managed. For example, better predictions would allow us to use fewer backup gas or coal power plants, which would otherwise be required to remain in reserve.
Another method of managing the differences between solar and wind supply and electricity demand is to use batteries. Batteries too have to make complex decisions based on likely electricity demand and supply. Here, a subset of AI called reinforcement learning can be used to make these decisions automatically for us, based on historic data.
Globally, transportation accounts for about a quarter of energy-related carbon emissions. In comparison to the electricity sector, however, transportation has not made significant progress in lowering carbon emissions. The rapid uptake of electric vehicles and development of autonomous vehicles could dramatically change this picture in the next decade.
With autonomous vehicles, complex decisions can be made using the same reinforcement learning subset of algorithms that make decisions for batteries, as discussed previously. For instance, decisions based upon the best time to charge a robo-taxi can be made using reinforcement learning. It may seem simple: charge when the battery is low, but this may not be optimal based upon the carbon intensity of the grid, or the customer demand at a particular time.
Another application of AI that could transform road transport is ‘platooning’. This is where trucks can travel very close together to reduce air resistance. By relying on AI, rather than on the skill of the driver, to make decisions about speed and position on the road, fuel efficiency can be increased, potentially reducing the challenge of electrifying long-distance road freight.
Buildings and Cities
It is possible today for buildings to consume almost no energy. However, two problems exist — the heterogeneity of buildings and inertia. Buildings can vary wildly in design and construction and so decarbonisation strategies may be different depending on the type of building. Machine learning and AI can be used to select strategies which are optimal for specific buildings, and can help to implement those strategies with smart controllers.
It is possible to improve the electricity, heating and cooling demands of a building using AI and machine learning technologies. This reduces the need for complex thermodynamic simulations, with a focus on data-driven algorithms. Once a single building has been optimised, it is possible to use the same algorithm on a different building by employing a process called transfer learning. Transfer learning takes a pre-trained algorithm and is used as a starting point for another task. This can make it easier for many buildings to be modelled whilst only training on a smaller subset.
This transfer learning approach can be used for the controlling of energy demands, such as air conditioners and heating. By predicting the temperature, or the occupancy of buildings, smart thermostats, for example, can be used to reduce the total energy required for the same level of comfort.
Climate change poses substantial financial risks to global assets. However, it can be hard to identify which stocks will be the most impacted. Most of the current financial system focuses on quarterly or yearly performance, but this fails to take into account the long-term impacts of changes to the climate.
Climate investment involves the investing of money in low-carbon assets. ML could have potential in the way that these stocks are chosen by analysing features of the stocks involved for portfolio selection and investment timing. For example, the monitoring of annual reports using ML could give advance notice of an undervalued or overvalued stock.
Another approach is climate analytics, which aims to predict the financial effects of climate change. This involves the analysis of investment portfolios, funds and stock in companies to pinpoint areas which have a higher risk due to climate change. Natural language processing, which is able to analyse text documents could, for example, identify climate risks.
Policy can play a vital role in all of the industries mentioned above, for example, by targeting subsidies or taxes in the appropriate way. However, AI can also be used to help decision makers explore the various policies themselves.
Increasingly, agent-based models such as MUSE, developed at the Sustainable Gas Institute at Imperial College, are looking at the long-term energy system. These models can utilise AI to explore a large amount of variables.
A particular example is the setting of a carbon tax over a multi-year time period. Through the use of a long-term model, a large amount of tax strategies can be trialled to get an optimal outcome, as per the wishes of the decision maker.
Alexander Kell is a research assistant at the Sustainable Gas Institute (SGI) at Imperial College London, where he works as part of the MUSE modelling team. Alexander joined the SGI in 2020 following a PhD in computer science from Newcastle University, where he developed a novel agent-based model to better understand energy transitions from a national perspective. Alexander previously studied a masters in electrical and electronic engineering at University College London
For further information see the following publications which helped to inspire this article:
- Rolnick, David, et al. “Tackling climate change with machine learning.” arXiv preprint arXiv:1906.05433 (2019).
- Kell, Alexander, A. Stephen McGough, and Matthew Forshaw. “Segmenting residential smart meter data for short-term load forecasting.” Proceedings of the Ninth International Conference on Future Energy Systems. 2018.
- Pettit, Jacob F., et al. “Increasing performance of electric vehicles in ride-hailing services using deep reinforcement learning.” arXiv preprint arXiv:1912.03408 (2019).
- Sachs, Julia, et al. “An agent-based model for energy investment decisions in the residential sector.” Energy 172 (2019): 752-768.
- Kell, Alexander JM, A. Stephen McGough, and Matthew Forshaw. “Optimizing carbon tax for decentralized electricity markets using an agent-based model.” Proceedings of the Eleventh ACM International Conference on Future Energy Systems. 2020.