THANK YOU FOR SUBSCRIBING
How AI can Impact Energy Sector
By Nazim Osmancik, Group Head of Economics & Fundamentals, Centrica PLC
Looking back, the journey to this point was not smooth as far as I witnessed. When I picked up coding during my days at Macalester College two decades ago, I became fascinated by what computer science could achieve. I remember dreaming of building a massive neural network that would fully automate the operations of a large organisation one day. While I knew such ideas were aspirational at the time, I was struck by how little enthusiasm there was for this technology outside academic research. This was the case even in organisations that took pride in using modelling and analytics to increase value or manage risk. Methods like neural networks were too cumbersome and required too much effort to be practically useful.
Who knew that akin to humans, AI simply needed lots of information and time to learn and master something. As computational capability advanced and the amount of digital data increased, AI started to work, unlocking possibilities that were in the realm of science fiction a decade earlier. AI systems are now routinely deciding on the ads you see on a website and recognising your face to log into phone apps. They can help diagnose medical conditions, increase the output of a dairy farm by identifying what makes animals “happier”, and park your car.
In the energy sector, and electricity in particular, there were several trends that paved the way for AI to be uniquely valuable. Falling costs of renewable energy, ascent of on-site distributed energy, emergence of IoT, and digitalisation of data disrupted the business models built around large centralised supply. The intermittent nature of renewable generation increased price volatility and created challenges in system operations. In this environment, predicting supply, recognising patterns and responding to them rapidly are key to maximising the value our flexible generation assets such as grid-scale batteries.
Meanwhile, on the demand side, energy customers have been asking for more insight and control over their energy use.
The recent revolution has been in the digitalisation of data and utilisation of algorithms that can ‘learn The recent revolution has been in the digitalisation of data and utilisation of algorithms that can ‘learn
The information they need is available from smart meters and sensors, which are already in their billions and growing rapidly. Centrica alone has sold nearly 2.5 million smart thermostats and other types of sensors under its Hive brand. Harnessing the information from these devices with AI supports us in serving the changing needs of our customers. Our Hive thermostats give our residential energy customers complete control over their energy use. Our Hive Link service combines information from sensors and machine learning to understand the customers’ routine over a two-week period and can alert the person caring for them if that routine is broken. Social care is one of the biggest challenges facing economies and we believe this is a good example of where technology has a significant role to play. Our BoilerIQ system predicts boiler failures before they happen and help schedule engineer visits to prevent them, so no one wakes up to a cold house. In the near future, as electric cars become more commonplace, scheduling charging and alternative uses of the car battery will be managed by AI, creating opportunities in demand side response and optimising how we use the power grid. Meanwhile, for businesses, the use of energy data is already well-beyond energy procurement. For our business customers, we have an Energy Insight service that provides intelligence driven by data from self-powered sensors to optimise site performance, deal with potential equipment failures before they happen, and reduce energy inefficiencies.
Beyond developing new products and solutions for our customers, AI also enabled us to innovate and be more efficient at what we do internally to support our commercial propositions. One of our subsidiaries, Io-Tahoe, which provides solutions that enable organisations to extract meaningful relationships from their own data via machine learning was born out of our own internal needs.
In my division, forecasting prices and flows play a crucial role as we develop Centrica’s view on energy markets and global macroeconomics for strategic and tactical insight. Machine learning is enabling us to measure, understand, and improve forecasts, not only our own but also those we get from external organisations. When we first applied machine learning techniques to our forecast data we were astonished with the results. We discovered that we had strengths in forecasting certain indicators that we were not aware of and that each forecast had a distinct ‘shelf life’ which helped optimise the time spent on quality control before publication. Most importantly, machine learning helped us improve the forecast performance dramatically by learning from past ‘mistakes’ in a systematic and adaptive manner.
Another application we implemented is in the automation of report writing. One of our tasks is to monitor market developments, identify key drivers, and communicate this to other functions for reporting purposes. Historically, this consumed a considerable amount of time. Typically, multiple analysts had to monitor different markets and write up their analyses in their own style, which then required editorial work to consolidate. Using supervised learning as well as other analytical techniques we managed to automate the analysis and produce reports in plain language that describe what happened in a certain market and why. This frees up time for our analysts to focus on complex questions that bots cannot yet tackle.
The examples I mentioned are just the tip of the iceberg in terms of the potential applications of AI and machine learning in making organisations more competitive and more efficient. The key to success will be early adoption and effective use of this technology to train it on data and ‘learn’ as the real value to the organisation is in what is learned, rather than the algorithms which are already well-developed and widely available open source