How Can Machine Learning Help Keep Electricity Networks Secure?
Panagiotis Papadopoulos, Associate Professor and UKRI Future Leaders Fellow at the University of Manchester
Electricity networks and the changing landscape
Electrical power systems worldwide are going through unprecedented changes motivated to a large extent by decarbonisation targets but also for various other technical, economic and social reasons. These changes affect most aspects of power system operation and planning: starting from generating electricity from primary energy sources, transmitting it over long distances, and finally distributing to consumers to be used.
The generation mix is changing drastically with more renewable generators being connected (e.g. onshore and offshore wind and solar) and conventional thermal units (e.g. coal) being decommissioned. The way we use energy is also changing in an effort to reduce carbon emissions in other sectors, by moving away from fossil fuels and electrifying parts of transport through Electric Vehicles (EVs), and heating through heat pumps. At the same time, due to the complex nature of power systems, being large complex nonlinear dynamical systems, electricity generation and demand needs to be continuously balanced, for the system to be stable. And also, we all expect very high reliability to meet our energy needs on demand.
In the UK, there are ambitious targets set to achieve net zero and some of the Future Energy scenarios of the National Energy System Operator (NESO) involve 50GW of offshore wind by 2030 and see up to 600,000 electric heat pumps/year by 2028 as well as the ban of petrol/diesel cars by 2035 with EVs being an alternative.
In a nutshell, the above changes cause increasing uncertainty and complexity in power system operation and planning. Increasing uncertainty largely comes due to the intermittent nature of renewable generation but also due to social dimensions related to how people use energy. For example, how will people (expect to) use and charge their EVs? In addition, market structures can also drive such behaviours. On the other hand, increasing complexity comes mainly due to the very different behaviour and technologies that these new devices use (interfaced through power electronics), governed to a large extent by complex control.
We also have more available measurement devices in modern power systems that can monitor the system and the demand in timescales from milliseconds to minutes/hours, providing valuable data. However, measurements can still be scarce, and new algorithms are needed to extract meaningful information from data, due to the complexity of the system. This is where AI can be helpful.
Importance of maintaining security
In this new paradigm, it is becoming increasingly important to ensure power system security, especially when considering the changing system dynamics which can introduce unknown modes of failure. Not only because power systems are part of critical infrastructure with provision of several other services relying on them (e.g. communications, water and gas supply, transport, etc.), but also due to increasing reliance on electricity as we are decarbonising energy (e.g. transportation and heating). Failing to do that can lead to power disruptions or in the worst-case blackouts, which can lead to huge economic costs, significant disruption to society and even loss of human life.
What it takes to run a power system
During this energy transition it is crucial to keep the pace of integrating low carbon technologies at the scale required but also to ensure we utilise them in a stable manner in operational timescales.
Technically speaking, planning (from days to years) and operation (from milliseconds to hours) of power systems requires solving a number of challenging problems related to optimisation, forecasting, and security assessment (i.e. checking against a number of external disturbances to ensure the system will remain stable), among others.
We need to forecast wind and solar generation as well as the demand in advance, to ensure adequate generation is dispatched in real time in a cost-efficient/optimal manner. And all this while ensuring the stability/security of the system.
How AI and data-driven methods can help
To this end, advances in Artificial Intelligence (AI) and more specifically Machine Learning (ML) domain are opening up avenues and possibilities for improving the above-mentioned aspects, including optimisation and security assessment. AI can also be particularly useful in utilising data and offer fast computation capabilities when dealing with large complex systems.
Notably AI is considered a state-of-the-art tool for forecasting of renewable generation which is crucial in fully utilising low carbon technologies and ensure the system remains balanced. It is also being considered as helpful for other applications such as predictive maintenance, visual inspection, reporting assistance, alarm management, and so on.
Finally, AI can be helpful in utilising data to improve situational awareness, provide decision support to system operators, and also enable automation, especially for cases when the timescales or complexity does not allow human interventions, e.g. in the millisecond timescales or when there is need to control millions of devices.
All these aspects are crucial in enabling the implementation of low/zero carbon technologies at the pace needed, as well as their efficient and secure utilisation in operational times.
Moving forward with AI for electricity networks
Due to the nature of the application to critical infrastructure, it is important to build trust in AI technologies and ensure implementation with proper failsafe mechanisms when considering implementations to electrical power systems. Data availability, quality and considerations around the need for exchange are also important to be kept in mind. Having said that, there is a very diverse range of potential applications around power systems which might have very different requirements in the above aspects, some stricter, some less so.