Integration, prediction and protection: AI's application in smart grids

With AI’s ability to unlock new efficiencies and manage complex energy networks, the focus now turns to transformative applications, emerging risks, and the technological leaps shaping the grid of the future.

While AI can unlock unprecedented predictive power, it also risks putting networks in the crosshairs of attack by bad actors.

Integration of Distributed Energy Resources (DERs)

One of AI’s most promising applications is its role in integrating Distributed Energy Resources (DERs), such as rooftop solar panels and home batteries into the grid.

By analysing data from countless small-scale energy producers, AI systems optimise energy distribution and consumption, contributing to grid stability and efficiency and easing the strain on grid infrastructure.

Advances in AI and wireless networks allow systems of these devices to be integrated in what is called the ‘Internet of Energy’ (IoE) - a digital ecosystem where smart devices, energy resources, and grids are interconnected. The IoE allows energy operators to efficiently coordinate energy generation, distribution, and consumption, ensuring optimal performance across the entire system.

The grid’s complexity makes it challenging to model devices and design control actions using traditional mathematical equations. What’s more, factors like consumer behaviour, electrical faults, weather changes, and tariffs do not follow clean mathematical patterns. AI can recognise patterns that cannot be captured by these equations and guide the design of control and operation.

DER-supplied electricity - dependent on variables like the weather - benefits from the predictive power to forecast impacts and identify periods of peak demand.

Associate Professor Hemanshu Pota specialises in AI utilisation, renewable energy integration and smart grid stability at the University of New South Wales (UNSW).

“AI facilitates integration by optimising the use of each device, [such as] batteries,” explains Pota. “Many parameters that matter for optimal expression are hard to model mathematically but AI can learn it and then help with optimal operation.”

AI also predicts intermittent power from individual generation assets and collections of assets like Virtual Power Plants (VPP), and analyses this against expected demand to balance supply and reduce the risk of power loss.

It can also be used to assess locations for renewables assets in a similar way, simulating demand and increasingly complex supply combinations to better support decision-making.

The Emergency Backstop Mechanism is another example of the dynamic control required by network service providers, which AI can enhance.

According to Struan Buchanan, national power, utilities and renewables consulting lead partner at professional services consulting firm Deloitte: “Existing grid infrastructure has various constraints that need to be managed from an engineering perspective, and AI has a significant role in dynamic load forecasting that enable operators and network service providers to forecast when limits may be exceeded for which controls need to be put in place, and for real-time load management.”

Advanced demand forecasting: Balancing supply and demand

Accurately predicting energy demand is critical for a reliable energy supply.

AI machine learning algorithms analyse historical and real-time data to forecast demand patterns accurately, allowing for better load management and resource allocation, ensuring a balanced supply and demand dynamic.

AI’s ability to handle dynamic data ensures the grid remains balanced even as consumer behaviours, such as EV use, shift energy consumption patterns. With smart meters (AMI 2.0) becoming more common, AI’s role in real-time balancing and forecasting is set to become even more critical.

According to Buchanan, AI is increasingly being leveraged to support reliable delivery of electricity to consumers. It is particularly useful for monitoring demand and consumption, managing the network, fault rectification and outage management. It’s also helpful for coordinating increasingly fragmented and dispersed distributed and consumer energy resources based on weather and load.

While there is strong uptake of AI in grid management, it is still relatively piecemeal, with significant variation across the market in how it is used.

Buchanan says it is critical that AI is trialled for specific use cases, as this will build buy-in and support from the industry.

“It often requires iteration to prove out the value, and finding the opportunities that are top of leadership’s mind as well as tracking that value, is important to building momentum and permission to invest.”

AI will play an increasingly key role in real-time balancing and forecasting (including simulation), and with the amount of information to be analysed rapidly growing, ‘grid-edge’ devices require significant computing power. These smart agents will enable significant scaling in network management and will be important in managing the complexity of communications between different devices and protocols.

Predictive maintenance and asset management

AI’s capability for predictive maintenance is another game-changer. Machine learning algorithms can predict equipment failures, allowing timely interventions that minimise downtime and reduce maintenance costs.

“The scale of maintenance and asset planning is significant for renewable asset operators and network managers,” Buchanan said.

Vegetation management is one of the most significant operating costs for network service providers.

“We see businesses seeking to leverage AI in computer vision from drones and satellites to more efficiently identify where and when work is required,” Buchanan said.

AI is also used in the autonomous and remote piloting of drones, and to classify and inspect assets.

Cybersecurity: Protecting a digital grid

With the digitisation of energy systems comes heightened cybersecurity risks.

The energy transition has led to a fragmented ecosystem with more devices to control, more data sources and more volatility to manage. As Buchanan notes, demand and supply can both change very quickly.

While AI can play a role in managing this new complexity, it also poses its own threat.

Deloitte has identified a landscape of new threats for AI that need to be managed by market participants.

In the energy sector, these threats include:

• Input injection: Attackers may manipulate data sources used by AI. Energy companies are often reliant on external data sources like weather forecasting, so malicious injection or interference with weather data is a concern.
• Excessive agency: AI systems can be used for a range of purposes, from informing human decisions to actively making decisions. As the energy system becomes more complex and volatile, AI is making more decisions, including workforce changes, which may mean a company becomes reliant on AI and can’t revert back to a manual process.
• Model poisoning: Training data may also be attacked in a way that influences the model. This is still an emerging area, but one to consider when thinking about how the training process occurs.

Pota is more optimistic: “The power industry is conservative and in general, very little data is transmitted for operational reasons.”

Smart meter data for billing purposes and long-term planning may be compromised - but these records are no different to any other records kept by any organisation.

“There is no more risk to data in power systems than for any other system,” Pota argued. “The biggest risk is people using AI who do not understand AI.”