Tech Insights

‘Brain’ Power: Artificial Intelligence Streamlines Grid Operations—Part 1

May 27, 2024 by Liam Critchley

Artificial intelligence can modernize the power grid but comes with risks. This article, the first in a three-part series, examines the benefits and drawbacks of an AI-enabled smart grid.

Artificial intelligence (AI) advancements are helping build a safer, cleaner, more efficient energy sector by enhancing smart grid operations and key components such as smart meters, fault detection, load balancing, and predictive models. However, integrating digital technologies can cause harm if not properly implemented or secured.


Concept of AI-enhanced smart grid.

Concept of AI-enhanced smart grid. Image used courtesy of Adobe Stock


Infrastructure Operational Awareness 

AI helps grid operators identify key information and trends to provide a clearer picture of their systems. Additionally, AI can detect system changes, even with incomplete or limited data, to provide operators with better situational and operational awareness.

AI can simultaneously measure data characteristics in real time and easily authenticate data sources. Authentication of smart grid data helps provide better grid awareness and resilience. It enables operators to identify, locate, and evaluate any unexpected events on the grid, such as grid inertia, event magnitude, and system breakup. AI provides a greater degree of control over the grid.

One real-world system providing awareness is the FNET/GridEye. Operated by the Power Information Technology Laboratory at the University of Tennessee, it is an AI-enabled wide-area frequency measurement network used to monitor grids worldwide. 


Video used courtesy of University of Tennessee/Oak Ridge National Laboratory


Using a frequency monitoring network and a frequency disturbance recorder sensor, FNET provides voltage magnitude, phase angle, frequency, and power quality measurements from power grids. This data is then sent to the data center via the Internet. The data can then be used to identify any localized events and provide a better awareness of grid functions. The system can also predict blackouts and grid instability and determine the impact of integrating renewable energy sources into the grid.


Physics-Based Modeling for Unexpected Weather Events

Physics-based models have been useful in many technical areas for solving complex problems. They categorize battery degradation mechanisms and failure modes in the wider energy space. However, in the power grid, the primary analysis area for physics-based modeling is predicting unexpected events, such as unforeseen and highly impactful weather systems affecting grid operation and stability.

Models focus on predicting impacts on energy resource distribution during or after a weather event. AI models can be used after a weather event to determine the amount of stored energy resources within the grid to distribute power to affected areas. 

Predicting events before they occur may be more effective than determining how to proceed post-event. Preventive measures based on known weather conditions can help protect the grid by allowing post-event measures to be tailored well in advance based on the most likely outcomes.

AI can improve weather prediction and the grid’s preparedness in extreme weather situations by determining which variable resources and loads to distribute based on the storm’s severity in different locales. Predicting weather patterns and adapting to weather events improves grid resilience.

Four major commercial weather models use machine learning. These models can forecast cyclones, hurricanes, and typhoons and determine the shape of the warm regions, the cloud head positioning, and the warm conveyor belt jet location. Machine learning helps grid operators understand the large-scale dynamics and storm development of hazardous weather systems to prepare for potential outages or disruptions based on the storm’s probable location and intensity.


Comparing weather prediction AI models.

Comparing weather prediction AI models. Image used courtesy of Nature


NVIDIA has developed two models, FourCastNet and FourCastNet2. FourCastNet uses Fourier Neural Operators alongside a vision transformer architecture, whereas FourCastNet2 is an extension of this model and uses spherical Fourier Neural Operators. Huawei’s Pangu-Weather model uses a 3D Earth-specific transformer with hierarchical temporal aggregation. The fourth model from Google DeepMind is called GraphCast. This model is based on graph neural networks. All commercial models have efficiently predicted weather patterns and can produce 10-day forecasts in just a few minutes.


Video used courtesy of Google DeepMind


Integrating AI into physics-based models increases their internal efficiency and helps create more efficient simulations and sophisticated weather models. The gained efficiencies also make the models more accessible, which promotes the operators’ better-informed decision-making.


Predictive Grid Maintenance

AI-enabled predictive maintenance can provide early warning signs that grid infrastructure will require downtime for maintenance, is about to fail, or is degrading and might need replacement.

Integrating renewable energy, coupled with increasing energy demands from users (including rapid charging of electric vehicles), increases load fluctuations and stresses the grid’s central components, making it more susceptible to blackouts. For example, one component under stress is switchgear, which distributes energy to microgrids. Operators can use switchgear for up to 40 years, but the equipment needs proper maintenance. Using predictive maintenance, AI can analyze switchgear and other components before they fail, making maintenance easier, less time-consuming, and less costly (with reduced downtimes).

Machine learning models can easily predict patterns to accurately determine the components’ lifetimes and when maintenance windows should be scheduled to avoid unexpected downtime.


Active Control of Energy Systems

Finally, AI gives human operators much better control over smart grid infrastructure during normal operation. In a hybrid approach, AI algorithms can control the energy system operation at machine-level speeds, but human operators dictate the process. AI could make energy operations more efficient and provide real-time control of energy infrastructure across the grid. 

While AI-assisted, human-operated processes give humans more control—and the ratio of human input vs. AI input can be changed to fit needs—there’s also the possibility of full AI autonomy, where AI makes all the decisions.


AI in grid operations 

AI in grid operations. Image used courtesy of Idaho National Laboratory


These AI-human operations can accommodate many external factors. One example is integrating renewable energy sources, which change the power grids’ frequencies. Adding renewables introduces inertial changes because it integrates DC-based energy into the AC-based grid. 

Monitoring and controlling these changes can provide strategies to change the grid's frequency. Using AI offers more control to the operators because neural networks can develop models predicting the frequency control performance for a given set of parameters, such as system inertia and system response.

The energy sector is unlikely to implement completely autonomous AI systems for actively controlling energy systems in the near future due to the potential for errors. Future models may be more robust and reliable. However, AI-assisted but human-driven support systems are a viable alternative for giving extra control to grid operators.


The Future of AI

AI architectures are improving smart grid operations and efficiency in several areas. Some processes are completely automated, while others use AI as a tool to make human decisions easier. Several commercial systems are already in place, but opportunities remain for wider adoption across other grid areas.

AI is impacting the grid in many areas during its journey toward digitalization. This article is the first in a three-part series on using AI in smart grid operations. The next article will examine AI’s role in cybersecurity and evaluate the risks within digitally connected smart grids.