//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>
The electrical grid is an amazing feat of human engineering due to the sheer number of power sources, transmission systems, vast distances crossed by cables, and the number of devices connected to the grid. To make the feat more impressive, the entire grid must be carefully maintained to ensure that as soon as someone turns on a light—any day or night of the year—it will shine. As such, engineers face many challenges requiring unique solutions that can be monitored from a central system.
To help network operators manage large, complex networks, tools that predict network usage not only provide operators with the ability to make long-term plans but can also reduce the amount of energy being wasted. However, due to uncertainties in traditional models, energy networks typically must have some amount of reserve power that can handle unexpected demand.
Thus, it is not uncommon to see as much as 20% of grid energy goes unused simply because of the inaccuracy of deterministic models. Furthermore, models used by network operators can quickly become out of date and/or contain incorrect information (such as improperly mapped network lines), which amplifies these uncertainties.
What are digital twins?
A digital twin is a digital representation of a physical system that allows engineers to test and monitor the physical system without actually interacting with it. While this may sound like a simulation, digital twins differ in that they do not represent generic models of a system; they are tied to a specific system in real life. Many of the complex digital twins in the electrical grid are a combination of physics-based models and artificial-intelligence models so that they can best represent real-life situations.
Digital twins connect to their physical counterparts via sensors and data links, and they attempt to recreate the system they are connected to with the use of AI. As time progresses, their accuracy grows. This enables predictive capabilities, such as routinely scheduled maintenance.
The use of real-time data and constant learning also makes digital twins a closed loop, which allows them to improve without human interference. Furthermore, these digital twins allow engineers to play around with systems to understand how they will behave. That’s particularly useful for education.
Overcoming challenges with data entry
One particular problem that power networks can face is incorrectly logged data. For example, a particular building will have a meter, and this meter will be connected to a transformer, but if the serial number on the meter is incorrectly logged or the connection diagram is incorrect, there is no way for traditional systems to identify this incorrect data. Worse, this incorrect data can cause errors to ripple through models, resulting in inaccuracies that are impossible to identify.
However, the use of AI and domain understanding can create intelligent data-logging systems that can spot information that doesn’t make sense. For example, an incorrectly entered serial number that results in a cable jumping over a river to connect a building to its local transformer would be spotted immediately. Digital twins cannot function with these inaccuracies, thus forcing engineers to focus on data-entry challenges.
Recognizing unseen relationships
A major advantage that AI possesses is the ability to recognize patterns between datasets that humans would see as unrelated. This makes AI extremely sensitive to minute changes in noise—and able to see patterns even through noise. In many cases, the output of the electrical-grid power-system models becomes the input to the AI models that detect adverse situations early and accurately.
In the case of digital twins, engineers can also choose a list of variables in the electrical power models that they believe affect power networks, allow the digital twin to start learning, and then observe how each input variable affects the output. This is the power of the digital twin as it integrates.
Exactly how AI arrives at conclusions has historically been a mystery due to the use of hidden layers. But recent advances in explanatory AI now allow engineers to interrogate each layer and observe exactly how different variables relate to each other. (For example, how do Variables A and B affect each other, and then how does this result affect Variable C?)
This pattern-recognition ability gives engineers insight into how different factors affect electrical grids while also helping to produce an accurate digital representation. Furthermore, the ability to more accurately represent power grids also allows engineers to reduce the amount of energy wasted through redundant power systems, as demand can be better determined ahead of time.
Benefiting from early-warning systems
The ability to accurately represent power grids while also being able to accept numerous data inputs, such as weather and global markets, allows network operators to identify and mitigate against potential risks.
At the same time, the ability of AI to recognize anomalies also allows network operators to investigate potential issues with the network and provide maintenance before larger systems fail. Replacing a dodgy cable is far cheaper than a blown substation.
When considering that reliability is by far the most important factor in a power grid, having an early-warning system can help provide energy stability, reduce energy costs, and improve network performance.
AI systems can also use physics-based grid simulators as part of a reinforcement-learning system to identify unique and optimal solutions that would not be possible with a conventional power-systems approach.
Getting away from deterministic models
Electrical grids are massively complex systems consisting of a vast amount of infrastructure that must respond to demand changes in real time. Trying to use deterministic models to represent future electrical grids will only result in more energy being wasted, as large amounts of redundancy will be needed to cover the increasing disparity between real networks and their models.
Digital twins that leverage AI to learn how various factors affect grid performance will provide network operators with a model that provides an accurate representation—one that will continue to improve over time.
Furthermore, engineers will be able to dissect learning models and understand how and why different variables correlate to each other. And the ability to have predictive maintenance combined with anomaly detection will provide network operators the opportunity to make better decisions, thus realizing a network that will always turn on the light no matter when the switch is flipped.