Building Trustable AI
Empowering the Electrical Grid
Built on top of an aging infrastructure and facing a declining workforce, the electrical grid is also increasing in complexity. Demand is escalating as data centers and consumer usage increase, in some cases, dramatically, taxing the grid’s capabilities. In addition, with other sources of electric power such as solar and wind being integrated into the grid, the technical challenges of managing this heterogeneous grid are also increasing.
ThinkLabs AI Copilot is an AI-first grid management application that assists with both grid planning and operation. Planning can be either short term or long term, while the real-time command and control center is the basis of operations. ThinkLabs AI Copilot provides a digital twin that models grid power flow behavior and can run agentic workflows. A physics-informed AI digital twin trained on large, synthetic datasets can perform more rapidly and better respond to unpredictable real-time grid scenarios than traditional engineering models.
ThinkLabs AI Copilot helps with decision making by running online real-time simulations and providing timely analyses. “The digital twin replicates the base power flow,” said Josh Wong, CEO of ThinkLabs AI. “It is science-driven and is faster, cheaper, and more data resilient than the traditional grid simulations.” ThinkLabs AI Copilot conforms to business and technical standards operating for utilities and is informed by human experience.
The agentic workflows provide a recommendation for a response to a situation such as lack of capacity. Human operators can accept the recommendation of the AI system or override it. “The AI is trained on millions of scenarios,” explained Wong, “but the human operator can incorporate additional details such as knowledge about impending weather and can use their experience to modify the advice given by ThinkLabs AI Copilot.” Existing grid management systems are not AI-native and are deeply rooted in legacy data systems, Wong pointed out. “ThinkLabs AI is not attempting to replace them but can supercharge them to be more proactive.”
The basis for trust in this AI-based application is its adherence to first-principles simulations, including building and testing the model and analyzing results. In addition, the standards for power flow, such as strict voltage limits, are built into the model. “Utilities are notoriously conservative,” observed Wong, “But the pain points are so high now that solutions are needed. Forecasting capability is widespread, but having an AI-enabled power flow with large scenario analyses is unique.”
The system is also designed to be verifiable. “If an engineer wants to run a physics audit, they can compare our results to a traditional engineering power flow,” said Wong. “This makes the results explainable, even on huge datasets, and allows the users to validate the recommendations.” An additional benefit is that engineers and planners no longer need to spend time running ad hoc analyses; with generative solutions from AI, they can take on a more strategic role that addresses long-term goals.
AI continues to make inroads into virtually every area of business and science. With careful planning and keeping humans in the loop, it’s possible to create AI systems that are accurate, improve outcomes, leverage knowledge effectively, and maintain security. This contributes to building the trust necessary to continue forward with AI projects.