Nimbus

Nimbus: Autonomous AI Prediction for Energy Management

Nimbus is a dynamic microgrid controller used to make island power access reliable, especially when renewable energy sources are under stress. Unlike traditional energy grid systems that wait for stress on power sources and blackouts to trigger, Nimbus proactively predicts when a crisis is approaching and automatically adjusts power consumption, saving communities from blackouts.

We started by analyzing the problem: island microgrids fail because they use fixed thresholds. When the battery hits a low percentage, they shed load suddenly. In reality, severe storms cause power loss to accelerate exponentially—and by the time the threshold triggers, it's too late.

Our solution tracks three metrics in real time: net energy flow (smoothed by exponential moving average), velocity (how fast power is being lost), and acceleration (whether that loss is speeding up). These signals feed a multi-cascade (prioritizing importance of energy consumption) control layer that throttles flexible loads gradually using PD controllers, rather than cutting them off.

The architecture respects a strict hierarchy: hospitals/vital sources get guaranteed power always, desalination plants throttle smoothly, residential areas reduce only in severe crises, and unnecessary loads shed first.

We ran into quite a few problems during the development of Nimbus. The hardest part was tuning the predictive window. Signal too early and you waste energy production rates; too late and you're no different from simple static systems. We also had to balance mathematical derivatives with real-world noise in sensor data. There is also the problem of getting island communities to trust our model.

The simplicity and robustness of the solution. Instead of complex machine learning, this project optimizes the elegance of physics. We're proud that the prioritization highlights important services (hospitals) while stability of energy for everyone else. The math works, buying precious stability for isolated communities.

We're planning on potential targeting of integration within actual island microgrids in the Pacific and Caribbean. Real-world testing will refine our tuning, and we'll explore adaptive priority plans that actually respond to community input. The framework could also generalize to other critical infrastructure, anywhere microgrids face similar fragility. We are also hoping that the insights acquired by the research and creation of this project would be able to be utilized in future developments.

Built With

• analysis
• claude
• infrastructure
• iot
• islands
• microgrids
• prediction
• remote
• renewable
• resilience
• sustainablility