Our project, Automated Synthesis for Trustworthy Autonomous Utility Services (ASTrA), aimed to create a new way of thinking about complex descision making in large scale smart cities. We wanted to help people and cities use energy and other utilities more efficiently and affordably while making the most of renewable sources like solar and wind power. Our goal was to develop tools and methods to design smart, reliable, and fair energy systems that can adapt to the unpredictable nature of renewable energy.

In particualr, we focused on creating decentralized control schemes that let different parts of a smart city's energy system—like solar panels, battery storage, and traditional power sources—work together seamlessly. Instead of a single central authority, our approach uses a system that allows each part to make its own decisions based on a shared goal, like increasing the use of renewable energy. This method is highly scalable, meaning it can work for both small towns and large metropolitan areas which can help achieve ambitious goals like zero-net energy for entire cities.

Key Accomplishments and Innovations: 

1. Advanced Tools for Complex Systems: A significant part of our work involved developing sophisticated computational tools. We created a novel solver, PolyARBerNN, that can handle complex mathematical problems involving polynomial equations. This solver is essential for designing and verifying safety-critical systems. By using new techniques inspired by neural networks, we made this solver significantly faster than previous methods, achieving a speedup of up to 100 times. This allows us to more efficiently ensure that our descision making systems are both effective and safe.

2. Enhancing Safety and Reliability: Ensuring that descision making systems are safe is our top priority. We developed several new methods to verify the reliability of systems controlled by neural networks.

- We created a new framework, NNSynth, that uses machine learning to speed up the process of designing descision making for complex systems. This approach not only makes the design process faster but also allows us to incorporate human preferences, which is crucial for systems that interact with people.

- We also developed a tool, BERN-NN-IBF, that performs a rigorous check on neural network controllers to ensure they behave predictably and safely. This tool is more accurate and efficient than existing methods, helping to build trust in these new technologies.

- We developed an innovative method to make sure that descision making systems with multiple sensors operating on the edge (e.g., smart home devices, IoT) don't compromise safety for energy savings. Our framework, which uses a "controller shield" as a low-power safety monitor, allows us to optimize energy use without risking the system's safe operation.

3. City-Scale Energy Simulation: To test our ideas, we built a publicly available smart city energy simulator using an open-source tool called EnergyPlus. This simulator allows us to model multiple buildings and their interactions, giving us a realistic testbed for evaluating our algorithms. It lets us see how our new methods would perform in a city-wide setting, helping us refine our approach for real-world applications.

Broader Impacts:

Our research is creating a new generation of computational tools and methods for designing distributed descision making systems. While our project focused on energy management in smart cities, these tools have broader applications in areas like robotic swarms and smart manufacturing. By making these tools publicly available and integrating our findings into workforce development programs, we are helping to train the next generation of STEM students and engineers, preparing them for the challenges of an increasingly automated and sustainable future.

Last Modified: 08/06/2025
Modified by: Yasser Shoukry