I’d like to address a crucial issue regarding Phase 2 of the CityLearn Challenge: runtime limitations.
In Phase 2, we’re tasked with controlling 9 buildings over three months, including public and private evaluations. However, the 30-minute runtime limit poses challenges, especially for algorithms like Model Predictive Control (MPC). These algorithms need more time, especially on systems with limited CPU resources (e.g., 2 CPUs).
In real-world applications, optimizing nine buildings over three months doesn’t require such tight constraints. The current time limit could limit the diversity of control strategies and discourage the use of valuable algorithms.
I kindly suggest extending the runtime limit to encourage diverse control strategies and inclusive participation. This change would benefit all participants and foster innovation.