Dynax: a differentiable dynamic energy simulator for inverse inference, optimal control and end-to-end learning
Yangyang Fu, Zheng D. O’Neill · Building Simulation Conference proceedings · 2023
This paper presents a differentiable simulator for dynamic energy systems to support gradient-based tasks, such as inverse inference, optimal control and end-to-end learning. The differentiable simulator is built on an open-source auto-differentiation (AD) platform specifically designed for deep learning communities.With the inherent differentiability and parallelization across hardware accelerators, this differentiable simulator can benefit the digital twin era by leveraging modern computation hardware accelerators to perform real-time inference and learning-based control.