Abstract: A disclosed system for dynamically controlling energy performance operations in a facility includes: energy sources, energy sources controllers, and a centralized controller. The centralized controller can: receive signals indicating performance of the energy sources in real-time, retrieve at least one model that is iteratively trained using machine learning techniques and training data including (i) at least a portion of the received signals and (ii) decisions made by the centralized controller, provide at least a portion of the received signals as input to the model, and receive, as output from the model, control operations for one or more of the energy sources for a predetermined period of time, and return the control operations to respective controllers of the one or more energy sources for execution during the predetermined period of time.