Abstract: Variants of the system includes an industrial facility, job scheduling services, and a set of control agents configured to manage power and cooling resources. The method includes determining a current system state, optionally predicting a future system state, determining control setpoints, and controlling the facility based on the setpoints. In a data center implementation, the control agents receive physical state information from infrastructure components and job data from the job scheduling services. Using this information, the control agents predict future power and heat loads associated with scheduled compute jobs. The predictions are generated by a state approximator. Based on the predicted resource demands, the control agents determine control setpoints for cooling and power infrastructure components. The job scheduling services can allocate, delay, or reschedule jobs based on predicted thermal and power constraints.