A computer-implemented method, according to one embodiment, includes: setting a target power demand corresponding to a consumer, and performing a process. The process includes: determining an actual power demand presented to the utility by the consumer based on a reward table, determining a current error, determining whether the actual power demand is adjustable in a direction that reduces the current error, reducing the current error by adjusting the actual power demand in response to determining that the actual power demand is adjustable in the direction that reduces the current error, and modifying the target power demand in response to determining that the actual power demand is not adjustable in the direction that reduces the current error. Determining the actual power demand includes measuring the power demand over a period of time in accordance with a process for measuring stability of power demand received from the utility.