Predicted environment conditions are received for time steps within a prediction horizon N. Information is accessed regarding expected system responses generated by a system model, each expected system response related to a different combination of control parameters and starting environment conditions applied to the model.An optimal response is designated for each combination of starting environment conditions for each time step from time step N−1 to the current time step based in part on the expected system response and the predicted environment conditions, each optimal response in a time step based at least in part on an optimal response determined for a later time step.For the current time step, for starting environment conditions substantially similar to the current environment conditions, the combination of control parameters that results in the system response designated as the optimal response is determined from the expected system responses and is applied to the system.