Optimization
With Reward Table Statement
A new statement type is available for Optimization goals: With Reward Table. The statement can be added within a Summation soft constraint set. It allows a piecewise linear reward function to be applied to satisfaction variables within the objective to maximize the satisfaction of the soft constraints. Conceptually this can be thought of as applying a penalty function on violations. For example, it could be used to minimize the sum of squared violations. Applying an appropriate reward function within a Summation objective tends to improve the solution quality by “smoothing out” violations at a smaller magnitude over longer time periods whereas the standard Summation solution tends to concentrate larger violations on a few time periods. The reward function is expressed in a table slot that is provided as an argument to the With Reward Table statement. For more information, see With Reward Table in Optimization.
Improved Percent Satisfaction Reporting
Previously there were cases in which the percent satisfaction reported for a goal could be higher than the actual percent satisfaction from the user’s perspective. This is because RiverWare skips the addition of a constraint to the optimization problem when it will not change the solution. This most often occurs when a lower priority constraint has the same left-hand-side as a higher priority constraint that was already frozen. In these cases, the satisfaction reporting would only report on the constraints that were formally added to the optimization problem at that priority; it would not account for the constraints that were omitted. Thus it was possible for a goal to report 100% satisfaction even though, from the users perspective, there were constraints at that priority that were violated. The percent satisfaction reporting for these cases has been improved in the following manner:
• For Summation soft constraint objectives, RiverWare now reports the average satisfaction as if the skipped constraint(s) were included, i.e., percent satisfaction is defined from the user’s perspective.
• For Maximin and Repeated Maximin soft constraint objectives, if a constraint was skipped and would have been violated at that priority, the satisfaction will be flagged with a tilde (~) to indicate that the satisfaction from the user’s perspective is actually lower than what is being reported. For example a satisfaction reported as ~100% will indicate that there were actually violations at that priority.
This improved satisfaction reporting applies to diagnostic messages, the Priority-oriented Optimization Solution Analysis Tool (POSAT) and optimization analysis CSV files exported from POSAT.
Tailwater Base Value Infeasibility Fixed
In some cases, Tailwater Elevation values calculated in Simulation were causing an infeasibility in Optimization. This occurred for the Opt Linked or Input method when an upstream Power Reservoir Tailwater Base Value was linked to a downstream Pool Elevation, and the upstream reservoir dispatched in Simulation while the downstream reservoir did not dispatch. The resulting Tailwater Base Values were based on earlier time step Tailwater Elevation values and, due to the link, resulted in Pool Elevation values downstream that were inconsistent with the flows. This behavior has been modified for the Opt Linked or Input and Opt Base Value Only methods in the Optimization Tailwater category.
Revised: 07/05/2022