[Online Optimization] [Implementation] [Output Report] [User Guide]

After an online optimization, a detailed automatic report is generated by the OptControlCentre. It contains the following performance data of the NLP solver in four plots:

Time Plot

The total time for the optimization (including system identification, optimization and writing results files) and the pure CPU s for the optimization are plotted. These values even contain the time needed for intermediate result plots. As MATLAB is an interpreter (much slower than a pre compiled C++ or FORTRAN program) the total time is significantly higher than the pure CPU seconds shown in the same plot.

Iterations Plot

The number of iterations (blue - number of iterations, black - number of evaluations of the objective, red - number of evaluations of the constraints) for each cycle are plotted together with the optimization status. As green underlying bar represents “Optimal Solution Found”, a red bar represents that the optimization failed (e.g. requested accuracy could not be  achieved). The OptControlCentre tries to recover from a failed optimization, so this could happen in an intermediate state.

Objective value / Constraint violation and KKT error Plot

The value of the objective function is shown for each completed optimization. The constraint violation and the maximal KKT error are plotted in another plot. The later values depend largely on the settings for the desired accuracy by the user.

The report is automatically generated after an online optimization. It can be shown at any time later through the online-commands:

This report shows the online optimization of the Tennessee Eastman Challenge process (Base Case). 300 cycles have been completed. IPOPT (warmstart features were used) needed around 6 CPU seconds for each optimization with over 10,000 optimization variables.