-The optimisation target is given by the squared
difference between experimental and simulation data,
relatively to the 3 conversion curves (CO, HC and H2)
-Once the LO_optim.bwf model is set and the target
functions are built, the integral value Y_MEAN_INT_DX
for all the three target curves is defined as output variable
(SD_CO, SD_HC, SD_H2)
-The global optimisation objective is given by the
minimisation of global_error = (SD_CO+SD_HC+SD_H2)
-A different approach has been considered:
-After 718 designs calculated by NSGAII (global_error=50),
multi-objective scheduler MOGT is run (minimise SD_CO,
SD_HC, SD_H2 as 3 different objectives)
-MOGT starts from a good configuration (#718) and sets 3
constraints on the values of the 3 objectives, that should
be less than the ones of the starting configuration
-A last approach has been considered:
-Simplex mono-objective algorithm is used from random
DOE (particular efficient in problems that are not highly
non-linear)
-An example of integration of AVL Boost with modeFrontier
has been shown
-The test case was relative to the calibration of 6 kinetic
parameters in a catalytic conversion reaction
-Different optimisation algorithms has been used to minimise
the difference between the experimental and the simulated
conversion curves
No comments:
Post a Comment