Parameter
Description
Range
Recommended value
p
No. of complexes
-
m
No. of points in a complex
2n + 1
q
No. of points in a sub-complex
n + 1
b
No. of evolution steps taken by each complex before shuffling
2n + 1
The algorithmic parameters of the SCE algorithm, their feasible range and recommended values are shown in Table 1.2.
If one complex is chosen in SCE and the number of points in the complex as well as the sub-complex are set equal to n + 1, the local search simplex method is obtained as a special case.
No. of complexes
Number of complexes p applied in the SCE algorithm. This is the most important parameter of the SCE algorithm. Sensitivity tests show that the dimensionality of the optimisation problem (No. of calibration parameters) is the primary factor determining the proper choice of p. In general, the larger value of p is chosen the higher the probability of converging into the global optimum but at the expense of a larger number of model simulations (the number of model simulations is virtually proportional to p), and vice versa. One should choose p to balance the trade-off between the robustness of the algorithm and the computing time.
No. of points in complex
Number of points in a complex. A recommended value for this parameter is 2n + 1 where n is the number of optimisation parameters.
No. of points in sub-complex
Number of points in a sub-complex. A recommended value for this parameter is n + 1 where n is the number of optimisation parameters.
No. of evaluation steps by each complex
Number of evaluation steps taken by each complex before complex shuffling. A recommended value for this parameter is 2n + 1 where n is the number of optimisation parameters.
Apply default parameters
If the Apply default parameters option is selected, the default algorithmic parameters in Table 1.2 are used.
