Algorithmic parameters

The algorithmic parameters of the PSE algorithm, their feasible range and recommended values are shown in Table 1.3.

Population size

Population size s applied in the PSE algorithm. This parameter is important for the convergence properties of the algorithm. In general, the larger value of s is chosen the higher the probability of converging into the global optimum but at the expense of a larger number of required model evaluations. One should choose s to balance the robustness of the algorithm and the comput­ing time. The proper choice of s depends on the dimensionality of the prob­lem.

No. of points in simplex

Number of points in a simplex. A recommended value for this parameter is n + 1 where n is the number of optimisation parameters.

Reflection step probability

The probability of performing a reflection step of the simplex. In this case the new point is found by reflecting the worst point of the simplex in the centroid of the remaining points. A recommended value for this parameter is in the range 0.70-0.90.

Contraction step probability

The probability of performing a contraction step of the simplex. In this case the new point is found as the mean between the worst point of the simplex and the centroid of the renaming points. A recommended value for this parameter is in the range 0.05-0.20. The sum of the reflection and contraction probabilities should be less than one. The remaining portion (1-p_r-p_c) is assigned a mutation probability with a recommended value in the range 0.01-0.10.