Geoscience Reference
In-Depth Information
probable to get trapped into local minima. Popular global search methods
are the so-called population-evolution-based search strategies such as the
Shued Complex Evolution
1
(SCE-UA) and the Genetic Algorithm.
4
This
study deals with significant improvement of the robustness and eciency
of the SCE-UA. Thus, a brief introduction to the SCE-UA algorithm is
first presented in the next section. This is followed by the two proposed
enhancements and then the comparison of the enhanced SCE-UA with its
original counterpart on popular test functions is presented.
2. The Shued Complex Evolution
The SCE-UA combines the best features of “multiple complex shu
ing”
and “competitive evolution” based on the downhill simplex search method.
5
The use of multiple complexes and their periodic shuing provide a more
effective exploration of different promising regions of attraction within the
search space. This effective exploration is coupled with evolution of each
simplex, which is provided by the simplex search method. This “competitive
evolution” of the simplexes provides effective exploitation within the search
space. Thus, the SCE-UA achieves a superior balance between exploration
and exploitation as compared to other population-evolution-based search
strategies. For a lucid explanation on the details of the algorithm, the reader
is referred to Ref. 6.
A number of studies have been conducted to compare the SCE-UA and
other global and local search procedures for model calibration.
1
-
3
,
7
These
studies have demonstrated that the SCE-UA method is an effective and
ecient search algorithm.
3. The Proposed Enhancements
Various population-evolution-based search strategies, including the SCE-
UA use a random data generator to generate the initial population. As the
search proceeds, the population converges toward an optimum in one of
the many possible regions of attraction. If this region of attraction does not
contain the global optimum, then the search converges to a local optimum.
The reason for such local minimum convergence could be insuciently large
initial population size or an initial population that is not well spread in the
search space.
Thus, with the aim of having an initial population of points that are
well spread in the search space, a scheme to strategically locate the initial
