The research was conducted using the hybrid evolutionary algorithm which has been applied previously for
the predictive modelling of cyanobacteria growth in a variety of
lakes and rivers worldwide. This algorithm is designed
to evolve over time to improve the fitness between model results
and observations by combining genetic programming (GP) for optimizing
the model structure and differential evolution (DE) for
optimizing model parameters. GP is an evolutionary
algorithm, in which the genetic population consists of computer
programs of different sizes and shapes. Parse
trees represent computer programs. These programs are subsequently
evaluated by means of “fitness” function. Fitter programs
are selected for recombination by using arithmetic and logic operators
(such as crossover, mutation and reproduction) in order to
create the next generation. This step is iterated for continual generations
until the termination criterion has been satisfied. Differential evolution is an evolutionary algorithm designed
for parameter optimization, which extracts
differential information (i.e., distance and direction to global optimum)
from the current population of solutions to guide the search
for the global optimum. It can outcompete the other optimization
methods in convergence speed and robustness. By comparison
with conventional optimization algorithms, the DE could make it a
self-organizing scheme, and does not depend on additional probability
distributions.