Chapter 4 . Applications of Evolutionary Computation
Microcystis = 0 cells
ELSE
IF T >= 5.67 °C AND T <= 15.7 °C THEN
Microcystis = 3,000 cells
ELSE
Microcystis = 100,000 cells.
61
This model successfully predicted both the tImmg and magnitude of
Microcystis on unseen data from the same lake environment and was assessed as
having a physically plausible interpretation.
A second example of role discovery involved evolving a set of roles to predict
the habitat density of a spatially-distributed marsupial (Whigharn 2000). This
work used a grammar-based GP system to show that spatially-explicit roles could
be evolved that predicted the location and density of a marsupial, based on
surveyed information and a set of spatial data. The resulting model was used to
question the currently accepted horne range for these marsupials. The evolved
roles allowed expressions to be constructed that could not be easily formed using
other techniques (McKay et al. 1997).
Several other role-based systems, based on using aGA, have been created.
GARP (Stockwell and Peters 1999) allowed the explicit integration of
geographical data and a GA to discover role sets that expressed spatial knowledge.
This system has been used to automate the predictive spatial modelling of the
distribution of species of plants and animals. BEAGLE (Fox et al. 1994), aGA for
constrocting logical expressions, was used to generate roles that could predict
presence/absence of the duck species Aythya Jerina on gravel pit lakes in southern
Britain. GAFFER (Jeffers 1999) allowed the discovery of roles for numerical
prediction and classification, and was designed to apply to real data sets where
litde background knowledge of relationships between variables was available.
GAFFER was applied to discover roles that characterised habitat features
determining the abundance of individual aquatic species.
4.4.5
Modelling Individual and Cooperative Behaviour
Modelling and understanding individual behaviour in an environment is of
fundamental ecological interest. Work has been done using GP to evolve foraging
strategies based on a model of the Anolis lizard (Koza et al. 1992). This work
aimed at understanding what was an optimal foraging strategy, based on the
abundance of insects, the velocity of the lizard, and the spatial relationship
between the lizard and insects. The work demonstrated that for a set of insect
abundance, lizard velocities and spatial placement the system evolved a sequence
of progressively improved strategies. The models were expressed as a
mathematical function that included decision points, based on an if-then function,
Microcystis = 0 cells
ELSE
IF T >= 5.67 °C AND T <= 15.7 °C THEN
Microcystis = 3,000 cells
ELSE
Microcystis = 100,000 cells.
61
This model successfully predicted both the tImmg and magnitude of
Microcystis on unseen data from the same lake environment and was assessed as
having a physically plausible interpretation.
A second example of role discovery involved evolving a set of roles to predict
the habitat density of a spatially-distributed marsupial (Whigharn 2000). This
work used a grammar-based GP system to show that spatially-explicit roles could
be evolved that predicted the location and density of a marsupial, based on
surveyed information and a set of spatial data. The resulting model was used to
question the currently accepted horne range for these marsupials. The evolved
roles allowed expressions to be constructed that could not be easily formed using
other techniques (McKay et al. 1997).
Several other role-based systems, based on using aGA, have been created.
GARP (Stockwell and Peters 1999) allowed the explicit integration of
geographical data and a GA to discover role sets that expressed spatial knowledge.
This system has been used to automate the predictive spatial modelling of the
distribution of species of plants and animals. BEAGLE (Fox et al. 1994), aGA for
constrocting logical expressions, was used to generate roles that could predict
presence/absence of the duck species Aythya Jerina on gravel pit lakes in southern
Britain. GAFFER (Jeffers 1999) allowed the discovery of roles for numerical
prediction and classification, and was designed to apply to real data sets where
litde background knowledge of relationships between variables was available.
GAFFER was applied to discover roles that characterised habitat features
determining the abundance of individual aquatic species.
4.4.5
Modelling Individual and Cooperative Behaviour
Modelling and understanding individual behaviour in an environment is of
fundamental ecological interest. Work has been done using GP to evolve foraging
strategies based on a model of the Anolis lizard (Koza et al. 1992). This work
aimed at understanding what was an optimal foraging strategy, based on the
abundance of insects, the velocity of the lizard, and the spatial relationship
between the lizard and insects. The work demonstrated that for a set of insect
abundance, lizard velocities and spatial placement the system evolved a sequence
of progressively improved strategies. The models were expressed as a
mathematical function that included decision points, based on an if-then function,