Geography Reference
In-Depth Information
is preserved. However, it may have problems when the fitness values differ very
much. For example, if the best chromosome fitness takes 90 % of the roulette
wheel, chances for the other chromosomes to be selected would be quite low. Rank
selection first sort out chromosomes in the population according to their fitness
values and then compute selection probabilities according to their ranks rather
than fitness values. Hence, a uniform scaling across the population is introduced
in the rank-based selection with no influence of super-individuals. Nonetheless, this
method may lead to slower convergence because the best chromosomes do not differ
much from other ones. In tournament selection,
n
individuals are selected randomly
from the population and compete against each other. Winners with highest fitness
value will be chosen. Although tournament selection has high efficiency because no
fitness scaling or sorting is needed, diversity of the population may be lost.
After the chromosomes are selected, crossover and mutation operations are
conducted to create new chromosomes that potentially have higher fitness values.
Crossover operation can mainly be classified into three categories based on the
above encoding methods, including one-point crossover, two-point crossover and
uniform crossover. As the name implies, if one-point crossover is used, one gene
(value of certain determinant of trip purpose) or one segment of genes (values
of certain determinant) in a chromosome will be exchanged with the gene(s) of
the same locus in another chromosome so as to generate two offspring. In two-
point crossover, parents will exchange two genes (value of two determinants of trip
purpose) or two segments (values of two determinants), so and so forth. Mutation
operation can be conducted by inverting several continuous genes or change value of
one/several gene(s) in the same chromosome in order to produce new chromosome
(Holland
1992
). Similar to the choice of selection approach, either of the above
methods for crossover and mutation operation can be used, which largely depends
on the representation of chromosome.
Through the above procedures, a new population is generated, evaluated, and
modified. After hundreds or thousands of iterations, optimal classification models
can be obtained.
13.4
Field Study and Data Collection
To test the feasibility and validity of the genetic algorithm for identifying activity
type and trip purpose proposed in the previous section, three types of data are
needed. Firstly, GPS tracking data should be collected. Respondents need to be
recruited to carry the GPS data logger wherever they go during the survey days. Sec-
ond, GIS land use data like road network and Point of Interest (e.g. typical buildings
such as shopping mall, school, hospital and park etc.) need to be collected. Third,
to evaluate the accuracy of prediction by the proposed method, respondents are also
required to provide activity diaries as well as their socioeconomic characteristics.
Search WWH ::
Custom Search