Page 282 - Design and Operation of Heat Exchangers and their Networks

268 Design and operation of heat exchangers and their networks

Genetic algorithm is a kind of stochastic algorithm based on the theory of
probability. In application this method to a stagewise superstructure model,
the search process is determined by stochastic strategy. The global optimal
solution for the synthesis of heat exchanger networks can be obtained at
certain probability. The search process begins with a set of initial stochastic
solutions, which is called “population.” Each solution is called
“chromosome,” the chromosome is composed of “gene,” and the “gene”
stands for the optimal variables of heat exchanger networks, for example,
the mass flowrates of cold streams and hot streams.
There are two kinds of calculation operation in the genetic algorithm:
genetic operation and evolution operation. The genetic operation adopts
the transferring principle of probability, selects some good chromosomes
to propagate at certain probability, and lets the other inferior chromosomes
to die; thus, the search direction will be guided to the most promising
region. With a stochastic search technique, they can explore different
regions of the search space simultaneously and hence are less prone to ter-
minate in local minimum. The strength of the genetic algorithm is the
exploration of different regions of the search space in relatively short com-
putation time. Furthermore, multiple and complex objectives can easily be
included. But genetic algorithm provides only a general framework for solv-
ing complex optimization problem. The genetic operators are often
problem-dependent and are of critical importance for successful use in prac-
tical problem. Specifically, to the synthesis problem of heat exchanger net-
works with multistream heat exchangers, an approach for initial network
generation, heat load determination of a match within superstructure should
be given. Some operators such as crossover operator, mutation operator,
orthogonal crossover, and effective crowding operators are appropriately
designed to adapt to the synthesis problem. Another difficulty for genetic
algorithm application is the treatment of constraints. During the genetic
evolution, an individual of the population may turn into infeasible solution
after manipulated by genetic operators, which will lead to failure to find a
feasible solution during evolution, especially for the optimization problem
with strict constraints. Hence, some strategy should be contrived for con-
straints guarantee in genetic computation.

6.5.2 Simulated annealing algorithm
Another effective algorithm used to solve large-scale combinatorial optimiza-
tion problems is the simulated annealing algorithm (Kirkpatrick et al., 1983).