Optimization of heat exchanger networks using genetic algorithms

Most thermal processes encountered in the power industry (chemical, metallurgical, nuclear and thermal power stations) necessitate the transfer of large amounts of heat between fluids having different thermal potentials. A common practice applied to achieve such a requirement consists of using heat exchangers. In general, each current of fluid is conveniently cooled or heated independently from each other in the power plant. When the number of heat exchangers is large enough, however, a convenient arrangement of different flow currents may allow a considerable reduction in energy consumption to be obtained (Linnhoff and Hidmarsh, 1983). In such a case the heat exchangers form a 'Heat Exchanger Network' (HEN) that can be optimized to reduce the overall energy consumption. This type of optimization problem, involves two separates calculation procedures. First, it is necessary to optimize the topology of the HEN that will permit a reduction in energy consumption to be obtained. In a second step the power distribution across the HEN should be optimized without violating the second law of thermodynamics. The numerical treatment of this kind of problem requires the use of both discrete variables (for taking into account each heat exchanger unit) and continuous variables for handling the thermal load of each unit. It is obvious that for a large number of heat exchangers, the use of conventional calculation methods, i.e., Simplexe, becomes almost impossible. Therefore, in this paper we present a 'Genetic Algorithm' (GA), that has been implemented and successfully used to treat complex HENs, containing a large number of heat exchangers. As opposed to conventional optimization techniques that require the knowledge of the derivatives of a function, GAs start the calculation process from a large population of possible solutions of a given problem (Goldberg, 1999). Each possible solution is in turns evaluated according to a 'fitness' criterion obtained from an objective equation. This equation must completely describe the optimization problem to be handled, i.e., maximization or minimization. The best solutions are then retained and Genetic operators such as crossover and mutation are then applied in order to reproduce a new population of solutions that have a better fitness than the previous ones. These processes of crossover, mutation and selection are repeated until a suitable convergence criterion is able to stop the procedure. It is important to point out that GAs handle a coded form of each possible solution (for instance binary coded solutions) that represent the individuals, i.e., chromosomes of a population, instead of handling the solution to the problem itself. In order to carry out the synthesis of HEN we have implemented two different coded populations; one population is used to code for the topology of the HEN and the second for the heat load handled by each heat exchanger (Lewin et al., 1998). Ck is a coefficient used to adjust the degree of penalty. This approach has been used to treat several HEN problems taken from the open literature. In general the results obtained with the proposed algorithm are in excellent agreement with those obtained by using conventional techniques, i.e., Simplexe. We have found that the use of GAs also permits other satisfactory solutions corresponding to different heat exchanger topologies and thermal load distributions to be obtained. Further, we were able to handle HENs containing more than 15 heat exchanges, that were impossible to solve using conventional methods. However, it is important to point out that the proposed technique is not appropriate to handle HENs that require the division of currents. (author)