Abstract :
In this paper, an improved application of the genetic algorithm to large scale distribution systems planning is discussed. The genetic algorithm can find a sub-optimal solution of a large scale combinatorial optimization problem, by simulating the adoptive nature of natural genetics, faster than the simulated annealing method and more accurately than the existing heuristic approximate solution algorithm. However, from the author´s experiences in applying the algorithm to the distribution loss minimization problem, calculation results are accurate enough for small scale systems, but are not so accurate for large scale systems. The reason for this may be that so-called “implicit parallelism” does not work well for large scale problems because of the huge number of combination of solutions. Therefore this paper discusses improvements of the simple genetic algorithm needed to refine the accuracy for large scale power systems planning problems. The following operators and improvements are tested and discussed in this paper: (1) crossover between strings with high fitness function values, (2) number of crossover points, (3) mutation by burst of bits, (4) reduction of infeasible strings, (5) selection of initial strings, and (6) change of distribution of fitness function values. The performances of these improvements are compared with each other through comparative numerical results
