Artificial immune systems in evolutionary method for optimal distribution network design

Planning of urban distribution networks is a complex problem. When the problem is being solved, it is commonly translated into combinatorial optimization problems, like single and multiple depot vehicle routing problems (MDVR-P). Since such optimization problems are NP-hard, their exact solving is practically impossible. Evolutionary algorithms have been successful in solving those problems. Despite their relatively high efficiency, expected progress is related to obtaining better stability and lesser dependency on parameters. Moreover, there are several opposed criteria that the distribution networks must satisfy. For instance, there are limits on allowable voltage drops and conductor capacities. An algorithm for successful planning of the distribution networks should implement all of these criteria. In this paper a model of the evolutionary algorithm for solving the MDVRP is shown, based on analogy between the MDVRP and the distribution network planning problem. The algorithm includes an improvement inspired by artificial immune systems' techniques. Some specifics of the distribution network design are implemented in the immune operator which is a part of the evolutionary algorithm, instead of applying them in post-optimization phase. Some practical examples have been used to investigate the performance of the proposed algorithm.