The genetic algorithm has proven itself to be a fairly good optimization algorithm. Despite its many successful applications, there is a lack of theoretical understanding of why it works. In this paper, Vose-Liepins model (the so called infinite population model) is used to derive lower and upper bounds for the probability of success of the genetic algorithm in finding the global optimal solution within a prescribed number of generations. This is carrie dout by aggregating the Markov chain. Our bounds take into account some general fitness function landscapes.

This paper evaluates different optimization techniques for operations of multi-reservoir systems. Performance of linear programming (LP), non-linear programming (NLP), dynamic programming (DP) and genetic algorithm (GA) has been assessed on the basis of the objective function values achieved and the execution time required. Sensitivity to GA parameters; Crossover probability, Mutation probability, and Population size has also been carried out to evaluate the GA performance. Application of LP, DP, NLP using LINGO and GA to a four reservoir system with linear objective function is presented. The modified four-reservoir problem (with non-linear objective function) has also been solved using DP and GA. A generic model based on LINGO for the optimization of reservoir systems operation has been developed which is transportable with minimal changes to any reservoir system. The results obtained from various techniques have been compared. For both the problems, the state trajectories for all the reservoirs match closely. A distinct practical advantage of the GA approach is that discretization of state variables is not required as in DP, which is capable of solving non-convex, non-linear and discontinous objective and constraint function. The GA technique does not become computationally bounded for problems of large size. On the contrary, DP and NLP are likely to be computationally bounded for problems of moderate size.

This paper explores the application of a multiple colony ant systems meta-heuristic method called COMPETants in railroad yard operations optimization. Some slight modifications are proposed to make the original algorithm more suitable for use in the railroad yard context. The problem proposed here can be classified as a Full Truckload Pickup and Delivery Problem with Time Windows and Capacity Constraints (PDPTWC) and we aim to attend a set of transportation requests with minimal cost. Implementation issues and the first computational results are presented.

Le problème d’affectation de cellules aux commutateurs d’un réseau mobile est un problème NP-Complet ayant une complexité exponentielle. Notre article propose l’application de la méta-heuristique relativement récente d’Optimisation par Colonie de Fourmis (Ant Colony Optimization) à la résolution de ce problème. Notre méthode est relativement efficace tant en termes de qualité des solutions fournies qu’en termes du temps d’exécution de l’algorithme.