Catálogo de publicaciones - libros

Purpose of the paper is to introduce a methodology for a parameter-free multi-criterion optimization of water distribution networks. It is based on a two-level approach, with a population of inner multi-objective genetic algorithms (MOGAs) and an outer simple GA (without crossover). The inner MOGAs represent the network optimizers, while the outer GA – the meta GA – is a supervisor process adapting mutation and crossover probabilities of the inner MOGAs. The hypervolume metric has been adopted as fitness for the individuals at the meta-level. The methodology has been applied to a small system often studied in the literature, for which an exhaustive search of the entire decision space has allowed the determination of all Pareto-optimal solutions of interest: the choice of this simple system was done in order to compare the hypervolume metric to two performance measures (a convergence and a sparsity index) introduced on purpose. Simulations carried out show how the proposed procedure proves robust, giving better results than a MOGA alone, thus allowing a considerable ease in the network optimization process.

It has been a controversial issue in the research of cognitive science and artificial intelligence whether signal-type representations (typically connectionist networks) or symbol-type representations (e.g., semantic networks, production systems) should be used. Meanwhile, it has also been recognized that both types of information representations might exist in the human brain. In addition, symbol-type representations are often very helpful in gaining insights into unknown systems. For these reasons, comprehensible symbolic rules need to be extracted from trained neural networks. In this paper, an evolutionary multi-objective algorithm is employed to generate multiple models that facilitate the generation of signal-type and symbol-type representations simultaneously. It is argued that one main difference between signal-type and symbol-type representations lies in the fact that the signal-type representations are models of a higher complexity (fine representation), whereas symbol-type representations are models of a lower complexity (coarse representation). Thus, by generating models with a spectrum of model complexity, we are able to obtain a population of models of both signal-type and symbol-type quality, although certain post-processing is needed to get a fully symbol-type representation. An illustrative example is given on generating neural networks for the breast cancer diagnosis benchmark problem.

This paper presents a methodology to generate representations for isolated handwritten symbols, modeled as a multi-objective optimization problem. We detail the methodology, coding domain knowledge into a genetic based representation. With the help of a model on the domain of handwritten digits, we verify the problematic issues and propose a hybrid optimization algorithm, adapted to needs of this problem. A set of tests validates the optimization algorithm and parameter settings in the model’s context. The results are encouraging, as the optimized solutions outperform the human expert approach on a known problem.

This paper investigates the use of a multi-objective genetic algorithm, MOEA, to solve the scheduling problem for aircraft engine maintenance. The problem is a combination of a modified job shop problem and a flow shop problem. The goal is to minimize the time needed to return engines to mission capable status and to minimize the associated cost by limiting the number of times an engine has to be taken from the active inventory for maintenance. Our preliminary results show that the chosen MOEA called GENMOP effectively converges toward better scheduling solutions and our innovative chromosome design effectively handles the maintenance prioritization of engines.

This paper presents a new search procedure to tackle multi-objective traveling salesman problem (TSP). This procedure constructs the solution at-tractor for each of the objectives respectively. Each attractor contains the best solutions found for the corresponding objective. Then, these attractors are merged to find the Pareto-optimal solutions. The goal of this procedure is not only to generate a set of Pareto-optimal solutions, but also to provide the infor-mation about these solutions that will allow a decision-maker to choose a good compromise solution.

The problem of computing spanning trees along with specific constraints is mostly NP-hard. Many approximation and stochastic algorithms which yield a single solution, have been proposed. In this paper, we formulate the generic multi-objective spanning tree (MOST) problem and consider edge-cost and diameter as the two objectives. Since the problem is hard, and the Pareto-front is unknown, the main issue in such problem-instances is how to assess the convergence. We use a multiobjective evolutionary algorithm (MOEA) that produces diverse solutions without needing a priori knowledge of the solution space, and generate solutions from multiple tribes in order to assess movement of the solution front. Since no experimental results are available for MOST, we consider three well known diameter-constrained minimum spanning tree (dc-MST) algorithms including randomized greedy heuristics (RGH) which represents the current state of the art on the dc-MST, and modify them to yield a (near-) optimal solution-fronts. We quantify the obtained solution fronts for comparison. We observe that MOEA provides superior solutions in the entire-range of the Pareto-front, which none of the existing algorithms could individually do.

In this paper, we experiment with a combination of innovative approaches to rule induction to encourage the production of interesting sets of classification rules. These include multi-objective metaheuristics to induce the rules; measures of rule dissimilarity to encourage the production of dissimilar rules; and rule clustering algorithms to evaluate the results obtained. Our previous implementation of NSGA-II for rule induction produces a set of cc-optimal rules (coverage-confidence optimal rules). Among the set of rules produced there may be rules that are very similar. We explore the concept of rule similarity and experiment with a number of modifications of the crowding distance to increasing the diversity of the partial classification rules produced by the multi-objective algorithm.

The design of large scale water distribution systems is a very difficult optimisation problem which invariably requires the use of time-expensive simulations within the fitness function. The need to accelerate optimisation for such problems has not so far been seriously tackled. However, this is a very important issue, since as MOEAs become more and more recognised as the ‘industry standard’ technique for water system design, the demands placed on such systems (larger and larger water networks) will quickly meet with problems of scaleup. Meanwhile, LEM (Learnable Evolution Model’) has appeared in the Machine Learning literature, and provides a general approach to integrating machine learning into evolutionary search. Published results using LEM show very great promise in terms of finding near-optimal solutions with significantly reduced numbers of evaluations. Here we introduce LEMMO (Learnable Evolution Model for Multi-Objective optimization), which is a multi-objective adaptation of LEM, and we apply it to certain problems commonly used as benchmarks in the water systems community. Compared with NSGA-II, we find that LEMMO both significantly improves performance, and significantly reduces the number of evaluations needed to reach a given target. We conclude that the general approach used in LEMMO is a promising direction for meeting the scale-up challenges in multiobjective water system design.

This papers proposes an enhanced Particle Swarm Optimization algorithm with multi-objective optimization concepts to handle constraints, and operators to keep diversity and exploration. Our approach, PESDRO, is found robust at solving redundancy and reliability allocation problems with two objective functions: reliability and cost. The approach uses redundancy of components, diversity of suppliers, and incorporates a new concept called Distribution Optimization . The goal is the optimal design for reliability of coherent systems. The new technique is compared against algorithms representative of the state-of-the-art in the area by using a well-known benchmark. The experiments indicate that the proposed approach matches and often outperforms such methods.

Water systems often allow efficient water uses via water reuse and/or recirculation. Defining the network layout connecting water-using processes is a complex problem which involves several criteria to optimize, frequently accomplished using Water Pinch technology, optimizing freshwater flowrates entering the system. In this paper, a multiobjective optimization model considering two criteria is presented: (i)the minimization of freshwater consumption, and (ii) the minimization of the cost of the infrastructure required to build the network that make possible the reduction of freshwater consumption. The optimization model considers water reuse between operations and wastewater treatment as the main optimization mechanism. The operation of the Cuernavaca city water distribution system was analyzed under two different operation strategies considering: leak reduction, operation of wastewater treatment plants as they currently operate, operation of wastewater treatment plants at design capacity, and construction of new infrastructure to treat 100 % of discharged wastewater. Results were obtained with MDQL a multiobjective optimization algorithm based on a distributed reinforcement learning framework, and they were validated with mathematical programming.