Innovative models and methodologies for multi-objective optimization, decision making and knowledge discovery, with applications

In the last recent years, optimization has become increasingly focused on multi-objective paradigms. Today, there typically exists the main requirement of simultaneously optimizing multiple (conflicting) aspects, while satisfying more and more challenging requirements. Therefore, it has been observed a constant development of progressively more sophisticated models and methodologies, particularly aimed at faithfully representing the preferences expressed by human experts on the priority to be assigned to each objective to optimize. The non-lipschitz relation between the weights of the objectives and the performance of the corresponding Pareto-optimal solution have motivated the development of sophisticated techniques for preferences representation, based, e.g., on fuzzy set theory. On the other hand, performance requirements concerning both the search and the subsequent selection of particular Pareto-optimal solutions which represent compromises closer and closer to the preferences, have given rise to peculiar evolutionary techniques, typically combined with multi-criteria decision-making approaches. In parallel, data mining techniques are continuously applied to the large amounts of data available today in order to search for hidden information, which may be extremely valuable, and play a key role if properly modeled and then integrated into optimization problems. Within the described scenario, this thesis presents some innovative models and methodologies based on techniques combining analytical approaches and Computational Intelligence methods, aimed at knowledge discovery and representation, and the consequent multi-objective optimization and multi-criteria decision-making in two areas: i) optimization of energy dispatch in the next-generation power network (the so-called smart grid); ii) risk analysis and management in the workplace. In particular, in relation to the first area, this thesis firstly describes an a priori technique for the resolution of a multi-objective version of the network-constrained economic dispatch problem. It is then described a module for automatic prioritization of objectives, based on artificial neural networks. In relation to the second area, this thesis presents three works concerning, namely, the automatic profiling of workers in terms of their risk sensibility, the subsequent classification of the workers in such profiles, and, finally, an innovative multi-objective model of the personnel assignment problem, specifically aimed at achieving the best trade-off between the minimization of the cost, the maximization of the gratification deriving from the task being performed, and the maximization of the safety in the workplace.