Neural Network-Based Methods for the Management and Control of Complex Systems

In response to the growing complexity of modern infrastructures and the escalating demand for enhanced reliability, efficiency, and safety, advanced methods for managing and controlling complex systems have become increasingly critical. artificial intelligence and neural networks have emerged as powerful tools to address these challenges by offering efficient solutions for optimizing performance and mitigating risks. This thesis investigates neural network-based methods for the management and control of complex systems, focusing on practical applications in critical fields such as water distribution and autonomous marine navigation. The primary objective is to develop advanced strategies for uncertainty management and efficiency improvement in these systems. The first contribution presents an artificial intelligence-driven methodology for leak detection in water distribution networks, employing neural network regularization techniques to optimize sensor placement. This novel approach in the hydraulic context reduces the number of required sensors while maintaining high fault detection performance, leading to significant cost savings and improved network management. The second contribution involves modeling nonlinear systems using neural networks integrated with a Bayesian approach for uncertainty quantification. Incorporated into a stochastic model predictive control schema, this method allows to account for the model epistemic uncertainty, resulting in more reliable and efficient control actions. The proposed methodology was tested \textit{in-silico} with the ship maneuvering process, demonstrating promising results. In summary, the methodologies developed in this thesis not only advance the management of uncertainty and enhance the efficiency of complex systems but also offer scalable solutions with significant potential for adaptation in a wide range of critical applications.