Adaptive optimization and learning over networks

Adaptive optimization and learning over networks

Ours is a connected world. Technological (power, communications and transportation) networks play a leading role in our daily lives. To cope with the technological challenges posed by the modern society, contemporary networked systems have became more flexible, involved and autonomous. Power and transportation networks are now equipped with batteries and have access to renewable sources of energy. This evolution has opened the door to more energy efficient operation schemes and to better user’s experience. However, it has also made the design, management and operation of the networks more difficult. Successful execution of those tasks requires a detailed modeling and analysis of the network and its terminals. It also calls for adopting up-to- date optimization tools. While significant progress has been achieved in the last years, existing solutions still suffer from several weaknesses including: extremely simple network models, separate design of optimization and monitoring tasks, and suboptimal use of the network state information. This project aims to deal with such problems using a holistic approach. The power and transportation networks at hand are modelled as complex dynamic systems, where cognitive capabilities allow both nodes and controllers to make decisions about the network operation; and where optimization and monitoring are designed jointly (sharing objectives and considering the coupling between the two tasks) and robustly (considering the uncertainty and spatio-temporal variability of the network state information). The design of the schemes to operate the network will be accomplished by using contemporary tools in the fields of robust, stochastic and dynamic optimization, as well as reinforcement learning, distributed inference and network theory.