Time Decomposition Methods For Optimal Management Of Energy Storage Under Stochasticity

The development of energy storage paves the way to innovative methods to manage energy at a local scale. Micro grids are a novel kind of electrical grids with local production (renewable and waste energy), local demand, local storage and an Energy Management System (EMS). A wide literature already studies EMS implementations in micro grids but the produced methods are not exhaustively framed and compared. One of the main difficulty in micro grids energy management is to handle the different dynamics of electrical devices. Current variations are lighting fast, solar power changes quickly, different kind of storage react at different paces and batteries ageing is a slow process. We study a mathematical framework and algorithms, based on multistage stochastic optimization theory and Dynamic Programming, to model and solve energy management problems in micro grids with time decomposition methods. In the first part of this thesis, Contributions to time decomposition in multistage stochastic optimization, we present a general framework to decompose temporally large scale stochastic optimization problems into smaller sub-problems. We then classify multiple existing resolution methods inside this framework. In the second part, stochastic optimization of energy storage for management of micro grids, we compare different methods presented in the first part on realistic applications. First we control a battery and a ventilation in a subway station recovering subways braking energy with four different algorithms. Then we present how these results could be implemented on a real micro grid. We implement a fast online control method to stabilize the voltage in a simulated islanded DC micro grid connecting solar panels, an electrical load and two sorts of energy storage: a battery and a supercapacitor. Finally we apply our time decomposition framework to a problem of long term aging and energy management of a storage in a micro grid. This last chapter introduces a framework to model time decomposition of micro grids hierarchical control architectures, as well as two algorithms to solve temporally large scale stochastic optimization problems. In the third part, Softwares and experiments, we present DynOpt.jl, a Julia language package developed to produce all the results of this thesis and more. Then we study an application of this software to the control of a real test bed: the energy aware temperature regulation of a real house in the equipment named 'Sense City'. (author)