Energy exchange in a network of microgrids

This PhD thesis consists of 3 parts, all related the problem of energy management in a network of microgrids. In the first part we discuss peer-to-peer (P2P) energy trading among a network of microgrids. Firstly, we adapt a preference-based mechanism into the energy exchange model of a network of microgrids in the framework of P2P energy trading in the distribution system. We design an alternative direction method of multiplier (ADMM) based distributed algorithm to realize such P2P energy transactions, considering the coupling constraints of virtual energy trading and physical constraints on the connecting edges between any two microgrids within the network. The preference mechanism applied directly on the energy trading indirectly gives rise to a customized power flow within the network of microgrids by adjusting the preference values, for example to create several clusters of microgrids where each cluster ends up operating in an islanding mode. Two classical distribution networks are thoroughly examined: the first one is used to analyze the effect of the preference mechanism on the customized power flows; the second one is used to evaluate the convergence performance of the proposed distributed algorithm considering different time and size scales. In the second part we discuss the forecast of renewable energy, especially wind power forecasts, required by each local agent when solving its local optimization problem in the designed distributed framework of P2P energy trading. Wind power forecasts are investigated in two different scales where the first scale is for a single wind farm while the second scale is a region where multiple wind farms are included. Such scales are comparable to a network of microgrids in the distribution network when the distribution system operator (DSO) is considered as well. We first discuss the wind power forecast for a single wind farm using machine learning algorithms including shallow neural networks and deep neural networks of convolutional neural networks. A comprehensive comparison is carried out to evaluate all methods with four different datasets of four wind farms. We also discuss the distributed reconciliation of wind power forecasts in a two-layer hierarchy where the top layer corresponds to the system operator of a region and the bottom layer includes all the individual wind farms within the region. An ADMM-based algorithm is proposed to achieve the aggregation consistency that the aggregated forecast at the top layer is equal to the sum of all the individual forecasts at the bottom layer. Finally, in the last part of the thesis we discuss possible communication solutions involved in the practical implementation of P2P energy trading. Communication techniques play an important role in such an implementation. Each agent collects local data from remote terminal units and communicates information to the DSO or to other agents, which may require a hybrid communication solution including many different communication techniques in different segments. We only focus on one candidate communication solution, which is power line communication. We propose an impulsive noise mitigation method to cope with one of the challenging obstacles in the application of power line communication and evaluate it in ideal and frequency selective fading channels. All the parts and proposed methods are evaluated and analyzed using mathematical tools and extensive simulations.