Abstract :
[en] As more PhotoVoltaic (PV) units are being installed, some low-voltage (LV) distribution networks have already at- tained their maximum hosting capacity, i.e. the maximum amount of distributed energy resources that they can accommodate during regular operations without suffering problems, such as overvoltages. This thesis presents dif- ferent solutions to increase the hosting capacity and to prevent the disconnection of PV units, without resorting to network reinforcements.
The first action to increase the hosting capacity of an LV network with a high penetration of PV panels is to better balance production on its three phases. Indeed, the hosting capacity is reduced by the imbalance caused by the currents consumed by single-phase household appliances, or produced by single-phase distributed generation units. To do so, the Distribution System Operators (DSOs) have to know to which phase of their networks the houses and the PV units are connected. In other words, they need to know the phase identification of the meters. In this thesis, we propose two novel algorithms to identify the phases of the smart meters. Both algorithms solely rely on the correlation between the voltage measurements of the smart meters to solve the phase identification problem. The first one is dedicated to the identification of three-phase smart meters and uses graph theory as well as the notion of maximum weight spanning tree to associate the smart meters that have the most correlated voltages. The second algorithm overcomes the main drawback of the first one, its inability to identify both single-phase and three-phase smart meters. Our algorithm improves the quality of the clustering by taking into account the underlying structure of the LV distribution networks beneath the voltage measurements without a priori knowledge on the topology of the network. The performance of this algorithm is compared using real measurements to those of the constrained k-means clustering method, which has been previously used for phase identification.
The second action to increase the hosting capacity is to control the power flows inside the network. This is called Active Network Management. We propose a distributed scheme that adjusts the reactive and active power output of inverters to prevent or alleviate the overvoltage problems that might arise from the integration of photovoltaic panels in LV networks. The proposed scheme is model free and makes use of limited communication between the controllers, in the form of a distress signal, only during emergency conditions. It prioritizes the use of reactive power, while active power curtailment is performed only as a last resort. The behaviour of the scheme is studied using balanced three-phase dynamic simulations on a single low-voltage feeder and on a larger network composed of 14 low-voltage feeders. This control scheme is then extended to the case of unbalanced three-phase four-wire distribution networks with single- and/or three-phase inverters. It works by first partitioning the inverters into four groups, three for the single-phase inverters (one for each phase), and one for the three-phase ones. Each group then independently applies a distributed algorithm similar to the one previously presented. Its performance is compared to those of two reference schemes, an on-off algorithm that models the default behaviour of PV inverters when there is an overvoltage, and the other one based on an unbalanced OPF.
Finally, as a way to implement active network management at the level of LV distribution, we introduce the concept of electricity prosumer communities, which are groups of people producing, sharing and consuming electricity locally. We focus on building a rigorous mathematical framework in order to formalise sequential decision-making problems that may soon be encountered within electricity prosumer communities. After introducing our formalism, we pro- pose a set of optimisation problems reflecting several types of theoretically optimal behaviours for energy exchanges between prosumers. We then discuss the advantages and disadvantages of centralised and decentralised schemes and provide illustrations of decision-making strategies, allowing a prosumer community to generate more distributed electricity (compared to commonly applied strategies) by mitigating overvoltages over a low-voltage feeder. We fi- nally investigate how to design distributed control schemes that may contribute to reaching (at least partially) the objectives of the community, by resorting to machine learning techniques to extract, from centralised solution(s), decision-making patterns to be applied locally.
