Accurate forecasts of electrical substations are mandatory for the efficiency of the Advanced Distribution Automation functions in distribution systems. The paper describes the design of a class of machine-learning models, namely neural networks, for the load forecasts of medium-voltage/low-voltage substations. We focus on the methodology of neural network model design in order to obtain a model that has the best achievable predictive ability given the available data. Variable selection and model selection are applied to electrical load forecasts to ensure an optimal generalization capacity of the neural network model. Real measurements collected in French distribution systems are used to validate our study. The results show that the neural network-based models outperform the time series models and that the design methodology guarantees the best generalization ability of the neural network model for the load forecasting purpose based on the same data.

The increasing level of Dispersed Generator (DG) connected to the distribution network lead to think about new integration solutions. In this paper, three decentralized control strategies are investigated and compared on a network. The three strategies are: local reactive power compensation, voltage tracking mode, and a mix of these two strategies. Losses and grid congestions will be the key factors to evaluate the controls strategies. The three strategies will be tested on the study case with an increasing DG rate. In the last part, the strategies will be evaluated on a Power Hardware In the Loop simulation and a real solar inverter.

This thesis investigates the potential contributions of flexibilities in Low Voltage Smart Grids. These networks are intrinsically different than Medium and High Voltages networks, so that the control of LV flexibilities cannot be directly taken from MV and HV networks, and new methods should be developed. The contribution of these flexibilities is studied through two main benefits: improved network operation and peak shaving. The first benefit focuses on maintaining the critical variables within the admissible constraints. The objective is to manage the network closer to its limits, reducing the need for margins, and therefore the need for upscaling. This is especially true in case of significant insertion of distributed generations or electric vehicles. The studied flexibility is the coordinated management of decentralized generation (active and reactive powers, phase switch) and a tap changer. The second benefit concerns the reduction of the peak consumption, either at the transformer, either at the national level. The studied flexibility is the shedding of electric heating for a short time, followed by a rebound when the heating is turned back on.

This paper presents a method allowing to choose the best phase for connecting a single phase DG. The aim is to find which connection phase induces the lowest losses while avoiding overrun constraints. In that order, some weigthing function are created. A Smart Metering system has been considered, that save and transmit data consumption from customers to the operator. Various consumption periods have been considered to take into account winter and summer critical period. Finally a study case presents a LV feeder with a gradualy increasing insertion rate of DGs. A comparison is done between cases with and without optimal connection.

This paper investigates a formulation of the Optimal Power Flow (OPF) problem, which aims to find the predictive optimal operational state of a low voltage smart grid. The optimization problem includes distributed generations (DG), distributed storage (DS), and smart buildings. Active and reactive power management is done through DG and DS control, and soft load shedding. It takes into account networks constraints, through a full AC power flow calculation. The low-voltage distribution network is modeled as a three-phased four-wire system. Several objective functions are described. Finally, a case study is solved using KNITRO [1], a robust and efficient solver for large, nonlinear and constrained problems, after being modeled in GAMS [2].