Jean-Yves Dieulot

Electric vehicle (EV) charging demand introduces significant variability in power systems, requiring forecasting approaches capable of representing both aggregated consumption trends and stochastic charging behaviors. While machine learning methods often provide strong predictive performance, they generally require large datasets and substantial computational resources. This paper proposes a stochastic framework based on compound Poisson and Cox processes to model EV charging demand using real charging station data collected at one-minute resolution. The proposed methodology jointly models charging-event arrivals, charging duration, and charging power through probabilistic distributions calibrated from historical observations. A compound homogeneous Poisson process (CHPP) and a double stochastic compound Poisson process (Cox process) are investigated and compared for the generation of synthetic EV charging profiles and short-term forecasting applications. The framework is validated using 1863 charging sessions recorded at a workplace charging infrastructure composed of 37 charging terminals. Monte Carlo simulations are performed to generate synthetic daily charging profiles and evaluate the capability of the models to reproduce key operational indicators, including daily energy consumption and peak grid power demand. The CHPP process achieves average forecasting errors up to 0.8% for daily energy and 6.2% for maximum grid power demand. The results show that Poisson-based stochastic models can generate diverse and realistic charging profiles while requiring only limited historical data and having low computational complexity. The proposed approach provides an interpretable and computationally efficient probabilistic framework for EV charging demand forecasting, synthetic profile generation, and power system operational studies. Stochastic compound Poisson processes may therefore constitute a valuable tool to support the ongoing electrification of mobility and the digital transformation of future smart grids and smart cities.

Physics-informed neural networks (PINNs) have emerged as a promising alternative to purely data-driven neural networks (NNs) for surrogate modeling, particularly in data-scarce scenarios. This study evaluates the performance of hybrid-PINNs against traditional NNs for modeling the adsorption step of a Direct Air Capture (DAC) process. As the complexity of the modeled system increases, larger datasets and longer computational times are required for numerical methods. Therefore, the study aims to develop approaches that minimize data requirements while maintaining accuracy, which is crucial for efficient modeling of complex physical systems. While both AI models can achieve high accuracy with abundant data, the advantages of hybrid-PINNs become more evident as data becomes scarce. In the intermediate and low-data regimes, the physics constraints embedded in hybrid-PINNs significantly improve generalization and predictive accuracy. For extreme low-data conditions, a curriculum learning strategy is implemented, progressively enforcing physics constraints to mitigate underfitting and enhance model stability. Despite these benefits, hybrid-PINNs exhibit a computational cost approximately one order of magnitude higher than traditional NNs as enforcing physics constraints increases training complexity. The results suggest that PINNs hold potential for modeling complex multi-physics problems in DAC and beyond, provided challenges related to gradient balancing and computational efficiency are addressed.

Physics-Informed Neural Networks (PINNs) have emerged as a promising alternative to purely data-driven neural networks (NNs) for surrogate modeling, particularly in data-scarce scenarios. This study evaluates the performance of hybrid-PINNs against traditional NNs for modeling the adsorption step of a Direct Air Capture (DAC) process. As the complexity of the modeled system increases, larger datasets and longer computational times are required for numerical methods. Therefore, the study aims to develop approaches that minimize data requirements while maintaining accuracy, which is crucial for efficient modeling of complex physical systems. While both AI models can achieve high accuracy with abundant data, the advantages of hybrid-PINNs become more evident as data becomes scarce. In the intermediate and low-data regimes, the physics constraints embedded in hybrid-PINNs significantly improves generalization and predictive accuracy. For extreme low-data conditions, a curriculum learning strategy is implemented, progressively enforcing physics constraints to mitigate underfitting and enhance model stability. Despite these benefits, hybrid-PINNs exhibit a computational cost approximately one order of magnitude higher than traditional NNs, as enforcing physics constraints increases training complexity. The results suggest that PINNs hold potential for modeling complex multi-physics problems in DAC and beyond, provided challenges related to gradient balancing and computational efficiency are addressed.

With the integration of power converters into the power grid, it becomes crucial for the Transmission System Operator (TSO) to ascertain whether they are operating in Grid Forming or Grid Following modes. Due to intellectual properties, classification can only be performed based on non-intrusive measurements and models, such as admittance at the PCC. This classification poses a challenge as the TSO lacks precise knowledge of the actual control structures and algorithms. This paper introduces a novel classification algorithm based on Convolutional Neural Networks (CNN), capable of detecting patterns in sequential data. The proposed CNN utilizes a new architecture to separate admittances along the d and q axes, and a decision layer allows to determine the correct converter mode. The performance of the proposed CNN model was assessed through two tests and compared to the traditional feedforward model. The proposed CNN architecture demonstrates significant classification capabilities, as it is able to identify the control mode of the converter even when its control structure is not part of the training dataset.

The service life of equipment is generally linked to degradation factors depending on its operating conditions, including the rate of use and the frequency of the switching modes. The novel operating mode management proposed in this paper takes into account equipment lifetime in addition to all the previously mentioned requirements. This algorithm does not rely only on real-time data, as is traditionally presented in the literature, but also integrates predictive operating data. Therefore, it can be considered as a hybrid operating mode management as it embeds both predictive and event data, which yields improved results with respect to traditional event-driven management. This allows to optimize a criterion over a finite horizon, and, hence, an optimal sequence of the switching times of the different components of an energy system are generated. While the proposed approach is considered to be generic, it is illustrated by the production of green hydrogen from renewable sources. In order to ensure the operating safety and energy efficiency of the system, the objective is to maximize the life duration of the electrolyzer and the batteries by avoiding excessive stored quantities. Simulations using data obtained from a laboratory platform which replicates the process at a smaller scale highlight the effectiveness of the proposed approach.