Abstract
With technological advancement and the urgency to decarbonize energy consumption habits, smart grids have gained special prominence in recent years, highlighting the importance of the massive integration of endogenous renewable sources and decision-making tools, like forecasting tools. The relevance and accuracy of the forecast make it possible to add a contribution to energy management tools in residential communities, from the point of view of end-users and the distribution network operator. This work presents the development of a short-term hybrid forecasting model, combining Long-Short Term Memory (LSTM) model forecast with the Holt-Winters forecast model, where the ability of the LSTM stands out in capturing the complex temporal patterns of historical time series, while Holt-Winters deals with trends and seasonality of historical data. Combining these models results in an intelligent hybrid system capable of efficiently dealing with the complexity inherent to renewable energy. Then, the forecasted results from load and solar generation are introduced on the home energy management model considering a small residential community, showing the relevance of accurate forecasted results tools to assist in the making decisions processes.

Abstract
[No abstract available]

Abstract

Abstract
Microgrids are poised to become the building blocks of the future control architecture of electric power systems. As the number of controllable points in the system grows exponentially, traditional control and optimization algorithms become inappropriate for the required operation time frameworks. Reinforcement learning has emerged as a potential alternative to carry out the real-time dispatching of distributed energy resources. This paper applies one of the continuous action-space algorithms, proximal policy optimization, to the optimal dispatch of a battery in a grid-connected microgrid. Our simulations show that, though suboptimal, RL presents some advantages over traditional optimization setups. Firstly, it can avoid the use of forecast data and presents a lower computational burden, therefore allowing for implementation in distributed control devices.

Abstract
Large-scale datasets are essential for training deep learning models in medical imaging. However, many of these datasets contain poor-quality images that can compromise model performance and clinical reliability. In this study, we propose a framework to detect non-compliant images, such as corrupted scans, incomplete thorax X-rays, and images of non-thoracic body parts, by leveraging contrastive learning for feature extraction and parametric or non-parametric scoring methods for out-of-distribution ranking. Our approach was developed and tested on the CheXpert dataset, achieving an AUC of 0.75 in a manually labeled subset of 1,000 images, and further qualitatively and visually validated on the external PadChest dataset, where it also performed effectively. Our results demonstrate the potential of contrastive learning to detect non-compliant images in large-scale medical datasets, laying the foundation for future work on reducing dataset pollution and improving the robustness of deep learning models in clinical practice. © 2024 IEEE.

Abstract