Abstract

The evaluation of open-ended responses in serious games presents a unique challenge, as correctness is often subjective. Large Language Models (LLMs) are increasingly being explored as evaluators in such contexts, yet their accuracy and consistency remain uncertain, particularly for smaller models intended for local execution. This study investigates the reliability of five small-scale LLMs when assessing player responses in \textit{En-join}, a game that simulates decision-making within energy communities. By leveraging traditional binary classification metrics (including accuracy, true positive rate, and true negative rate), we systematically compare these models across different evaluation scenarios. Our results highlight the strengths and limitations of each model, revealing trade-offs between sensitivity, specificity, and overall performance. We demonstrate that while some models excel at identifying correct responses, others struggle with false positives or inconsistent evaluations. The findings highlight the need for context-aware evaluation frameworks and careful model selection when deploying LLMs as evaluators. This work contributes to the broader discourse on the trustworthiness of AI-driven assessment tools, offering insights into how different LLM architectures handle subjective evaluation tasks.

Abstract

Energy transition is a major goal for humanity to mitigate climate change. Transportation is one of the sectors that should contribute the most to reducing greenhouse gas emissions. Shore Side Energy is proposed as one of the main approaches to reduce the emissions of vessels when they are docked. Nevertheless, several improvements should be made in ports to achieve optimal energy management, considering the needs of vessels. To achieve this goal, accurate forecasting tools need to be used. In the present paper, a forecasting framework to be used in ports is described, and some results are presented. The proposed framework will be used in a European project, Shift2DC.

Abstract

This paper presents a probabilistic forecasting approach tailored for low voltage (LV) substations, offering short-term predictions for three crucial variables: voltage, reactive power, and active power. These parameters play a vital role in the resilience of distribution systems, especially in the presence of Distributed Energy Resources (DERs). Evaluation with simulated data shows that active and reactive power forecasts degrade notably with higher EV penetration, whereas voltage forecasting experiences less degradation across all scenarios.

Abstract

We propose a comprehensive approach to increase the reliability and resilience of future power grids rich in distributed energy resources. Our distributed scheme combines federated learning-based attack detection with a local electricity market-based attack mitigation method. We validate the scheme by applying it to a real-world distribution grid rich in solar PV. Simulation results demonstrate that the approach is feasible and can successfully mitigate the grid impacts of cyber-physical attacks.

Abstract

This paper presents the ALAMO vision, which is a research project that aims at developing technologies to facilitate the planning and operation of power grids with very high penetration of Distributed Energy Resources, while at the same time assuring the privacy of the main actors on the demand side. The ALAMO project will address outstanding research challenges related to the development of accurate forecasting algorithms based on Federated Learning (e.g., forecasting in- front and behind-the-meter PV production); as well as challenges related to producing sharped and well-calibrated quantifications of epistemic and aleatoric uncertainty for such forecasting models. Finally, use-cases will be carefully crafted to understand how FL forecasts and uncertainty estimates can be incorporated into operational planning and operation tools. The use-cases will be demonstrated in virtual and physical testbeds in Portugal, USA, and Brazil.

Abstract

This proposal aims to develop more accurate federated learning (FL) methods with faster convergence properties and lower communication requirements, specifically for forecasting distributed energy resources (DER) such as renewables, energy storage, and loads in modern, low-carbon power grids. This will be achieved by (i) leveraging recently developed extensions of FL such as hierarchical and iterative clustering to improve performance with non-IID data, (ii) experimenting with different types of FL global models well-suited to time-series data, and (iii) incorporating domain-specific knowledge from power systems to build more general FL frameworks and architectures that can be applied to diverse types of DERs beyond just load forecasting, and with heterogeneous clients.

Abstract

There are many benefits from the accurate forecasting of Arctic sea ice, however existing models struggle to reliably predict sea ice concentration at long lead times. Many numerical models exist but can be sensitive to initial conditions, and while recent deep learning-based methods improve overall robustness, they either do not utilize temporal trends or rely on architectures that are not performant at learning long-term sequential dependencies. We propose a method of forecasting sea ice concentration using neural circuit policies, a form of continuous time recurrent neural architecture, which improve the learning of long-term sequential dependencies compared to existing techniques and offer the added benefits of adaptability to irregular sequence intervals and high interpretability.

Abstract:

The problem of appliance recognition is one of the most relevant issues in the field of Non-Intrusive-Load-Monitoring; its importance has led, in recent years, to the development of innovative techniques to try to solve it. The use of methods such as V-I trajectory, Fryze Theory Decomposition, and Weighted Recurrence Graph have proved effective in recognizing both single (Single Label) and multiple active appliances (Multi-Label). This paper presents a new way of approaching the problem by unifying Single Label and Multi-Label learning paradigms. The proposed approach exploits feature extraction techniques that allow the detection of both activated/deactivated appliances and all active appliances given aggregate current signal. We evaluate the proposed approach on a PLAID dataset. The obtained results indicate combining single-label and multi-label learning strategies for appliance recognition provides improved classification results with an F-score of 0.91.

Abstract

Over the years, an enormous amount of research has been exploring Deep Neural Networks (DNN), particularly Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) for estimating the energy consumption of appliances from a single point source such as smart meters – Non-Intrusive Load Monitoring (NILM).
However, most of the existing DNNs models for NILM use a single-task learning approach in which a neural network is trained exclusively for each appliance. This strategy is computationally expensive and ignores the fact that multiple appliances can be active simultaneously and dependencies between them. In this work, we propose UNet-NILM for multi-task appliances’ state detection and power estimation, applying a multi-label learning strategy and multi-target quantile regression. The UNet-NILM is a one-dimensional CNN based on the U-Net architecture initially proposed for image segmentation. Empirical evaluation on the UKDALE dataset suggests promising performance against traditional single-task learning.

Abstract

This paper pursues the question of how seasons of the year affect disaggregation performance in Non-Intrusive Load Monitoring.
To this end, we select the dishwasher, a common household appliance that may exhibit usage cycles depending on the user. We utilize an auto-correlation function to detect usage patterns of dishwashers in each season. Then, we examine the dissimilarity across each season with the help of the Keogh Lower Bound measure. Finally, we conduct a disaggregation study using the REFIT dataset and relate the outcome to the dissimilarity across seasons. Our findings indicate that in cases where energy consumption shows similarity throughout seasons, the performance of load disaggregation approaches can be positively affected.

Abstract

Large scale renewable energy source (RES) integration planned for multiple power grids around the world will require additional resources/reserves to achieve secure and stable grid operations to mitigate the inherent intermittency of RES. In this paper, we present formulations to understand the effect of fast storage reserves in improving grid reliability under different cost functions. Our formulations not only aim to minimize imbalance but also maintain state-of-charge (SoC) of storage. The proposed approaches rely on a macroscopic supply-demand model of the grid with total power output of energy storage as the control variable. We show that accounting for system response due to inertia and local governor response enables a more realistic quantification of storage requirements for damping net load fluctuations. Simulation case studies are embedded in the paper by using datasets from the Elia TSO in Belgium and BPA in the USA. The numerical results benchmark the marginal effect on reliability due to increasing storage size under different system responses and associated cost functions. Further we observe myopic control of batteries proposed approximates deterministic control of batteries for faster time scale reserve operation.

Resources

Preprint