Jose Nuno Fidalgo

Understanding the complex dynamics of the sugar–biofuel market is challenging due to the multifaceted effects of economic, agronomic, and policy‐related factors, with sugar and ethanol fuel competing for the same sugarcane input. In this paper, we propose a comprehensive eXplainable Artificial Intelligence (XAI) framework to predict and interpret ethanol price formation in Brazil, leveraging a rich, multi‐source agroeconomic panel database spanning nearly two decades of indicators across economic, fiscal, energy, and global commodity domains. Our approach integrates machine learning, exploratory data analysis, and XAI techniques to uncover the key drivers behind price fluctuations. Forecasting experiments demonstrate that deep learning networks show superior performance at capturing temporal dependencies and nonlinear trends from the data, while interpretability tools such as Shapley Additive Explanations and Permutation Feature Importance reveal gasoline prices, sugarcane quality, and VAT taxation as primary drivers, with inflation and exchange rates exerting moderating influences. We show that the proposed framework provides interpretable insights to support evidence‐based decision‐making for policymakers and industry actors in the sugar–biofuel sector.

Short-term load forecasting (STLF) models routinely integrate weather variables, but the operational value of this information depends on forecast horizon, climate context, and thermal regime. This paper proposes a validation-weighted thermal ensemble for STLF that combines direct weather-informed models with residual thermal correction models. The framework is evaluated using hourly load, temperature, and irradiance data from five European countries over 2015--2019, considering 1-, 6-, and 24-hour horizons. Candidate models are trained on 2015--2017, validation-based adaptation is performed in 2018, and final out-of-sample testing is conducted in 2019. The experimental design compares direct feature integration, learned thermal thresholds, thermal residual correction, hard regime-aware selection, regime-weighted objectives, and validation-weighted soft ensembles under a unified framework. The results show that direct thermal integration is generally preferable for 6- and 24-hour horizons, whereas residual thermal correction provides useful adjustments at very short horizons. The proposed soft ensemble outperforms the strongest direct thermal model in all 15 country--horizon combinations, yielding an average MAPE gain of 0.0375 percentage points and an average relative MAPE reduction of about 3.3\%. The gains are confirmed under WAPE, and 13 of the 15 pairwise tests remain significant after Benjamini--Hochberg correction. An additional rolling-origin robustness check reproduces positive MAPE and WAPE gains in all tested cases. These findings support validation-weighted combinations of direct and residual thermal structures as a robust and computationally light strategy for operational forecasting workflows supporting balancing, reserve scheduling, congestion management, and flexibility planning.

Daylight Saving Time (DST) has long been justified as a measure to reduce electricity consumption by better aligning human activity with daylight. However, its effectiveness under current energy-use patterns remains uncertain. This study provides a comprehensive assessment of the impact of DST on electricity-related needs for lighting and climatization across multiple European countries.The analysis is based on a unified methodological framework that evaluates the number of hours requiring artificial lighting and thermal conditioning, using empirical solar irradiance and temperature data. By focusing on the extensive margin, the approach enables consistent comparisons between DST and non-DST scenarios under alternative legal time configurations.The results show that DST reduces lighting needs, but these reductions are modest when realistic irradiance data are taken into account. In contrast, DST increases climatization requirements, primarily due to higher heating demand. As these effects operate in opposite directions, the overall impact of DST on electricity-related needs is limited. Additionally, the analysis indicates that adopting a uniform legal time across the European Union may be suboptimal from an energy perspective, because the alignment between civil time and environmental conditions varies with longitude.Overall, the findings indicate that the energy implications of DST are relatively small and context-dependent, and that legal time settings should account for regional differences.

Accurate prediction of peak electricity demand at the substation level is critical for power system planning and operation. This study proposes an interpretable panel data framework with fixed effects to forecast substation-level peak demand in Portugal, combining robust event-based peak identification with climate-aware demand modeling.Compared to black-box machine learning approaches, in this application the proposed model maintains competitive predictive performance while providing clear insights into the structural drivers of demand. Results show that panel-data fixed effects specifications outperform Random Forest and Gradient Boosting methods when persistent heterogeneity across substations is explicitly accounted for.The study further introduces temperature-related variables capturing deviations from local climatic norms and short-term thermal dynamics around peak events. The results reveal a clear seasonal asymmetry: winter peak demand is significantly influenced by sustained temperature dynamics, whereas summer demand is largely insensitive to temperature effects within the observed range of conditions.Overall, the findings highlight the importance of explicitly modeling structural heterogeneity and demand responses in electricity demand forecasting. By combining interpretability with robust predictive performance, the proposed approach supports more informed operational forecasting and infrastructure planning decisions in modern distribution networks

Current residential electricity tariffs often combine a flat energy price with a fixed charge linked to contracted power, resulting in electricity bills that are weakly responsive to changes in consumption. This lack of proportionality reduces incentives for energy savings and may undermine demand-side efficiency.This paper proposes a novel Power-Conditioned Pricing (PCP) tariff, in which unit energy prices depend on the power level at which electricity is consumed. By associating higher prices with higher consumption intensity, the proposed tariff introduces progressivity while preserving transparency and regulatory feasibility. The tariff is calibrated to ensure revenue neutrality with respect to the current tariff for each contracted power level.Two complementary calibration strategies are analysed: a profile-based approach using representative regulatory load profiles, and an empirical approach based on statistical distributions derived from real consumer data. To assess consumer responsiveness, electricity bills are evaluated under both vertical and horizontal consumption adjustment models.Results show that bill elasticity increases from values between 0.43–0.73 under the current tariff to values close to unity under PCP, while maintaining revenue neutrality across contracted power levels. These findings suggest that power-conditioned pricing constitutes a promising alternative to current residential tariff structures, better aligned with energy-efficiency and conservation objectives.

Technical loss estimation in distribution networks is fundamentally limited by the temporal resolution of metering data. Because Joule losses scale with the square of instantaneous current, coarse sampling suppresses short-term variability and leads to systematic underestimation. This paper develops a unified analytical framework that quantifies this sampling-interval bias and expresses its magnitude as a function of intra-interval power variance. The framework is validated through controlled singleline experiments, real 1-minute residential measurements, and nearly one million power-flow simulations on three European medium-voltage feeders. Across these networks, replacing 1minute with 60-minute data leads to a 5-15% underestimation of annual losses. To address this issue, we introduce a variability-based correction approach that reconstructs highresolution losses directly from coarse AMI measurements using simple statistical indicators. The method is lightweight, scalable, and compatible with existing metering infrastructures, enabling accurate loss estimation without sub-minute simulations and supporting advanced monitoring, planning, and operational analytics.

This article analyzes the effects of the current policy trends - high levels of distributed generation (DG) and grid load/capacity ratio - on network efficiency. It starts by illustrating the network losses performance under different DG and load/capacity conditions. The second part concerns the simulation of network investments with the purpose of loss reduction for diverse system circumstances, including the impact of DG levels, energy cost, and discount rate. The attained results showed that DG, particularly large parks, have a negative impact on network efficiency: network losses tend to intensify with DG growth, under the current regulation. Furthermore, network investments in loss reduction would have a small global impact on network efficiency if the DG parks’ connection lines are not included in the grid concession (not subjected to upgrade). Finally, the study determines that it is preferable to invest sooner, rather than to postpone the grid reinforcement for certain conditions, namely for low discount rates.

In Europe, clean distributed generation, DG, is perceived as a crucial instrument to build the path towards carbon emission neutrality. DG already reached a large share in the generation mix of several countries and the reduction of technical losses is one of its most mentioned advantages. In this scope, this paper discusses the weaknesses of this postulation using real networks. The adopted methodology involves the power flow simulation of a collection of real networks, using 15 min real measurements of loads and generations for a whole year. The clustering of similar cases allows identifying the situations that cause higher losses. A complementary objective of this research was to define an approach to mitigate this problem in terms of identifying the branches that, if reinforced, most contribute to losses reduction. The results obtained confirm the rationality of the proposed methodology.