Napoleon Bezas
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Napoleon Bezas has a total of 44 co-authors. Here are the top co-authors by number of shared publications:
- Dimitrios Tzovaras
Centre for Research and Technology Hellas (CERTH)
8 shared publications - Dimosthenis Ioannidis
Information Technologies Institute, Centre for Research and Technology Hellas, Thessaloniki, Greece
8 shared publications - Christos Timplalexis
Centre for Research and Technology Hellas, Information Technologies Institute, Thessaloniki, Greece
3 shared publications - Paschalis A. Gkaidatzis
Information Technologies Institute, Centre for Research & Technology, 57001 Thessaloniki, Greece
3 shared publications - Alexios Papaioannou
Management Science and Technology Department, Democritus University of Thrace, 65404 Kavala, Greece
2 shared publications - Angelina D. Bintoudi
Centre for Research and Technology—Hellas, Information Technologies Institute, P.O. Box 60361, 5700 Thessaloniki, Greece
2 shared publications
Latest Publications
- Open Access
- 25 March 2026
- Published by Elsevier in SSRN Electronic Journal
Time Series Foundation Models for PV Power Forecasting: Benchmarking and Physics-Weather-Aware Agentic Correction
Time Series Foundation Models (TSFMs) offer a new approach for forecasting tasks via zeroshot generalization. At the same time, the advances in LLM reasoning capabilities show promise towards enhancing forecasting pipelines. In the present study, focusing on the photovoltaics (PV) power generation domain, prominent TSFMs are benchmarked against established task-specific supervised models across real-world datasets. The results demonstrate that TSFMs, particularly TabPFN and Chronos2, outperform supervised models in data-sparse settings. To address frequent forecast deviations of TSFMs caused by transient meteorological events, this work introduces a PV Physics-Weather-Aware Reasoning & Correction Agent (PV-PWARCA). This proposed framework employs an LLM as a semantic arbitrator correcting stochastic outputs utilizing deterministic physical laws and meteorological context. Experimental results demonstrate the agent’s efficiency and efficacy, particularly in volatile PV generation regimes, by arbitrating between model outputs and physical constraints. PV-PWARCA maintains inference latency (1.5–3s) comparable to zero-shot TSFMs, while delivering an average NRMSE reduction of 4.3% to 15.8%.
- 25 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessTime Series Foundation Models for PV Power Forecasting: Benchmarking and Physics-Weather-Aware Agentic Correction
Time Series Foundation Models (TSFMs) offer a new approach for forecasting tasks via zeroshot generalization. At the same time, the advances in LLM reasoning capabilities show promise towards enhancing forecasting pipelines. In the present study, focusing on the photovoltaics (PV) power generation domain, prominent TSFMs are benchmarked against established task-specific supervised models across real-world datasets. The results demonstrate that TSFMs, particularly TabPFN and Chronos2, outperform supervised models in data-sparse settings. To address frequent forecast deviations of TSFMs caused by transient meteorological events, this work introduces a PV Physics-Weather-Aware Reasoning & Correction Agent (PV-PWARCA). This proposed framework employs an LLM as a semantic arbitrator correcting stochastic outputs utilizing deterministic physical laws and meteorological context. Experimental results demonstrate the agent’s efficiency and efficacy, particularly in volatile PV generation regimes, by arbitrating between model outputs and physical constraints. PV-PWARCA maintains inference latency (1.5–3s) comparable to zero-shot TSFMs, while delivering an average NRMSE reduction of 4.3% to 15.8%.
- Open Access
- 13 February 2025
- Published by MDPI AG in Sustainability
- Vol. 17 (4) , 1551
MAS-DR: An ML-Based Aggregation and Segmentation Framework for Residential Consumption Users to Assist DR Programs
The increasing complexity of energy grids, driven by rising demand and unpredictable residential consumption, highlights the need for efficient demand response (DR) strategies and data-driven services. This paper proposes a machine learning-based framework for DR that clusters users based on their consumption patterns and categorizes individual usage into distinct profiles using K-means, Hierarchical Agglomerative Clustering, Spectral Clustering, and DBSCAN. Key features such as statistical, temporal, and behavioral characteristics are extracted, and the novel Household Daily Load (HDL) approach is used to identify residential consumption groups. The framework also includes context analysis to detect daily variations and peak usage periods for individual users. High-impact users, identified by anomalies such as frequent consumption spikes or grid instability risks using IsolationForest and kNN, are flagged. Additionally, a classification service integrates new users into the segmented portfolio. Experiments on real-world datasets demonstrate the framework’s effectiveness in helping energy managers design tailored DR programs.
- 13 February 2025
- Published by MDPI AG in Sustainability
- Vol. 17 (4) , 1551
Open AccessMAS-DR: An ML-Based Aggregation and Segmentation Framework for Residential Consumption Users to Assist DR Programs
The increasing complexity of energy grids, driven by rising demand and unpredictable residential consumption, highlights the need for efficient demand response (DR) strategies and data-driven services. This paper proposes a machine learning-based framework for DR that clusters users based on their consumption patterns and categorizes individual usage into distinct profiles using K-means, Hierarchical Agglomerative Clustering, Spectral Clustering, and DBSCAN. Key features such as statistical, temporal, and behavioral characteristics are extracted, and the novel Household Daily Load (HDL) approach is used to identify residential consumption groups. The framework also includes context analysis to detect daily variations and peak usage periods for individual users. High-impact users, identified by anomalies such as frequent consumption spikes or grid instability risks using IsolationForest and kNN, are flagged. Additionally, a classification service integrates new users into the segmented portfolio. Experiments on real-world datasets demonstrate the framework’s effectiveness in helping energy managers design tailored DR programs.
- Open Access
European Union Innovation Efficiency Assessment Based on Data Envelopment Analysis
Though much attention is dedicated to the development of its research and innovation policy, the European Union constantly struggles to match the level of the strongest innovators in the world. Therefore, there is a necessity to analyze the individual efforts and conditions of the 27 member states that might determine their final innovative performance. The results of a scientific literature review showed that there is a growing interest in the usage of artificial intelligence when seeking to improve decision-making processes. Data envelopment analysis, as a branch of computational intelligence methods, has proved to be a reliable tool for innovation efficiency evaluation. Therefore, this paper aimed to apply DEA for the assessment of the European Union’s innovation efficiency from 2000 to 2020, when innovation was measured by patent, trademark, and design applications. The findings showed that the general EU innovation efficiency situation has improved over time, meaning that each programming period was more successful than the previous one. On the other hand, visible disparities were found across the member states, showing that Luxembourg is an absolute innovation efficiency leader, while Greece and Portugal achieved the lowest average efficiency scores. Both the application of the DEA method and the gathered results may act as viable guidelines on how to improve R&I policies and select future investment directions.
Open AccessEuropean Union Innovation Efficiency Assessment Based on Data Envelopment Analysis
Though much attention is dedicated to the development of its research and innovation policy, the European Union constantly struggles to match the level of the strongest innovators in the world. Therefore, there is a necessity to analyze the individual efforts and conditions of the 27 member states that might determine their final innovative performance. The results of a scientific literature review showed that there is a growing interest in the usage of artificial intelligence when seeking to improve decision-making processes. Data envelopment analysis, as a branch of computational intelligence methods, has proved to be a reliable tool for innovation efficiency evaluation. Therefore, this paper aimed to apply DEA for the assessment of the European Union’s innovation efficiency from 2000 to 2020, when innovation was measured by patent, trademark, and design applications. The findings showed that the general EU innovation efficiency situation has improved over time, meaning that each programming period was more successful than the previous one. On the other hand, visible disparities were found across the member states, showing that Luxembourg is an absolute innovation efficiency leader, while Greece and Portugal achieved the lowest average efficiency scores. Both the application of the DEA method and the gathered results may act as viable guidelines on how to improve R&I policies and select future investment directions.
- 10 June 2022
- Published by Springer Nature
- p. 301-313
Probabilistic Quantile Multi-step Forecasting of Energy Market Prices: A UK Case Study
- 10 June 2022
- Published by Springer Nature
- p. 301-313
Probabilistic Quantile Multi-step Forecasting of Energy Market Prices: A UK Case Study
- Open Access
- 3 August 2021
- Published by MDPI AG in Preprints.org
Utilizing Hybrid Feature Extraction, Tree-Based Learning and Data-Lightweight Model Re-Training for Accurate Short-Term and Day-Ahead Residential Load Forecasting
Residential load forecasting is one of the most important tasks of the overall supply management process in electrical grids, since it enables smart grid services such as demand response (DR). Hence, several approaches for accurate residential load forecasting have been proposed in the relevant literature. However, most of the existing methods focus on the forecasting performance and neglect other aspects of the problem like training time and model size (i.e. memory usage). In this paper, we introduce a new model for both short-term and day-ahead residential load forecasting. The model synthesizes an heterogeneous feature set, which is constituted by both automatically-selected lagged values from the load time series and manually-extracted temporal features. Then, the tree-based algorithm light gradient boosting machine (LGBM) is fed with the constructed feature set and used as a regression model. Finally, a data-lightweight strategy is used for retraining the proposed model, which leads to both high forecasting accuracy and low training times. The proposed model has been extensively evaluated on a large real-world residential load dataset. The experimental results indicate that the proposed model achieves both higher forecasting performance and lower training times and model sizes compared to state-of-the-art solutions.
- 3 August 2021
- Published by MDPI AG in Preprints.org
Open AccessUtilizing Hybrid Feature Extraction, Tree-Based Learning and Data-Lightweight Model Re-Training for Accurate Short-Term and Day-Ahead Residential Load Forecasting
Residential load forecasting is one of the most important tasks of the overall supply management process in electrical grids, since it enables smart grid services such as demand response (DR). Hence, several approaches for accurate residential load forecasting have been proposed in the relevant literature. However, most of the existing methods focus on the forecasting performance and neglect other aspects of the problem like training time and model size (i.e. memory usage). In this paper, we introduce a new model for both short-term and day-ahead residential load forecasting. The model synthesizes an heterogeneous feature set, which is constituted by both automatically-selected lagged values from the load time series and manually-extracted temporal features. Then, the tree-based algorithm light gradient boosting machine (LGBM) is fed with the constructed feature set and used as a regression model. Finally, a data-lightweight strategy is used for retraining the proposed model, which leads to both high forecasting accuracy and low training times. The proposed model has been extensively evaluated on a large real-world residential load dataset. The experimental results indicate that the proposed model achieves both higher forecasting performance and lower training times and model sizes compared to state-of-the-art solutions.
- Open Access
Incentive-Based Demand Response Framework for Residential Applications: Design and Real-Life Demonstration
In 2020, residential sector loads reached 25% of the overall electrical consumption in Europe and it is foreseen to stabilise at 29% by 2050. However, this relatively small increase demands, among others, changes in the energy consuming behaviour of households. To achieve this, Demand Response (DR) has been identified as a promising tool for unlocking the hidden flexibility potential of residential consumption. In this work, a holistic incentive-based DR framework aiming towards load shifting is proposed for residential applications. The proposed framework is characterised by several innovative features, mainly the formulation of the optimisation problem, which models user satisfaction and the economic operation of a distributed household portfolio, the customised load forecasting algorithm, which employs an adjusted Gradient Boosting Tree methodology with enhanced feature extraction and, finally, a disaggregation tool, which considers electrical features and time of use information. The DR framework is first validated through simulation to assess the business potential and is then deployed experimentally in real houses in Northern Greece. Results demonstrate that a mean 1.48% relative profit can be achieved via only load shifting of a maximum of three residential appliances, while the experimental application proves the effectiveness of the proposed algorithms in successfully managing the load curves of real houses with several residents. Correlations between market prices and the success of incentive-based load shifting DR programs show how wholesale pricing should be adjusted to ensure the viability of such DR schemes.
Open AccessIncentive-Based Demand Response Framework for Residential Applications: Design and Real-Life Demonstration
In 2020, residential sector loads reached 25% of the overall electrical consumption in Europe and it is foreseen to stabilise at 29% by 2050. However, this relatively small increase demands, among others, changes in the energy consuming behaviour of households. To achieve this, Demand Response (DR) has been identified as a promising tool for unlocking the hidden flexibility potential of residential consumption. In this work, a holistic incentive-based DR framework aiming towards load shifting is proposed for residential applications. The proposed framework is characterised by several innovative features, mainly the formulation of the optimisation problem, which models user satisfaction and the economic operation of a distributed household portfolio, the customised load forecasting algorithm, which employs an adjusted Gradient Boosting Tree methodology with enhanced feature extraction and, finally, a disaggregation tool, which considers electrical features and time of use information. The DR framework is first validated through simulation to assess the business potential and is then deployed experimentally in real houses in Northern Greece. Results demonstrate that a mean 1.48% relative profit can be achieved via only load shifting of a maximum of three residential appliances, while the experimental application proves the effectiveness of the proposed algorithms in successfully managing the load curves of real houses with several residents. Correlations between market prices and the success of incentive-based load shifting DR programs show how wholesale pricing should be adjusted to ensure the viability of such DR schemes.
- Open Access
A Tri-Layer Optimization Framework for Day-Ahead Energy Scheduling Based on Cost and Discomfort Minimization
Over the past few decades, industry and academia have made great strides to improve aspects related with optimal energy management. These include better ways for efficient energy asset management, generating great opportunities for optimization of energy distribution, discomfort minimization, energy production, cost reduction and more. This paper proposes a framework for a multi-objective analysis, acting as a novel tool that offers responses for optimal energy management through a decision support system. The novelty is in the structure of the methodology, since it considers two distinct optimization problems for two actors, consumers and aggregators, with solution being able to completely or partly interact with the other one is in the form of a demand response signal exchange. The overall optimization is formulated by a bi-objective optimization problem for the consumer side, aiming at cost minimization and discomfort reduction, and a single objective optimization problem for the aggregator side aiming at cost minimization. The framework consists of three architectural layers, namely, the consumer, aggregator and decision support system (DSS), forming a tri-layer optimization framework with multiple interacting objects, such as objective functions, variables, constants and constraints. The DSS layer is responsible for decision support by forecasting the day-ahead energy management requirements. The main purpose of this study is to achieve optimal management of energy resources, considering both aggregator and consumer preferences and goals, whilst abiding with real-world system constraints. This is conducted through detailed simulations using real data from a pilot, that is part of Terni Distribution System portfolio.
Open AccessA Tri-Layer Optimization Framework for Day-Ahead Energy Scheduling Based on Cost and Discomfort Minimization
Over the past few decades, industry and academia have made great strides to improve aspects related with optimal energy management. These include better ways for efficient energy asset management, generating great opportunities for optimization of energy distribution, discomfort minimization, energy production, cost reduction and more. This paper proposes a framework for a multi-objective analysis, acting as a novel tool that offers responses for optimal energy management through a decision support system. The novelty is in the structure of the methodology, since it considers two distinct optimization problems for two actors, consumers and aggregators, with solution being able to completely or partly interact with the other one is in the form of a demand response signal exchange. The overall optimization is formulated by a bi-objective optimization problem for the consumer side, aiming at cost minimization and discomfort reduction, and a single objective optimization problem for the aggregator side aiming at cost minimization. The framework consists of three architectural layers, namely, the consumer, aggregator and decision support system (DSS), forming a tri-layer optimization framework with multiple interacting objects, such as objective functions, variables, constants and constraints. The DSS layer is responsible for decision support by forecasting the day-ahead energy management requirements. The main purpose of this study is to achieve optimal management of energy resources, considering both aggregator and consumer preferences and goals, whilst abiding with real-world system constraints. This is conducted through detailed simulations using real data from a pilot, that is part of Terni Distribution System portfolio.
- 23 March 2021
- Published by Springer Nature in Computing
- Vol. 104 (1) , 209-250
An interdisciplinary approach on efficient virtual microgrid to virtual microgrid energy balancing incorporating data preprocessing techniques
- 23 March 2021
- Published by Springer Nature in Computing
- Vol. 104 (1) , 209-250
An interdisciplinary approach on efficient virtual microgrid to virtual microgrid energy balancing incorporating data preprocessing techniques
- 7 December 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 321-326
Design and Real-life Deployment of a Smart Nanogrid: A Greek Case Study
Microgrids are becoming one of the most promising building blocks for smart grids gaining an increasing attention in the last decade. From critical infrastructures such as hospitals, military bases, desolated or hard to reach areas to universities, research centers, and other industrial infrastructures, basic microgrid principles have been applied presenting enhanced reliability and added value business services to their operation. Recently, such application principles have been further scaled-down, presenting an even more challenging paradigm at single building infrastructure, the nanogrid. Nevertheless, even in such occasions, optimal operation is restricted by the commercial product services, that aren’t always easy to configure or extend towards enabling additional functionalities. This manuscript presents step by step the transformation of a smart home infrastructure into a flexible yet safe nanogrid that allows various control strategies in real-life operation while respecting the fail-safe mechanisms of the deployed commercial assets while introducing additional safety measures. After two years of operation and configuration, by only optimising the energy storage system and evaluating overall performance for almost half a year, promising results demonstrate the untapped potential of such deployments.
- 7 December 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 321-326
Design and Real-life Deployment of a Smart Nanogrid: A Greek Case Study
Microgrids are becoming one of the most promising building blocks for smart grids gaining an increasing attention in the last decade. From critical infrastructures such as hospitals, military bases, desolated or hard to reach areas to universities, research centers, and other industrial infrastructures, basic microgrid principles have been applied presenting enhanced reliability and added value business services to their operation. Recently, such application principles have been further scaled-down, presenting an even more challenging paradigm at single building infrastructure, the nanogrid. Nevertheless, even in such occasions, optimal operation is restricted by the commercial product services, that aren’t always easy to configure or extend towards enabling additional functionalities. This manuscript presents step by step the transformation of a smart home infrastructure into a flexible yet safe nanogrid that allows various control strategies in real-life operation while respecting the fail-safe mechanisms of the deployed commercial assets while introducing additional safety measures. After two years of operation and configuration, by only optimising the energy storage system and evaluating overall performance for almost half a year, promising results demonstrate the untapped potential of such deployments.
- Open Access
Benchmark Comparison of Analytical, Data-Based and Hybrid Models for Multi-Step Short-Term Photovoltaic Power Generation Forecasting
Accurately forecasting power generation in photovoltaic (PV) installations is a challenging task, due to the volatile and highly intermittent nature of solar-based renewable energy sources. In recent years, several PV power generation forecasting models have been proposed in the relevant literature. However, there is no consensus regarding which models perform better in which cases. Moreover, literature lacks of works presenting detailed experimental evaluations of different types of models on the same data and forecasting conditions. This paper attempts to fill in this gap by presenting a comprehensive benchmarking framework for several analytical, data-based and hybrid models for multi-step short-term PV power generation forecasting. All models were evaluated on the same real PV power generation data, gathered from the realisation of a small scale pilot site in Thessaloniki, Greece. The models predicted PV power generation on multiple horizons, namely for 15 min, 30 min, 60 min, 120 min and 180 min ahead of time. Based on the analysis of the experimental results we identify the cases, in which specific models (or types of models) perform better compared to others, and explain the rationale behind those model performances.
Open AccessBenchmark Comparison of Analytical, Data-Based and Hybrid Models for Multi-Step Short-Term Photovoltaic Power Generation Forecasting
Accurately forecasting power generation in photovoltaic (PV) installations is a challenging task, due to the volatile and highly intermittent nature of solar-based renewable energy sources. In recent years, several PV power generation forecasting models have been proposed in the relevant literature. However, there is no consensus regarding which models perform better in which cases. Moreover, literature lacks of works presenting detailed experimental evaluations of different types of models on the same data and forecasting conditions. This paper attempts to fill in this gap by presenting a comprehensive benchmarking framework for several analytical, data-based and hybrid models for multi-step short-term PV power generation forecasting. All models were evaluated on the same real PV power generation data, gathered from the realisation of a small scale pilot site in Thessaloniki, Greece. The models predicted PV power generation on multiple horizons, namely for 15 min, 30 min, 60 min, 120 min and 180 min ahead of time. Based on the analysis of the experimental results we identify the cases, in which specific models (or types of models) perform better compared to others, and explain the rationale behind those model performances.