Basic Project Information:
Principal Investigator: Prof. D.Sc. Jonatan Lerga
Project Budget: 60.091,35 €
Project Title (in Croatian): Duboko učenje u upravljanju pametnim energetskim sustavima
Project Acronym: AI-ENERGO
Project Code: uniri-mzi-25-1
Scientific Field of the Project: Technical Sciences
Scientific Area: Computer Science
Project Start Date: 01 October 2025
Project End Date: 30 September 2029
Research Team: izv. prof. dr. sc. Nicoletta Saulig, izv. prof. dr. sc. Rene Prenc, izv. prof. dr. sc. Vedran Kirinčić, izv. prof. dr. sc. Ivan Volarić, izv. prof. dr. sc. Nino Krvavica, prof. dr. sc. Peter Kovács, dr. sc. Ana Vranković Lacković, Anna Maria Mihel, Lucija Žužić, Boris Gašparović, Luka Škrlj, Matko Glučina
Abstract:
As a key component of modern energy management systems, energy demand forecasting plays a crucial role in efficient resource allocation, grid stability, and the integration of renewables. Traditionally, classical techniques such as ARIMA and SARIMA have been used for energy demand prediction due to their simplicity and interpretability. However, these conventional methods often fail to capture dynamic interactions and nonlinear dependencies present in contemporary energy systems.
With recent advancements in Artificial Intelligence (AI), particularly in Machine Learning (ML) and Deep Learning (DL), energy demand forecasting has undergone a significant transformation. AI models that leverage real-time data, socioeconomic factors, and external variables have demonstrated superior performance compared to traditional approaches, not only in predicting energy demand but also in optimizing energy distribution, managing renewables, and reducing operational costs.
Furthermore, the combination of AI and advanced technologies, such as smart grids and the Internet of Things (IoT), has further enhanced energy efficiency and system reliability across various industries. These technologies enable real-time monitoring and control of energy consumption, leading to better decision-making and more sustainable energy usage.
This project aims to apply state-of-the-art AI techniques to analyze data from Croatia’s Energy Management Information System, with the objective of improving its efficiency and adaptability. The research will take into account the specific characteristics of the Croatian energy system, addressing critical challenges such as data quality, measurement gaps, computational complexity, and privacy concerns.
By tackling these challenges, the project seeks to unlock the full potential of AI in enhancing the efficiency and sustainability of Croatia’s energy infrastructure, ultimately contributing to a more resilient and intelligent energy management system.
Project Objectives:
C.1. Improving energy demand forecasting using artificial intelligence
C.2. Detection and processing of inaccurate and incomplete data in the national Energy Management Information System
C.3. Proposals for upgrading the Croatian Energy Management Information System using artificial intelligence
State of the Art and Motivation:
Energy demand forecasting is a key element of modern energy system management, enabling efficient resource allocation, grid stability, integration of renewable sources, and cost reduction. For decades, traditional techniques such as econometric models or Autoregressive Integrated Moving Averages (ARIMA) [1] have been widely used due to their simplicity. However, these approaches face limitations in handling dynamic interactions, nonlinear dependencies, and the growing complexity of contemporary energy systems [2].
Recent studies demonstrate that Artificial Intelligence (AI)-based methods effectively address these challenges. For instance, [3] examined the shortcomings of ARIMA models, highlighting their inability to predict sudden changes in energy consumption caused by extreme weather conditions or shifts in energy policies. Machine Learning (ML)-based approaches can model complex patterns in data and incorporate a variety of influencing factors, including social behavior, economic activity, and meteorological conditions [4]. According to recent research, AI techniques outperform classical statistical models in terms of both accuracy and scalability, as well as in analyzing large volumes of real-time data [5].
Energy management and consumption monitoring have been transformed by combining AI with Internet of Things (IoT) technologies. AI algorithms can process real-time energy consumption data generated by IoT devices to predict and optimize energy use [6]. Integrating AI with smart grid technology has been shown to improve energy efficiency, reduce losses, and increase system stability [7]. Furthermore, the accuracy and resilience of short- and long-term energy demand forecasts have been enhanced through hybrid AI models that combine AI with traditional approaches [8].
Challenges related to transparency and interpretability of AI models have been mitigated by the emergence of Explainable AI (XAI), which enables decision-makers to understand model predictions and trust the results [9]. XAI methods such as SHAP and LIME identify the main factors influencing energy demand forecasts, supporting informed decision-making in energy planning [10].
Recent research also highlights the potential of AI to support renewable energy integration by predicting resource availability and improving grid management [11]. To enhance forecast accuracy for renewable energy production, AI models leverage historical, geographical, and meteorological data [12].
In conclusion, AI-driven energy demand forecasting offers unprecedented opportunities to develop robust, efficient, and sustainable energy systems. AI has the potential to significantly impact the future of energy management by addressing current challenges, particularly in the Croatian energy sector, where the application of AI remains limited.
Methodology:
1. Data Collection (WP1):
The project leader has secured access to the Croatian National Energy Management Information System, with the Faculty of Engineering, University of Rijeka, formalizing cooperation through an agreement with the competent agency. The system provides data on energy consumption in public-sector buildings (e.g., those owned by cities, counties, the Government of the Republic of Croatia, schools, and other public institutions). In addition to energy consumption data, other relevant datasets—such as meteorological, demographic, and regulatory information—will be collected to improve the effectiveness of the developed AI solutions.
2. Data Preprocessing and Transformation (WP2):
Data in the national energy management system originate from various sources. Besides remote automated measurements, much of the data is entered manually (e.g., from monthly utility bills), which introduces errors. Consequently, data are marked as “validated” only after verification by experienced operators.
Despite this verification step, incomplete records and outliers remain common and must be detected and processed. For AI model training, the data will undergo preprocessing steps including scaling, normalization, dimensionality adjustments, feature extraction, and identification of inter-feature relationships. Additional decomposition and evaluation of various training approaches (e.g., raw time-series, time-frequency transformations, or decomposed components) will also be carried out.
3. AI Model Development and Training (WP3):
Following data preparation, state-of-the-art AI architectures will be selected and trained, with a focus on hybrid methods that combine statistical approaches with deep learning. To maximize prediction accuracy and precision, different hyperparameter optimization strategies will be applied.
4. Performance Analysis of AI Models (WP4):
The trained models will be validated and tested on historical energy consumption data from the national energy management system. Performance will be evaluated using standard metrics such as MSE, MAE, accuracy, precision, F1-score, and others, and compared with existing AI solutions reported in the scientific literature. Additionally, the computational efficiency of the developed AI models and their potential integration into the national energy management system will be assessed.
5. Additional Aspects of AI-Driven Energy Management (WP5):
Beyond technical and computational considerations, other relevant aspects will be addressed, including modeling user behavior, developing strategies to reduce peak loads, addressing data privacy and security concerns, and ensuring compliance with energy regulations.
By following this structured approach, the project will contribute to the development of a next-generation energy management system that is more efficient, sustainable, and adaptive to dynamic energy demand patterns.
References:
[1] Mystakidis, Aristeidis, Paraskevas Koukaras, Nikolaos Tsalikidis, Dimosthenis Ioannidis, and Christos Tjortjis. ”Energy Forecasting: A Comprehensive Review of Techniques and Technologies.” Energies 17, no. 7 (2024): 1662.
[2] Singh, Rajesh, Kuchkarbaev Rustam Utkurovich, Ahmed Alkhayyat, G. Saritha, R. Jayadurga, and K. B. Waghulde. ”Machine Learning Applications in Energy Management Systems for Smart Buildings.” In E3S Web of Conferences, vol. 540, p. 08002. EDP Sciences, 2024.
[3] Wang, Xinlin, HaoWang, Binayak Bhandari, and Leming Cheng. ”AI-empowered methods for smart energy consumption: A review of load forecasting, anomaly detection and demand response.” International Journal of Precision Engineering and Manufacturing-Green Technology.
[4] Kumar, Chandan, and Prakash Chittora. ”Secure and Privacy-Preserving Framework for IoTEnabled Smart Grid Environment.” Arabian Journal for Science and Engineering 49, no. 3 (2024): 3063-3078.
[5] Guo, Jiaxun, Manar Amayri, Wentao Fan, and Nizar Bouguila. ”A scaled dirichlet-based predictive model for occupancy estimation in smart buildings.” Applied Intelligence (2024): 1-16.
[6] Huang, Zhendai, Zhen Zhang, Cheng Hua, Bolin Liao, and Shuai Li. ”Leveraging enhanced egret swarm optimization algorithm and artificial intelligence-driven prompt strategies for portfolio selection.” Scientific Reports 14, no. 1 (2024): 26681.
[7] Zhao, Pei, Shaojun Zhang, Paolo Santi, Dingsong Cui, Fang Wang, Peng Liu, Zhaosheng Zhang et al. ”Challenges and opportunities in truck electrification revealed by big operational data.” Nature Energy (2024): 1-11.
[8] Ribeiro, Matheus Henrique Dal Molin, Ramon Gomes da Silva, Sinvaldo Rodrigues Moreno, Cristiane Canton, Jos´e Henrique Klein¨ubing Larcher, Stefano Frizzo Stefenon, Viviana Cocco Mariani, and Leandro dos Santos Coelho. ”Variational mode decomposition and bagging extreme learning machine with multi-objective optimization for wind power forecasting.” Applied Intelligence 54, no. 4 (2024): 3119-3134.
[9] Boresta, Marco, Diego Maria Pinto, and Giuseppe Stecca. ”Bridging operations research and machine learning for service cost prediction in logistics and service industries.” Annals of Operations Research (2024): 1-27.
[10] Clement, Tobias, Hung Truong Thanh Nguyen, Nils Kemmerzell, Mohamed Abdelaal, and Davor Stjelja. ”Beyond explaining: XAI-based Adaptive Learning with SHAP Clustering for Energy Consumption Prediction.” arXiv preprint arXiv:2402.04982 (2024).
[11] Huang, Chicheng, Serhat Y¨uksel, and Hasan Din¸cer. ”A Novel Fuzzy Model for Knowledge-Driven Process Optimization in Renewable Energy Projects.” Journal of the Knowledge Economy (2024): 1-33.
[12] Lakhdar, Laib, and Toufik Tayeb Naas. ”The Machine Learning for Predicting Gas Turbine Performance in Naval Vessels.” ITEGAM-JETIA 10, no. 49 (2024): 187-193.