Resilient Machine Learning-Based Forecasting of Electricity Demand in the Context of Climate Change - A Case Study of The Dynamic Weather Patterns of The Nordic Countries

Aug 4, 2026

Upload by Asher

This thesis compares traditional econometric forecasting models, especially SARIMAX, with machine learning approaches, including Random Forest and Gradient Boosting, for electricity demand prediction in Norway, Sweden, and Finland. SARIMAX provides strong baseline results when time-series data is stationary, while machine learning models perform better in more complex conditions by incorporating climate-change variables. Gradient Boosting delivers the highest resilience and accuracy under climate-influenced demand patterns. Results show that adding climate variables improves accuracy and ensemble strategies combining SARIMAX with Gradient Boosting achieve the lowest MSE and RMSE, validated through time-series cross-validation. The study recommends broader climate inputs and advanced AI methods such as RNNs and LSTMs to capture temporal dependencies and long-term dynamics, supporting more adaptive energy planning and policymaking.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages91

Reading Time3 h 49 min

AccessFree to download

Abstract

Acknowledgment

List of Figures

Master’s Thesis 2024 30 ECTS School fo Economics and Business Resilient Machine Learning-Based Forecasting of Electricity Demand in the Context of Climate Change. A Case Study of The Dynamic Weather Patterns of The Nordic Countries. Samuel Kwesi Kumi Master of Science in Applied Economics and Sustainability Acknowledgment This thesis embodies my efforts and the collaboration and support of many individuals. I am pleased to express my heartfelt gratitude to all of them. First and foremost, I would like to express my sincere gratitude to my supervisor, Prof. Olvar Bergland, for his continuous support of my study and his patience, motivation, and immense knowledge. Without your guidance, I could not have imagined having the right tools to complete this body of work. Special love to my family, especially Joana and Josephine, for holding me down to the very end. To Edwin, thank you for the support and advice. I could not have made it without you. I would also like to thank Biney and Andoh-Appiah for their insightful comments and challenging questions, encouraging me to widen my research from various perspectives. I am deeply grateful to the European Network of Transmission System Operators for Electricity (ENTSO-E) for providing comprehensive and accessible data and their valuable feedback, which significantly enhanced the depth and quality of my work. Finally, I would like to acknowledge the support of my colleagues and friends, who have constantly motivated me throughout this journey. Their camaraderie and assistance have been integral to the completion of this thesis. I am profoundly grateful to everyone who has contributed in one way or another. Thank you. Abstract This study explores the performance comparison of traditional econometric models, specifically SARIMAX, against machine learning models, Random Forest, and Gradient Boosting in forecasting electricity demand across Norway, Sweden, and Finland. SARIMAX models demonstrated robust baseline performance, particularly effective when time series data exhibited stationarity. In contrast, machine learning models excelled in more complex scenarios, accounting for climate change variables. Gradient Boosting showed superior resilience and accuracy in managing climate-influenced demand patterns. Key findings indicate that integrating climate variables into forecasting models enhances accuracy, with ensemble methods (combining SARIMAX and Gradient Boosting) yielding the lowest Mean Squared Error (MSE) and Root Mean Squared Error (RMSE) . This underscores the value of leveraging diverse modeling techniques for more robust predictions. The study also employed time series cross-validation to ensure model robustness, with Gradient Boosting consistently outperforming Random Forest in error metrics. The research highlights the necessity of combining traditional econometric models with advanced machine learning techniques to improve forecast accuracy, especially under the complexities introduced by climate change. It recommends incorporating a broader range of climate variablesand exploring advanced AI methods, such as Recurrent Neural Networks (RNNs) and Long ShortTerm Memory Networks (LSTMs), for better capturing temporal dependencies and long-term patterns. These findings support the development of more adaptive and precise electricity demand forecasting models, essential for effective energy management and policymaking in the face of evolving climate conditions. List of Figures Figure 1: The trend of Global Surface Air Temperature Anomalies from 1940 to 2023 4 Figure 2: Global Distribution of Population with Access to Electricity in 2020 7 Figure 3: Fall in EU coal and gas electricity generation resulting in power sector 9 emissions reduction by 19% Figure 4: A comparative bar chart of electricity consumption from fossil fuels, nuclear, 10 and renewable sources across different countries in 2022, showcasing the diversity in energy mixes world

FAQs

Which models are compared for forecasting electricity demand?

The study compares SARIMAX with Random Forest and Gradient Boosting models across Norway, Sweden, and Finland.

How do the models differ in performance under changing climate conditions?

SARIMAX performs robustly when time-series data is stationary, while machine learning models excel when climate-related variables make forecasting scenarios more complex.

What is the main finding about using climate variables and ensembles?

Integrating climate variables improves prediction accuracy, and ensemble methods combining SARIMAX with Gradient Boosting produce the lowest MSE and RMSE values.

Aug 4, 2026

Upload by Asher