As the world moves toward more sustainable energy systems, the heating sector, one of the largest energy consumers, must adapt. District heating offers a system-level solution that enables efficient energy use and easier integration of renewable and recycled heat sources. Accurate heat load profiles are essential for effective long-term operation, planning and expansion of district heating systems. This thesis aimed to improve long-term heat load forecasting by examining current and future demand drivers and comparing advanced data-driven techniques to the industry-standard regression approach. The influencing factor analysis confirmed temperature as the dominant driver of heat demand, with time-based features reflecting behavioural patterns and emerging socio-economic factors expected to become increasingly important. For the methodological comparison, three models were developed: a baseline regression model based on daily average temperatures using polynomial regression, an enhanced regression model incorporating hourly data and a two-layer structure to capture intra-day variation, and a machine learning model based on XGBoost. While the regression models offered transparency and ease of modification, XGBoost consistently outperformed both across all evaluation metrics. It more accurately captured seasonal and behavioural patterns and provided more realistic results under simulated future climate scenarios. These findings highlight the potential of advanced data-driven methods in developing robust and adaptable heat load forecasts for future district heating systems.

Se filen