Predicting summer indoor temperatures in Nordic apartments considering heatwaves forecasts

Abstract

With global ambient air temperature rise, more intense, frequent, and much longer heatwaves are expected. This is associated with high indoor temperatures and overheating risks in apartments in the Nordic region. This study investigates the prediction of indoor temperatures in Nordic apartment buildings during summer heatwaves using outdoor weather forecasts to forecast indoor overheating risks. A comprehensive dataset of hourly indoor temperatures from over 20,000 apartments in the Helsinki region between 2018 and 2021 was used. The outdoor hourly parameters, including air temperature, humidity, and solar irradiance, were integrated with limited apartment characteristics such as size, number of rooms, and age. Three machine learning models—Long Short-Term Memory (LSTM), XGBoost, and Multivariate Linear Regression (MLR)—were employed to predict indoor temperatures for the next 24, 48, and 72 h. The performance of the models was evaluated using mtrics e.g., Mean Absolute Error (MAE) and Mean Squared Error (MSE. The XGBoost model achieved the highest accuracy with an MAE of 0.23 °C and MSE of 0.12 for the 24-h prediction. While LSTM showed superior performance under high-temperature conditions, compared to XGBoost, it could not capture hourly changes in indoor temperature and follow its pattern. The results provide insights into the challenges of predicting indoor temperatures in residential buildings during extreme heat events and highlight the importance of including outdoor weather forecasts for improved model accuracy. Its findings offer a novel approach to maintaining comfortable indoor thermal conditions and early forecasting of possible overheating in Nordic apartments during hot summers.