Predictive and correlational analysis of heating energy consumption in four residential apartments with sensitivity analysis using long Short-Term memory and Generalized regression neural network models

• Evaluated heating energy predictions using LSTM and GRNN for four residential units. • Solar irradiance and CO 2 levels had the biggest impact on heating energy usage. • Introduced analysis of settings, behavior, and energy use to improve efficiency. • This study conducted an impact factor analysis on the seven input parameters. The aim of this study is to explore several approaches to analyze how local weather conditions, indoor CO 2 levels, and façade opening ratios affect the heating energy usage of a residential structure. To achieve this, the study uses two techniques: long short-term memory and Generalized Regression Neural Network methods. By applying these methods, the study suggests methods to predict the impact factors and evaluate the strength of their correlation with the actual heating energy consumed by the building. The study used both LSTM and GRNN algorithms to forecast the performance of heating energy usages in residential buildings using mechanical and natural ventilation systems. The results described that both models had low average error rates, ranging from 3.36% to 6.12%. However, the LSTM model had a better correlation with measured data. The examination of impact factor indicated that outside thermal and humidity factor had the most primarily influences for heating energy usage, while other environmental factors also significantly affected the residential building’s performance. Solar irradiance, wind velocity, and façade opening ratio had limitations in influencing heating performance because occupants may find it challenging to adjust ventilation rates in extreme weather conditions. Additionally, these factors could not affect heating energy consumption independently.