Comparative evaluation of ANN and NARX models for localized air temperature prediction in Al Ain, UAE

Typical Meteorological Years (TMY) and Test Reference Year (TRY), while commonly used in building energy modeling software, often fail to accurately represent localized weather conditions due to their reliance on averaged data from remote sources. This can lead to inaccuracies in energy demand assessments, particularly in regions with extreme climates. To address this limitation, this study proposes a novel framework for generating synthetic localized urban climate weather profiles that integrate both local context and site-specific microclimatic conditions.

Focusing on Al Ain, UAE, a city with a hot, arid climate and high reliance on cooling energy, this paper combines Neural Network modeling, building energy simulation, and microclimate modeling to quantify the influence of localized weather data on building energy performance. The proposed five-stage framework utilizes a Neural Network model trained on long-term weather records to predict localized air temperature. This data is then used to generate a weather file compatible with the microclimate modeling software ENVI-met. The resulting microclimatic parameters are then fed back into the weather file generator, creating a dataset compatible with the building energy modeling software DesignBuilder.

This iterative process allows for the exchange of boundary conditions between microclimate and building energy models, resulting in a highly localized weather profile that reflects the unique meteorological phenomena of the site. By utilizing this refined weather data, the study demonstrates a more accurate assessment of building energy performance and highlights the potential for significant energy savings compared to using standard TMY data.