Sparsity and mixing effects in deep learning predictions of temperature and humidity.

This article focuses on applying Long Short-Term Memory (LSTM) deep learning models to predict indoor environmental conditions—specifically temperature, relative humidity, and velocity magnitude—in a ventilated confined space modeled as a rectangular room with an air conditioning (AC) unit. Using high-resolution computational fluid dynamics (CFD) simulations based on the compressible Navier–Stokes equations, the study generated data to train and validate the LSTM model, examining how data sparsity and turbulent flow physics affect prediction accuracy. Results indicate that the LSTM model predicts velocity magnitude robustly even at higher data sparsity, while temperature and relative humidity predictions remain accurate only at lower sparsity levels, with errors increasing notably at sparser sampling intervals, especially in complex airflow regions. The study highlights that environmental conditions shortly after AC activation can slow respiratory droplet evaporation, potentially increasing airborne pathogen transmission risk, and suggests that improved data granularity or model refinement is needed for precise forecasting in turbulent mixing zones to optimize HVAC system performance and indoor air quality management.