First Exploration on Applying the LSTM Model to Calculate the Bioaerosol Trajectory

Abstract

The pandemic of COVID-19 raises worldwide consideration from the designers and managers of closed environments. Bioaerosol transmission is a critical transmission route, thus it is essential to calculate the bioaerosol trajectories rapidly. One of the artificial intelligence methods, the Long Short Term Memory (LSTM) model, is suitable for handling and predicting significant events with very long intervals and delays in time series, which might be ideal for calculating bioaerosol trajectories in closed environments. In this study, a lightweight single-layer LSTM model was first adopted to explore the application of the calculation of the bioaerosol trajectories. The performance of the LSTM model was assessed using the mean squared error (MSE) as an indicator. The supervised learning training of the LSTM model was carried out for multiple groups of bioaerosol spatial data extracted from solving the motion equation on the bioaerosol. When the motion equation without the Discrete Random Walk (DRW) model generated the training set, the LSTM model could better predict the trajectory and demonstrated a certain transferability. However, when the motion equation with the DRW model developed the training set, the calculated trajectories showed a significant derivation from those obtained by directly solving the motion equation. The unpredictability of the DRW model might cause the phenomenon. This study provides an application prospect for the application of the LSTM model on the acceleration of the trajectory calculation for the early warning and rapid design.

Keywords: LSTM model, bioaerosol trajectory, Lagrange method, DRW model, Discrete phase model