Multi-Zonal Modeling in a Residential Apartment Using Physics Informed Long Short-Term Memory Approach
ALOK KUMAR THAKUR, Sameer Patel, Indian Institute of Technology Gandhinagar
Abstract Number: 280
Working Group: Indoor Aerosols
Abstract
Most IAP research focusing on modeling and material balance assumes well-mixed conditions, which is usually not the case, especially in bigger spaces. Studying the inter-zonal transport of pollutants and their governing factors provides critical insights into the fate and transport of pollutants from emission zones to different zones in a multi-zonal indoor space. The current work focuses on predicting PM concentrations in different zones of a residential apartment using PM levels measured by a monitor in one zone. The governing parameters here include inter-zonal flow and deposition rates. The work consists of two parts - (a) estimation of inter-zonal flow rates between different zones and (b) prediction of spatial PM concentrations using a deep learning model.
Inter-zonal flow rates are obtained using different optimization algorithms where initial conditions are based on the experimentally obtained datasets. The derived flow and deposition rates served as critical inputs for the subsequent stage, including predicting PM concentration in different zones from PM levels obtained from a single zone using a physics-informed long short-term memory approach (PI-LSTM) model. The ordinary LSTM models require a sufficient dataset to get trained and fail to adhere to the laws of physics. In contrast, PI-LSTM will require fewer datasets and will depend on physics constraints to improve the accuracy and reliability of the obtained results.
PM2.5 and CO2 transport experiments were performed to obtain the model's training and testing datasets. Validation of the results against the experimentally obtained data was done. Deposition rates for three zones are found using experimental datasets. Further, different optimization algorithms are tested to estimate the optimum value of the interzonal flow rate.