Abstract: Hydrogen sulfide (H2S) is a key odorous pollutant in solid waste treatment facilities, often leading to odor nuisance and environmental complaints. Therefore, accurate prediction of H2S is essential for odor pollution risk prevention and control. Based on hourly monitoring data of H2S concentrations collected from three stations at a solid waste treatment site in Shanghai from 2018 to 2021, this study proposed a hybrid model combining Empirical Mode Decomposition (EMD) and Long Short-Term Memory (LSTM) networks, namely the EMD-LSTM model, for predicting H2S concentrations at the base. The analysis showed that while the average temperatures at the three sites were similar, wind speeds differed significantly. H2S concentrations fluctuated considerably across all sites, with the southern boundary site, adjacent to the municipal solid waste landfill operation area, exhibiting significantly higher H2S concentrations than the other sites. Using wind speed, wind direction, temperature, humidity, air pressure, and ammonia concentration as input features, the EMD-LSTM model achieved accurate prediction of H2S concentrations. The prediction results demonstrated that the model performed well in forecasting concentrations for the next hour, with acceptable levels of mean absolute error (MAE: 1.71-13.13 μg/m3) and coefficient of determination (R2： 0.74-0.84) across the three boundary sites. The model performed best at the northern boundary site, where emissions were relatively stable (MAE: 1.71 μg/m3). The results verify the effectiveness and generalization capability of the model, providing technical support for precise early warning of odor pollution and emergency management decision-making at solid waste disposal bases.

Key words: deep learning, hydrogen sulfide, empirical mode decomposition, long short-term memory network, prediction