Fast prediction of odor concentration along pig manure chain based on machine learning: Monitoring 20 instead of over 100 odorous substances

Abstract

Pig production is the main source of odor emissions. Owing to the complex composition of odors and the time-consuming and laborious process of measuring odor concentration (OC), identifying odorous substances, and quantifying odor concentration are technical bottlenecks in odor reduction. This study innovatively constructed a machine learning (ML) model to predict OC based on key odorous components. Over 400 gas samples were collected from the whole pig manure management chain in different regions, and 128 odor components were determined. The prediction results showed that extreme random tree regression ML had superior predictive performance for OC, with a better determination coefficient (R² > 0.8) and fewer features. 20 key odor substances out of over 100 components were the important features contributing to OC based on the Shapley additive explanation. Dimethyl sulfide, ammonia, hydrogen sulfide, ethyl sulfide, acetylene, and hexaldehyde were found to have the most significant impact on OC. This method can be easily extended to other types of farms, such as cattle and chicken, and provides a scientific basis for the research and development of qualitative and quantitative odorous substances and equipment for the rapid determination of odor concentration.