关键词: 牛粪沼渣, 含氨臭气, 原位自净化, 好氧发酵, 机器学习

Abstract:

The burgeoning livestock industry in China has led to the annual generation of livestock manure reaching 3.05 billion tons, highlighting the urgent need for effective waste management strategies. Considering a 70% collection efficiency, approximately 2.135 billion tons of manure require annual processing. The anaerobic digestion process, a common treatment method, generates an additional 130 million tons of biogas residue, which is challenging to manage because of its high organic content, high moisture content, and tendency to emit foul odors. Aerobic fermentation presents a sustainable solution by transforming residual organics into humic substances, thereby enhancing soil properties and reducing reliance on chemical fertilizers. However, odororous emissions (primarily ammonia) from aerobic composting have limited the widespread adoption of this technology. This study investigated the use of mature cow manure biogas residue, rich in nitrifying bacteria, to purify ammonia-containing odors emitted during the early stages of composting, thereby retaining nitrogen within the compost and reducing investment and operational costs. Machine learning, particularly Multilayer Perceptron (MLP) models, has revolutionized traditional odor monitoring and management by effectively capturing the complex relationships between odor components, reducing dependency on domain knowledge, and integrating diverse data sources for comprehensive analysis and prediction. This study employs an MLP model, along with SHAP (SHapley Additive exPlanations), to predict the interactive effects of ammonia inlet concentration, airflow rate, and absorption column operation time on ammonia removal efficiency, providing a reference for practical applications of this technology. This study used fresh cow manure biogas residues obtained from a dairy farm in Tianjin, China. After 48 days of aerobic fermentation until maturity, mature cow manure residue was obtained. Mature cow manure biogas residue, characterized by physicochemical properties such as pH, moisture content, Volatile Solids (VS) to Total Solids (TS) ratio, carbon and nitrogen content, and carbon-nitrogen ratio, was used as an adsorbent for ammonia odor removal. The experimental setup included a gas cylinder, gas mixing chambers, flowmeters, gas absorption bottles, and absorption column filled with mature biogas residue. Dry ammonia gas was used to simulate ammonia-containing odors, mixed with air, and passed through the absorption column for adsorption and biological purification using mature biogas residue. The experiment was conducted at room temperature using two sets of identical devices by varying the effective volume and the ammonia inlet concentration. The low-concentration group demonstrated a high removal efficiency for simulated ammonia odors, reaching 99.19% on the first day and stabilizing at approximately 97.60% after 13 d of operation. The system showed resilience to concentration fluctuations, maintaining removal efficiencies between 96.85% and 99.25% even when the concentration increased to 200 mg∙m⁻³. However, further increases in concentration led to a decline in efficiency due to potential toxic effects on microbes. The high-concentration group showed an initial removal rate of 45.1%, which increased to approximately 95.60% after 10 d, indicating system adaptation to the simulated ammonia odor. Stable operation resulted in removal rates over 98.04% for ammonia concentrations between 100 and 500 mg∙m⁻³. At 800 mg∙m⁻³, the removal rate dropped to 77.62% but recovered to over 90.44% within four days. A further increase to 1200 mg∙m⁻³ led to a sustained decrease in removal efficiency, suggesting an exceedance in the processing capacity of the absorption column. The MLP model, based on the experimental data, effectively predicted the ammonia removal performance, with an R² value of 0.95. The model predictions were accurate, with most errors less than 5%, indicating their suitability for predicting the ammonia removal rates. The SHAP analysis revealed the importance of the input features, with the mature biogas residue volume, pH, ammonia inlet concentration, airflow rate, and absorption column operation time influencing the output. The SHAP waterfall diagram and scatter plots highlight the interactions between features, such as the impact of airflow rate and ammonia concentration on the absorption column operation time and the consequent effects on the ammonia removal efficiency. The findings of this study underscore the potential of mature cow manure biogas residue as an effective medium for ammonia odor removal during aerobic composting. The observed high removal efficiency suggests that mature biogas residue can serve as a cost-effective and environmentally friendly alternative to traditional chemical treatments. The presence of nitrifying bacteria within the residue plays a crucial role in the biological purification process by converting ammonia into non-volatile ammonium compounds, thus reducing odor emissions. The resilience of the system to varying ammonia concentrations is particularly noteworthy. The ability to maintain high removal efficiencies despite fluctuations in inlet concentration indicates that the mature biogas residue can adapt to different operational conditions, making it suitable for real-world applications in which ammonia emissions may vary. However, the observed decline in efficiency at extremely high concentrations highlights the need for careful monitoring and control of the inlet conditions to prevent microbial toxicity and ensure optimal performance. The integration of machine learning models, specifically the MLP model, provides a powerful tool for predicting and optimizing ammonia removal processes. The high accuracy of the model predictions, as evidenced by the R² value of 0.95, demonstrates its potential for use in real-time monitoring and control systems. By leveraging the SHAP analysis, this study offers valuable insights into the relative importance of different input features, enabling more informed decision-making and process optimization. The study concluded that mature cow manure biogas residue could rapidly initiate and effectively remove simulated ammonia odors when maintained at a moisture content of 55%‒60%. Under conditions of 30‒120 s empty bed residence time and ammonia concentrations of 50‒500 mg∙m⁻³, the removal rates remained stable at 95.47%‒100.00%, demonstrating the shock resistance of the system. However, excessive loads can lead to microbial toxicity and reduced removal rate. The MLP model, with an R² value of 0.95, can predict system performance and optimize ammonia removal by adjusting airflow rates, concentrations, and acclimation times, reducing experimental costs and providing optimal parameters for odor self-purification during aerobic fermentation. Future research will focus on improving the completeness and efficiency of odor treatment during aerobic fermentation, enhancing machine learning model capabilities, and exploring intelligent monitoring systems integrated with various sensing technologies and smart algorithms for the real-time monitoring and regulation of odor emissions. The implications of this research extend beyond the immediate context of ammonia odor removal. By demonstrating the efficacy of mature cow manure biogas residue and the utility of machine-learning models, this study contributes to the broader field of sustainable waste management and environmental protection. These findings highlight the potential of integrating biological treatment methods with advanced data analytics to address complex environmental challenges, paving the way for more sustainable and efficient waste-management practices.

Key words: cow manure biogas residue, ammonia odor, in-situ self-purification, aerobic fermentation, machine learning predictions