关键词: 混合模型, 机理模型, 随机森林回归, 厌氧发酵, 甲烷产量

Abstract:

As a mass and energy transfer process involving microorganisms, the complexity, time variability and severe nonlinearity of anaerobic fermentation process greatly reduced the accuracy and practicability of the prediction models established on pure mechanism or pure data. In order to solve this problem, taking the methane production process of cow dung fermentation as the research object, a hybrid model for predicting gas production by anaerobic fermentation was proposed. Firstly, the clear kinetic equations of hydrolysis, acidification hydrogen production and acetic acid production were established, and then the data-driven model of methane production and liquid-gas mass transfer process was constructed combined with Random Forest Regression(RFR). And the Sparrow Search Algorithm(SSA) was used to optimally estimate the super parameters of RFR algorithm. Finally, two sub-models, the kinetic model of the fermentation mechanism and the random forest regression model, were cascaded to form a hybrid model for the whole process of anaerobic fermentation biogas production prediction.The hybrid model was used to simulate the anaerobic digestion experiment of cow manure. The results showed that the methane production rate using the hybrid model simulation was in good agreement with the experimental data, and the prediction accuracy of the mechanism and data-driven mixed model(SSA-RFR-ADM) was close to the prediction accuracy of the complex pure mechanism model. The mean squared error of the model prediction was 0.003 035. The proposed hybrid model effectively overcame the limitation of single model, and could be used as a proxy model of complex pure mechanism model, providing correct theoretical guidance for practical anaerobic fermentation methane production process.

Key words: hybrid model, mechanism model, random forest regression, anaerobic fermentation, methane yield