Anaerobic Food Waste Treatment Technology Performance Enhancement And Dynamic Subsidy Decision Analysis

Food waste(FW)is an alarming global problem.Most FW ends up in landfills,where it is converted into CH₄,CO₂,and other harmful greenhouse gases.In the next25 years,FW problems will be especially acute in Asia,where population growth and economic development are increasing rapidly.Anaerobic Digestion(AD)technology is a popular method for treating FW.However,the anaerobic FW treatment industry in China is confronted with many challenges,included poorly designed government subsidy mechanisms and low technology performance issues.These issues threaten the sustainable operation of industrial-scale anaerobic FW treatment facilities.To address these issues,this thesis focused on three areas.Firstly,technology performance analysis–defined based on innovation and efficiency–was conducted.Regarding innovation,natural language processing was used to analyze text data from3,186 patents to identify technology trends.Regarding efficiency,data envelopment analysis(DEA)and stochastic gradient boosting(SGB)were used to analyze 373 AD projects in Germany and the USA and identify the determinants of these projects'efficiency.Secondly,based on the technology performance results,two AD facilities co-digesting food waste in Hainan and Shenzhen were selected as case studies.Several years of time-series data from these projects were analyzed in-depth.Thirdly,based on the operational bottlenecks identified in the case studies,a subsidy decision analysis model based on machine learning was constructed.This was done in order to provide decision support for setting a reasonable subsidy amount to support these projects.The results were as follows:(1)Firstly,Emerging technology included pre-treatment,biomethane utilization(for instance upgrading technology such as“carbon dioxide absorbing”),and digestate utilization.Secondly,although results were different for Germany and the USA,the common factors affecting AD technology performance included pre-treatment steps(for example,the total number of steps),pre-treatment type(e.g.crushing or slurrying),anaerobic digester type(e.g.single stage AD or plug flow),and whether co-digestion occurred.(2)The Hainan project had poor economic performance and required subsidy:unit costs ranged between 4.7-20.5 RMB/m³,but the biomethane price only ranged from 2.7-3.0 RMB/m³,which was insufficient to cover costs.(3)Random Forest and XGboost machine learning models provided the best performance in modelling future biomethane output,with the best out-of-sample R² scores ranging from 0.83 to 0.86.These models were used to predict biomethane output one month into the future in order to provide a dynamic required subsidy recommendation.Based on this model,the required subsidy was estimated to range between 1.10-4.35 RMB/m³.In addition to these findings,the entire code behind the analysis conducted in this paper is entirely replicable,open-source,and available freely online.This means that researchers can build upon the analysis and adapt it to local conditions involving anaerobic food waste treatment.