Classification of biogas digesters based on their substrate using biogas patternsand machine learning (a case study on livestock manures of cow and chicken)

Agricultural waste management in the right way can contribute to saving energy and recyclingvaluable organic material, but unfortunately, agro-waste is a main source of environmentalpollution. Livestock manure is a large portion of agricultural waste that can be treated throughanaerobic digestion (AD) as sustainable waste management. As waste type is an importantfactor in the AD process, this study focuses on the potential of classifying the biogas patternsof cow and chicken manures as substrates using e-nose. The batch digesters were fed by threesubstrates including cow manure, chicken manure, and a combination of them in a ۱ :۱ ratio.The digesters of chicken manure and the combination produced high biogas on the first day butdeclined the production rate by over days and failed less than ۹ days. The digesters of cowmanure produced biogas at a stable rate for more than ۱۱ days. The biogas patterns wereintroduced to PCA and LDA. PCA plot showed that the Mq-۱۳۶ sensor, which is highsensitivity to H۲S and NH۴, was the most effective sensor to separate the digesters. Theconfusion matrix of the PCA-LDA model showed an overall classification accuracy of ۷۷.۶%with P value: ۰.۰۰۰۰۱, and recall of ۱۰۰% for the classifier of the digesters of chicken manure.Also, the biogas patterns were classified by different algorithms and validated by the k-foldcross-validation method. The SVM model presented the best accuracy of ۷۵.۹% by ۵-fold crossvalidation.

Keywords:

biogas , digester , cow manure , chicken manure , e-nose , classification , k-fold