Evaluation of artificial neural network models for online monitoring of alkalinity in anaerobic co-digestion system

Abstract

Compared to pH monitoring during the anaerobic digestion process, alkalinity as an indicator could provide earlier warning for instability of digestion process, which is very important for efficient operation of biogas digesters, especially for multiple feeding substances. However, the online monitoring of alkalinity is still unavailable until now. In this study, available online measured parameters such as pH, oxidation and reduction potential (ORP), and electrical conductivity were selected as inputs, and the soft sensor method based on artificial neural network (ANN) was applied for alkalinity modeling to develop an online monitoring strategy. The dataset was obtained from a 6 month continuously operating anaerobic co-digestion system of cow manure, corn straw, and fruit and vegetable waste, and splited randomly by cross-validation. The results show that the optimum ANN model for total alkalinity prediction is 3-2-1 structure based on back propagation-feedforward neural network. The constructed ANN model was proved to be reliable through the predictive accuracy analysis and sensitivity analysis. The coefficient of determination (R2) of 0.9948 was obtained. ORP is the most significant model factor with the highest sensitivity degree. The online alkalinity monitoring may effectively prevent the failure of anaerobic digestion process and improve the anaerobic digestion efficiency practically.