Self learning bioreactors design development and deployment

Abstract

Biogas production (BGP) through anaerobic digestion (AD) is an complex, nonlinear biological process where understanding its dynamics is crucial for effective newlineprocess control and optimization. This thesis presents a comprehensive approach to newlineadvancing the monitoring, control, optimization, and predictive modeling of BGP newlinethrough AD. The research incorporates both time series-based and machine learningbased methods to develop robust and accurate predictive models for biomethane newlineproduction. newlineA MATLAB-based dynamic simulation model was developed to predict biogas newlineyield over time, applicable to both batch and continuous processes with various newlinefeedstocks and conditions. The model achieved 92% accuracy, with a deviation of less newlinethan ±7.6% compared to literature values. It was statistically validated (plt0.05), newlinemaking it particularly valuable in identifying stable and optimized loading rates, newlinewhich are essential from an industrial perspective. Despite its strengths, the dynamic newlinemodel relies on numerous assumptions, such as kinetics and death rates, prompting newlinethe need for a data-driven approach to enhance its capabilities. newlineTo overcome the limitations, the research expanded into the scope of data-driven newlineapproaches. A machine learning (ML) based biogas predictive model was developed, newlineemploying a classification and regression algorithms (CAR): k-Nearest Neighbors, newlineDecision Tree, Gaussian Process Regression, Ensemble of Trees, and Support Vector newlineMachine. Two datasets were utilized, with Dataset-1 containing 10 inputs and newlineDataset-2 containing 5 inputs. Among these, SVM achieved the highest accuracy, with newlineR² values of 91% for Dataset-1 and 87% for Dataset-2. Statistical analysis indicated no newlinesignificant difference across datasets (p=0.377), demonstrating that accurate newlinepredictions could be made with fewer inputs. Key factors influencing model newlinepredictions included loading rate and retention time. This machine learning model newlineunderscores the potential of integrating artificial intelligence (AI) to control and newlineoptimize the AD process,