Artificial Intelligence and Data Science Applications in Anaerobic Digestion: Monitoring and Optimization Strategies

Abstract

Abstract Anaerobic digestion (AD) is a significant biological process that transforms organic waste into renewable biogas and nutrient-dense digestate. This can be returned as fertilizer, which promotes waste management and renewable energy production. However, typical monitoring relies on offline laboratory measurements of critical factors, including volatile fatty acids (VFAs) and chemical oxygen demand (COD), leading to reactive process management, instability, high costs, and suboptimal methane yields. Many anaerobic digestion plants operate based on experience rather than precise control strategies, making them highly vulnerable to acidification and spoilage due to fluctuations in pH, temperature, and organic loading rate. This overview investigates how artificial intelligence (AI), specifically machine learning (ML)-driven soft sensors and digital twins, infer difficult-to-measure variables from existing online sensors, enabling early instability detection and yield optimization. Key technical contributions include the unification of classical ML models (e.g., support vector machine (SVM), random forest) with deep learning for VFA prediction; hybrid physics-ML methods that use an AD model (ADM); and implementation of digital twins that achieve high accuracy (R2 > 0.94 for CH4 and pH). Current shortcomings include limited poor model portability, lack of AI interpretability, data quality issues, and low adoption of federated/reinforcement learning for autonomous control.