Title : Predicting the biogas production in an electrical voltage-applied reactor using real-time data monitoring and deep learning
Abstract:
Modeling is widely used to understand anaerobic digestion (AD) and predict biogas production. However, the modeling of electrical voltage (EV)-applied reactors has been scarcely explored. This study introduces two strategies to enhance the predictability of biogas production rate (BPR) in EV-applied reactors using machine learning (ML) and deep learning (DL) techniques. In Strategy 1, the BPR was predicted using randomly generated data for operational parameters, while Strategy 2 combined this approach with real-time data monitoring of key chemical properties. Two reactors, a control and an EV-applied reactor were equipped with biochemical sensors to monitor pH, electrical conductivity (EC), and oxidation-reduction potential (ORP). Three ML models (random forest, extreme gradient boosting, and support vector regression) and one DL model (a combined convolutional neural network and long short-term memory (CCL)) were implemented. The CCL model outperformed the ML models, achieving a coefficient of determination (R²) >0.85 and a mean absolute error (MAE) <0.1, compared to average R² values <0.80 and MAE >0.2 for the ML models. Notably, in the control reactor, the difference in R² for the CCL model between the two strategies was minimal. However, in the EV-applied reactor, the R² difference exceeded 15%, emphasizing the significant impact of real-time monitoring data. Feature importance analysis further highlighted the critical role of EC in optimizing AD performance under EV application. This study demonstrates the effectiveness of integrating a DL model with real-time data monitoring to improve biogas prediction accuracy and reduce system failure risks.
Keywords: Biogas production rate; Deep learning; Electrical voltage application; Real-time data monitoring; Prediction.
