A multivariate multi-headed deep LSTM model: a new approach to enhance the prediction of methane production in the biogas plant

Biogas is a sustainable method for converting diverse waste sources into energy. It is of utmost importance to develop a new approach for enhancing the prediction of methane production in the biogas plant. This study introduces a standalone multi-headed neural network architecture called multivariate multi-headed long short-term memory (MM-LSTM) to improve the prediction accuracy of methane production from biogas digesters. The proposed model treats each input variable independently, employing them as distinct LSTM network models (heads) to improve the prediction accuracy. In the MM-LSTM approach, the outputs of these independent LSTM models are concatenated to forecast high methane production. The present study evaluated the proposed model's predictive performance using sensor data from a biogas digester and compared it to state-of-the-art models and single-headed long short-term memory (LSTM) neural network architectures. Based on the experimental results, the MM-LSTM model achieves a prediction accuracy of 99.52%, which is best among all the existing models. The experimental results suggest that the proposed prediction model can be effectively used in biogas plants for biogas production processes and for advancing sustainable energy production.