Generalization of the LSTM-based SMFC Energy Prediction Model to New Deployment Conditions

Determine the generalization performance of the Long Short-Term Memory (LSTM) deep learning model with quantile regression that predicts soil microbial fuel cell voltage, current, and power, when applied to microbial fuel cells deployed under conditions different from the single SMFC deployment used for training and validation.

Background

The paper presents a deep learning approach using LSTM models with quantile regression to predict SMFC voltage, current, and power across multiple time horizons. Due to challenges in data collection, the model was trained and validated on data from a single SMFC deployment.

The authors explicitly note that this constraint leaves the model’s out-of-domain generalization unassessed. Establishing whether the model’s predictive performance transfers to SMFCs operating under different deployment conditions is crucial for real-world applicability across diverse environments.