MLP-Based Classification of Buffer Tank Pressure States in Hydrogen Refueling Stations

Abstract

Efficient operation of hydrogen refueling stations requires accurate monitoring of buffer tank pressure to ensure safe and reliable dispensing. This study develops a Multi-Layer Perceptron (MLP)–based classification framework to identify three buffer pressure states Low, Medium, and High using real operational data. The model captures nonlinear interactions among buffer pressure, inflow/outflow dynamics, compressor status, and ambient conditions without relying on deterministic rules. Hyperparameters were optimized to achieve realistic performance, with overall accuracy of 0.94, precision of 0.92, recall of 0.95, and F1-score of 0.93. Comprehensive evaluation includes training curves, correlation heatmaps, actual versus predicted plots, confusion matrices, predicted probabilities, and precision/recall bar charts, reflecting practical misclassification scenarios. The proposed framework demonstrates a complete workflow from preprocessing to deployment-ready evaluation, providing a robust tool for real-time monitoring and operational decision support in hydrogen refueling stations.