A Data-Driven Based Framework of Model Optimization and Neural Network Modeling for Microbial Fuel Cells

Abstract

Microbial fuel cells (MFCs) are devices that transform organic matters in wastewater into green energy. Microbial fuel cells systems have strong nonlinearity and high coupling, which involves control science, microbiology, electrochemistry and other disciplines. According to the requirements of microbial fuel cell system for model robustness and accuracy, we designed a comprehensive model optimization framework. Firstly, the influence of uncertain parameters on system was analyzed by combining global sensitivity analysis with uncertainty analysis. In accordance with analysis results, the uncertain parameters were optimized. Secondly, based on the optimized stochastic model, a simplified model was proposed by combining variable selection with neural networks. The results shown that the proposed framework can deeply analysis the influence of uncertain parameters on output, and provide theoretical basis for experimental research. It fully simplifies the original MFCs model, and has guiding significance for other types of fuel cells.