Parameter identificaton of solid oxide fuel cell using teaching-learning based collective intelligence

The accurate extraction of internal operating parameters associated with multi-physicochemical processes forms the basis for precise modelling of solid oxide fuel cells (SOFCs), which serves as the foundation for predicting performance degradation and estimating the lifespan of SOFCs. In this work, a novel integration of the teaching-learning based algorithm (TLBO) and collective intelligence (CI), referred as the teaching-learning based collective intelligence algorithm (TLBO-CI), is introduced. This algorithm utilizes diverse characteristic patterns, including current-voltage (I-V) curves and electrochemical impedance spectroscopies (EIS), to enhance the overall identification of degardation process.Experimental data was gathered from a 3-cell SOFC short stack during a 640-hour durability test. The proposed parameter identification algorithm employs a collective intelligence framework, wherein sub-identifiers are based on particle swarm optimization (PSO) and individually tasked with processing specific formats of cell characteristics and identifying parameters from their degradation-time-dependent changes. Post-identification iterations, all sub-identifiers undergo an assessment to evaluate their prediction accuracy with respect to current, voltage, and impedance and their degradation rate respectively.Within the TLBO framework, poor sub-identifiers are designated as "students," whereas the best sub-identifier assumes the role of "teacher." The students engage in a process of crossover and learning from the teacher sub-identifier, thereby augmenting the overall precision of the sub-identifier collection across current, voltage, and impedance metrics.Compared to several state-of-the-art algorithms previously employed for parameter identification in SOFCs, the proposed TLBO-CI algorithm achieves superior optimization of both the prediction of output current and voltage decline, as well as the increase in internal impedance, with a high degree of precision quantified by mean squared error (MSE) and less computation time.