Dynamic SOFC Temperature Estimation with Designed Experiments and Time-Series Model Identification

This paper presents the development of ARX-type (autoregressive with extra input) time-series models for the dynamic estimation and prediction of the SOFC stack maximum temperature. Experiment design aspects, model identification as well as filtering are discussed, and practical results obtained on a 10 kW SOFC system are presented. Data-based time-series models, whose parameters are identified directly from system measurements provide an alternative modeling approach compared to models based on physical first principles. Although physical models are very useful during the system design, they often become nonlinear and complex by structure, meaning that their application to control development or in embedded system software can be impractical. Often at least significant model simplification is required. Time-series models can be directly created as linear discrete-time models, which means that their utilization in control design as well as deployment into a modern embedded computational environment is straightforward.