Abstract
To quickly and accurately obtain the multi-dimensional state in a sealing reactor is critical for the safety and high performance of industrial system. In this work, a generic machine learning based multi-dimensional soft sensor and a long-term calibration scheme is proposed by using the case of steam reforming solid oxide fuel cell (SR-SOFC) system. Firstly, based on a validated SR-SOFC system model, the temporal-spatial temperature distribution (TSTD) characteristics are analyzed and a TSTD characteristic model is constructed by Multivariable Linear Regression. Then, the central node temperature of the stack is estimated by applying the Least Square Support Vector Machine and the SR-SOFC stack temperature distribution is accurately obtained by explaining the TSTD characteristics model with the central node temperature. In addition, the temperature distribution is calibrated by Stochastic Gradient Descent algorithm to eliminate the estimation error resulting from stack degradation in a long term. The simulation results show that the proposed method can obtain the SR-SOFC stack temperature distribution in time and effectively, the average error is less than 1 K. The proposed estimation strategy can be easily expanded to other scenarios with similar operation conditions.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call