Data-driven modeling for fixed-bed intermittent gasification processes by enhanced lazy learning incorporated with relevance vector machine

An enhanced lazy learning approach incorporated with relevance vector machine (ELL-RVM) is proposed for modeling of the fixed-bed intermittent gasification processes inside UGI gasifiers. The online measured temperature of produced crude gas plays a dominant role during gasification processes. However, it is difficult to formulate the dynamics of gasifier´s temperature via first principles due to the complexity of UGI gasification process, especially severe changes in the temperature versus infrequent manipulation of the gasifier and noise in the temperature data collected from practical fields. Noticing that the changes of some input variables of UGI gasification process are small but impactful, a novel weighted-neighbour selection method, which is based on minimizing dynamic cost functions for different outputs coordinately, is adopted to enhance the lazy learning approach. The sparseness and short test time of RVM is fully utilized in design and implementation of the proposed online modeling algorithm under the Bayesian learning framework. The effectiveness of ELL-RVM for modeling UGI gasification processes is verified by a series of experiments based on the data collected from practical fields.