Joint prediction of multiple thermal variables for flow instability based on extreme learning machine model

The coupling of multiple thermal variables in two-phase boiling natural circulation system under rolling motion can result in complex flow instability. To predict the important thermal variables of flow instability systems under rolling motion, a multiple thermal variables and time series joint forecast method based on extreme learning machine neural networks model was proposed. Both flow rate and heating wall temperature were taken into consideration. The extreme learning machine model was trained using measured data. Both single-step-ahead and multi-step-ahead predictions were conducted and the influence of hidden note number on forecast performance was studied. Simulation experiment results show that multiple variables joint prediction based on extreme learning machine can produce better forecast performance than single variable forecast method and this advantage is more significant in prediction with more steps ahead. The method can be generalized to conditions with more thermal variables and is proved to be an effective real-time forecast method of thermal variables in flow instability systems. (authors)