DYNAMIC MODEL LEARNING BASED ON SLOW PROCESS FEATURE TRANSFER

In order to realize cross domain learning of dynamic feature extraction, a dynamic model learning based on slow process feature transfer is proposed. A transfer slow feature analysis technique was proposed to transfer the dynamic model learned from the non-homologous process, so as to improve the prediction performance of the target process. Two weight functions were further dynamically updated to quantify the mobility from the source domain to the target domain at each time. In addition, an improved Bayesian algorithm scheme was designed to obtain the parameters under the probability distribution considering the corresponding uncertainty. A simulation example, a public dataset and an industrial case show the effectiveness of the proposed method.