An Online Soft-sensor Modeling Method Including Process Variable Time-delay Estimation

We propose a novel method based on fuzzy curve analysis (FCA) in order to effectively introduce delay information into the soft-sensor model and track real-time process dynamics. The proposed method can estimate the process time delay parameter set, which is achieved offline and is then used to reconstruct the whole modeling sample set. When new input samples are available, a time difference Gaussian process regression (TDGPR) model is employed for current time online predictions based on historical variable values collected at certain moments. The proposed method is applied to a real debutanizer column process, and its effectiveness and accuracy are verified by the simulation results.