Abstract:
The problem of increasing the accuracy of a soft sensor (SS) with multiple output variables is considered. It is shown that the introduction of a vector time series predictor of errors ensures that the dynamic interdependence of process components is taken into account and allows increasing the accuracy of the SS. The construction of an error predictor with several outputs is performed using vector autoregressive models and a set of distributed lag autoregressive models, the optimal structures and parameters of which are found by numerical methods. A comparison of the proposed approach to constructing a multivariate SS with traditional methods based on the sequential construction of one-dimensional SS by output in the quality control system of the target product (light diesel fraction) of an industrial complex distillation column is carried out. The effectiveness of the proposed approach is also demonstrated for the class of adaptive SS.
Keywords:soft sensor with multiple outputs, error predictor with multiple outputs, vector autoregressive model, autoregressive model with distributed lag, adaptation, complex distillation column
First page:
PDF פאיכ