A unified recursive just-in-time approach with industrial near infrared spectroscopy application

Abstract

Time varying and non-linearity issues are commonly seen in soft senso development. Recently, just-in-time approach has been widely used to address the non-linearity problem in near infrared (NIR) spectroscopy modeling. However, to the best of the authors' knowledge, the time varying problems in just-in-time (JIT) framework are rarely discussed and the adaptation strategy for the local models in JIT approach remains an open issue. In this paper, a new model updating approach is proposed which can adjust to process changes by merging the traditional recursive algorithm in the JIT framework. The advantage of the presented approach is that it can solve both time varying and non-linearity issues simultaneously under the JIT framework. The performance of the method has been tested on a spectroscopic dataset from an industrial process. By comparison with traditional PLS, locally weighted PLS and several other on-line model updating strategies, it is shown that the proposed method achieves good performance in the prediction of fuel properties.