A SEVA soft sensor method based on self-calibration model and uncertainty description algorithm

• Identify and reconstruct a faulty input sensor by PCA before making prediction.
• JIT learning are improved by a new selection method and ensemble learning.
• Uncertainties of the prediction model are expressed by ICP algorithm.
• Generate five types of outputs for a soft sensor.
• A wastewater treatment case study is presented.

Soft sensors are widely used to estimate process variables that are difficult to measure online. However, due to poor quality of input data and deterioration of prediction model as time passes, soft sensors make poor performance. We have been constructing a principal component analysis (PCA) model before performing a prediction. Furthermore, the just-in-time (JIT) learning model has been improved and served as prediction model for self validating (SEVA) soft sensors. The proposed soft sensor not only carries out internal quality assessment but also generates multiple types of output data, including the prediction values (PV), input sensor status (ISS), validated measurement (VM), output sensor status (OSS) and the uncertainty values (UV) which represent the credibility of soft sensors' output. The effectiveness of the proposed SEVA soft sensors is demonstrated through a case study of a wastewater treatment process.