A Framework of Soft Sensor Systems with Machine Learning

A Soft-sensor is a means or a method to predict response variables that are difficult to predict by using the data of variables that can be easily obtained[1]. There have been increasing technical demands on improving the accuracy of soft sensors and reducing development complexity since the soft sensor is more cost-effective and easier to collect data than hardware sensors. However, few systematical methods to select the optimal set of features for building soft-sensor models have been proposed, although feature selection is the most essential factor to improve the development quality of the soft sensors. Therefore, this thesis is to present a systematic method for generating soft-sensor models to enhance the model accuracy by measuring similarities of soft-sensor models and selecting the best feature set from the similarity analysis. The proposed method utilizes ML technologies to build soft-sensor models and presents an algorithm to build soft-sensor models by reusing existing and similar soft-sensor models to improve the accuracy of the soft-sensor models.