Soft Sensor Modeling Method of Dynamic liquid Level Based on Improved KS Algorithm

Duing to the problem that the sample data of the soft-sensor modeling in the oilfield production process has influence on the modeling quality, this paper proposes a subspace-based Kennard-Stone algorithm. Firstly, according to the characteristics of oil production data, production process parameters are divided into subspace of different working conditions. Then, the method of similarity of spatial data is used to replace the original Euclidean distance calculation method to complete the sample selection of KS algorithm. Finally; the soft-sensing model of the dynamic liquid level is constructed based on the sample selected and BH-LSSVM algorithm. Experimental results show that compared with random sample selection and KS algorithm based on Euclidean distance, this method can improve the prediction quality of the model, and save the modeling time.