Research on soft sensing method of pumping unit system efficiency based on LLE and SVR

The simplified calculation of system efficiency is paramount for evaluating the energy consumption of pumping units, serving as a critical technical parameter. This evaluation is essential for implementing cost-saving measures and objectively assessing the effectiveness of energy-saving technologies applied to pumping units. Addressing the limitations of current calculation methods, which often require numerous complex or difficult-to-obtain parameters, this study proposes a novel approach that integrates Locally Linear Embedding (LLE) and Support Vector Regression (SVR) to develop a soft sensor model for estimating system efficiency. Kernel Principal Component Analysis (PCA) is employed, utilizing a kernel mapping method within the PCA framework, to effectively manage non-linear data patterns. By projecting the data into a high-dimensional feature space through non-linear function transformations, the model effectively captures the inherent non-linear distribution characteristics. Subsequently, the SVR algorithm is implemented to construct a soft sensor model capable of accurately predicting pumping unit system efficiency. Testing results demonstrate that this LLE-SVR-based soft sensor model outperforms traditional modeling approaches, exhibiting enhanced measurement precision and improved generalization capabilities. The application of this model ultimately leads to a reduction in the monthly testing workload for pumping units and a corresponding decrease in production and operational costs within the oilfield.