Oil Production Measurement Method for Pumping Wells Based on Multi-Source Information Fusion

Abstract Measuring oil well production is a fundamental task in oilfield operations and development. Precise measurement results are vital for diagnosing well anomalies, assessing field development, and optimizing well operating conditions. Currently, the majority of oilfields depend on metering stations for production measurement, a costly and labor-intensive approach with limited automation. Similarly, oil measurement methods based on indicator diagrams also face challenges due to substantial measurement errors. To improve the efficiency and accuracy of oil production measurement for pumping wells, we propose an intelligent measurement approach based on multi-source information fusion, integrating electrical parameters, indicator diagram data, and well production parameters. The intelligent measurement model incorporates two distinct feature extraction channels, taking electrical parameters and indicator diagram data from a single well cycle as inputs. The first channel uses a convolutional neural network (CNN) enhanced by a squeeze-and-excitation (SE) attention mechanism to extract 2D image information of load and displacement during the pumping well's operation. The second channel employs a gated recurrent unit (GRU) to capture 1D temporal information from the electrical parameter signals. Subsequently, well production parameters, including stroke length, stroke rate, and pump diameter, are combined with features extracted from both channels. This fused data is then analyzed by a fully connected network to enable intelligent oil measurement. Validation results using actual production data from 5,317 data samples across 13 wells indicate that the proposed intelligent measurement method achieves favorable accuracy in liquid production measurement, with a root mean square error (RMSE) of 0.3512 on the test set, a mean absolute percentage error (MAPE) of 4.51%, and a Mean Absolute Error (MAE) reaching 0.0299 under typical conditions, including normal operation, gas interference, and insufficient liquid supply. Compared to models based solely on electrical parameters or indicator diagrams, the multi-source information fusion model demonstrates significantly higher accuracy and robustness across various operational scenarios.