Hybrid model for dynamic fluid level measurement in oil wells

Abstract

Real-time monitoring of dynamic fluid levels in oil wells is crucial for ensuring production efficiency and safety. Traditional acoustic signal-based dynamic fluid level measurement methods often encounter significant noise interference, leading to inaccurate measurements. This paper proposes a hybrid model that combines machine learning and deep learning models to address this issue. First, raw audio data is preprocessed with wavelet transform to minimize noise. Then, a lightGBM classifier classifies the data into low- and high-noise data classes based on waveform features. Finally, for low-noise data, YOLOv7 is employed for target detection to evaluate fluid levels, as the imaging characteristics of such data are more precise; for high-noise data, the CNN-LSTM time series model is utilized, leveraging historical production data to forecast fluid levels, as image-based methodologies are inadequate. Unlike conventional techniques, which are limited to analyzing ideal low-noise waveforms for dynamic fluid level measurements, this hybrid model offers superior accuracy and resilience in fluid level measurements. It also broadens the applicability of acoustic-based dynamic fluid level assessment in oil wells. Consequently, this advanced hybrid approach for measuring dynamic fluid levels surpasses traditional methods, significantly contributing to blowout prevention, production strategy optimization, and overall enhancement of oil well management safety and efficiency.