MFM-MSCL: Multi-scale Self-supervised Contrastive Learning for Multiphase Flow Meters Under Harsh Production Conditions

Accurate multiphase flow measurement is vital for effective oilfield production management, yet environmental fluctuations and sensor drift often hinder the performance of flow meters. Traditional calibration methods, which rely on manual adjustments, fail to address the dynamic, multi-scale temporal dependencies of flow data under harsh conditions. In this work, we propose MFM-MSCL, a self-supervised contrastive learning framework designed to capture multi-scale temporal relationships in flow data. By employing contrastive learning, our model learns to identify and distinguish between significant temporal patterns across multiple time scales, without requiring large labeled datasets. This ability enables the model to adapt to complex variations in flow characteristics, such as pressure, temperature, and fluid composition, while mitigating sensor drift. The multi-scale learning aspect ensures that both short-term fluctuations and long-term trends in the flow data are effectively captured, leading to improved model robustness and calibration accuracy. We demonstrate the effectiveness of MFM-MSCL through simulated and field test results, showing its superior performance over traditional calibration techniques in maintaining measurement accuracy under challenging operating conditions. This approach significantly reduces the need for manual recalibration and offers a more reliable solution for real-time flow measurement in oilfield operations.