Automatic Detection of Hydraulic Fracturing Events by Real-Time Data Mining
Shuai Zhang, Mao Sheng, Leifeng Meng, Kankan Bai, Jian He, Shouceng Tian · International Petroleum Technology Conference · 2025
Abstract Accurate, quick, reliable event detection is essential to Intelligent hydraulic fracturing. The real-time detection of events mostly relies on the expert knowledge. Although the machine learning algorithms had been used for events detection, the accuracy is still not stable that depending on the data quality and amount. This paper proposed an alternatively threshold-based algorithms with special sliding windows by integrating feature mining with expert knowledge from time-series data. The 378 stages of historical fracturing data were collected. More than 5 experts were organized to divide and label the six types of fracturing events involving wellhead pressure testing, plug ball seating, pad stage, proppant laden stage, flushing, and fracture closure stage from second-point data. The correlation between fracturing events and data features was built by using 10-seconds sliding window and forward difference to extract the slope features of pumping curves and their corresponding threshold values of pumping pressure, rate, and proppant concentration. Furthermore, the wavelet analysis was used to identify the fracture closure time from wellhead pressure after the fracturing is completed. Finally, the threshold-based algorithms by integrating feature mining with expert knowledge from time-series data were proposed to handle the fast detection of fracturing events. The model was tested on 200 stages of data and performed a high accuracy in fracturing events detection. Particularly, the accuracy of fracturing events detection ranges from 92.7% to 99.6%. The accuracy of Fracture Closure Pressure were 93.8%.