Intelligent Diagnosis for Effectiveness of Data-Knowledge Mixed-Driven Fracturing Ball Seat Setting
Real-time diagnosis of the effectiveness of the bridge plug ball seat setting is a key step in the staged fracturing of horizontal wells.If the ball seat setting fails,follow-up operations cannot proceed normally.Currently,manual observation of wellhead pressure changes is primarily relied upon,making it difficult to quickly and accurately identify key characteristics.To address this,a combination of expert qualitative judgment and quantitative feature mining of setting data was implemented.Sliding window data was segmented to form 5792 sets of labeled data.A long short-term memory(LSTM)neural network,using a two-dimensional input of wellhead pressure and displacement,was selected.An intelligent diagnosis model for evaluating the effectiveness of the fracturing ball seat setting was established,utilizing an under-sampling balanced dataset to improve the model's prediction accuracy.The results show that the setting data exhibits a clear three-stage characteristic:a steep rise,a steep drop,and a gentle rise in wellhead pressure.If the wellhead pressure lacks any of these stage characteristics,it indicates an invalid setting.The wellhead pressure slope exhibits a wide distribution range,making it difficult to form explicit rules for accurate diagnosis.Artificial intelligence technology is used to learn the valid/invalid setting data characteristics from various wellhead pressure forms,producing diagnosis results per second with an accuracy of 96.8% for the test set and 84.3% for the validation set.The findings are expected to provide a method for real-time and automatic diagnosis of the effectiveness of the bridge plug ball seat setting.
万方数据
维普
