Abstract
The drilling operation procedures are complicated and its risks are high. The unsafe behavior of well site personnel and the unsafe state of equipment and materials are the main causes of drilling accidents. At present, these are mainly supervised by drilling supervisors. The supervisors, who's supervising means are single, cannot achieve full coverage of on-site personnel, equipment and materials. In order to realize intelligent identification and warning of drilling operation risks, the intelligent risk identification and warning model for typical drilling operation scenes and its application are carried out.
First of all, considering the influence of different environmental conditions, the approach of automatically generating image dataset based on machine learning is proposed, and the typical scene sample image database is established. Meanwhile, the typical scene risk identification model based on YOLOv5 algorithm is designed and established by introducing feature aggregation, loss function and attention mechanism, and the algorithm model is trained and tested by using neural network method. In addition, based on the risk identification of drilling operation, the approach of risk warning and feedback is put forward. Finally, a set of ablation experiments are designed to test the performance of the improved algorithm models in drilling well sites.
By using the approach of automatically generating image dataset based on machine learning, the foreground and background images can be automatically fused, and the standardized collection and classified storage of well site video image data are realized, saving a lot of manpower labeling costs. With the use of the risk identification model of typical scenes, typical risks can be automatically identified, with the mAP of 90.3% and the response time of less than 2 seconds. Three ways of mobile phone short message, well site speaker and screen pop-up reminder have been developed to timely send the identified risks to relevant personnel. Through intelligent risk identification and processing, the operation risk is reduced, the operation quality is guaranteed, and the supervision efficiency and effect are improved significantly.
The intelligent risk identification and warning models of typical drilling operation scenes are innovatively established by using the approach of combining the drilling operation risk identification theory and artificial intelligence technology, which solves the problem of intelligent risk identification and warning of typical drilling operation scenes, and provides theoretical and practical basis for the development of digital supervision management in the drilling operation.
