BiGRU network drilling working condition identification method combined with attention mechanism

Abstract

The invention discloses a BiGRU network drilling working condition identification method in combination with an attention mechanism, and the method employs a large amount of actual data generated during drilling as a basis, carries out a series of data enhancement methods, and solves a problem that the number of samples is small to the greatest extent. In order to weaken a black box effect during machine learning, a decision tree recognition model guided by a semi-empirical formula combined with actual data is designed to assist manual data annotation. For sequential well drilling information, a two-way gate control unit network capable of extracting sequence features is used for assisting an attention mechanism to train a model, and strict parameter adjustment is carried out. Finally, data of a single well which does not participate in training is used, and generalization ability testing of the model is carried out on the premise that transfer learning and pre-training are not carried out; comparison experiments prove that the method ensures feasibility and universality under the condition of high accuracy in drilling working condition time sequence prediction, and has practical application value.

Description

technical field [0001] The invention relates to the technical field of oil and gas field exploitation, in particular to a method for identifying drilling conditions with a bi-directional gate control unit BiGRU network combined with an attention mechanism. Background technique [0002] In the context of the times, China has paid more and more attention to oil exploration, and the demand for oil has continued to rise. Drilling engineering serves the development process of oil fields. The rapid operation of drilling construction provides economic benefits and high timeliness. Monitoring and management during drilling operations pose enormous challenges. Restricted by traditional manual management methods, in the drilling construction stage, the management efficiency, decision-making accuracy and response speed are still at a low level, and the drilling supervision ability cannot keep up with the rapid development of drilling technology. Fast and accurate decision-making respo...

AI Technical Summary

Problems solved by technology

[0004] All in all, the following problems have not been fully resolved in the rare research on the identification of drilling conditions: 1. The number of samples for training is small, and the obtained model has a good effect on paper, but the actual effect has not been rigorously demonstrated, so it is difficult to apply it to practice Production

2. The network model is very simple, no further parameter adjustment work has been done, and the upper limit of the neural network has not been reached

3. There is no generalization ability test, such as whether the same experimental results as during training can be obtained under different data characteristics is still unknown

4. There is no targeted design for data with very time-series characteristics such as drilling information

5. There is no restriction on the black box effect during machine learning training, and the interpretability of the model is poor

Images