| Abstrakt: |
In the drilling exploitation of hot dry rock for geothermal energy, whirling is one of the main lowefficiency conditions that affect the efficiency of Polycrystalline Diamond Compact (PDC) bit in horizontal well drilling. Realizing automatic and real-time identification with whirling is of great significance to save non-productive time and ensure drilling safety and benefit. In this paper, we first constructed a real-time drilling mechanical specific energy (MSE) model combined with while-drilling testing to reflect the real-time drilling conditions. The MSE model is used to normalize multi-source parameters. Secondly, we constructed a Back Propagation (BP) artificial neural network (ANN), and then the normalization effect verification, optimization of network parameters, and identification effect verification of whirling were carried out. The final results show that the established MSE model has a favourable effect on data normalization, which could also reduce the complexity of the required network model, and shorten the training time by 20-30 s/step. The optimal algorithm is Trainscg, whose optimal number of hidden layer nodes is 5, and the optimal maximum number of iteration steps is 1000. The established ANN model can accurately identify whirling based on MSE, the accuracy is about 0.94, and the average relative error is 1.3%. The method established in this paper provides a reference for the automatic identification of various low-efficiency conditions based on MSE. [ABSTRACT FROM AUTHOR] |
