Lithology Identification Based On Time-Frequency Image Of Vibration In Nearly Real Time

Complex formation lithology while drilling identification is the key technology for wellbore stability control,rock fragmentation optimization and complex risk reduction during drilling,which can ensure safe and rapid drilling and effectively prevent downhole risks.The traditional identification of lithology based on on-site cuttings and logging data analysis has obvious hysteresis,and is affected by factors such as drilling fluid fluid intrusion and logging interpreter's business level.The recognition accuracy is limited;based on drilling tools Although the log response lithology identification method satisfies the low delay requirement,the application is limited by the applicable environment and maintenance cost,and the high economic cost and the harsh wellbore environment limit its promotion.However,the downhole drilling tool vibration data is a high-density ancillary data generated during the drilling process.It has the advantages of low-latency and low-cost that can be acquired in real time.Its high frequency band can feedback the lithology characteristics of the formation.Therefore,how to effectively use the downhole vibration data Identifying the lithology of the formation is of great significance.In view of the above problems,this paper proposes to use the vibration data of the drill bit as a reasonable identification signal source,extract the vibration time-frequency characteristics into time-frequency images,and combine the advantages of Mobilenet and Res Net structure to construct a convolutional neural network algorithm.Complex formation lithology(conglomerate,sandstone and mudstone)while drilling identification model,and carry out related model accuracy verification and model prediction results interpretation work.Specifically complete the following work:The original sample contains 1410 lithology data of conglomerate,sandstone and mudstone.The original vibration signal is filtered by Butterworth(BHPF)and the timefrequency image is extracted by using short-time Fourier transform(STFT)transform.The enhancement technology expands the training sample set by a factor of five,using Mobile Script and Res Net as the base unit to implement specific models using python scripts and the MXNET framework,and completing iterative processes such as training,tuning,and evaluation.In order to improve the trustworthiness of decision-making results,the gradient-weighted class-activated thermal localization map and t-distribution neighborhood embedding technology are used to interpret the decision-making basis and spatial clustering decision-making results of the model.The final verification test shows that the single-sample decision time of the model is 10 ms,the model occupies 49.38 KB of storage space,the test macro precision rate is 74.3%,and the macro recall rate is 72.3%.This paper takes the lithology identification demand in the oil drilling process as the entrance.The first is to provide low-latency and low-cost lithology judgment methods for ensuring safe and rapid drilling.The second is to explore whether cutting-edge deep learning technology can be used to improve and optimize.Difficulties in the oil field,exploring the feasibility and applicability of the application of industrial big data in the petroleum field.