Application of PSO-LSTM Neural Network Based on Improved Wavelet Denoising to Predict Drilling Rate of Penetration

Accurate prediction of the rate of penetration (ROP) is one of the important factors to improve the drilling efficiency of oil fields. At present, most of the established ROP prediction models have low accuracy, because the traditional noise reduction methods are generally used to process the drilling data. At the same time, the influence of wellbore friction and well deviation on WOB is not considered, and only the ground WOB is qualitatively used as the input parameter of the ROP prediction model. In this paper, seven deep directional wells in the central part of Bohai Bay, China are selected as the data source. Firstly, the drilling data are cleaned and screened, and the improved wavelet denoising method is adopted to effectively eliminate the noise interference while ensuring the data quality. Then the ground drilling pressure is corrected to improve the rationality of the data. The random forest feature selection method is used to analyze the correlation of various factors affecting the penetration rate and reduce the redundancy of the model. Finally, the particle swarm optimization algorithm is used to optimize the LSTM neural network mechanical drilling speed prediction model. The influence of traditional wavelet noise reduction method, ground drilling pressure and LSTM model hyperparameters on the error of mechanical drilling speed prediction model is discussed. The results show that the average error of the prediction model of the rate of penetration is reduced after the noise reduction method and the ground drilling pressure are modified. The optimized LSYM prediction model of the rate of penetration has higher prediction accuracy than other models, and the prediction accuracy reaches 87.3 %, which further verifies the superiority and effectiveness of the prediction model of the rate of penetration established in this paper.