Accurate prediction of ROP is the premise and key to increase drilling speed. Artificial intelligence provides technical means for accurate ROP prediction. In this paper, the influence of borehole trajectory and drilling parameters on ROP is considered, and the prediction model of ROP is established by using PSO-BP neural network. Firstly, the wavelet filtering method is used to reduce the noise of the measured drilling data, and the input parameters of the ROP prediction model are selected according to the correlation analysis by mutual information. Secondly, BP neural network model improved by particle swarm optimization is used to establish the prediction model of ROP. Finally, drilled X-Wells is taken as a case study. The results show that the prediction accuracy of PSO-BP model is 98.33%, the average absolute percentage error is 0.0479%, and the root-mean-square error is 0.567. The prediction accuracy of PSO-BP model is improved by 4%. This paper provides reference and guiding for using artificial intelligence technology to optimize drilling and reduce cost and increase efficiency.
Rate of penetration (ROP) is one of the important indexes to evaluate drilling efficiency, and accurate prediction of ROP is a prerequisite for realizing drilling optimization (Ahmed et al, 2019; Alkinani, et al, 2019; Alsaihati, et al, 2022). Current research on drilling rate prediction mainly falls into two categories: mechanism model prediction and artificial intelligence prediction (Hegde, et al, 2017; Barbosa, et al, 2019; Najjarpour, et al, 2021). Artificial intelligence can fully excavate the inherent rock breaking law contained in the field drilling data, and has higher accuracy than the mechanism model, which has become the hot direction of the current drilling rate prediction.
Many scholars have conducted a series of researches on intelligent prediction of drilling rate and optimization of drilling parameters. Bilgesu (1997), was the first scholar to study the intelligent prediction of mechanical drilling rate. In 1997, he applied artificial neural network for the first time to predict ROP, used experimental data and field data for training and verification, and achieved high prediction accuracy. Eskandarian et al. (2017), used a one-way multi-layer perceptron model to predict ROP, which took weight on bit, mud density, well inclination, azimuth angle, pump pressure, plastic viscosity and other training input parameters. Li et al. (2021), established a ROP prediction model based on BAS-BP neural network, which proved that the error between the predicted value and the actual measured value of the BAS-BP mechanical penetration rate model was minimal, and it had good convergence and searching ability. SHA et al. (2022), proposed a ROP prediction model based on principal component analysis, and introduced chaotic variation niche particle swarm optimization (NCPSO) to optimize the BP neural network to improve the convergence speed and accuracy of the model. TANG et al. (2023), proposed a new model for predicting ROP based on principal component analysis algorithm (PCA) optimized BP neural network. The PCA-BP model can evaluate the rationality of factors affecting the rate of penetration in real time, and provide guidance for improving the rate of penetration. Feng et al. (2024), proposed a depth sequence ROP prediction method based on long short-term memory (LSTM) neural network in order to predict the ROP with high accuracy with a more efficient method.
