基于主成分分析的NCPSO-BP机械钻速预测

Prediction of NCPSO-BP ROP based on principal component analysis

沙林秀 胥陈卓

SHA Linxiu XU Chenzhuo

西安石油大学电子工程学院

School of Electronic Engineering, Xi’an Shiyou University, Xi’an 710065, Shaanxi, China

目前常用的机械钻速预测理论模型仅通过相关性、贡献度来筛选模型输入参数，没有积极挖掘随钻采集的复杂属性间关系，导致信息缺乏完整性。为了最大化保留复杂属性间线性关系，提出了一种基于主成分分析的钻速预测模型，并引入混沌变异的小生境粒子群算法(NCPSO)优化BP神经网络，提高模型的收敛速度与精度。首先，采用主成分分析法根据不同的方差贡献度对高维钻井数据进行降维、降噪；其次，建立智能优化算法-神经网络钻速预测模型，利用混沌变异的小生境粒子群算法的训练结果为BP神经网络权值、阈值赋予初值，以此建立机械钻速预测模型；最后，在不同输入维度进行对比分析NCPSO-BP模型与PSO-BP，GA-BP和标准BP的机械钻速预测结果。研究结果表明，在8维、10维输入的情况下，NCPSO-BP机械钻速模型的预测精度平均提高了59%，训练速度平均提高了26.3%，为日益复杂的钻井环境下机械钻速精确预测提供了理论基础。

The currently and commonly used theoretical models predicting rate of penetration (ROP) only screen the model input parameters through correlation and contribution, which lacks active exploration on the relationship between complex attributes collected while drilling, resulting in a lack of completeness of information. In order to maximize the preservation of the linear relationship between complex attributes, a drilling speed prediction model based on principal component analysis was proposed. And by optimizing the BP neural network with niche particle swarm algorithm after introducing chaotic variation, the convergence speed and the accuracy of the model were improved. Firstly, with the help of the principal component analysis method, the dimension and the noise of the high-dimensional drilling data were reduced according to different variance contributions. Secondly, an intelligent optimization algorithm-neural network drilling speed prediction model was established, and using the training results from the niche particle swarm algorithm after introducing chaotic variation to give initial values to the weights and the thresholds of the BP neural network, so as to establish the ROP prediction model. Finally, the ROP prediction results between NCPSO-BP model and PSO-BP, as well as the results between GA-BP and standard BP in different input dimensions were compared and analyzed. The results show that in the case of 8-dimensional and 10-dimensional inputs, the prediction accuracy of the NCPSO-BP ROP model was increased by an average of 59%, and the training speed was increased by an average of 26.3%. The NCPSO-BP ROP model can provide a theoretical basis for accurate prediction of ROP in increasingly complex drilling environments.

10.13639/j.odpt.2022.04.017