基于改进BP算法的摩阻因数预测方法研究  

Research on Friction Coefficient Prediction Based on Improved BP Algorithm

作　　者：申静波[1] 闫铁[2] 李井辉[1,2] 孙丽娜[1] 

机构地区：[1]东北石油大学计算机与信息技术学院,黑龙江大庆163318 [2]东北石油大学油气钻井技术国家工程实验室,黑龙江大庆163318

出　　处：《计算机技术与发展》2018年第1期164-168,共5页Computer Technology and Development

基　　金：国家科技重大专项(2016ZX05020-006);国家自然科学基金(51374077)

摘　　要：在钻井工程设计与实钻过程中,恰当指定摩阻因数是准确预测摩阻、扭矩的前提条件,摩阻因数受众多因素影响且具有不确定性,很难利用普适的数学公式明确表达摩阻因数及其影响因素之间的关系。针对摩阻因素的特点,提出了采用改进的BP神经网络对钻柱力学分析中的摩阻因数进行计算的方法。首先研究BP算法的原理和数学表示,然后结合预测实际将增加动量项、自适应学习率等方法对其进行改进,最后根据摩阻因数的内涵建立以改进BP算法为基础的摩阻因数预测模型。实验结果表明,利用改进BP神经网络能够有效实现摩阻因数的准确预测,解决了钻井过程中普遍存在的摩阻因数个体差异问题。In the process of drilling engineering design and drilling, the reliable friction coefficient should be provided to predict the drag and torque accurately. The friction coefficient is influenced by many factors with uncertainty, so it is difficult to express the relationship between friction coefficient and its factors. According to the characteristics of friction coefficient,we propose a method for computing friction coeffi- cient in drilling string mechanics analysis by BP neural network. Firstly, we research the principle and mathematical representation of BP al- gorithm, and then improve it with addition of momentum and adaptive learning rate combined with the actual situation. Finally, the friction coefficient prediction model based on improved BP algorithm is established under the connotation of friction coefficient. The simulation shows that the improved BP neural network can be used to improve the prediction accuracy of friction coefficient, which solves the individual difference of friction coefficient in the drilling process.

关 键 词：神经网络 摩阻因数 BP算法 摩阻扭矩 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]