摘要
针对标准BP神经网络预测连续管疲劳寿命时容易陷入局部极小值和训练时间过长的缺点,利用有动量的梯度下降法、拟牛顿算法和一步正割算法分别对BP神经网络进行优化。拟牛顿算法优化后的BP神经网络性能最佳。利用拟牛顿算法优化后的BP神经网络预测连续管疲劳寿命,并与标准试验结果进行对比研究。研究结果表明:拟牛顿算法优化后的BP神经网络预测结果与标准试验结果最小相对误差率为1.7%,最大相对误差率为3.6%,满足工程精度要求。同时利用优化改进的标准BP神经网络预测方法,提出连续管疲劳寿命区间预测。预测结果表明,所有的预测样本都处于合理的预测范围之内,证明了优化后BP神经网络预测连续管疲劳寿命区间的可行性。所得结果可为连续管的疲劳寿命预测提供参考。
When the standard BP neural network is used to predict the fatigue life of coiled tubing,local minimum is easily fallen into and training time is too long.To solve these problems,the momental gradient descent method,quasi-Newton algorithm and one-step secant algorithm were used to optimize the BP neural network respectively.The BP neural network optimized by the quasi-Newton algorithm had the best performance,and was used to predict the fatigue life of coiled tubing,which then compared with the standard test results.The research results show that the minimum relative error rate between the predicted results of quasi-Newton algorithm optimized BP neural network and the standard test results is 1.7%,and the maximum relative error rate is 3.6%,meeting the engineering accuracy requirements.Meanwhile,the optimized and improved standard BP neural network prediction method was used to conduct fatigue life interval prediction of coiled tubing,indicating that all prediction samples are within a reasonable prediction range,proving the feasibility of the optimized BP neural network to predict the fatigue life interval of coiled tubing.The research results provide reference for predicting the fatigue life of coiled tubing.
作者
窦益华
张佳强
李国亮
韦亮
曹银萍
Dou Yihua;Zhang Jiaqiang;Li Guoliang;Wei Liang;Cao Yinping(Mechanical Engineering College,Xi an Shiyou University;Well Test Company of CNPC Xibu Drilling Engineering Company Limited)
出处
《石油机械》
北大核心
2023年第10期144-149,共6页
China Petroleum Machinery
基金
国家自然科学基金青年基金项目“附加牵引力与减阻效应的大位移水平井连续管极限下入机理研究”(52004215)。
关键词
连续管
疲劳寿命预测
BP神经网络
拟牛顿算法
方法优化
寿命区间
标准试验
coiled tubing
fatigue life prediction
BP neural network
quasi-Newton algorithm
method optimization
life interval
standard test
