应用带有非线性连接权的神经网络识别水泥胶结质量

Identification of Cement Quality by Neural Network with Nonlinear Connected Weights

为了解决石油测井中水泥胶结质量识别误差大的问题,采用改进的双发双收补偿式声波测井仪进行声幅测量,综合利用首波幅度信息来消除仪器的倾斜偏心、泥浆对声信号的衰减以及仪器灵敏度变化等引起的不可避免的误差.研究并阐述了优越于统计识别方法的带有非线性连接权的神经网络方法,能使2层神经网络具有3层BP网络的功能,且消除了隐含层的麻烦,并简化了运算、提高了运行速度.在具体应用中,先进行声幅刻度以全面取得合格样本,再构造神经网络进行学习训练以得到非线性权值和阈值等参数,然后在测井过程中自动进行水泥胶结质量的识别.通过实际测井应用表明,应用人工神经网络方法能实时进行水泥胶结质量的识别,其识别正确率远高于相对幅度法,效果显著.

There are some problems in the present cement bond log of oil casing-well engineering, such as big unavoidable errors, incorrect interpretation result by the amplitude-compare method based on the traditional statistics theory. To solve the above problems, the improvement-type compensation sonic instrument with dual transmitters and dual receivers is adopted to measure the sonic amplitude, and the errors, which stem from the tool's inclination-eccentricity, sonic signal attenuation in mud, and sensitivity change, can be eliminated by use of integrated head wave amplitude values. The method of artificial neural network (ANN) with nonlinear connected weights superior to that of statistics theory is studied. This method can replace the three-layer error back-propagation (BP) algorithm, so the implied layer is removed, the calculating is simplified, and the operated speed is increased. In concrete application, sonic amplitude values must be calibrated to acquire samples, and then the ANN is constructed and trained to acquire the nonlinear connected weights and thresholds. Finally, the cement quality can be automatically identified in logging process. An application example shows that the method of ANN can real-time identify the cement quality, the identification accuracy is much better than that of the amplitude-compare method, and the application effect is very notable.