一种可提高试井模式识别准确度的改进HT-BP方法

在用BP神经网络进行试井问题模式识别的过程中 ,若将现场试井数据简单地以点对的方式送入BP神经网络进行识别 ,则与训练模式无法对应 ;若在用BP神经网络学习和识别曲线之前进行归一化处理 ,则会引起曲线的尺度变化和空间位移。结合试井问题对Hough变换进行了改进 ,提出了与改进Hough变换相应的比例变换和空间压缩方法。用比例变换和空间压缩方法及改进Hough变换法对试井曲线进行预处理后再送入BP神经网络 ,可以大大改善识别能力。实例说明 ,改进Hough变换与BP神经网络相结合的方法 (MHT BP)对试井模式识别的准确性较高。

A MODIFIED Hr-BP METHOD TO IMPROVE THE PRECISION OF WELL TESTING MODEL IDENTIFICATION

During the process of well testing model identification (WTMI) by using Backward Propagation Neural Network (BPNN), most data in field tests surely cannot match those gotten from type curves if co ordinates are fed. When the traditional normalization procedure is carried out just before the training of BPNN and the recognition using BPNN, the size of a curve will be changed and displaced in space. Hough transform is tailored for well testing, and a method of size changing and space compressing (SCSC) compatible with the modified Hough transform (MHT) is given. All curves are processed by SCSC and MHT before they are input into BPNN. Several case studies verify the high performance of BPNN coupled with MHT (MHT BP) when it is used for WTMI.