BP神经网络隐式法在测井数据处理中的应用

现有神经网络方法对时间向量序列数据的处理是通过单点进行的,割裂了数据间的关联性.为此,利用隐式曲线的构造原理,通过对时间向量序列的变换,提出了一种整体预测时间向量序列的测井数据的方法.神经网络隐式整体预测方法的步骤是:①将数据变换为封闭曲线,构造约束点以简化神经网络的输入与输出;②利用神经网络的隐式方法,通过智能学习和仿真模拟,得到封闭的预测曲线;③经过变换得到最终的预测曲线.实验证明了该方法的有效性.

Processing of well log data based on backpropagation neural network implicit approximation

The previous methods based on neural network are difficult to predict the water saturation of next year based on the data of the given years. A new method combining the neural networks with the principle of implicit curve can effectively handle the above problem. First, the vector data of every year are mapped into a closed curve, and a virtual explicit function is constructed on the constraint points. Then, the explicit function is approximated by a backpropagation neural network. Finally, the isoline of the neural network is extracted from the simulation surface. The predicted data can be obtained by the inverse mapping of the isoline. Some experiment results verified the effectiveness of this method.