基于ATD-BP神经网络的页岩气产量预测方法

储层改造是页岩气开发的关键步骤,根据储层改造数据进行页岩气井产量预测,对后续施工优化有重要指导意义。然而,储层改造数据与气井产量间呈非线性相关关系,不适用于传统线性预测方法。且储层改造数据存在有效数据较少、噪声数据占比较大、维数较高等问题,不适用于受噪声影响较大的传统BP神经网络非线性预测方法。由此,本文提出一种页岩气储层改造产量预测方法,首先利用自适应阈值去噪(adaptive threshold denoise,ATD)算法去除噪声,再运用BP神经网络对储层改造数据进行非线性拟合,得到页岩气井产量预测模型。实验表明,相比传统的BP神经网络,本文所提方法能够有效提高预测的准确率和稳定性。

Shale gas production prediction method based on adaptive threshold denoising BP neural network

Reservoir reconstruction is a key step in the process of development of shale gas. It is of great practical significance to predict the production of shale gas wells according to the data of reservoir reconstruction. However, there is a nonlinear relationship between the data of reservoir reconstruction and gas well production, which is not suitable for the traditional linear prediction method. And there are many problems in the reservoir reconstruction data, such as less effective data, larger noise data, higher dimension, and so on, which is not suitable for the traditional BP neural network prediction method. Thus, this paper presents a prediction method of shale gas reservoir production, firstly using adaptive threshold denoising (ATD) algorithm to remove noise, and then nonlinear modeling using BP neural network. Experiments show that compared with the traditional BP neural network, the proposed method can effectively improve the prediction accuracy and stability.