论文部分内容阅读
储层物性参数的估算是油气勘探开发的一个重要环节，然而物性参数的横向变化很难从分布稀疏的井孔资料插值外推得到精确描述，综合利用地震和测井资料可得到较精确的物性分布图像。传统地，将地震和测井资料综合来进行储层参数描述的方法有线性回归法，这种方法由于假定函数关系简单，因而很难得到准确的物性参数。后来Ｄｏｙｅｎ〔５〕发展了协同克里金法，这种方法可得到较准确的物性参数，但处理过程复杂，需作大量的统计研究。人工神经网络是一种高度非线性系统，可逼进任意函数，因此可以综合地震和测井资料，利用人工神经网络来描述储层参数。我们采用前馈神经网络模型，包含输入层、输出层和隐伏中间层，隐伏中间层的个数以及各隐伏中间层的单元数可调，输出层为测井导出的和实测的储层物性参数集，输入层为地震数据集另加位置坐标。为保证全局最优，我们采用模拟退火算法来训练网络。用文献〔５〕的例子进行了试验，表明该方法其结果与协同克里金法相当，但处理过程却简单得多。
Estimation of reservoir physical parameters is an important part of oil and gas exploration and development. However, lateral changes of physical parameters are difficult to be accurately described by extrapolation of sparse borehole data interpolation. Comprehensive utilization of seismic and well logging data can obtain more accurate physical properties Distribution image. Traditionally, the method of integrating seismic and well logging data to describe reservoir parameters has a linear regression method. This method is difficult to get accurate physical parameters because of the simple function assumption. Later Doyen [5] developed a collaborative kriging method, which can obtain more accurate physical parameters, but the process is complex and requires a large amount of statistical research. Artificial neural network is a highly nonlinear system that can be used to enter arbitrary functions. Therefore, the seismic data and logging data can be used to describe reservoir parameters by using artificial neural network. We use the feedforward neural network model, including the input layer, the output layer and the hidden intermediate layer, the number of hidden intermediate layers and the number of units of each hidden intermediate layer is adjustable. The output layer is derived from the well logging and measured physical properties of the reservoir Set, input layer for the seismic data set plus location coordinates. To ensure global optimization, we use simulated annealing algorithm to train the network. The experimental results of [5] show that the result of this method is equivalent to that of the collaborative kriging method, but the treatment process is much simpler.