论文部分内容阅读
一些油藏（例如试井解释）系统的偏微分方程模型，经过变换能化为非线性函数项级数。级数的每一项均为地层参数θ的复杂非线性函数。级数的项数ｎ与模型结构有关，可称为模型的结构参数。把级数中的函数看成非线性神经元，来建立油藏系统的函数型连接人工神经网络模型。用系统辨识理论中的Ｆ检验法确定网络模型的结构参数ｎ，用多步广人梯度学习算法估计网络模型的权系数。地层参数是试井解释的依据，要求其估值应具有唯一性，而上述函数为多峰函数，在极值点处关于θ的变化很敏感，使问题更为困难，现有迭代法和遗传算法均未奏效。一种新型的遗传算法解决了这个问题。应用表明用上述方法建模有很高的精度，能求出地层参数的唯一估值。
In some reservoirs (for example, well test interpretation), the system of partial differential equations can be transformed into non-linear function series after transformation. Each of the series is a complex non-linear function of formation parameter θ. The number of series n is related to the model structure and can be called the structural parameter of the model. The functions in the series are treated as nonlinear neurons to establish the functional connection artificial neural network model of the reservoir system. The F-test of system identification theory is used to determine the structural parameters of the network model n, and the weight coefficients of the network model are estimated by the multi-step and wide-ranging gradient learning algorithm. Stratigraphic parameters are the basis for well test interpretation and require that their estimates should be unique. The above function is a multimodal function, sensitive to the change of θ at the extremum point, making the problem more difficult. Existing iteration methods and genetic The algorithm did not work. A new type of genetic algorithm solves this problem. Applications show that modeling with the above method has a high degree of accuracy and can yield unique estimates of formation parameters.