摘要
对测井资料进行必要的预处理和合理的取舍后，与地质参数结合，建立起正确的样本集。设计ＢＰ网神经网络，修改神经元作用函数，调整神经元的权值递推式，对样本进行网络训练，在网络系统误差小于允许误差时，网络训练结束。使用得到的网络模型参数，就能得到所需结果。输入测井资料、计算储层参数，计算结果与岩心分析结果相比较，其误差很小。在测井所处理的储层参数与储层产能对应关系很差的情况下，将储层参数与储层产能挂钩，设计ＢＰ网络进行样本训练，训练过程中修整了步长调整因子和平滑因子，得到网络模型参数后进行储层产能评价，效果很好，精度较高。对汉明网络的结构、神经元的权值、域值和输出函数作了改进，使其适用于输入为连续值的模式识别，经实际资料处理证实，该网络有较强的模式识别能力，并见到良好效果。
After the essential pretreatment and reasonable choice of the logging data have been done,a correct sample set may be established by combining the logging data with geological parameters.For designing the BP neural network,it is necessary to correct neuron action function,to adjust the weighted value recursion formula of neurons and to carry out the network training for the samples.When the system error of the network is smaller than the allowable error,the network training will be ended.By use of the network model parameters obtained,the result needed will be got.Reservoir parameters can be calculated through inputting log data and by comparing the calculated result with core analysis result,the error of the former is very small.When the parameters calculated by logging are badly related with reservoir productivity,it is necessary to link the reservoir parameters with reservoir productivity to design BP network to carry out sample training.After the model parameters have been got through correcting step width adjustment factor and smoothing factor in the process of training,the effect of evaluating reservoir productivity is very good with high accuracy.The structure,neuron's weighted value,domain value and output function of Hamming network were improved to suit for the pattern recognition with successive input values.It is proved by treating practical data that such a network has stronger pattern recognition ability,a good result being obtained.
出处
《天然气工业》
EI
CAS
CSCD
北大核心
1997年第5期23-26,共4页
Natural Gas Industry
关键词
神经网络
测井解释
储集层
预处理
测井资料
Nerve network,Log interpretation,Reservoir,Production calculation,Error,Comprehensive evaluation.
