改进遗传神经网络在洛带气田测井解释中的应用

洛带气田蓬莱镇组的主要储产层为浅水三角洲沉积，含油气资源比较丰富，但地质情况非常复杂，测井物性参数解释十分困难。文章针对本地区储层非均质性强、岩性复杂，用常规测井方法难以解释的特点，在测井解释时选用了神经网络计算法，并且应用遗传算法（ＧＡ）对传统的BP神经网络进行了改进。在此基础上，利用多种测井数据及岩心分析资料作为网络模型的学习样本，通过网络的学习、训练，建立了渗透率等参数的神经网络计算模型，应用此模型计算了新样本的预测物性参数值。计算结果与实际岩心分析结果相比较后表明，遗传BP算法的学习速度和计算精度均令人满意。

APPLICATION OF BP NEURAL NETWORK IMPROVED BY GENETIC ALGORITHM IN LOG INTERPRETATION IN LUODAI GAS FIELD

The main reservoirs and pay zones of Penglai zhen Formation, Upper Jurassic, in Luodai gas field belong in shallow water delt a facies and are rich in hydrocarbon resources. However, the log interpretation of petrophysical parameters was very difficult because of complicated geological conditions. The BP neural network improved by genetic algorithm was used for lo g interpretation, because the conventional logging methods were difficult of int erpretation owing to the strong heterogeneity and complex lithology of reservoir . For this reason, by taking several log and core analysis data as the learning sample of network model, the neural network model of permeability and so on was established through the learning and training of network, then the petrophysical parameters of new sample might be predicted by applying the model. By comparing the calculation results with the core analysis results, it was shown that the i teration speed and calculation accuracy are satisfactory.