遗传神经网络在煤质测井评价中的应用

Application of Genetic Neural Network to Coal Quality Evaluation Based on Log Data

煤层气储层具有很强的非均质性和各向异性,使得测井资料解释结果具有多解性、模糊性和不确定性。提出了将遗传算法和神经网络相结合的方法,利用遗传算法优化神经网络的连接权值及阈值,从而提高网络训练精度和煤层气储层评价精度。该方法避免了标准BP算法易陷入局部最小和遗传算法局部搜索能力较差的缺点,提高了运算速度。介绍了利用遗传算法优化网络连接权值及阈值的步骤和煤质参数预测步骤。通过选取学习样本、确定网络结构、归一化处理数据,建立了基于GA-BP神经网络的煤层气储层煤质测井评价模型。对26个样本数据的分析对比表明,该算法具有较高的预测精度和较快的运算速度。10多口井的实际应用表明,GA-BP神经网络模型预测结果与煤心测试数据匹配很好,且与体积模型计算结果具有良好的一致性。

Coalbed methane reservoir log data interpretation results often show multi-solutions, ambiguity and uncertainty due to its heterogeneity and anisotropy. Put forward is a method to improve network training accuracy and coalbed methane reservoir evaluation accuracy by combining genetic algorithm and neural network. This method uses genetic algorithm to optimize neural network connection weights and threshold. It increases computing speed by avoiding its disadvantages that standard BP algorithm is apt to trap in local minimal solution, and genetic algorithm is weak at the locally searching capability. Introduced is the process for optimizing neural network connection weights and threshold and coal quality parameters forecast. Established is a coal quality log evaluation model based on GA-BP neural network, learning-samples selection and network structure determination, and data normalizing. Comparative analysis of 26 samples shows that this algorithm has higher accuracy and faster processing speed. Practical applications in more than 10 wells indicate that the prediction results of GA BP method match well with coal core test data, and have good consistency with volume model calculation results.