基于测井数据的砂岩型铀矿异常识别BP神经网络方法应用

Application of BP Neural Network Method for Anomalous Identification of Sandstone-type Uranium Deposit Based on Logging Data

为了快速有效的获取砂岩型铀矿矿集区铀矿异常分布信息,以砂岩型铀矿异常的测井响应特征为理论依据,利用BP神经网络强大的非线性映射和学习能力,以已知铀矿矿化层信息为学习样本,构建3层BP(back propagation)神经网络模型。对松辽盆地大庆长垣南端某铀矿矿集区铀矿钻孔测井数据进行异常层和矿化层的识别提取,并将模型识别结果与已知矿化层信息进行分析对比。结果表明:BP神经网络模型识别准确率达86.55%,效果较好,矿化层的识别结果同已知矿化层信息重合度高,同常规的铀矿异常识别方法相比更加接近铀矿异常分布的形态。此方法能快速有效的获取未知孔的异常信息、降低人为解释工作带来的误差,具有较高的准确性,优势明显。BP神经网络技术应用于铀矿勘察工作中具有良好的前景。

To effectively and efficiently obtain the information of sandstone-type uranium anomaly in the concentration area of uranium ore, a three-layered back propagation(BP) neural network was established with available uranium ore logging data based on the logging response characteristics of the sandstone-type uranium anomaly because of the nonlinear mapping and learning ability of BP neural network. The network was used to identify and extract the information of the anomaly and mineralization layers based on the uranium ore logging data of a uranium ore area in the southern end of the Daqing Placanticline of the Songliao Basin, and the identification results of model were compared with the known mineralized layer information. Results show that the identification of BP neural network model reached 86.55% in matching rate, and was highly coincident with the information of the known ore layer and closer to the abnormal distribution of the uranium ore than the conventional identification method. It was shown that the method was able to effectively and efficiently obtain anomaly information of unknown drilling and reduce the mistake of human interpretation, and has high accuracy and obvious advantages. Therefore, BP neural network has a great prospect in application for uranium exploration.