Abstract Accurate delineation of hydrocarbon seepage has significant implications in accentuating hydrocarbon migration pathways and assessing seal integrity thereby alleviating drilling hazards. Knowledge of migration pathways and understanding about the source/reservoir plays a vital role in successful evaluation of hydrocarbon seepage. Many researchers have reported events of gas leakage in Poseidon area; however, it has never been investigated in detail to confirm the origin of the gas leakage and their migration pathways. In this study, an attempt has been made to decipher the relationship between shallow gas migration expressions such as pockmarks, mud-volcanos, direct hydrocarbon indicators (DHIs), push-downs, amplitude blanking, with the reservoir present in the study area. Adopted approach includes development of a chimney probability cube (CPC), in which extracted seismic attributes are optimally combined using a non-linear multi-layer perceptron (MLP) network. During the training and testing phase of the 3-layer MLP network, 0.3-0.25 normalized RMS error and

中文翻译：

---

使用多属性分析和人工神经网络方法从 3D 地震数据中检测气烟囱及其与澳大利亚西北大陆架波塞冬深部储层的联系

摘要 碳氢化合物渗漏的准确描绘对于强调碳氢化合物运移路径和评估密封完整性从而减轻钻井危险具有重要意义。对迁移路径的了解和对源/储层的了解在成功评估油气渗漏方面起着至关重要的作用。许多研究人员报告了波塞冬地区的气体泄漏事件；然而，对于气体泄漏的来源及其迁移途径，从未进行过详细调查。在这项研究中，试图破译浅层气体运移表达式之间的关系，例如麻点、泥火山、直接烃指示剂 (DHI)、下推、振幅消隐与研究区存在储层之间的关系。采用的方法包括开发烟囱概率立方 (CPC)，其中使用非线性多层感知器 (MLP) 网络优化组合提取的地震属性。在 3 层 MLP 网络的训练和测试阶段，0.3-0.25 归一化 RMS 误差和