【作者】 沈永红；

【导师】 匡建超； 王玉兰；

【作者基本信息】 成都理工大学 ， 应用数学， 2007， 硕士

【摘要】 油气储层预测是地球物理勘探的一个重要任务，预测的准确与否将直接决定勘探开发的成功与失败，历来受到油气勘探开发专家的重视。而寻求一种较好的油气储层预测方法则是油气储层预测的重要环节。近年来，有不少研究人员提出了许多油气储层预测方法，如模式识别方法、神经网络方法、模糊模式识别方法、灰色遗传过程识别方法等。诚然这些方法在实际油气勘探开发中都取得了一定的效果，但是又存在着诸如有效属性选取、网络优化以及隶属度函数的确定等问题。目前，利用地震资料进行油气储层预测的各种方法适用条件不同，预测效果也有差异，各有优点和局限性。但无论采取何种预测方法，都与对大量观测和收集的数据的处理、归纳、分类等相联系，因此必然涉及到不完整、不精确知识等概念。RS理论是一种新的处理模糊和不确定性知识的数学工具，仅凭观测数据本身删除冗余信息，提取有效属性。但其容错能力与推广能力相对较弱，且只能处理量化数据，而信息熵可以作为知识分类时属性重要性的量度。为此本文在分析、归纳现有研究结果的基础上提出了一种新的属性离散化方法。小波神经网络集小波分析与神经网络两者的优点，具有较强的容错能力与推广能力以及更强的函数逼近能力和识别能力。因此本文将充分利用RS理论、信息熵、小波神经网络三者的优点，主要围绕油气储层预测中的重要环节来组建一种新的预测模型，并以伊朗卡山地区Aran构造带为例来对该模型进行可行性与有效性分析。更多还原

【Abstract】 The oil gas reservoir prediction is a important task in the geophysical exploration domain, the prediction accuracy will directly determine the success or failure of exploration development, so it always is valued by the oil gas exploration development expert. Therefore it is important part to seek a better method for predicting the oil gas reservoir.In recent years, many researchers have proposed some methods for predicting the oil gas reservoir, such as Pattern Recognition, Neural Networks, Fuzzy Pattern Recognition, Gray Process of Genetic Recognition and so on. Indeed, these methods have obtained the certain effect in the actual oil gas exploration development, But there is also some questions such as the effective attribute selection, the network optimization as well as a subordination function determination and so on.At present, the applicable conditions are different that the methods using seismic data to reservoir prediction methods, the prediction results are different, respectively has own advantages and limitations. But no matter what predicting methods, all with relate to a large number of data which was observed and collected, which need to be processed or inducted or classified. Therefore it inevitably involves to the concept like incomplete or inaccurate knowledge and so on. The RS theory is new mathematical tool for dealing with the fuzzy and uncertain knowledge mathematical instrument, only depends on observation data itself to delete the redundant information, withdraws the effective attribute. But the fault-tolerant ability and the promoted ability are relatively weak, and only can process the quantification data, and information entropy can be used as a measure of the attributes importance when the knowledge is classified. So this paper gives a new discretization method. Wavelet Neural Network has the advantages of both of wavelet analysis and neural network, with strong and fault-tolerant ability and the promoted ability as well as greater recognition and ability to function approximation. Therefore, this paper will make full use of three advantages of the RS theory, the information entropy and the wavelet neural network, and set up a new prediction model that mainly aim at the important part in the oil gas reservoir prediction, and analyzes the feasibility and effectiveness of the model as an example of the Aran belt of the Kashan area in Iran.更多还原