利用神经网络方法确定薄差层剩余油的分布

Determination of Remaining Oil Distribution in Thin and Poor Reservoir by Useing ANN Method

根据密闭取芯检查井资料和地质分析方法 ,通过人工神经网络 (ANN)模式预测 ,即利用ANN方法可以确定薄差储层可动剩余油。首先输入形成剩余油的主要参数 ,然后通过网络的不断学习 ,最后输出判别精度较高的含油饱和度、含水率或水淹级别等参数。该方法的技术关键是输入参数类型的确定 ,它涉及剩余油的形成机制和分布规律等问题。在深入探讨高含水油田开发后期剩余油成因类型的同时 ,还在诸多的剩余油影响因素中 ,确定了利用神经网络判别单井、单层剩余油的参数 ,即井点砂体类型、井点所处位置、注水井砂体类型、注水井距和注水时间。将研究方法应用在大庆长垣萨尔图油田北二区东部三次加密试验区 ,预测薄差层的水淹分布状况 ,对解决三次加密调整井区存在的问题很有成效。同时指出了对该识别方法产生影响的因素。

Artificial neural network pattern recognition technique (ANN), as a simulation and abstraction of human being's brain thoughts, can be used to recognize and classify objectives by imitating the transmission manner of nerve cell. The most popular ANN model at present is the error back propagation, which trains the nerve network by back propagation algorithm. A typical back propagation nerve network has three layer feed forward structure, consisting of input layer, cryptic layer and output layer. In this study, the ANN method is applied to recognize the remaining oil of thin and poor reservoir of Daqing oilfield, associated with the data of sealing coring inspection well and development geological method. The process is first to input the known parameters related to the formation and distribution of remaining oil; then select the suitable mathematic algorithm for calculation, and finally to obtain the parameters, including the accurate oil saturation, water bearing and water flooding degree. However, the key of this technique is to determine the input parameters which are related to formation mechanism and distribution of remaining oil. The authors analysed the development conditions and producing status of the thin and poor reservoir of Xing 2 1 Jian 29 well, which is located at Xingshugang, a typical district of oil field of Daqing. The results show that geological factor and the development factor are both important affecting the distribution of remaining oil. The remaining oil is usually distributed in the districts of sand bodies with discontinuous growth or incomplete injection production. The main parameter of ANN for recognizing the remaining oil of single well and single stratum is sand body type. The recognition model of water flooding degree and oil saturation is established by the nerve network training. This model was tested by the data of other sealing coring inspecting wells, and the average error was 8.4 %, which indicates that the recognition model is good in use. The authors applied this model to densified well pattern testing district in Sa'ertu oil field of Daqing to analyse and interpret the water out degree of perforated reservoir, which could predict the water out distribution of thin and poor reservoirs effectively.