关键词: 低渗透砂岩, 孔隙结构评价, 线性判别分析, 有监督自组织映射

Abstract: The pore structure determines the storage and seepage capacity of the reservoir, and controls the initial production capacity of the oilfield and even the distribution of the remaining oil in the middle and late stages of development. However, at present, the characterization of the pore structure mostly stays in the core scale. To up-scale characterization of pore structure, pore structure types of cores are divided based on pore and throat size, fluid mobility and physical property parameters obtained from high-pressure mercury intrusion, nuclear magnetic resonance, and conventional physical property analysis experiments, and are used as learning supervised samples. LDA (linear discriminant analysis) algorithm is used to excavate log curves sensitive to pore structure features, and a nonlinear log prediction model is established in collaboration with SSOM (supervised self-organizing map) algorithm to evaluate pore structure of  low-permeability sandstones  in the fourth member of J oilfield in Bohai Bay basin. The results show that there are six types of pore structures developed, among which, microfractures are developed in the type Ⅰ pore structure, which is characterized by low porosity and relative high permeability. The type Ⅱ pore structure is mainly developed in fine sandstone and consists of residual intergranular pores and secondary dissolved pores. The type Ⅲ pore structure is mainly composed of secondary dissolution pores and a few residual intergranular pores, which are common in siltstone. The type Ⅳ pore structure is mainly composed of secondary dissolved pores, and pore-filled carbonate cements are commonly observed. The type Ⅴ pore structure is characterized by extensive carbonate cementation, and the pore system is dominated by micropore throats. The volume fraction of argillaceous matrix in type Ⅵ pore structure is high, and the phenomenon of relatively high porosity and low permeability often occurs. The overall accuracy rate of the LDA-SSOM prediction model attains 86.20%. In terms of the accuracy rate in blind tests, which stands at 82.67%, it outperforms prediction models like LDA-BP (back propagation) (77.00%), LDA (65.67%), and SSOM (73.33%). Moreover, it is capable of fulfilling the objective of escalating the research scale regarding pore structure.

Key words: low-permeability sandstones, pore structure evaluation, linear discriminants analysis, supervised self-organizing map