The study area is located on the east side of the middle part of xx Depression in Zhujiangkou Basin. The Zhuhai Formation in the study area is a semi-anticlinal fault-nose structure controlled by the fault, with a near E-W strike, and the structure tilts north, east and south [15] [16]. The trap area of Zhuhai Formation is small, but its closing range is large. All faults in the oilfield are normal faults, with strikes mainly including near E-W strike, S-N strike, NEE-SWW strike and NW-SE strike (Figure 1).

During the sedimentation of Zhuhai Formation, the subaqueous distributary channel came from the southeast provenance, developed fan delta front to the northwest, and extended from the delta inner front controlled by river course to the delta outer front controlled by sea wave [17]. In general, due to the direct entry of the fan delta into the sea/bay, massive amounts of terrigenous debris are accumulating rapidly nearby. Therefore, the sedimentary sand body is well developed, and the sand body has a high space superposition rate. It can be found from the sedimentary facies of the study area that the interlayer is very well developed and the reservoir heterogeneity is strong during the well profile interpretation (Figure 2). Therefore, it is of great practical significance to establish a geological model which can well reflect the heterogeneity and lithologic characteristics of reservoirs for the analysis of the spatial distribution of sand bodies and the reservoir combination of the Zhuhai Formation in the study area.

The model was created by software of Petrel 2016, and the key parameters of drilling stratification, lithology type, and sedimentary microfacies type were usedtobuild the model. The lithologic curve of these wells that participate in the simulation is the result of the standardization of the logging curve. As the lithofacies are mainly divided into sandstone and mudstone in interpretation results, and that sandstone is also divided into reservoir sandstone and dry layer, hierarchical simulation is used to depict lithofacies distribution in the simulation process. Firstly, the distribution of sandstone facies and mudstone facies is simulated, and then the distribution of reservoir sandstone and dry layers in sandstone facies is simulated. The fluid phase model established by sand mudstone facies constraint can well reveal the characteristics of plane distribution and vertical superposition rules of reservoir sandstone, dry layer and mudstone.

The variogram is the key to stochastic modeling [18]. Thereinto, the parameter variation is the parameter describing the spatial correlation state of the regionalized variable, whose size reflects the distance between the variables in spatial correlation, and further reflects the degree of heterogeneity of variables in different directions. The size of the variable range and the main direction will directly affect the results of the stochastic simulation, so it is necessary to analyze the variogram in detail before the simulation. Based on the sedimentary background and provenance direction of the study area, it can be get that the main direction of the variogram is SEE-NWW, with the main range of 900 m, the second range of 600 m and the vertical range of 5 m, and other structural characteristic parameters of the variogram through a detailed analysis on the variogram. Based on the variogram analysis, the relative content and thickness distribution of sand-mudstone facies on well point are counted, and their relative content and vertical variation rules on each small layer plane are restrained. Based on the above data analysis, the sequential indicator method is used to simulate the lithofacies distribution probability model (Figure 3). In the lithofacies model of sandstone facies, relative content and thickness distribution characteristics of reservoir sandstones and dry layer well point and the variogram under constraints of sandstone facies are analyzed. Then finally, the fluid phase model can be simulated using the sequential indicator method under the constraints of sand-mudstone facies. Similarly, the fluid phase model of Indicator Kriging interpolation using the established process is established, so as to lay a foundation for the simulation and comparative analysis of reservoir spatial distribution in the study area by these two models in the next step.

As can be seen from the fluid phase models of the sequential indicator simulation and Indicator Kriging (Figure 4 and Figure 5): the results of Indicator Kriging interpolation shows an obvious smoothing effect, which ignores the random noise generated between wells, thus cannot well describe the heterogeneity of the underground reservoir; The results of sequential indicator simulation have eliminated the excessive smoothing effect, and the continuity of several lithologies is also good, thus they can fully depict the heterogeneity of underground reservoirs and the spatial distribution rules of lithofacies; The comparison of the lithofacies percentages of the model between sequential indicator simulation and Indicator Kriging interpolation shows that the result of sequential indicator simulation is closer to the percentage of lithofacies of well point statistics (Figure 6 and Figure 7). Whether the simulation results are consistent with the hard data of the well point is one of the most important indexes to test whether the model is reasonable or not; As also can be seen in Figure 7 and Figure 8, not only the sequential indicator simulation method can better describe the uncertainty between wells, but also the mudstone interlayer that is simulated by its simulation results is more in line with the geological understanding in Figure 2; Generally speaking, when the number of well location is fewer, the sequential indicator simulation method is better than the Indicator Kriging interpolation in characterizing the reservoir, and also can better reflect the uncertainty of the underground reservoir.

Conventional geological model test methods include thinning testing, new well testing and dynamic verification, etc. [19] [20]. Thinning testing is used in this paper to test the model, due to limitations in conditions and data. After removing some of the drilled wells, the lithofacies of the study area are simulated under the condition that the original simulation parameters are not changed, and the petrographic data of removed well locations are extracted from the simulation results to compare with the petrographic data of known well locations.

Well Z2 is randomly removed from the wells in the study area, and then the lithofacies of the study area is simulated using sequential indicator simulation method [4] (Figure 9 and Figure 10). The figure shows the comparison of the thinning results with the original fluid phase results. As can be seen from the figure, there is little difference between the simulation results of Z2 well after thinning and those of the original fluid phase; In the original fluid facies simulation results, there are 23 mudstone facies grids, 14 dry layer facies grids, 163 reservoir facies grids in well Z2, while in the simulation results after thinning, there are 30 mudstone phase grids, 20 dry layer facies grids and 150 reservoir grids in well Z2, with a coincidence rate of over 80%.