Abstract:

The Ordovician reservoir of the Tahe oilfield is a complex fracture and vuggy karst carbonate reservoir which experienced multiphase karst and tectonic disruption. This is very difficult to identify the types of karst cave by conventional log data. The main reservoir space include karst cave, vug and cracks, which control distribution and accumulation of the reservoir fluid. It is very important for petroleum development that the different reservoir types are divided and distinguished. In this paper, based on six key wells of electric imaging logging (FMI), and the core sample characteristics, combined with the corresponding conventional logging data, the carbonate reservoir of the Tahe oilfield is divided into four categories：unfilled cave，collapsed cave, cave-filling of sand shale, solution fracture type. According to the conventional well logging response and result of four kinds of reservoir division, seven kinds of logging information GR, RD, RS, K3 (of the absolute value of lateral resistivity difference from bottom), AC, DEN and CNL are taken as inputs of neural network learning samples. By training neural network model of reservoir division, the neural networks model of reservoir type’s identification is established. Based on this network model, take TK604 well as an example to test the classification in the Tahe areas 6 and 7. Real data processing has proved that this method is effective. Through the actual data processing proved that the conventional log multiparameter constraint can be used to judge karst reservoir type, and initially semi-quantitative evaluation standard is formed, showing preliminarily that the method of the neural network method based on the imaging logging is suitable for complex carbonate reservoir in Tahe oilfield.

Key words: Tahe oilfield, carbonate rock, karst reservoir, neural network, log response