The development of bottom water reservoir is affected by strong bottom water, low height of oil column, high viscosity of formation crude oil, and complex distribution of separation interlayer, etc. In order to solve the shortcomings of current conventional methods, such as large amount of multi⁃factor data analysis and application limitations, a new method based on BP neural network data mining algorithm for predicting recoverable reserves of horizontal wells in bottom water reservoir is proposed, and its reliability is analyzed by means of mathematical model. Aiming at the static and dynamic data of bottom⁃water reservoir, the effective hidden information is fully mined, and a data⁃driven model for the prediction of recoverable reserves of bottom⁃water reservoir is constructed on the basis of the establishment of basic data set. The results of practical application show that this method can realize the quantitative analysis of the influencing factors and the limit of technical parameters of horizontal well development in bottom⁃water reservoir, the maximum error of recoverable reserves prediction is less than 8%, and the fitting effect is good, which can be further applied to the research of production performance, development limit and well location design of horizontal Wells in bottom⁃water reservoir.

Extra high water cut has been the main development stage of bottom⁃water reservoir, due to water cut highly with great escalating rate in the initial. At present, the relative permeability curve endpoint has become a serious distortion. This paper presents a method to calculate the performance of water cresting with two phases flowing using the streamlines theory, good for solving the existing problems in the common method, as no pertinence, ineffectiveness and less affecting factors considered. Meanwhile, optimized relative permeability curve was calculated based on the research coupling genetic algorithm with bottom⁃water drive model. The result of numerical calculation based on optimized relative permeability curve showed that the histogram of technical reserves was matched basically with the real practice. This research could provide some instructional meaning to the application of dynamic prediction and developing the oilfield in bottom⁃water reservoir.