A water absorption profile prediction method based on a small sample condition

Abstract

The invention discloses a water absorption profile prediction method based on a small sample condition, which collects multi-source data of an oilfield block to be analyzed. Connectivity analysis andgrey relational analysis were carried out to determine the static and dynamic factors affecting water intake profile, and the normalized treatment was carried out to form a standard water intake profile small sample library. The cost function of integrated multi-task is established by using the sublayer as the machine learning unit, and the generalized model adapting to the prediction of water absorption of each sublayer is obtained by using the gradient descent algorithm. Relying on the limited water injection profile data of water injection wells, further parameter fine-tuning and personalized learning is carried out, the water injection prediction model adapted to the water injection splitting law of water injection wells is established, and the continuous dynamic prediction of water injection profile is realized based on the model. Based on the machine learning theory under the condition of small samples, the method realizes the accurate splitting of water injection quantity and the prediction of water injection profile, has important significance for recognizing the distribution of underground remaining oil, and is the basis for realizing the stratified production allocation and injection allocation of an intelligent oilfield.

Description

technical field [0001] The invention belongs to the field of oil and gas field development, and in particular relates to a water absorption profile prediction method based on small sample conditions. Background technique [0002] Waterflooding is the main technology of China's oilfield development. Continuing to improve water flooding recovery in the late period of high water cut and ultra-high water cut period is still one of the main directions for improving recovery. Although it is very difficult, it has wide adaptability. However, most of China's oilfields are continental oilfields, with complex depositional rhythm and serious reservoir heterogeneity, resulting in particularly prominent interlayer and intralayer contradictions in the development process, which in turn leads to the tongue of injected water in the direction of the production well on the plane. Phenomenon and the phenomenon of vertically protruding into the high-permeability layer, so that the injected wate...

AI Technical Summary

Problems solved by technology

The existing methods mainly have the following three problems: 1. They cannot accurately reflect the dynamic changes of the water absorption profile data, and consider less dynamic factors, which are not comprehensive enough; Profile prediction, there is little research on the prediction of water absorption profile of injection wells without water absorption profile data; 3. Using machine learning methods that rely on big data training to solve the problem of small sample learning, the prediction accuracy of water absorption profile is low

Images