A Comprehensive Machine Learning Algorithm Based Methodology for Intelligent Water Injection Optimization - Case Study of D Field in Asia

Abstract In the oilfield development by water flooding, lower injection rate results in insufficient reservoir energy. Conversely, a higher injection rate usually results in an early breakthrough and rapid water cut development, which causes serious invalid and low-efficiency water cycling. The development of fine separated layer water injection technology which improves the water injection effect can better adapt to multi-layered reservoir and can alleviate the interlayer conflicts. Meanwhile, more accurate injection allocation is required to achieve good performance. Fine optimization of suitable water injection rate of each well and interval is critical for effective and successful water injection. This paper presents a fast and intelligent water injection allocation method based on comprehensive machine learning algorithm. The evaluation indexes of injection performance are established by integrating dynamic production data, such as injection-production ratio, water consumption rate, water cut, injection intensity, daily oil production capacity. According to the evaluation indexes, it is necessary to determine the water injection adjustment to achieve better effects. Random forest algorithm (RF) is applied to determine the allocation type for wells and intervals, such as strengthened, controlled, and unaltered wells and intervals. Subsequently, the multi-objective simulated annealing algorithm (MOSA) is chosen to specify the water injection rate and proportion for each well and interval that needs to be adjusted. This intelligent water injection optimization methodology is formed to realize scientific and reasonable water injection allocation. The water injection plan optimized through intelligent methodology was successfully implemented in a pilot project in the D oilfield which is a multi-layered sandstone reservoir in Asia. The selection and optimization of plan is as the goal of high oil and liquid production and low water cut. Dynamic data of the pilot, collected over eight months after applying this optimal plan, indicate that the single well oil production capacity increased from 1.42 to 1.5 tons per day, reflecting an increase of 5.6%. The water cut decreased from 96.33% to 95.93%, showing a slight decrease of 0.4 percent without any subsequent increase. The injection-production ratio improved from 1.04 to 1.05 with an increase of 0.01. The intelligent water injection optimization method has yielded favorable development effects. The new method promotes reasonable allocation and water injection development to fine and intelligent. It also provides a methodology and reference case for reservoir and production engineers to develop other similar fields.