Portal de Periódicos da CAPES

Production Optimization at the Kuparuk River Field Utilizing Neural Networks and Genetic Algorithms

1999; Linguagem: Inglês

10.2118/52177-ms

R. F. Stoisits, Kelly D. Crawford, D. J. MacAllister, Michael D. McCormack, Akanni S. Lawal, David O. Ogbe,

Drilling and Well Engineering

Abstract The Kuparuk River Field, located on the North Slope of Alaska, contains approximately 6.4 billion barrels of oil. Total cumulative production to date is approximately 1.4 billion barrels, and current average daily production is 270,000 barrels of oil. The total number of wells in the field is approximately 900. Oil, water, and gas are separated at three Central Processing Facilities. Gas is compressed at each facility for use in gas lift operations and for miscible and immiscible water alternating gas pressure support projects. Kuparuk is production facility limited, which imposes gas, and water handling constraints on the field. Allocation of wells to the production facilities as well as allocation of lift gas are key elements in maximizing oil production rate. An optimization program utilizing formation gas oil ratio and incremental gas oil ratio computations for well and lift gas allocations is currently in use. The objective of this work was to develop an optimization model which would be an improvement over the current model and also be superior to commercially available optimization software. This required the development of a field production model which included well performance, surface line, and facility models. For each iteration in the optimization model oil rate from the field production model is computed. Two major impediments to the rigorous optimization of large producing fields are the effectiveness of the optimization algorithm and the cost of computing the state of the objects being optimized. In order to successfully solve for an optimal solution to this large highly nonlinear allocation problem a technique based on genetic natural selection process is used. In order to address the problem of excessive computation time neural networks were developed to replace the surface line hydraulics simulation. The new optimization program was evaluated using twenty-seven days of field data. The average daily production rate based on the 27 days of information was 273,000 bopd. The program was run in predictive mode and yielded an average 272,000 bopd for these 27 days. This indicated the field production model was reasonably accurate. The program was then run in optimization mode to generate recommendations for allocation of wells to production and allocation of lift gas. Average forecasted optimized rate for the 27 days was 292,000 bopd, which is a 7% increase in production over the current optimization forecast method.