Abstract—In this paper, a new approximate gradient method is proposed for constrained reservoir production optimization. The new algorithm method is gradient-free, which is a compromised solution to finite-difference method. To get a quick evaluation of the gradient, all parameters are perturbed at one time stochastically and the calculated gradient is also stochastic. Based on the relationship between gradient and direction derivative, we construct a new search direction with the stochastically generated perturbation vector. It is proved that the stochastic gradient is always an uphill direction, ensuring that a better solution can be found along the stochastic gradient direction. Besides, projected gradient method is incorporated into the new algorithm to deal with constraints in production optimization. A comparison is made between the new algorithm and simultaneous perturbation stochastic approximation (SPSA) algorithm using a synthetic reservoir case. The results show that the new method outperforms SPSA in constrained production optimization problem. After optimizing the production strategy for a synthetic reservoir, the economic benefit improves about 20%.

Index Terms—constrained reservoir production optimization, SPSA algorithm, approximate gradient, directional derivative, projected gradient method

Cite: Shaolei Wei, Linsong Cheng, Wenjun Huang, and Hao Gu, "A New Approximate Gradient Algorithm Applied in Constrained Reservoir Production Optimization," Journal of Industrial and Intelligent Information, Vol. 2, No. 3, pp. 194-199, September 2014. doi: 10.12720/jiii.2.3.194-199