Application of Particle Swarm Optimization for Enhanced Cyclic Steam Stimulation in a Offshore Heavy Oil Reservoir

Xiaolin Wang,Xun Qiu

doi:10.5121/ijitmc.2013.1204

Abstract

Three different variations of PSO algorithms, i.e. Canonical, Gaussian Bare-bone and Levy Bare-bone PSO, are tested to optimize the ultimate oil recovery of a large heavy oil reservoir. The performance of these algorithms was compared in terms of convergence behaviour and the final optimization results. It is found that, in general, all three types of PSO methods are able toimprove the objective function. The best objective function is found by using the Canonical PSO, while the other two methods give similar results. The Gaussian Bare-bone PSO may picks positions that are far away from the optimal solution. The Levy Bare-bone PSO has similar convergence behaviour as the Canonical PSO. For the specific optimization problem investigatedin this study, it is found that the temperature of the injection sCO2 composition in the injection gas, and the gas injection rates have bigger impact on the objective function, while steam injection rate and the liquid production rate ha ve less impact on the objective function.

Highlights

Particle Swarm Optimization (PSO) algorithm is co-operative, population-based global search swarm intelligence mataheuristics developed by James Kennedy and Russell C
Three different types of Particle Swarm Optimization Algorithm are tested on a large scale reservoir simulation problem
It is found that all types of PSO algorithm are effective in terms of searching for optimal objective functions

Summary

Introduction

Particle Swarm Optimization (PSO) algorithm is co-operative, population-based global search swarm intelligence mataheuristics developed by James Kennedy and Russell C. PSO algorithm has been successfully used as a high efficient optimizer in numerous area. In a Canonical PSO, the movement of each particle is determined by the weighted sum of three components: particle’s previous movement direction (inertia), the distance between current position and the local best position (cognition influence) and the distance between current position and the global best position (social influence). Kennedy and Mendes (2004) suggested a fully informed particle swarm where each particle’s movement is affected by all its neighbors instead of just the local optima of itself and the global optima.

Objectives

Results

Conclusion