Um algoritmo genético para otimização do controle de redes de escoamento de petróleo.

Abstract

Real-time control of complex and large-scale oil pipeline networks is complicated by several reasons, among them (a) reliability of data acquisition and communication systems, (b) strict time limits between data acquisition and decision of control action, (c) operational constraints of a large number of pipeline devices, such as tanks, pumps, valves and pipes, (d) multi-objective control, involving economic, operational, environmental and institutional objectives and constraints. In this work we propose a Genetic Algorithm (GA) to solve the problem of optimizing the control of complex pipeline networks in real time. Centered in the pumping system, the optimization algorithm uses domain based knowledge to reduce the search time and to improve the quality of the solution. The control objectives are reduction of costs with consumption of energy and risks to the environment, at the same time that production and the operational security levels are maintained. Standard GA crossover and mutation operators were modified to prevent early convergence and to speed up search in promising search space areas, giving to the traditional ‘blindness’ of the genetic operators an insight of the best way to apply them in order to generate better descendents. Seeding was used to overcome the problem of delivering a suitable solution on time. The GA introduces an evaluation function weighted over time. This function minimizes the possible loss that one may have due to uncertainty of the production forecast over time. The results showed that the GA provides better solutions than ad hoc procedures for pipeline network operation. Our experiments have shown an average reduction on cost of 5,45%. Moreover, we verified that additional gains (16,92% in our experiments) can be achieve increasing the amount of computational resources available.