A learning-based artificial bee colony algorithm for operation optimization in gas pipelines

Abstract

The operation optimization of compressors is crucial for powering natural gas transportation and minimizing the gas consumption of the compressors themselves. In the literature, continuous control variables are typically discretized to cope with the curse of dimensionality by traditional dynamic programming methods and meta-heuristics, such as genetic algorithms and ant colony optimization. To provide a more accurate prediction, we developed a learning-based artificial bee colony (ABC) algorithm by integrating deep reinforcement learning. The merits of this innovation lie in two folds: 1) introduces function approximation to address challenges posed by the continuous state associated with gas consumption; and 2) improves the basic ABC's search capacity and reduces the risk of converging into local optima. Furthermore, the technology of multi-label classification is employed in the function approximation method to support the simultaneous optimal control of compressors in all stations, which can significantly enhance decision efficiency. Computational studies on real data demonstrate that the proposed method outperforms existing methods in the literature in terms of gas consumption.