Intelligent nonlinear model predictive control of gas pipeline networks

In gas pipeline networks, the set-points should be carefully tuned to minimize the fuel consumption of compressor stations and meet the network requirements. In practice, the real demand has some variations over the forecasted one and consequently utilizing an appropriate controller to minimize the fuel consumption and manage the network variations is inevitable. The model predictive control is a great choice for systems with long delay such as gas networks. In this paper, an intelligent nonlinear model predictive control of a gas pipeline plant is proposed. It models the plant in fully transient state by a multi-layer perceptron neural network. The prediction power of the neural network is used to predict the plant output over a receding horizon. Initially, the network is trained offline and is then paralyzed with the real plant for online training. The proposed strategy consists of two main stages. In the first stage, the compressor set-points are optimized in the open loop condition considering the forecasted demand over a receding horizon and the resulting output pressures are chosen as the reference trajectories for the closed loop system. In the second stage, the controller is applied to compensate the demand variations. The optimization task is carried out using particle swarm optimization gravitation search algorithm (PSOGSA). Numerical results confirm the accuracy and robustness of the proposed controller in the presence of demand variations, noise and uncertainties.