Abstract In a natural gas network the gas pressure decreases continuously due to friction. The compressor stations are placed in strategic locations of networks to make up for the lost pressure. This ensures the delivery of time-varying customer demands in desirable pressure. Howbeit, the compressor's operational costs constitute the major portion of a network costs. Different approaches can be used to find the optimum operation of the gas network compressor's when the demand profiles are known from the historical data. However, the demand profiles may become different from their long time average. For this case, a novel on-line predictive control scheme is presented in this work. This scheme finds the near optimum operation of the compressor more easily and in much less computational time, while eliminating the necessity of re-solving the optimization problem. The proposed strategy utilizes two multi layered neural networks (MLNNs) for on-line prediction and control tasks. The NN predictor is used for on-line prediction of highly nonlinear dynamics of the gas network in transient state. The on-line NN controller uses the prediction information to find the near optimum control inputs (rotational speeds of the compressors) to provide the new desired demands. This can be achieved by tracking the previously obtained optimum outlet pressures of the network. The controller results are validated with another controller which uses the particle swarm optimization (PSO) algorithm as the optimizer. To investigate the controller operation in performing the optimization task, its simulation results are compared with global optimum results, which assert its suitable performance. To investigate the robustness of the control scheme, the network outlet demand flow rates are changed in two different scenarios. Moreover, the performance of the controller is evaluated in the presence of noise and disturbances, which confirms its efficiency and accuracy.

中文翻译：

---

基于神经网络的天然气管道瞬态预测控制器设计

摘要 在天然气管网中，气体压力由于摩擦而不断降低。压缩机站位于网络的战略位置，以弥补失去的压力。这可确保以理想的压力满足随时间变化的客户需求。然而，压缩机的运行成本构成了网络成本的主要部分。当需求曲线从历史数据中得知时，可以使用不同的方法来找到气体网络压缩机的最佳运行。但是，需求概况可能会与它们的长期平均值有所不同。对于这种情况，本文提出了一种新颖的在线预测控制方案。该方案更容易且以更少的计算时间找到接近最佳的压缩机运行，同时消除了重新求解优化问题的必要性。所提出的策略利用两个多层神经网络 (MLNN) 进行在线预测和控制任务。NN 预测器用于在线预测瞬态气体网络的高度非线性动力学。在线 NN 控制器使用预测信息来寻找接近最佳的控制输入（压缩机的转速）以提供新的所需需求。这可以通过跟踪先前获得的网络最佳出口压力来实现。控制器结果通过另一个控制器进行验证，该控制器使用粒子群优化 (PSO) 算法作为优化器。为了研究控制器在执行优化任务时的操作，将其仿真结果与全局优化结果进行比较，断言其合适的性能。为了研究控制方案的稳健性，网络出口需求流量在两种不同的情况下发生变化。此外，在存在噪声和干扰的情况下评估控制器的性能，这证实了其效率和准确性。