Online Optimization of Heated-Oil Pipeline Operation Based on Neural Network System Identification

Traditional optimization methods of heated oil pipeline operation are usually offline or nondynamic, where steady-state models or complex and inaccurate physical models are utilized. To achieve online optimization with high adaptability and timeliness, this paper deploys a neural network (NN) to identify and imitate a real thermal and pressure response system instead of solving complicated model equations. The NN is trained and built by pipeline history data obtained from the Supervisory Control and Data Acquisition (SCADA) system. For a certain pipeline, real-time operating parameters like pressure and temperature are predicted by the NN once the inputs are given. Three different neural networks, a backpropagation neural network (BPNN), radial basis function neural network (RBFNN), and general regression neural network (GRNN), are compared to demonstrate their error-control abilities when predicting those parameters. Then the parameters are sent to particle swarm optimization and differential evolution (PSO-DE) optimization units to calculate the costs. The NN identification and PSO-DE optimization are applied to the Rizhao–Yizheng digital long crude oil pipeline. Experiments demonstrate that the three NNs have similar prediction accuracies, and the RBFNN and GRNN have better prediction stability than BPNN. Most of the prediction errors of the flow, outlet pressure, and inlet temperature can be controlled below $50{\mathrm{m}}^3/\mathrm{h}$, 0.1 MPa, and 0.1°C, respectively, and the online optimization is achievable. The maximum reduction of the total energy cost is 10.75%, and the saving effect is significant.