Energy consumption prediction plays an important role in pipeline operation regulation and energy management. Accurate energy consumption prediction is helpful to make important decisions, including unit commitment, batch scheduling, load dispatching, energy consumption target setting, etc. The energy consumption of crude oil pipeline is mainly the electrical energy of pump unit. The average annual electrical energy consumption of China’s crude oil pipelines accounts for more than half of the annual operating cost of pipelines. Therefore, the prediction of electrical energy consumption of crude oil pipelines is critical.
The energy consumption prediction of crude oil pipelines is very complicated. Firstly, it depends on the variables related to operation parameter, crude oil physical property parameter, environmental parameter and equipment parameter. Secondly, its nonlinearity is strong. Thirdly, the available samples are too little. Through the study on the monthly operation data collected by the Supervisory Control And Data Acquisition (SCADA) system and energy consumption analysis, the turnover and the electrical energy consumption is selected as input variable and output variable, respectively. The support vector machines (SVM) is introduced to predict the monthly electric energy consumption of crude oil pipelines driving oil pumps. However, the generalization capability of SVM is highly dependent on appropriate parameter setting, such as penalty coefficient and kernel parameter. The selection of the optimal parameters is critical to achieving good performance in the learning process. Therefore, in order to improve the generalization ability, GridSearchCV was adopted to optimize the hyperparameters of SVM.
Taking a crude oil pipeline from Qinhuangdao City, Hebei Province to Fangshan District, Beijing as an example, the actual operation data for four consecutive years (48 months) are used for this study. The data are divided into training set and test set by stratified sampling method, which consist of 28 samples and 20 samples respectively. The mean absolute percentage error (MAPE), root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R2) on the test set are 3.42, 21.64, 14.31 and 0.94 respectively. Compared with other five state-of-the-art prediction methods in predictive accuracy, the result shows that GSCV-SVM has the best performance in the case of small samples, and the prediction results are in good agreement with the actual data.