Accurate short-term energy consumption prediction is crucial to energy management of crude oil pipelines. Based on the prediction results of energy consumption, some vital decisions such as energy consumption target setting, scheduling optimization and unit combination can be implemented effectively. The energy consumption of crude oil pipelines covers all aspects of the pipeline transportation system, among which the electricity consumption of the pump units is the most extensive. The electricity consumption of the pump units by far accounts for the major part of the energy consumption of the crude oil pipelines. Therefore, it is necessary to accurately predict the energy consumption of the pump units, so as to have an overall assessment for the energy    consumption of the pipeline system. At present, there is a wealth of methods that can be used to predict the energy consumption of    crude oil pipelines. In traditional prediction methods, there are many limitations that make the prediction results deviate from the    actual energy consumption. Generally speaking, the neglect of noise interference and the lack of in-depth research on the nonlinear    characteristics of the data are the most common problems. The above factors complicate the energy consumption prediction of crude    oil pipelines and make the prediction accuracy unsatisfactory. In order to solve the shortcomings of the traditional prediction meth   ods, a novel hybrid prediction method is proposed for the short-term energy consumption prediction. The proposed hybrid method    is based on the decomposition technique, stratified sampling, a modified particle swam algorithm and a back-propagation neural    network. The proposed model consists of four parts: the data preprocessing module, the optimization module, the prediction module    and the evaluation module. The decomposition technique is adopted to eliminate the redundant noise and extract the major features    of the original data. The stratified sampling method is used to divide the data set to avoid the sampling bias of random sampling. The    back-propagation neural network optimized by the modified particle swarm optimization algorithm is regarded as a predictor. Based    on three crude oil pipelines located in China, the proposed prediction model is evaluated by comparing the predicted results with the    actual data. The mean absolute percentage errors of the evaluation indicators are 4.02   %   , 3.58   %   and 3.88   %   respectively. Compared with    several popular machine learning methods and the prediction modules in SPS software, the proposed prediction method has excellent    prediction accuracy and generation ability, which can be used for short-term energy consumption prediction of crude oil pipelines.