Due to the addition of the drag reducer in refined oil pipelines for increasing the pipeline throughput as well as reducing energy consumption, the classical method based on the Darcy-Weisbach Formula for precise pressure loss calculation presents a large error. Additionally, the way to accurately calculate the pressure loss of the refined oil pipeline with the drag reducer is in urgent need. The accurate pressure loss value can be used as the input parameter of pump scheduling or batch scheduling models of refined oil pipelines, which can ensure the safe operation of the pipeline system, achieving the goal of energy-saving and cost reduction. This paper proposes the data-driven modeling of pressure loss for multi-batch refined oil pipelines with the drag reducer in high accuracy. The multi-batch sequential transportation process and the differences in the physical properties between different kinds of refined oil in the pipelines are taken into account. By analyzing the changes of the drag reduction rate over time and the autocorrelation of the pressure loss sequence data, the sequential time effect of the drag reducer on calculating pressure loss is considered and therefore, the long short-term memory (LSTM) network is utilized. The neural network structure with two LSTM layers is designed. Moreover, the input features of the proposed model are naturally inherited from the Darcy-Weisbach Formula and on adaptation to the multi-batch sequential transportation process in refined oil pipelines, using the particle swarm optimization (PSO) algorithm for network hyperparameter tuning. Case studies show that the proposed data-driven model based on the LSTM network is valid and capable of considering the multi-batch sequential transportation process. Furthermore, the proposed model outperforms the models based on the Darcy-Weisbach Formula and multilayer perceptron (MLP) from previous studies in accuracy. The MAPEs of the proposed model of pipelines with the drag reducer are all less than 4.7% and the best performance on the testing data is 1.3627%, which can provide the calculation results of pressure loss in high accuracy. The results also indicate that the model`s capturing sequential effect of the drag reducer from the input data set contributed to improving the calculation accuracy and generalization ability.