In this paper, improved variational mode decomposition (IVMD) and support vector machine（SVM）are combined to predict the vibration response trend of the pipeline. Firstly, the decomposition mode of IVMD is determined by the mutual information criterion, overcoming the shortcoming of VMD which selects the decomposition parameters blindly. IVMD is used to decompose the vibration series of generating units and pipelines into several intrinsic mode functions which are used as the input and output of the SVM prediction model, respectively. Secondly, the optimal parameters of the SVM model corresponding to each modal component are determined by particle swarm optimization (PSO), and each component is predicted separately. Finally, the prediction results of each modal component are reconstructed to obtain the predicted value of the original series. Taking the No.2 pipeline of a pumping station as the research object, the three models of IVMD-SVM, PSO-SVM and BP neural network are adopted to predict the vibration response trend of pipelines, and the prediction results are compared and analyzed. The results show that the predictive value obtained by IVMD-SVM method is closer to the real value; moreover, the error is smaller and the calculation accuracy is higher. This method has certain utilization value for forecasting the vibration trend of pipelines and similar engineering structures.