The transient process of gas-liquid two-phase CO2 flow can occur in carbon capture and storage pipelines. Large measurement errors exist when Coriolis mass flowmeters are used to measure the mass flowrate of CO2 under such conditions. To solve this problem, a method for mass flowrate correction based on a gated recurrent unit (GRU) neural network is proposed. Since the GRU is suitable for dynamic process prediction, the GRU model is trained by using the collected datasets from a CO2 gas-liquid two-phase flow rig and optimized by using a grid search method combined with the K-fold cross-validation. The optimized GRU model is evaluated in terms of measurement accuracy and generalization capability by using eight groups of datasets under typical experimental conditions. The GRU model is compared with the least squares support vector machine (LS-SVM) model. Experimental results show that the GRU model could achieve better results than the LS-SVM model. The output of the GRU model can follow the change of CO2 mass flowrate in the steady state after the transient process, and relative error is within ±5% .