Abstract: Accurately predicting consumption of tail seal grease is very important to tail seal, shield tunnelling safe? ty and cost control. Therefore, the K-fold cross-validation and ANN neural network have been used to improve the conventional particle swarm optimization algorithm, so as to allow the IPSO algorithm to automatically find the optimum neuron hyper-parameters of ANN neural network and build the IPSO-ANN time series model for the slurry shield tail seal grease consumption. In the Jinan Yellow River Tunnel project, based on the seal grease consumption of the east and west lines and the influencing factors that are obtained through double screening, the 3 strategies including hybrid training, east line separate training and west line separate training have been established that are used for model training, and the grease consumption in construction section have been predicted and analyzed. As the results suggest, the IPSO-ANN model is effective in finding the neural network model that has the optimum neuron hyper-parameters; under different training strategies, the average prediction accuracy of optimum model is above 80%, the prediction accuracy under the hybrid training strategy is 85.142%, and there is also good stability,so this model is useful for predicting tail seal grease consumption.