Abstract: In order to establish a tunnel TBM (full-face tunnel boring machine) penetration rate prediction model, this paper is based on 100 sets of rock mechanics and TBM excavation parameters obtained through indoor tests and field records in the Pahang-Selangor tunnel in Malaysia. First, 17 TBM penetration rate (PR) prediction models were established by using statistical regression analysis, artificial neural network, machine learning and ensemble learning. Then, a new normalized multi-index model evaluation method is proposed. Each evaluation index of the model, in turn, to convert the orientation consistency, normalized, summation and ranking, and the normalization method and the existing ranking method are used to compare and analyze the performance of the 17 prediction models, the results show that: (1) The multi-algorithm fusion improves the prediction ability of artificial neural networks and classification and regression tree models, but slightly reduces the prediction ability of support vector regression models; (2) Classification and regression tree model and ensemble learning model based on classification and regression tree have the best prediction ability and are more suitable for PR prediction; (3) The proposed normalization method provides a quantitative way for multi-index comprehensive evaluation between different models, and overcomes the shortage of traditional ranking method that cannot accurately identified when the difference of model prediction ability is relatively small. The study results provide theoretical basic for reasonable evaluation of TBM roadway construction period and estimation of engineering cost.