The deep learning prediction model of advancing speed is proposed based on the analysis of Pearson correlation coefficient between shield tunneling parameters and advancing speed, so as to solve the problem that the advancing speed of EPB shield in complex soft strata cannot be predicted and controlled by theoretical method. In the proposed model, particle swarm optimization (PSO) algorithm is applied to optimize the weight and bias value of BP network to overcome the shortcomings of traditional BP neural network based on gradient descent algorithm, such as easily falling into local minimum value and large prediction error. The geological and shield tunneling parameters are selected as input values while advancing speed is determined as output. Based on the result that penetration has highest and positive correlation with advancing speed, a case study on a tunnel project of Hengqin to Zhuhai Airport HJZQ2 Bid Section is conducted to check performance of proposed model. The measured data is used to establish advancing speed prediction model. The result displayed that prediction error is basically controlled within ±4 mm/min (error within 10%) which is obtained via PSOBP deep learning model with two hidden layers. The predicted error meets requirement of engineering project which verifies effectiveness and applicability of the proposed model.