The invention discloses a shield tunneling machine parameter optimization method based on a deep neural network. According to the method, a neural network algorithm is utilized to establish a forwardmodel for predicting the tunneling parameters of the shield tunneling machine under complex geological conditions; and if the expected output value cannot be obtained at the output layer, the quadratic sum of the output and the expected error is taken as an objective function, back propagation is carried out, partial derivatives of the objective function to each neuron weight are solved layer by layer, a step quantity of the objective function to a weight vector is formed as a basis for weight modification, and network learning is finished in a weight modification process. When prediction is carried out by utilizing real data, the prediction value of the model is consistent with the change rule of original data, the average error is within 12%, the model structure is light in weight, the generalization ability meets the field construction requirements, a set of effective scheme is provided for shield tunneling machine tunneling parameter prediction, and certain practical value is achieved.