Abstract: In response to the prominent problem of mud and water inrush in deep and long tunnels for national water network construction, as well as the challenges of difficult coring and long time consumption in directional drilling-based advanced geological prediction, A method for identifying the integrity of surrounding rock based on directional drilling parameters was proposed.The Pearson correlation coefficient was used to quantify the sensitivity between drilling parameters and the surrounding rock integrity coefficient, and an identification index system that can reflect the integrity of surrounding rock was determined.A surrounding rock integrity identification model was established by cascading a stacked autoencoder (SAE) neural network with a Softmax classifier, and engineering application and verification were conducted using a composite directional drill in the Yangtze River-to-Hanjiang River Water Diversion Project.The research results show that: the accuracy of the model's identification results on the test set is 95.1%, with a root mean dquare error (RMSE) of 0.19 and a mean absolute percentage error (MAPE) of 5.7%, which is significantly better than the BP-Softmax identification model.The constructed identification model has strong feature learning capability, which can realize the identification of tunnel surrounding rock integrity, improve the applicability of directional drilling technology in the advanced geological prediction, and enhance the prevention and control capability of mud and water inrush disasters in deep and long tunnels.