To obtain the response of the supporting structure of the foundation pit nearby water and thus evaluate the safety of the supporting structure，this paper establishes a random finite element model of the foundation pit nearby water by considering the spatial variability of the soil.The paper also transforms the feature units whose parameters conform to the random field distribution in the model into images to construct multi-channel grayscale image data and inputs the data into the constructed convolutional neural network (CNN) for training.In addition，the paper studies the random finite element calculation results of the foundation pit nearby water，predicts the deformation responses of the foundation pit nearby water with different random field parameters，and then obtains the response of the supporting structure of the foundation pit nearby water with statistical significance.The results show that the analysis method of foundation pit nearby water based on deep learning and random field can predict the response probability distribution of the supporting structure accurately.