To ensure the safety of embankment construction，it is necessary to predict the settlement of embankment in different construction stages of breakwater to adjust the construction schedule and working procedure. The traditional settlement prediction methods mainly include Terzaghi consolidation theory，curve fitting method，and BP neural network. The prediction accuracy of Terzaghi consolidation theory and curve fitting method is low，and BP neural network needs a large number of samples to approximate the optimal solution. Aiming at these problems，this paper proposes a prediction method of breakwater settlement in different construction stages based on convolutional neural networks（CNN）. This method is applied to predict the settlement and rate of the embankment in the Dagukou port area of Tianjin port during the construction stage，and the safety risk level mapped by the settlement rate is analyzed by the prediction results，to provide actionable guidance for the actual construction period. The results show that the convolution neural network can accurately predict the settlement rate，and can analyze and guide the results of safety risk level according to the prediction results.