Usually, because the crack shape of the dam is very irregular, its information is difficult to obtain. In engineering applications, the prediction of complex cracks is more demanding and difficult. At present, many crack prediction models require a large amount of data and are vulnerable to external interference. The traditional gray prediction model (GM (1,1)) has a large prediction error due to the fixed background value. In this paper, Singer chaotic map is introduced into Sparrow Search Algorithm (CSSA), so that the background value of the grey prediction model can be generated dynamically. Therefore, we can improve the accuracy and adaptability of the prediction model. Firstly, we verify the improvement of the improved Sparrow Search Algorithm in accuracy and search speed through MATLAB. Secondly, the dynamic parameters are used to generate dynamic background values in the prediction model. Experiments show that the CSSA greatly improves the accuracy of crack prediction.