Abstract: Groundwater depth prediction plays an important role in regional water resources management and utilization ecological environment protection and economic and social development. Groundwater is affected by many factors,and its dynamic change has the characteristics of non-stationary,random and hysteresis. In order to accurately predict the depth of shallow groundwater,five prediction models,including multiple linear regression,gray GM（1,1）,gray GM（1,1）based on Markov chain optimization,BP neural network and BP neural network based on genetic algorithm optimization,are selected to take Zhaozhou County,Heilongjiang Province as an example,and the data from 1980 to 2009 are taken as training samples. Data from 2010 to 2019 are used as test samples,precipitation,evaporation,groundwater exploitation and early stage water level are used as input layers,and groundwater depth is used as the output layer. Absolute error,relative error,mean absolute error,mean absolute percentage error,mean square error and root mean square error are selected as evaluation indexes. The results show that:The average absolute error of BP neural network model optimized by genetic algorithm is 0.13m,the average absolute percentage error is 1.58%,the mean square error is 0.02,and the root mean square error is 0.15. The prediction accuracy is high and the fitting effect is good. Compared with the other four models,it can better simulate the dynamic change of groundwater depth. It provides reference for rational development and utilization of groundwater in Zhaozhou County. The optimization of genetic algorithm improves the training efficiency and stability of BP neural network,and the Markov chain theory makes up for the lack of fluctuation of gray GM（1,1）.The combined prediction model combines the two models,complementing each other’s advantages and significantly improving the prediction performance. Compared with the single model,the prediction result is more accurate,which can provide a new idea for the establishment of groundwater depth prediction model.