Abstract The global ecological environment is fragile, and large-scale coal mining has accelerated the loss of water resources, but it is difficult to quantify its impact. Studies have found that large-scale coal mining has a particularly large impact on the underflow zone of rivers, but blind mining affects environmental damage and loses the water volume of the river. The lack of a compensation mechanism for water damage caused by coal mining often exacerbates the province’s fragile water environment. This paper analyzes the prediction of the impact of coal mining on the underflow zone of the river. First of all, determine the mining area of the mine and collect and collect data. Then, the lost water volume was calculated as a predicted score using the double integro-differential equation. Finally, the GA-BiLSTM model is proposed to predict the lost water capacity. The experimental results show that the mean absolute error (MAE), root mean square (RMSE), and prediction pass rate are 11.78/%, 24.87/%, and 92/%, respectively, and the average relative error is only 9.98%. Compared with BP, SVM, LSTM, Bi-LSTM neural network model has better prediction accuracy. The experimental results also show that the model has good reliability and practicability, which can provide a new idea and method for the new analysis of the impact of coal mining on the underflow zone of the river.