Abstract
Mine water influx is a significant geological hazard during mine development, influenced by various factors such as geological conditions, hydrology, climate, and mining techniques. This phenomenon is characterized by non-linearity and high complexity, leading to frequent water accidents in coal mines. These accidents not only impact coal production quality but also jeopardize the safety of mine staff. In order to better predict the amount of water surging in mines and to provide an important basis for mine water damage prevention work, based on the time series data of mine water influx from January 2020 to February 2023 in Northern Guizhou Province Longfeng Coal Mine, the BP-ARIMA prediction model was established by combining the BP neural network model and ARIMA autoregressive sliding average model, It also predicted the mine influx for a total of 6 months from July 2022 to February 2023, and compared the prediction results with four models, namely, BP neural network model, ARIMA autoregressive sliding average model, traditional method of Large well method, and GM(1,1) grey model, and used the absolute relative error as the calculation of model accuracy. The results show that the established BP-ARIMA(3,1,1) prediction model is much closer to the actual value, with an average absolute relative error of 1.02% and a maximum absolute relative error of 3.036%, and the goodness of fit R² was 0.93, which is much better than the other four single models, and substantially improves the prediction accuracy of mine water influx. Furthermore, utilizing the BP-ARIMA model, future predictions for mine water influx in Longfeng Mine were made, offering a scientific foundation for effective prevention and control measures.