不同时间序列模型在岩溶山区矿井涌水量预测中的应用

矿井涌水量预测的精度对于煤矿开采安全有着至关重要的作用。文章以老鹰山煤矿为例，分析降雨与矿井涌水量的相关关系，结果表明：同期月及前第1个月降雨量与涌水量相关性具有逐渐减弱的趋势，而与前第2个月至第5个月的相关性有逐渐升高的趋势；基于矿井涌水量及降雨量，建立了单因素季节性时间序列SARIMA模型及多元季节性时间序列SARIMAX模型对矿井涌水量进行预测，预测结果表明：两种模型91.7%的预测值达到B级探明的矿井涌水量，预测精度均较高，SARIMAX模型预测结果的MAPE为18.57%，小于SARIMA模型的25.27%，预测精度更优。

Application of different time series models to the prediction for mine water inflow in karst mountainous areas

Coal resources are one of the important mineral resources in China. In the process of coal mining, due to the complex hydrogeological conditions in the mining area, and ineffective water exploration and discharge, accidents of mine water inrush occasionally occur, which may seriously restrict the safe production of coal resources. According to statistics, from 2000 to 2017, there were 1,173 accidents of coal mine flood in China, with 4,760 deaths. Therefore, the prediction reliability of mine water inflow plays a vital role in the safety of coal mining. A time series model is specifically designed to simulate and predict a time-sequential, time-varying, and interrelated data series. Most time series models require that the data must be stationary and the time series must follow a normal distribution. Taking Laoyingshan coal mine as an example, this study establishes a model of Seasonal Auto-Regressive Integrated Moving Average (SARIMAX model) and a model of Seasonal Auto-Regressive Integrated Moving Average with eXogenous factors (SARIMA model), compares the fitting and prediction results of these two models, and evaluates their adaptability in prediction of mine water inflow in karst mountainous areas. Based on the monthly average rainfall and monthly average water inflow from October 1994 to December 2014, a SARIMA model for univariate seasonal time series and a SARIMAX model for multivariate seasonal time series have been established. To establish a corresponding mathematical model, it is necessary to perform a parameter significance test on each model, analyze the model fitting goodness and model fitting accuracy, and determine the optimal model. The test parameters can be selected from the coefficient of determination R2 of the sample, the Nash efficiency coefficient (NSE), the mean absolute percentage error (MAPE), the deviation, the root mean square error (RMSE), the AIC value, the BIC value and other indicators to test. Since NSE, RMSE, R2, and MAPE standards are correlated in some degree, the NSE, AIC and BIC values are selected as the criteria for validating the quality of the model.