以晋祠泉域为例,在泉域驱动因素的基础上,将泉域岩溶水系统概化为一个多输入-双输出的系统,输入项为当年及前7 a的降水、岩溶水开采量和采煤排水量;输出项为晋祠泉总排泄量,即出流量和侧排量。然后采用多元线性回归和BP神经网络2种随机模型,建立泉域各输出项与各输入项的关系,再假定无岩溶水开采和无采煤排水2种情景,在这2种情景下模拟泉域的排泄量,对比分析采煤和岩溶水开采对晋祠泉出流量的影响。研究结果表明采用多元线性回归模型,在1956—1994年多年平均条件下,岩溶水开采后晋祠泉排泄量减少0.42 m³/s,采煤排水后出流量减少0.23 m³/s,人类开采使总排泄量减少0.65 m³/s;采用BP神经网络模型,1956—1994年多年平均条件下,岩溶水开采后泉域排泄量减少0.30 m³/s,采煤排水后出流量减少0.27 m³/s,人类开采使总排泄量减少0.65 m³/s。20世纪80年代后采煤活动加剧,2种随机模型均反映出采煤排水对泉域有严重的影响。

In an attempt to explore the influence of coal mining and karst water exploitation on Jinci spring, the discharges in spring area are simulated by multiple linear regression model and back propagation (BP) neural network model under two scenarios (in the absence of karst water exploitation and in the absence of coal mining drainage, respectively) for comparison. In line with the study area’s driving factors, the Jinci spring area is generalized as a multi-input system with dual output. Precipitation, karst water exploitation amount, and coal mining drainage amount in the current year and previous seven years are selected as input; the total discharge of Jinci spring, inclusive of outflow and side discharge, is taken as output. The relations between each input and output are established. Under the annual average conditions of 1956-1994, 1) the results of multiple linear regression model manifest that karst water exploitation had reduced the discharges of Jinci spring area by 0.42 m³/s, coal mining had cut the outflow by 0.23 m³/s, in total of 0.65 m³/s; 2) while the results of BP neural network model shows that karst water exploitation reduced the discharges by 0.30 m³/s,coal mining by 0.27 m³/s,and the total influence was 0.65 m³/s, which was nonlinear. Both stochastic models reflected severe impact of coal mining on Jinci spring area since coal mining and karst water exploitation exacerbated after the 1980s.