干旱内陆区自然-人工条件下地下水位动态的ANN模型

根据我国干旱内陆区自然-人工条件下地下水系统的特点，建立了甘肃省石羊河流域下 游地下水位动态的人工神经网络模型，采用附加动量法和学习速率自适应调整策略对反向传播算法（BP）进行改造，以提高计算速度。该模型以前期地下水位、降雨量、蒸发量、地表来水量、灌溉面积、灌水定额、人口数量作为输入变量，采用缺省因子检验法分析了上述各个因子对地下水位影响的敏感性，模拟了不同灌溉发展面积及地表来水条件下地下水位动态。结果表明：研究区人类活动及地表来水是影响地下水位动态的主要因子，灌溉面积的扩大及地表来水的减少会使地下水位持续下降。模型具有较高的精度，可以较好地定量描述地下水位动态与上述各因子之间的响应关系；研究结果可应用于该地区地下水系统的管理。

ANN model for simulating dynamic variation of groundwater under the condition of natural human activity in arid inland area

The ANN model for simulating the dynamic variation of ground water in Shiyang River basin, Gansu Province, China, is established. The additional momentum method and self adaptive tactic for training are adopted to accelerate the calculation speed of BP algorithm. The input factors of the model are initial groundwater level, precipitation, evaporation, surface water coming from outside region, irrigation area, irrigation schedule and population. A sensitivity analysis is conducted and the dynamic variation of groundwater level is simulated according to the amount of irrigation area and coming surface water. The result shows that the model well expresses the relationship between groundwater level and other factors with sufficient high accuracy. The sensitivity analysis demonstrates that human activities and reduction of surface water coming from outside region are the main causes leading to the descending of local groundwater. The study result is useful for groundwater management of the area to be studied.