基于人工神经网络的晋祠泉水位模拟研究

Simulation of Jinci Spring Water Level on ANN Models

晋祠泉是太原市第二岩溶大泉,受太原市工农业生产大量开采岩溶水影响,该泉已于1994年4月断流。为了探索人类因素影响下的泉水水位变化趋势,采用前馈神经网络、动态递归神经网络、时延神经网络、非线性动态自回归神经网络、级联神经网络5种人工神经网络,结合14种训练算法构建晋祠泉水位预测模型,基于2013—2017年实测泉水位数据分析各种人工神经网络预测模型精度,结果表明:动态递归神经网络可用来对晋祠泉水位进行准确预测,traincgb、trainrp、traincgf、traincgp等算法效果比较理想。同时应用LSTM深度学习模型预报未来10 a的降水量,进而计算出降水入渗补给量等,并结合动态递归神经网络预测晋祠泉域未来水位变化,结果测定2019年晋祠泉水位可以超过复流最低水位802.59 m。

Jinci Spring is the second largest karst spring in Taiyuan City. Since 1960s, due to massive exploitation of karst water by industrial and agricultural production in Taiyuan City, the spring had been cut off in April 1994. In order to explore the change trend of spring water level under the influence of human factors, five artificial neural networks of Feed forward net, Elman net, Time delay net, Narx net and Cascade net were used to combine 14 kinds of artificial neural networks. The training algorithm built the prediction model of Jinci spring water level. Based on the analysis of measured spring water level data from 2013 to 2017, the accuracy of various artificial neural network prediction models was analyzed. The results show that Elman recurrent dynamic neural network neural network can be used to predict Jinci spring water level accurately. The effect ratio of traincgb, trainrp, traincgf and traincgp is good and ideal. In addition, in order to make the prediction re sult of Jinci spring water level more realistic and instructive, this paper used LSTM deep learning model to forecast precipitation in the next 10 years and recursive dynamic neural network Elman net to predict the future change of Jinci Spring water level. The result shows that Jinci spring water level in 2019 can exceed the lowest level of recovery flow 802.59 m.