时间序列分析在低渗透基岩地下水位预测中的应用  

Application of Time Series Analysis to Groundwater Level Prediction in low Permeability Bedrock

作　　者：吉子健 张明[1,2] 季瑞利 范志勇[1,2] JI Zijian;ZHANG Ming;JI Ruili;FAN Zhiyong(Beijing Research Institute of Uranium Geology,Beijing 100029,China;CAEA Innovation Center for Geological Disposal of High-Level Radioactive Waste,Beijing 100029,China)

机构地区：[1]核工业北京地质研究院,北京100029 [2]国家原子能机构高放废物地质处置创新中心,北京100029

出　　处：《世界核地质科学》2023年第S01期525-531,共7页World Nuclear Geoscience

基　　金：中国国家原子能机构核设施退役治理专项资助科研项目(编号:科工二司[2020]194号);中核集团2022年基础研究项目(编号:CNNC-JCYJ-202206)资助。

摘　　要：为深入掌握低渗透基岩地下水水位波动的宏观和微观特征,需要开展对地下水水位数据的拟合预测研究。以高放废物地质处置北山预选区新场地段低渗透基岩为研究对象,采用二次指数平滑法、ARMA模型和BP神经网络模型对地下水径流路径上的3口深部和3口浅部监测孔水位数据开展研究工作。对于地下水位宏观波动特征的解译,BP神经网络模型拟合精度要高于另外两种模型,且更加适用于具有明显非线性特征的水位序列;ARMA模型和二次指数平滑法模型拟合效果相当,但更适用于预测变化相对平缓的监测水位数据,且对于水位突变特征的拟合ARMA模型要优于二次指数平滑法模型。对于浅部监测孔地下水水位微动态的解译,BP神经网络模型优于二次指数平滑法模型和ARMA模型,预测的地下水位微动态变化周期和振幅基本与实测值一致。In order to gain more insight into the macro and micro-dynamic characteristics of groundwater level in the low-permeability bedrock of the Beishan,it is necessary to fit the monitored groundwater level data.The secondary exponential smoothing method model,ARMA model and BP neural network model were used to carry out the fitting study of groundwater level data from three deep monitoring boreholes and three shallow monitoring boreholes along the groundwater flow path.For the interpretation of macro characteristics of groundwater level,the BP neural network model had higher fitting accuracy than the other two models,and was more suitable for the groundwater level sequences with significant nonlinear characteristics.The ARMA model and the secondary exponential smoothing method model had comparable fitting effects,but were more suitable for fitting monitored groundwater level data with relatively smooth changes.The ARMA model was superior to the secondary exponential smoothing method model for fitting unexpected changes in groundwater level characteristics.For the interpretation of micro characteristics of groundwater level in shallow monitoring boreholes,the BP neural network model was better than the secondary exponential smoothing method model or the ARMA model,and the predicted groundwater level micro-dynamic variation period and amplitude were basically consistent with the monitored data.

关 键 词：时间序列分析 低渗透基岩地下水 高放废物 地质处置 地下水动态预测 

分 类 号：P641[天文地球—地质矿产勘探]