Development of measurement of soil hydraulic properties by using in situ permeability tests data

近年来，我国对深埋地下资源的挖掘和开发利用需求日益增加。这就需要预测多层含水层中地下水的行为，以促进其保护。准确确定各含水层的水力特性对正确预测地下水流具有重要意义。抽水试验通常在多层或有泄漏的含水层条件下进行。然而，很难“解析”地分析在这些条件下从泵送试验中获得的数据。本文提出了一种在复杂含水层条件下利用抽水试验资料估算含水层系数的新方法。土壤的水力特性、渗透系数和蓄水量是预测地下水动态的重要数据。通常进行泵送测试来确定这些特性。本文提出了一种利用抽水试验获得的沉降曲线评价土壤水力特性的新方法。在我们开发的方法中，利用了神经网络的模式匹配能力。神经网络被训练来识别作为输入的压降数据模式和作为输出的承压含水层中相应的水力特性。训练后的网络接收抽水试验数据作为输入模式，产生水力特性输出。利用在渗漏含水层或各向异性承压含水层中观测到的降水数据来评价我们提出的方法的有效性。

Recently, the demand for deep underground excavation and the development and utilization of its resources has increased. This raises a need to predict the behavior of the groundwater in multilayered aquifers in order to promote its conservation. The exact determination of hydraulic properties of each aquifers is very important for the correct groundwater flow prediction. Pumping tests are usually performed under the multilayered or leaky aquifer conditions. It is, however, difficult to analyze the data obtained from the pumping test under these conditions "analytically".In this research, a new method of estimating aquifer coefficients from pumping test data in a complicated aquifer conditions is proposed. The soil hydraulic properties, coefficient of permeability and storage are essential data to predict the behavior of groundwater. Pumping tests are usually performed to determine these properties. In this paper, a new approach to evaluate soil hydraulic properties from drawdown curves which are obtained by pumping tests has been developed. In our developed method the pattern-matching capability of a neural network is used. The neural network is trained to recognize patterns of drawdown data as input and corresponding hydraulic properties in the confined aquifer as output. The trained network produces output of hydraulic properties when it receives pumping test data as the input patterns. Drawdown data which are observed in a leaky aquifer or an anisotropic confined aquifer are used to evaluate availability of our proposed method.