State Equation of Groundwater Based on Artificial Neural Network
韩非,薛禹群,张永祥
DOI: https://doi.org/10.3969/j.issn.1006-7493.2001.04.014
2001-01-01
Abstract:It is well known that the density of groundwater in geological environment is not invariable. In general, groundwater temperature varies little and the solute concentration is not so high, the density of groundwater is close to a constant. In seawater intrusion case, total dissolved substance (TDS) of groundwater is less than 35g/L, and the density of groundwater changes linearly with the solute concentration and its influence on movement of groundwater can not be neglected. With the occurrence of the problems such as brine water intrusion, heat storage in aquifer, exploitation of terrestrial heat and pumping brine water in deep aquifer, TDS of the groundwater is as high as 1000g/L and the temperature reaches 60 ℃, and the range of variation of TDS is very extensive. For example, TDS can vary from 0.1g/L inland to 220g/L in coast in Laizhou Bay, and hydrogeochemical types of groundwater is complete, which involve fresh water, brackish water, salt water and brine water. Three states exist in relationships between the density and the TDS of groundwater, i,e., nearly constant in fresh water, strongly nonlinear in brine water, and linea state in partially salted water. Moreover, the density of groundwater is influenced by the components of solute and the density may be different with the same TDS. In the south coast of Laizhou Bay, hydrogeochemical components vary complicatedly because they are influenced by many factors such as cation exchange, dissolution and sedimentation of minerals. Therefore, it is difficult to obtain a traditional state equation which can completely satisfy conditions demonstrated above. We obtain the state equation of groundwater based on artificial neural network (ANN) by gathering every hydrogeochemical types of groundwater samples in the south coast of Laizhou Bay, gauging their density and TDS, by using neural network toolbox to design neural network model of back propagation in 3 layers, and then simulating the relationships between the density and the TDS of groundwater. It is shown that the equation based on ANN is a reasonable mapping between density and TDS of groundwater and it is straightforward in form and is easy to use. ANN has broad prospect in hydrogeological field because major geological environment in which groundwater exists and the variations of groundwater itself are very hard to understand fully. At present ANN is mainly applied in total evaluation and prediction of groundwater quality. Applications of the neural network toolbox can simplify the design and exertion of ANN considerably, save time for programming and improve working efficiency, and thus provide more opportunity to popularize ANN.