Application of Prediction of Water Content in Geophysical Exploration based on B

论文导读：erofknowngeoPhysicaltestingthewellPumPingtestdatatothegeophysicalParametersofaquiferdischargeumtswithsingle?holemodlecomposedofthetrainingsampleset，establishingtheerrBPNetworkforecastingmodelforregionalgeophysicalmethodsunknownundergroundwatercontentPredieti
Abstract: This researeh is adopting neural network technology for the use of a number of known geoPhysical testing the well PumPing test data to the geophysical Parameters of aquifer discharge umts with single?hole modle composed of the training sample set，establishing the err BP Network forecasting model for regional geophysical methods unknown under ground water content Predietion， this will be major imProvemen to the traditional water content Prediction.
Key words: artiflcial net works:integrated geophysical methods:water content Prediction
1 Introduction
Water resources is a human production and life indispensable natural resources, biological survival of the environmental resources, with the water crisis intensifies and water quality deterioration, water shortages he evolved into the world paid close attention to the resources and environmental problemsone. Groundwater is an important part of water resources, human indispensable natural resources, its life, industrial and agricultural production, and urban construction, social and economic development plays a significant role.
In the North China Plain region, due to the large-scale development of groundwater use in the past 40 years, has been formed around Shijiazhuang, Hengshui, Changzhou and other cities of the three regional funnel. With the continued exploitation of groundwater resources, the composition of the groundwater resources component has changed significantly. According to different model results, the more flow in the deep water resources and clay layer released water accounts for about 75%, about 15% of the released water sand laminated, lateral recharge in only 10%. Non-compensation means that the status quo adopted most of the water from the aquifer storage resources, and has caused serious land subsidence, which also allows us to re-examine the use of groundwater resources.
Based on geophysical methods to groundwater in situ testing capabilities, with high efficiency, low cost, continuous measurement of other advantages, the use of geophysical techniques to predict the water content of the aquifer is expected to solve the above problem has become the hot topic of current research. Therefore, geological modeling technologies and integrated geophysical study the combination of aquifer water content forecasting methods he emerged. Ground of integrated geophysical中国论文中心www.7ctime.com
techniques, three-dimensional distribution of the fine to identify aquifer and changes; to carr论文导读：technologyresearch,precisecomputationofaquifereffectiveporosity,permeability,theeffectivesaturation,shalecontent,andothergeologicalparameters;Finally,geophysicalparametermathematicalmodel,tocarryoutthemoisturecontentoftheaquifer,thepredictionprocessingsystem.2The
y out ground geophysical technology research, precise computation of aquifer effective porosity, permeability, the effective saturation, shale content, and other geological parameters; Finally, geophysical parameter mathematical model, to carry out the moisture content of the aquifer, the prediction processing system.2 The basic theory of artificial neural networks
The artificial neural network is an emerging cross-disciplinary. In the practical application of artificial neural networks, artificial neural network model using BP neural network (Back - Propagation Neural Network) . The BP algorithm consists of two parts: the forward pass and error back-propagation of information. In the process of the forward pass input information calculated from the input layer hidden layer, layer by layer to the output layer, each layer of neurons status affects only the state of the next layer of neurons. If the output layer does not he the desired output, the calculation of the change of error value of the output layer, and then turned to the back-propagation, the error signal through the network along the connection path to anti-pass back to modify the weight of the layers of neurons until they reach the desired goals. Figure 1. BP neural network learning rules: adjust the network weights and thresholds of the square of the network errors and the allest, it is to adjust the network weights and thresholds in the direction of steepest descent, [3].
BP model consists of multiple nodes of input layer, hidden layer and an output of the output layer, each node of individual neurons, one-way connection between adjacent layers. Traner function between the nodes for the S-shaped function, ie
The output of the layers of node摘自：毕业论文格式模板www.7ctime.com
s are calculated as follows:
Each node input is connected by the weights ω of the input information and the threshold θ. Where y is the node output, xi is the node to accept the information the ωij is related to the connection weights, θ is the threshold, n is the number of nodes.
3 Geophysical method
Geophysical methods are divided into the power law and elastic we method. Which the power law: (1) transient electromagnetic method (TEM), (2) conductivity imaging method (AMT), (3) induced polarization (IP), each method has its advantages and characteristics, such as: exploration of the direct current method to detect saline very invasive, and water distribution; seiic exploration method in the constructor, stratigraph论文导读：castingmodelselectionofinputneuronstothefollowingprinciples:First,theexistinggeophysicalequipmentcanbemeasured,ortheconversionparameters,practicalandobservability;second,tobematchedwiththecomprehensivegeophysicalmethodstostudythefinestructureoftheaquiferto
ic division, strata rich water and lithology contrast with a strong ability; electromagnetic sounding method has the absolute advantage of the exploration depth. Therefore, the response to different exploration tasks, a variety of geophysical methods optimized combination of comprehensive analysis, a key aspect of this research project, several geophysical methods he been used to extract the different combinations of the following prediction model:① In areas of shallow aquifer level for example, the Quaternary aquifer structure of the North China Plain region alone induced polarization method, which he come to the use of its measurement and inversion of resistivity, polarizability, half-life, decay ratelayer thickness parameters is sufficient to model prediction;
② For Aquifer of the fine-grained structure and the stratigraphic sequence, lithology characteristics can be taken to stimulate the pattern of the polarization and shallow high resolution seiic method or the transient electromagnetic method in addition to the forecast model can also aquifer structure of a more nuanced understanding of;
③ For the deeper aquifer structure of the Ministry, we recommend the use of frequency domain electromagnetic sounding method, the exploration depth of the electromagnetic method, conditional case with seiic exploration, can be the physical parameters of the deeper geological, which improve the prediction accuracy great help.
4 BP neural network model

4.1 Modeling selection of neurons

Predecessors he been confirmed by a large number of field and laboratory test, the subject of forecasting model selection of input neurons to the following principles: First, the existing geophysical equipment can be measured, or the conversion parameters, practical and observability ; second, to be matched with the comprehensive geophysical methods to study the fine structure of the aquifer to fully tap the received observation data resources; optimal combination, taking into account the complementarities between the input neurons use to oid or reduce redundancy; Fourth, for the protection of the prediction model has a broad generalization ability.
Based on the above principles and before the text of the discussion of integrated geophysical methods which provide the geophysical parameters the amount of the initial neural network prediction model's prediction input basic neural element selected is: ρ, h, st / 2, D,, ηs such as ground geophysical observation parameters论文导读：}hi:1996.JonathanB，FraklinAjo.Usingspatiallyintegratedcrosswellgeophysieorenvironmelltalsiteassesent2004JointAssemblyoftheCanadianGetPhysiealUnion，AmericanGeoPhysiea1Union，SoeietyofExplorationGeophysicists，andEnvironmentalandEngineeringGeophysicalSoeiety，17一21
(resistivity ρ, h, aquifer thickness, St. / 2 half-life D attenuation ηs polarization rate).
4.2 Normalized inpu源于：大学毕业论文格式www.7ctime.com
t neurons
Because each unit of data collected as the input neurons are inconsistent, the scale difference of the input neurons, there will be a large sample to eat all sample, and therefore must be the input layer data (0,1) normalized to eliminate the impact of the scale. Commonly used normalization formula is:中国免费论文网www.7ctime.com
The command statement: net.trainparam.epoehs = 1000:
② performance function value is less than the error indicator g. a1 the mse core 10 a,;
Command statement: net.trainParam.goal = o.000001:
③ training longer than a time limit time.
The subject application MATLAB7.O aquifer water content of the BP neural network prediction model for training, training results are shown in Figure 1.
6 Conclusion
In this paper research results show that the BP neural network model combined with the integrated geophysical techniques during aquifer water content forecasting has certain advantages and potential. But at the same time we also he to point out, this subject is still price segment, objective reason for the data collection stage a summary of research, there is still some. By the initial results of the project is summarized, and found there are still the following issues need further study and improve the future work.
Reference
DamnCisar，TimothyJNovotn Eleetromagnetic data一evaluation using a neural network :initial investigation-underground storage tanks〔J」.geoPhysieal，1996(2)，
R.O.Hansen.Feature recognition from Potelltial fields using neural networks[J]·geophysieal，1997(3).
[3]DongY,，Ma YK etc. Patterm recognition of the charaeteristies of AE source using neural:letworklJ].Proc.Of 14th WCNDT，NewDel源于：毕业生论文www.7ctime.com
hi:1996.
[4]Jonathan B，Fraklin Ajo.Using spatially integrated cross well geophysies for environmelltal site assesent[C] 2004 Joint Assembly of the Canadian Get Physieal Union，American GeoPhysiea1 Union，Soeiety of Exploration Geophysicists，and Environmenta land Engineering Geophysical Soeiety，17一21 May2004，Montreal，Canada.EosTrans.AGU，Joint Assembly Supplement，Abstract NS23A09，2004，85(17):86~88.