CN116628442B - Groundwater reserve change space downscaling method based on artificial neural network - Google Patents
Groundwater reserve change space downscaling method based on artificial neural network Download PDFInfo
- Publication number
- CN116628442B CN116628442B CN202310536452.0A CN202310536452A CN116628442B CN 116628442 B CN116628442 B CN 116628442B CN 202310536452 A CN202310536452 A CN 202310536452A CN 116628442 B CN116628442 B CN 116628442B
- Authority
- CN
- China
- Prior art keywords
- reserve
- groundwater
- variable
- neural network
- artificial neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 239000003673 groundwater Substances 0.000 title claims abstract description 152
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 85
- 230000008859 change Effects 0.000 title claims abstract description 83
- 238000000034 method Methods 0.000 title claims abstract description 77
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 139
- 230000007613 environmental effect Effects 0.000 claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 38
- 230000006870 function Effects 0.000 claims description 58
- 210000002569 neuron Anatomy 0.000 claims description 29
- 238000004422 calculation algorithm Methods 0.000 claims description 24
- 239000002352 surface water Substances 0.000 claims description 19
- 238000000354 decomposition reaction Methods 0.000 claims description 18
- 239000002689 soil Substances 0.000 claims description 17
- 238000004088 simulation Methods 0.000 claims description 14
- 230000008018 melting Effects 0.000 claims description 13
- 238000002844 melting Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 8
- 238000011160 research Methods 0.000 claims description 6
- 238000010219 correlation analysis Methods 0.000 claims description 4
- 238000006467 substitution reaction Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 230000004931 aggregating effect Effects 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 3
- 238000001556 precipitation Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 230000008020 evaporation Effects 0.000 claims description 2
- 238000001704 evaporation Methods 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 abstract description 8
- 238000012545 processing Methods 0.000 description 5
- 230000004913 activation Effects 0.000 description 4
- 238000004590 computer program Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 230000007774 longterm Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000010220 Pearson correlation analysis Methods 0.000 description 2
- 239000006185 dispersion Substances 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 239000000047 product Substances 0.000 description 2
- 239000005413 snowmelt Substances 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 238000010183 spectrum analysis Methods 0.000 description 2
- RDJLIBMVLWBMKX-UHFFFAOYSA-N 1-(3,4-dihydro-2h-chromen-2-ylmethyl)piperidine Chemical compound C1CC2=CC=CC=C2OC1CN1CCCCC1 RDJLIBMVLWBMKX-UHFFFAOYSA-N 0.000 description 1
- 108010076504 Protein Sorting Signals Proteins 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000003651 drinking water Substances 0.000 description 1
- 235000020188 drinking water Nutrition 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000013505 freshwater Substances 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000012466 permeate Substances 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The application relates to an underground water reserve change space downscaling method based on an artificial neural network, which comprises the following steps: based on the water balance principle, obtaining the groundwater reserve variable quantity with low spatial resolution from the land total water reserve monitored by GRACE satellites; acquiring an environmental variable with high spatial resolution, and determining an optimal lag time of the groundwater reserve variable quantity relative to the environmental variable; the environment variable is used as sample data, and the optimal model parameters of the artificial neural network model are determined based on a control variable method for multiple training modeling; and constructing a GRACE underground water reserve change space downscaling model according to the artificial neural network model based on the optimal model parameters and the optimal lag time so as to simulate the high-spatial-resolution underground water reserve change, and finally improving the precision of monitoring underground water by the GRACE satellite and improving the applicability of the space downscaling model in a complex small area.
Description
Technical Field
The application relates to the technical field of remote sensing digital image space downscaling and hydrology intersection, in particular to an underground water reserve change space downscaling method based on an artificial neural network.
Background
Groundwater (GWS) is the largest source of fresh water in the global hydrologic cycle, providing about 50% of drinking water worldwide. In recent years, extreme climate, population growth and over exploitation of groundwater resources cause serious consumption of groundwater resources, and grasping of dynamic changes of groundwater is crucial to water resource management and human survival, while long-term monitoring of dynamic changes of groundwater is a foundation for realizing sustainable management and protection of groundwater resources.
The GRACE satellite in the satellite remote sensing technology provides new possibility for realizing high-efficiency, large-range and time-continuous monitoring of the dynamic change of the underground water. But the coarse spatial resolution of GRACE satellite data also severely limits its application in regional scale groundwater monitoring. Particularly, the time lag effect of GRACE underground water relative to environmental variables is usually ignored in the aspect of space downscaling research based on a machine learning model, and because the research on the optimal parameter combination of a machine learning algorithm is not deep enough, in arid areas, the condition that the time required for water to permeate from the earth surface to an underground water aquifer is long can not be accurately monitored because of the large depth of the underground water aquifer from the earth surface, so that the dynamic change of underground water can not be accurately monitored in areas with complex terrains and climate environments.
Disclosure of Invention
The application provides an underground water reserve change space downscaling method based on an artificial neural network, which can improve the accuracy of GRACE satellite monitoring underground water and improve the applicability of a space downscaling model in a complex small area.
In a first aspect, the present application provides a method for spatial downscaling of groundwater reserve changes based on an artificial neural network, the method comprising: based on the water balance principle, obtaining the groundwater reserve variable quantity with low spatial resolution from the land total water reserve monitored by GRACE satellites; acquiring a high spatial resolution environmental variable corresponding to the groundwater reserve variation and determining an optimal lag time of the groundwater reserve variation relative to the environmental variable; the environment variable is used as sample data to determine the optimal model parameters of the artificial neural network model based on a control variable method for multiple training modeling; the optimal model parameters comprise one or more of the proportion of each sample participating in training, the number of neurons of an implicit layer, the network learning rate, the maximum iteration number, the initial range of weight values and the training number; and constructing a GRACE groundwater reserve change space downscaling model according to an artificial neural network model based on the optimal model parameters and the optimal lag time so as to realize simulation of high-spatial resolution groundwater reserve change.
Optionally, based on the water balance principle, the method for obtaining the groundwater reserve variable quantity with low spatial resolution from the land total water reserve monitored by the GRACE satellite includes: collecting land total water reserve variation monitored by a GRACE satellite, wherein the land total water reserve variation comprises variation of surface water, soil water, groundwater reserve, ice and snow melting water and vegetation canopy water; calculating historical variation of the surface water, the soil water, the ice and snow melting water and the vegetation canopy water in a preset time period in a research area, and describing by a formula (1):
ΔG(t)＝G(t)-G avg (1)，
wherein ΔG (t) represents the historical variation of a certain variable in the land total water reserve variation within a preset time period, G (t) represents the historical variation corresponding to a certain variable at the moment t, G avg Representing the average value of the historical variation of a certain variable in a preset time period; subtracting the land total water reserve variable quantity monitored by the GRACE satellite from the variable quantity of the surface water, the soil water, the ice and snow melting water and the vegetation canopy water based on a water quantity balance principle to obtain the underground water reserve variable quantity, wherein the underground water reserve variable quantity is described by a formula (2):
ΔGWS＝ΔTWS-ΔSWS-ΔSMS-ΔSWE-ΔCWS(2)，
where Δgws represents the groundwater reserve amount change, Δtws represents the land total water reserve amount change, Δsws represents the surface water reserve amount change, Δsms represents the soil water content change, Δswe represents the ice and snow melt amount change, and Δcws represents the vegetation canopy water content change.
Optionally, the acquiring the high spatial resolution environmental variable corresponding to the groundwater reserve variation and determining an optimal lag time of the groundwater reserve variation relative to the environmental variable comprises: decomposing the time sequence of the underground water reserve variable quantity and each environment variable quantity according to an integrated empirical mode decomposition algorithm to obtain a plurality of eigenmode functions and residual errors; wherein the environmental variables include precipitation, surface temperature, potential evaporation, normalized vegetation index, and soil moisture content; calculating the average period corresponding to each eigenmode function according to a fast Fourier transform method; calculating a variance contribution rate corresponding to each eigenmode function according to the eigenmode function, and describing through a formula (3):
wherein P is i Variance contribution ratio for ith eigenmode function, V i Is the variance of the ith eigenmode function, and n is the total number of eigenmode functions; based on the obtained average period and variance contribution rate corresponding to each eigenmode function, comparing the average period and variance contribution rate corresponding to each eigenmode function, selecting the eigenmode function with the average period closest to 1 year and the variance contribution rate larger as an alternative time sequence of the original obtained underground water reserve variable quantity and each environment variable; calculating correlations between the groundwater reserve variable amounts of different lag times and the substitute time sequences of the environmental variables according to a pearson correlation coefficient method, wherein the lag time corresponding to the highest correlation coefficient is defined as the optimal lag time, and describing by a formula (4):
Wherein CC represents a correlation coefficient, (-)>And->Represents the average value, x, of each set of data i And y i Is two sets of data for correlation analysis, n is the length of each set of data.
Optionally, the decomposing the time sequence of the groundwater reserve variable quantity and each environmental variable according to the integrated empirical mode decomposition algorithm to obtain a plurality of eigenmode functions and residuals, including: decomposing the time sequence of the groundwater reserve variable quantity and each environment variable based on an integrated empirical mode decomposition algorithm, and describing by a formula (5):
wherein x (t) represents the time series to be decomposed, i.e. the time series of the groundwater reserve variation and the respective environmental variable, IMF i (t) represents the ith eigenmode function, R (t) represents the residual, t is time, and n represents the total number of eigenmode functions.
Optionally, the determining the optimal model parameters of the artificial neural network model by using the environmental variable as sample data based on multiple training modeling by a control variable method includes: setting model parameters in the artificial neural network model; training the environment variable based on an artificial neural network model to obtain a plurality of groups of output data, and determining the optimal value of each model parameter by utilizing an error back propagation algorithm; and determining optimal model parameters of the artificial neural network model according to the prediction precision index.
Optionally, the setting the model parameters in the artificial neural network model includes: dividing the sample data of the environment variable into a training set, a verification set and a test set, and configuring three data sets according to different proportions to serve as input data of an artificial neural network model; and performing multiple experimental settings on the hidden layer neuron, the network learning rate, the maximum iteration number and the training number.
Optionally, the determining the optimal model parameters of the artificial neural network model according to the prediction precision index includes: according to the input data and the output data of each group of model parameters, respectively calculating the correlation coefficient, the root mean square error and the deviation corresponding to each model parameter; determining optimal model parameters of the artificial neural network model through the correlation coefficient, the root mean square error and the deviation; wherein the root mean square error and the deviation are described by the following formulas (11), (12), respectively:
wherein RMSE represents root mean square error, BIAS represents BIAS, x i Is the value of the ith input data, y i Is the value of the output data predicted by the ith model,/->And->Representing the average value of each set of data, n is the length of each set of data.
Optionally, the constructing a spatial downscaling model of the change in the ground water reserve of the GRACE according to the artificial neural network model based on the optimal model parameters and the optimal lag time to realize the simulation of the change in the ground water reserve of high spatial resolution includes: constructing a GRACE groundwater reserve variation spatial downscaling model with low spatial resolution and environmental variables by using the artificial neural network model; and applying the GRACE groundwater reserve change spatial downscaling model to the environment variable with high spatial resolution, and obtaining the groundwater reserve change with high spatial resolution.
Optionally, the constructing a GRACE groundwater reserve variation spatial downscaling model with low spatial resolution and environmental variables by using the artificial neural network model includes: aggregating the environmental variable pixels with high spatial resolution into environmental variables with low spatial resolution, wherein the environmental variables with low spatial resolution are consistent with the low spatial resolution of the GRACE groundwater reserve variable; training the environment variable with low spatial resolution by using the artificial neural network model, and constructing a GRACE groundwater reserve change spatial downscaling model with the environment variable with low spatial resolution and the GRACE groundwater reserve change quantity being nonlinear functions, wherein the GRACE groundwater reserve change spatial downscaling model outputs the simulated groundwater reserve change with low spatial resolution.
Optionally, the applying the GRACE groundwater reserve change spatial downscaling model to the environmental variable with high spatial resolution and obtaining the groundwater reserve change with high spatial resolution includes: inputting high-spatial-resolution environmental variables based on a GRACE groundwater reserve change spatial downscaling model with a nonlinear function, and obtaining groundwater reserve change with high spatial resolution through simulation; subtracting the GRACE groundwater reserve variable quantity from the groundwater reserve variable quantity with low spatial resolution obtained by simulation to obtain a prediction residual error with low spatial resolution; and interpolating the prediction residual error with low spatial resolution to high spatial resolution to obtain the prediction residual error with high spatial resolution, and performing addition operation on the simulated groundwater reserve variable quantity with high spatial resolution and the prediction residual error with high spatial resolution to obtain the groundwater reserve variable quantity with high spatial resolution after residual error correction.
In a second aspect, the present application further provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the steps of the above method are implemented when the processor executes the computer program.
In a third aspect, a computer readable storage medium is provided, on which a computer program is stored which, when executed by a processor, carries out the steps of the method described above.
The application has at least the following advantages:
according to the technical content provided by the embodiment of the application, the underground water reserve variable quantity with low spatial resolution is obtained, the underground water reserve with low spatial resolution and each environment variable are decomposed according to an integrated empirical mode decomposition algorithm, the correlation between GRACE underground water reserve variable with different lag time and each environment variable is calculated according to a Pearson correlation analysis method, the optimal lag time of the underground water reserve variable quantity relative to the environment variable is determined, a plurality of groups of artificial neural network model parameters are set according to a control variable method, further, based on artificial neural network model training sample data, the optimal model parameters of the artificial neural network model are determined according to three prediction precision indexes including correlation coefficient, root mean square error and deviation, finally, a GRACE underground water reserve variable spatial downscale model is obtained according to the optimal lag time and the optimal model parameters, the model is applied to the environment variable with high spatial resolution, the underground water reserve variable quantity data with high spatial resolution is obtained, and then the accuracy of GRACE underground water monitoring can be improved, and the applicability of the spatial downscale model in a complex small area is improved.
Drawings
FIG. 1 is a flow chart showing an artificial neural network-based method for spatial downscaling of groundwater reserve changes in an embodiment;
FIG. 2 is a schematic flow chart showing the acquisition of low spatial resolution groundwater reserve variation in an embodiment;
FIG. 3 is a schematic flow chart showing the determination of an optimal lag time in one embodiment;
FIG. 4 is a flow diagram illustrating determining optimal model parameters for an artificial neural network model in one embodiment;
FIG. 5 is a schematic diagram of a simulation flow for constructing a GRACE groundwater reserve variation spatial downscaling model to obtain high spatial resolution groundwater reserve variation in one embodiment;
FIG. 6 is a schematic flow chart showing the construction of a GRACE groundwater reserve change spatial downscaling model in one embodiment;
fig. 7 is a schematic structural diagram of a computer device in one embodiment.
Detailed Description
The present application will be described in further detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the singular is "a," an, "and/or" the "when used in this specification is taken to mean" the presence of a feature, step, operation, device, component, and/or combination thereof.
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present application more apparent, the embodiments of the present application will be described in detail below with reference to the accompanying drawings. However, as will be appreciated by those of ordinary skill in the art, in the various embodiments of the present application, numerous technical details have been set forth in order to provide a better understanding of the present application. However, the technical solutions claimed in the present application can be implemented without these technical details and with various changes and modifications based on the following embodiments. The following embodiments are divided for convenience of description, and should not be construed as limiting the specific implementation of the present application, and the embodiments may be combined with each other and cited with each other without contradiction.
Fig. 1 is a schematic flow chart of an underground water reserve change spatial downscaling method based on an artificial neural network according to an embodiment of the present application, as shown in fig. 1, the method may include the following steps:
s201, acquiring the underground water reserve variable quantity with low spatial resolution from the land total water reserve monitored by GRACE satellites based on the water balance principle;
s202, acquiring an environment variable with high spatial resolution corresponding to the underground water reserve variation, and determining the optimal lag time of the underground water reserve variation relative to the environment variable;
S203, using the environment variable as sample data and determining optimal model parameters of the artificial neural network model based on a control variable method for multiple training modeling;
the optimal model parameters comprise one or more of the proportion of each sample participating in training, the number of neurons of an implicit layer, the network learning rate, the maximum iteration number, the initial range of weight values and the training number;
s204, constructing a GRACE groundwater reserve change space downscaling model according to the artificial neural network model based on the optimal model parameters and the optimal lag time so as to simulate the groundwater reserve change volume with high spatial resolution.
Each step is specifically described in detail below:
s201, acquiring the underground water reserve variable quantity with low spatial resolution from the land total water reserve monitored by GRACE satellites based on the water balance principle;
in the present embodiment, as shown in fig. 1, it should be noted that the remote sensing satellite is used as an artificial satellite for the outer space remote sensing platform. The remote sensing technology using satellites as a platform is called satellite remote sensing. Typically, a remote sensing satellite may be in orbit for years. The satellite orbits can be determined as desired. The remote sensing satellite can cover the whole earth or any appointed area in a prescribed time, and can continuously remotely sense a certain appointed region on the surface of the earth when running along the geosynchronous orbit. All remote sensing satellites need to be provided with remote sensing satellite ground stations, satellite data obtained from a remote sensing bazaar platform can monitor agricultural, forestry, ocean, homeland, environmental protection, meteorological conditions and the like, and the monitoring and collection of land total water reserves are realized through remote sensing satellites GRACE. The spatial resolution obtained is generally 0.25 DEG x 0.25 DEG, and about 25km x 25km in the equatorial region. Groundwater reserve change data with a spatial resolution of 0.25 ° x 0.25 ° gives results with a relatively low spatial resolution when used to study hydrologic processes or water resource distribution in local areas.
In this embodiment, by the water balance principle, the groundwater reserve variation component is stripped from the terrestrial total water reserve monitored by the GRACE satellite according to the obtained terrestrial total water reserve, and the groundwater reserve variation with low spatial resolution is obtained.
S202, acquiring an environment variable with high spatial resolution corresponding to the underground water reserve variation, and determining the optimal lag time of the underground water reserve variation relative to the environment variable;
in this embodiment, according to the water balance principle, the groundwater reserve variable quantity with low spatial resolution is obtained, and the environment variable with high spatial resolution corresponding to the groundwater reserve variable quantity is obtained. Because the groundwater reserve variation has time lag in response to environmental variables, the relationship between groundwater reserve variation and environmental variables is tested at different times to obtain the optimal lag time of groundwater reserve variation with low spatial resolution relative to environmental variables with high spatial resolution.
S203, using the environment variable as sample data and determining optimal model parameters of the artificial neural network model based on a control variable method for multiple training modeling; the optimal model parameters comprise one or more of the proportion of each sample participating in training, the number of neurons of an implicit layer, the network learning rate, the maximum iteration number, the initial range of weights and the training number.
In this embodiment, it should be noted that, based on the artificial neural network model, the single hidden layer feedforward network model training data in the artificial neural network model is specifically adopted, and according to the control variable method, the data is trained, and according to the optimization target with the minimum error, the optimal model parameters of the artificial neural network model are obtained.
S204, constructing a GRACE groundwater reserve change space downscaling model according to an artificial neural network model based on the optimal model parameters and the optimal lag time so as to realize simulation of high-spatial resolution groundwater reserve change;
in this embodiment, it should be noted that, by peeling the groundwater reserve variation component from the land total water reserve monitored by the GRACE satellite according to the water balance principle, comprehensively considering the time lag effect of the groundwater reserve variation relative to the environmental variable and the optimal parameter setting of the artificial neural network model, the spatial downscaling is performed on the groundwater reserve variation based on the GRACE satellite, so as to obtain a spatial downscaling model of the GRACE groundwater reserve variation, which is more suitable for a arid small area with a complex environment and has higher simulation precision, and can realize the groundwater reserve variation simulation with high spatial resolution.
Referring to fig. 1 and 2, in some embodiments, obtaining a low spatial resolution groundwater reserve variation from a GRACE satellite monitored total terrestrial water reserve based on a water balance principle in S201 includes:
s2011, collecting land total water reserve variable quantity monitored by GRACE satellites, wherein the land total water reserve variable quantity comprises the variable quantity of surface water, soil water, underground water reserve quantity, ice and snow melting water and vegetation canopy water;
s2012, calculating historical variation of surface water, soil water, ice and snow melting water and vegetation canopy water in a research area within a preset time period, wherein the historical variation is described by a formula (1):
ΔG(t)＝G(t)-G avg (1)，
wherein ΔG (t) represents the historical variation of a certain variable in the land total water reserve variation within a preset time period, G (t) represents the variation corresponding to a certain variable at the time t, G avg Representing the average value of the historical variation of a certain variable in a preset time period;
s2013, subtracting the land total water reserve variable quantity monitored by the GRACE satellite from the variable quantity of surface water, soil water, ice and snow melting water and vegetation canopy water based on a water balance principle to obtain the groundwater reserve variable quantity with low spatial resolution, wherein the groundwater reserve variable quantity is described by a formula (2):
ΔGWS＝ΔTWS-ΔSWS-ΔSMS-ΔSWE-ΔCWS(2)，
Where Δgws represents the groundwater reserve amount change, Δtws represents the land total water reserve amount change, Δsws represents the surface water reserve amount change, Δsms represents the soil water content change, Δswe represents the ice and snow melt amount change, and Δcws represents the vegetation canopy water content change.
In this embodiment, it should be noted that the water balance principle means that the number of water circulation means that in a given arbitrary scale of time domain space, the motion of water (including phase change) has continuity, and balance is maintained in number. The basic principle of equilibrium is the law of conservation of mass.
And subtracting the historical variation of the surface water, the soil water, the ice and snow melting water and the vegetation canopy water in the research area in the preset time from the land total water reserve variation observed by GRACE, and obtaining the groundwater reserve variation with low spatial resolution according to the water balance principle.
In one example, the time t is taken as an example in 2009, G (t) represents a variation of surface water corresponding to 2009, and G avgof2004-2009 Representing the average value of the surface water in the period from 2004 to 2009, and subtracting the average value of the surface water variation corresponding to 2009 from the average value of the surface water variation in the period from 2004 to 2009 to finally obtain the historical variation of the surface water in the land total water reserve. According to the calculation method, the variation of the soil water, the ice and snow melting water and the vegetation canopy water is calculated respectively, and then the variation of the surface water, the soil water, the ice and snow melting water and the vegetation canopy water is subtracted according to the land total water reserve variation monitored by the GRACE satellite, so that the groundwater reserve variation with low spatial resolution is obtained.
Referring to fig. 1 and 3, in some embodiments, in S202, acquiring a high spatial resolution environmental variable corresponding to the groundwater reserve variation and determining an optimal lag time of the groundwater reserve variation with respect to the environmental variable includes:
s2021, decomposing a time sequence of groundwater reserve variable quantity with low spatial resolution and each environment variable according to an integrated empirical mode decomposition algorithm to obtain a plurality of eigenmode functions and residual errors;
wherein the environmental variables include precipitation, surface temperature, potential evapotranspiration, normalized vegetation index and soil moisture content;
s2022, calculating an average period corresponding to each eigenmode function according to a fast Fourier transform method;
s2023, calculating a variance contribution rate corresponding to each eigenmode function, and describing by a formula (3):
wherein P is i Variance contribution ratio for ith eigenmode function, V i Is the variance of the ith eigenmode function, and n is the total number of eigenmode functions;
s2024, based on the obtained average period and variance contribution rate corresponding to each eigenmode function, comparing the average period and variance contribution rate corresponding to each eigenmode function, and selecting the eigenmode function with the average period closest to 1 year and the larger variance contribution rate as a substitute time sequence of the original obtained groundwater reserve variable quantity and each environment variable;
S2025, calculating correlations between groundwater reserve variable amounts of different lag times and alternative time sequences of various environment variables according to a Pearson correlation coefficient method, wherein the lag time corresponding to the highest correlation coefficient is defined as the optimal lag time, and describing by a formula (4):
in the formula, CC represents a correlation coefficient,and->Represents the average value, x, of each set of data i And y i Is two sets of data for correlation analysis, n is the length of each set of data.
In this embodiment, the empirical mode decomposition (Empirical Mode Decomposition, EMD) is a time-frequency domain signal processing method that performs signal decomposition according to the time-scale characteristics of the data itself, without setting any basis functions in advance. The EMD has obvious advantages in processing non-stationary and non-linear data, is suitable for analyzing a non-linear non-stationary signal sequence, and has higher signal-to-noise ratio. Stationary signal: the distribution parameters (mean, variance, covariance, etc.) or the distribution law do not change over time. Non-stationary signal: the distribution parameters (mean, variance, covariance, etc.) or the distribution law change over time. The key point of the method is empirical mode decomposition, so that a complex signal is decomposed into a limited number of eigenmode functions (Intrinsic Mode Function, IMF), and each decomposed IMF component contains local characteristic information of different time scales of the original signal. Here, the time series of the low-spatial-resolution groundwater reserve variable quantity and each environmental variable quantity are decomposed through an integrated empirical mode decomposition algorithm, and a plurality of eigenmode functions and residual errors are obtained, so that the correlation of the low-spatial-resolution groundwater reserve variable quantity and each environmental variable quantity is conveniently analyzed.
In statistics, the Pearson correlation coefficient (Pearson correlation coefficient), also called Pearson product-moment correlation coefficient, abbreviated as PPMC or PCCs, is used to measure the correlation (linear correlation) between two variables X and Y, and its value is between-1 and 1.
In one example, the correlation between groundwater reserve variation at different lag times and substitution timing of each environmental variable is calculated by analyzing the time series of groundwater reserve variation and each environmental variable. Specifically, the different lag times can be zero lag, namely 0 month, 1 month, 2 months, 3 months, 4 months and 5 months, the correlation between each input variable and the GRACE groundwater reserve variable is tested according to the different lag times, and the lag time corresponding to the highest correlation coefficient is selected according to the calculated correlation coefficient to be determined as the optimal lag time.
Referring to fig. 3, in some embodiments, in S2021, decomposing the time series of the low spatial resolution groundwater reserve variable and each environmental variable according to the integrated empirical mode decomposition algorithm, to obtain a plurality of eigenmode functions and residuals includes:
Since in some embodiments the time series change of the change amount of the GRACE groundwater reserve and each environmental variable has a non-stationarity characteristic and includes a significant long-term change trend, the correlation between the change amount of the GRACE groundwater reserve and each environmental variable cannot be directly calculated, and therefore in this embodiment, the time series of the change amount of the GRACE groundwater reserve and each environmental variable is decomposed by using an integrated empirical mode decomposition method, so as to obtain a plurality of eigen mode functions and residuals. In this embodiment, a relatively stable annual change sequence is obtained by removing the long-term change trend in the time sequence, and then correlation analysis is performed on the change amount of the Grace groundwater reserves and each environmental variable based on different lag time. Wherein the integrated empirical mode decomposition method is described by equation (5):
wherein x (t) represents the time series to be decomposed, i.e., the time series of the low spatial resolution groundwater reserve variation and the respective high spatial resolution environmental variables, IMF i (t) represents the ith eigenmode function, R (t) represents the residual, t is time, and n represents the total number of eigenmode functions.
In some embodiments, referring to fig. 3, in S2022, an average period corresponding to each eigenmode function is calculated according to a fast fourier transform method;
In this embodiment, it should be noted that, based on the fft algorithm, the rotation factor is based onDecomposing a large discrete fourier transform into small discrete fourier transforms, the discrete fourier transforms being described by equation (6):
wherein x (N) represents the input discrete point sequence, and the length of x (N) is n=2 M If the length does not meet the multiple of 2, zero length can be added to enable the length to meet the condition, and M is a positive integer;
x (n) is divided into two groups according to parity, described by formula (7):
wherein, let n=2r and n=2r+1,
based on the symmetry and periodicity of the twiddle factor, the parity grouped discrete point sequence is reduced by sorting until the discrete fourier transform of N points is finally calculated with a set of two-point discrete fourier transforms, described by equation (8):
wherein A (k) and B (k) are described by formulas (9), (10):
according to the discrete Fourier transform of the N/2 points of A (k) and B (k), wherein X (k) is the discrete Fourier transform of the N points, the A (k) and the B (k) can be continuously decomposed by adopting the same method, and finally, the discrete Fourier transform of the N points is finally calculated by using a group of two-point discrete Fourier transforms;
Based on the obtained discrete Fourier transform result, performing spectrum analysis on the time sequence to obtain an amplitude spectrum f of X (k) k Each k corresponds to a discrete frequency value, the frequency of which, for a periodic sequence x (n), impacts at the period, the average period of x (n) being obtained from the impact values.
In this embodiment, the fast fourier transform (fast Fourier transform), that is, a general term of an efficient and fast calculation method for calculating a Discrete Fourier Transform (DFT) by a computer, is simply referred to as FFT. It is obtained by improving the algorithm of discrete Fourier transform according to the characteristics of discrete Fourier transform such as odd, even, virtual, real, etc. The method can greatly reduce the multiplication times required by a computer for calculating the discrete Fourier transform, and particularly, the more the number of transformed sampling points N is, the more remarkable the FFT algorithm calculation amount is saved.
Specifically, the average period corresponding to each eigenmode function obtained by decomposing the integrated empirical mode decomposition method is calculated by using a fast fourier transform method, and the discrete fourier transform of N points is finally calculated by using a group of discrete fourier transforms of two points. According to the obtained average period, performing spectrum analysis on the time sequence to obtain an amplitude spectrum f of X (k) k Each k corresponds to a discrete frequency value. And for the periodic sequence x (n), according to the occurrence of the impact at the period, finally obtaining the average period corresponding to each eigenmode function.
In some embodiments, referring to fig. 1 and 4, in S203, determining optimal model parameters of an artificial neural network model based on a control variable method for multiple training modeling using the environmental variables as sample data includes:
s2031, setting model parameters in the artificial neural network model;
in this embodiment, it should be noted that, according to the proportion of each sample involved in training, the number of neurons in the hidden layer, the network learning rate, the maximum iteration number, the initial range of the weight, the training number and the like, based on the controlled variable method, the model is trained for multiple times, and the optimal model parameters of the artificial neural network model are determined.
In one example, referring to table 1 below, sample data of an environment variable is divided into a training set, a validation set, and a test set, and three data sets are configured in different proportions. And configuring sample data in different proportions according to the sequence of training samples, verification samples and test samples, and performing experiment setting on hidden layer neurons, network learning rate, maximum iteration times and training times for multiple times. And respectively determining the parameter configuration of the artificial neural network model according to a control variable method to train so as to obtain corresponding output data.
Table 1
S2032, training the environment variables based on an artificial neural network model to obtain a plurality of groups of output data, and determining the optimal values of the model parameters by using an error back propagation algorithm;
specifically, based on a control variable method, training an artificial neural network model according to sample data configured in different proportions to obtain multiple groups of output data, and determining the optimal value of each model parameter through an error back propagation algorithm, wherein the specific principle is as follows:
and solving model parameters by adopting a single hidden layer feedforward network model in the artificial neural network model and using a Levenberg-Marquardt error back propagation algorithm. In neural networks, the most basic information processing unit is the M-P neuron model, which is responsible for receiving input signals from other n neurons, the input signals being transmitted through weighted connections, the information received by the neurons being z, z= Σw i x i Wherein x is i Representing input from the ith neuron, w i Representing the connection weight of the ith neuron, z represents the combination of the information of the neuron and the excitation of the neuron itself, and if z is greater than θ, the output y of the neuron is generated by the activation function process, where f represents an activation function, y represents output information, θ represents an excitation value, θ is a threshold value of a neuron, and the activation function is generally described by a Sigmoid function through a formula:
training set d= { (x) by setting 1 ，y 1 )，(x 2 ，y 2 )，...，(x m ，y m )}，x i ∈R d ，y i ∈R l I.e. the input samples are described by d attributes and the output is an l-dimensional real value vector. Assuming that the number of hidden layer neurons is q, the threshold value of the output layer neurons is represented by θ j The threshold value of the h neuron of the hidden layer is shown as gamma h Representing that the connection weight between the ith neuron of the input layer and the h neuron of the hidden layer is v ih The connection weight between the h neuron of the hidden layer and the j neuron of the output layer is w hj . The input received by the h neuron of the hidden layer is alpha h The j-th neuron of the output layer receives the input beta j . Alpha of it h 、β j The descriptions by the formulas are respectively:
wherein b is h The output of the h neuron is the hidden layer.
In one example, both hidden layer and output layer neurons use Sigmoid activation functions, training samples example (x k ，y k ) The output of the neural network isThat is, the predicted output expression of the neural network is:
the neural network is trained on the sample case (x k ，y k ) The mean square error is:
the mathematical model of the whole feedforward neural network is established, and the parameter matrix of the neural network is solved according to the optimization target with the minimum error.
Specifically, the parameters are adjusted in the negative gradient direction of the target based on the gradient descent strategy according to the back propagation algorithm, and the learning rate eta is given to obtain
Wherein w is hj The input value beta of the jth neuron is affected first j Then influence the output valuePost influence E k Finally, obtaining:
according to beta j Is defined as follows:
from the property f' (x) =f (x) (1-f (x)) of the Sigmoid function, it is possible to obtain:
and further get the w-related back propagation algorithm hj Is:
Δw hj ＝ηg j b h
according to the chain derivation rule, it is possible to obtain:
Δθ j ＝-ηg j
Δv ih ＝ηe h x i
Δθ h ＝-ηg h
wherein e h The expression of (2) is:
thus, according to the above algorithm, the optimal values of the respective model parameters can be determined therefrom.
In some embodiments, referring to fig. 4, S2033, optimal model parameters of the artificial neural network model are determined according to a prediction accuracy index.
In the embodiment, the optimal model parameters of the artificial neural network model are determined by selecting three prediction precision indexes, namely a correlation coefficient, a root mean square error and a deviation, and the correlation coefficient, the root mean square error and the deviation corresponding to each model parameter are calculated according to the input data and the output data of each group of model parameters; and determining optimal model parameters of the artificial neural network model through the correlation coefficient, the root mean square error and the deviation; the root mean square error and the deviation are described by the following formulas (11) and (12):
Wherein RMSE represents root mean square error, BIAS represents BIAS, x i Is the value of the ith input data, y i Is the value of the output data predicted by the ith model,and->Representing the average value of each set of data, n is the length of each set of data.
In this embodiment, it should be noted that the optimal model parameters are determined according to the correlation coefficient CC, the root mean square error RMSE, and the BIAS. The correlation coefficient may reflect a statistical indicator of how closely the correlation between the variables is. The correlation coefficient is calculated according to a product difference method, and the degree of correlation between two variables is reflected by multiplying the two dispersions on the basis of the dispersion of the two variables and the average value of the two variables. The root mean square error is the square root of the ratio of the square of the deviation of the predicted value from the true value to the number of observations n, and is used to measure the deviation between the observed value and the true value. Deviations, also known as apparent errors, refer to the difference between individual measured values and the average value of the measurement, which can be used to measure the accuracy of the measurement. And comparing the set multiple groups of model parameters with the correlation coefficient obtained by the Pearson correlation coefficient method and the calculated root mean square error and deviation, and finally determining the optimal model parameters of the artificial neural network model.
In some embodiments, referring to fig. 1 and 5, in S204, constructing a GRACE groundwater reserve change spatial downscaling model according to an artificial neural network model based on the optimal model parameters and the optimal lag time, to implement a simulation of a high spatial resolution groundwater reserve change, including:
s2041, constructing a GRACE groundwater reserve change space downscaling model with low spatial resolution and GRACE groundwater reserve change volume of environmental variables by using an artificial neural network model;
s2042, applying a GRACE groundwater reserve change space downscaling model to the environment variable with high spatial resolution, and obtaining groundwater reserve change with high spatial resolution;
in this embodiment, as shown in fig. 5 and 6, S2041, constructing a GRACE groundwater reserve change spatial downscaling model with low spatial resolution and environmental variables by using the artificial neural network model includes:
s20411, aggregating environment variable pixels with high spatial resolution into environment variables with low spatial resolution, wherein the environment variables with low spatial resolution are consistent with the low spatial resolution of GRACE groundwater reserve variable;
S20412, training the environment variable with low spatial resolution by using an artificial neural network model, and constructing a GRACE groundwater reserve change spatial downscaling model with the environment variable with low spatial resolution and the GRACE groundwater reserve change quantity being nonlinear functions, wherein the GRACE groundwater reserve change spatial downscaling model outputs the simulated groundwater reserve change quantity with low spatial resolution. In addition, because the artificial neural network model is sensitive to the numerical range of each variable, before training the model, all variables are standardized by using a maximum and minimum normalization method, namely, the numerical value is converted into between 0 and 1, so that the model can be trained later.
In this embodiment, as shown in fig. 5 and 6, S2042, applying the regression model to the environmental variable with high spatial resolution, and obtaining the groundwater reserve variation with high spatial resolution includes:
s20421, based on the GRACE groundwater reserve change space downscaling model with the nonlinear function obtained in the step S20412, inputting high-spatial-resolution environmental variables into the model, and simulating to obtain groundwater reserve change with high spatial resolution;
s20422, subtracting the original GRACE groundwater reserve variable quantity from the groundwater reserve variable quantity with low spatial resolution obtained by simulation in the step S20412 to obtain a prediction residual error with low spatial resolution;
S20423, interpolating the prediction residual error with low spatial resolution obtained in the step S20422 to high spatial resolution to obtain the prediction residual error with high spatial resolution, and performing addition operation on the groundwater reserve variable quantity with high spatial resolution obtained in the simulation of the step S20421 and the prediction residual error with high spatial resolution to obtain the groundwater reserve variable quantity with high spatial resolution after residual error correction.
In this embodiment, it should be noted that, a GRACE groundwater reserve change spatial downscaling model based on the optimal lag time and the optimal model parameters is constructed to implement simulation of groundwater reserve change data with high spatial resolution. The method comprises the steps of constructing a regression model of GRACE groundwater reserve change data and environmental variables under low resolution by using an artificial neural network model, and then applying the model to the environmental variables with high spatial resolution.
The method mainly comprises the steps of obtaining the underground water reserve variable quantity with low spatial resolution through the land total water reserve monitored by GRACE satellites according to the water balance principle, decomposing the underground water reserve variable quantity with low spatial resolution and each environment variable according to an integrated empirical mode decomposition algorithm, calculating the correlation between the GRACE underground water reserve variable quantity with different lag time and each environment variable according to a Pearson correlation analysis method, determining the optimal lag time of the underground water reserve variable quantity relative to the environment variable, setting a plurality of groups of artificial neural network model parameters according to a control variable method, further determining the optimal model parameters of the model according to artificial neural network model training sample data through three indexes of prediction accuracy such as correlation coefficient, root mean square error and deviation, finally obtaining a GRACE underground water reserve variable space downscale model according to the optimal lag time and the optimal model parameters, applying the model to the environment variable with high spatial resolution, obtaining underground water reserve variable quantity data with high spatial resolution, improving the underground water monitoring precision of the GRACE satellites, and improving the applicability of the space downscale model in a small area.
According to embodiments of the present application, there is also provided a computer device, a computer-readable storage medium.
As shown in fig. 7, is a block diagram of a computer device according to an embodiment of the present application. Computer equipment is intended to represent various forms of digital computers or mobile devices. Wherein the digital computer may comprise a desktop computer, a portable computer, a workstation, a personal digital assistant, a server, a mainframe computer, and other suitable computers. The mobile device may include a tablet, a smart phone, a wearable device, etc.
As shown in fig. 7, the apparatus 600 includes a computing unit 601, a ROM 602, a RAM 603, a bus 604, and an input/output (I/O) interface 605, and the computing unit 601, the ROM 602, and the RAM 603 are connected to each other through the bus 604. An input/output (I/O) interface 605 is also connected to bus 604.
The computing unit 601 may perform various processes in the method embodiments of the present application according to computer instructions stored in a Read Only Memory (ROM) 602 or computer instructions loaded from a storage unit 608 into a Random Access Memory (RAM) 603. The computing unit 601 may be a variety of general and/or special purpose processing components having processing and computing capabilities. The computing unit 601 may include, but is not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), as well as any suitable processor, controller, microcontroller, etc. In some embodiments, the methods provided by embodiments of the present application may be implemented as a computer software program tangibly embodied on a computer-readable storage medium, such as storage unit 608.
The RAM 603 may also store various programs and data required for operation of the device 600. Part or all of the computer program may be loaded and/or installed onto the device 600 via the ROM 602 and/or the communication unit 609.
An input unit 606, an output unit 607, a storage unit 608, and a communication unit 609 in the device 600 may be connected to the I/O interface 605. Wherein the input unit 606 may be such as a keyboard, mouse, touch screen, microphone, etc.; the output unit 607 may be, for example, a display, a speaker, an indicator light, or the like. The device 600 is capable of exchanging information, data, etc. with other devices through the communication unit 609.
It should be noted that the device may also include other components necessary to achieve proper operation. It may also include only the components necessary to implement the present application, and not necessarily all the components shown in the figures.
Various implementations of the systems and techniques described here can be implemented in digital electronic circuitry, integrated circuitry, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof.
Computer instructions for implementing the methods of the present application may be written in any combination of one or more programming languages. These computer instructions may be provided to a computing unit 601 such that the computer instructions, when executed by the computing unit 601, such as a processor, cause the steps involved in the method embodiments of the present application to be performed.
The computer readable storage medium provided herein may be a tangible medium that may contain, or store, computer instructions for performing the steps involved in the method embodiments of the present application. The computer readable storage medium may include, but is not limited to, storage media in the form of electronic, magnetic, optical, electromagnetic, and the like.
The above embodiments do not limit the scope of the application. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application are intended to be included within the scope of the present application.
Claims (10)
1. The underground water reserve change space downscaling method based on the artificial neural network is characterized by comprising the following steps of:
Based on the water balance principle, obtaining the groundwater reserve variable quantity with low spatial resolution from the land total water reserve monitored by the GRACE satellite;
acquiring a high spatial resolution environmental variable corresponding to the groundwater reserve variation and determining an optimal lag time of the groundwater reserve variation relative to the environmental variable;
decomposing the time sequence of the underground water reserve variable quantity and each environment variable according to an integrated empirical mode decomposition algorithm to obtain a plurality of eigenmode functions and residual errors; wherein the environmental variables include precipitation, surface temperature, potential evaporation, normalized vegetation index, and soil moisture content;
calculating the average period corresponding to each eigenmode function according to a fast Fourier transform method;
calculating the variance contribution rate corresponding to each eigenmode function;
based on the obtained average period and variance contribution rate corresponding to each eigenmode function, comparing the average period and variance contribution rate corresponding to each eigenmode function, selecting the eigenmode function with the average period closest to 1 year and the variance contribution rate larger as an alternative time sequence of the original obtained underground water reserve variable quantity and each environment variable;
Calculating correlations between the groundwater reserve variable quantities of different lag times and the substitution time sequences of all environment variables according to a Pearson correlation coefficient method, wherein the lag time corresponding to the highest correlation coefficient is defined as the optimal lag time;
the environment variable is used as sample data to determine the optimal model parameters of the artificial neural network model based on a control variable method for multiple training modeling;
the optimal model parameters comprise one or more of the proportion of each sample participating in training, the number of neurons of an implicit layer, the network learning rate, the maximum iteration number, the initial range of weight values and the training number;
and constructing a GRACE groundwater reserve change space downscaling model according to an artificial neural network model based on the optimal model parameters and the optimal lag time so as to realize simulation of high-spatial resolution groundwater reserve change.
2. The method for spatially downscaling groundwater reserve variation based on an artificial neural network according to claim 1, wherein the obtaining the groundwater reserve variation with low spatial resolution from the total terrestrial water reserve monitored by the GRACE satellite based on a water balance principle comprises:
collecting land total water reserve variation monitored by a GRACE satellite, wherein the land total water reserve variation comprises variation of surface water, soil water, groundwater reserve, ice and snow melting water and vegetation canopy water;
Calculating historical variation of the surface water, the soil water, the ice and snow melting water and the vegetation canopy water in a preset time period in a research area, and describing by a formula (1):
，
in the method, in the process of the invention,representing the historical variation of a certain variable in the land total water reserve variation within a preset time period,/for>Represents the corresponding variable quantity of a certain variable at the moment t, < >>Representing the average value of the historical variation of a certain variable in a preset time period;
subtracting the land total water reserve variable quantity monitored by the GRACE satellite from the variable quantity of the surface water, the soil water, the ice and snow melting water and the vegetation canopy water based on a water quantity balance principle to obtain the underground water reserve variable quantity, wherein the underground water reserve variable quantity is described by a formula (2):
，
in the method, in the process of the invention,indicating groundwater reserve change, +.>Indicating land total water reserve change, < >>Representing the change of the reserve of surface water, +.>Indicating the water content of the soilQuantity change amount->The change of the ice and snow melting water quantity is shown,indicating the water content variation of vegetation canopy.
3. The method for spatially downscaling groundwater reserve variation based on an artificial neural network according to claim 1,
the variance contribution rate corresponding to each eigenmode function is calculated and described by a formula (3):
，
In the method, in the process of the invention,is->Variance contribution of the individual eigenmode functions, +.>Is->Variance of individual eigenmode functions,/->Is the total number of eigenmode functions;
the correlation between the groundwater reserve variation of different lag time and the substitution time sequence of each environment variable is calculated according to the Pearson correlation coefficient method, and is described by a formula (4):
，
in the method, in the process of the invention,CCthe correlation coefficient is represented by a correlation coefficient, and/>represents the mean value of each group of data, +.>And->Is two sets of data for correlation analysis, +.>Is the length of each set of data.
4. The method for spatial downscaling of groundwater reserve variation based on artificial neural network according to claim 3, wherein decomposing the time series of groundwater reserve variation and each environmental variable according to the integrated empirical mode decomposition algorithm to obtain a plurality of eigenmode functions and residuals comprises:
decomposing the time sequence of the groundwater reserve variable quantity and each environment variable based on an integrated empirical mode decomposition algorithm, and describing by a formula (5):
，
in the method, in the process of the invention,representing the time sequence to be decomposed, i.e. the time sequence of the groundwater reserve variation and the respective environmental variable,/for each of the environmental variables >Indicate->Intrinsic mode function>Representing residual error,/->Is time, & lt>Representing the total number of eigenmode functions.
5. The method for spatially downscaling groundwater reserve variation based on an artificial neural network according to claim 1, wherein determining optimal model parameters of an artificial neural network model based on a controlled variable method multiple training modeling using the environmental variable as sample data comprises:
setting model parameters in the artificial neural network model;
training the environment variable based on an artificial neural network model to obtain a plurality of groups of output data, and determining the optimal value of each model parameter by utilizing an error back propagation algorithm;
and determining optimal model parameters of the artificial neural network model according to the prediction precision index.
6. The method for spatially downscaling groundwater reserve variation based on an artificial neural network according to claim 5, wherein said setting model parameters in the artificial neural network model comprises:
dividing the sample data of the environment variable into a training set, a verification set and a test set, and configuring three data sets according to different proportions to serve as input data of an artificial neural network model;
And performing multiple experimental settings on the hidden layer neuron, the network learning rate, the maximum iteration number and the training number.
7. The method for spatially downscaling groundwater reserve variation based on an artificial neural network according to claim 6, wherein determining optimal model parameters of the artificial neural network model according to a prediction accuracy index comprises:
according to the input data and the output data of each group of model parameters, respectively calculating the correlation coefficient, the root mean square error and the deviation corresponding to each model parameter;
determining optimal model parameters of the artificial neural network model through the correlation coefficient, the root mean square error and the deviation; wherein,
the root mean square error and the deviation are described by the following formulas (11) and (12), respectively:
，
，
wherein,representing root mean square error>Indicating deviation->Is->Value of the input data->Is->The value of the output data predicted by the individual model, +.>And->Represents the mean value of each group of data, +.>Is the number of each group
The length of the data.
8. The method for spatially downscaling groundwater reserve variation based on an artificial neural network according to claim 1, wherein the constructing a GRACE groundwater reserve variation spatially downscaling model based on the optimal model parameters and the optimal lag time according to an artificial neural network model to achieve simulation of high spatial resolution groundwater reserve variation comprises:
Constructing a GRACE groundwater reserve variation spatial downscaling model with low spatial resolution and environmental variables by using the artificial neural network model;
and applying the GRACE groundwater reserve change spatial downscaling model to the environment variable with high spatial resolution, and obtaining the groundwater reserve change with high spatial resolution.
9. The method for spatially downscaling groundwater reserve variation based on an artificial neural network according to claim 8,
the construction of the GRACE groundwater reserve variation spatial downscaling model with low spatial resolution and environmental variables by using the artificial neural network model comprises the following steps:
aggregating the environmental variable pixels with high spatial resolution into environmental variables with low spatial resolution, wherein the environmental variables with low spatial resolution are consistent with the low spatial resolution of the GRACE groundwater reserve variable;
training the environment variable with low spatial resolution by using the artificial neural network model, and constructing a GRACE groundwater reserve change spatial downscaling model with the environment variable with low spatial resolution and the GRACE groundwater reserve change quantity being nonlinear functions, wherein the GRACE groundwater reserve change spatial downscaling model outputs the simulated groundwater reserve change with low spatial resolution.
10. The method for spatially downscaling groundwater reserve variation based on an artificial neural network according to claim 9,
the application of the GRACE groundwater reserve variation spatial downscaling model to high spatial resolution environmental variables and obtaining groundwater reserve variation with high spatial resolution comprises:
a spatial downscaling model based on GRACE groundwater reserve variation with a nonlinear function;
inputting high-spatial-resolution environmental variables, and simulating to obtain groundwater reserve variable quantity with high spatial resolution;
subtracting the GRACE groundwater reserve variable quantity from the simulated low-spatial-resolution groundwater reserve variable quantity to obtain a low-spatial-resolution prediction residual; and interpolating the prediction residual error with low spatial resolution to high spatial resolution to obtain the prediction residual error with high spatial resolution, and performing addition operation on the simulated groundwater reserve variable quantity with high spatial resolution and the prediction residual error with high spatial resolution to obtain the groundwater reserve variable quantity with high spatial resolution after residual error correction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310536452.0A CN116628442B (en) | 2023-05-12 | 2023-05-12 | Groundwater reserve change space downscaling method based on artificial neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310536452.0A CN116628442B (en) | 2023-05-12 | 2023-05-12 | Groundwater reserve change space downscaling method based on artificial neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116628442A CN116628442A (en) | 2023-08-22 |
| CN116628442B true CN116628442B (en) | 2023-12-22 |
Family
ID=87591255
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310536452.0A Active CN116628442B (en) | 2023-05-12 | 2023-05-12 | Groundwater reserve change space downscaling method based on artificial neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116628442B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117172131A (en) * | 2023-10-12 | 2023-12-05 | 中国科学院地理科学与资源研究所 | Method and system for predicting high-resolution underground water level based on artificial neural network |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529164A (en) * | 2016-11-03 | 2017-03-22 | 清华大学 | Method and system for acquiring ground water storage variation value by combining GRACE satellite |
| CN110852474A (en) * | 2019-09-24 | 2020-02-28 | 广州地理研究所 | Land water reserve prediction method, device and equipment based on decision tree algorithm |
| CN110852473A (en) * | 2019-09-24 | 2020-02-28 | 广州地理研究所 | Neural network algorithm-based land water reserve prediction method and equipment |
| CN112907441A (en) * | 2020-12-29 | 2021-06-04 | 中央财经大学 | Space downscaling method based on super-resolution of underground water satellite image |
| CN114218853A (en) * | 2021-11-18 | 2022-03-22 | 中国空间技术研究院 | Method for improving water reserve resolution and precision based on machine learning space downscaling |
| CN114491967A (en) * | 2021-12-30 | 2022-05-13 | 中国科学院地理科学与资源研究所 | Land water reserve prediction method, device, equipment and storage medium |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106295194B (en) * | 2016-08-12 | 2018-06-26 | 中国水利水电科学研究院 | A kind of glacier based on basin principle of water balance stores variable evaluative method |
| CN111752934B (en) * | 2020-05-29 | 2023-04-28 | 首都师范大学 | Gravity satellite-based high-resolution aquifer water storage coefficient inversion method and system |
-
2023
- 2023-05-12 CN CN202310536452.0A patent/CN116628442B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529164A (en) * | 2016-11-03 | 2017-03-22 | 清华大学 | Method and system for acquiring ground water storage variation value by combining GRACE satellite |
| CN110852474A (en) * | 2019-09-24 | 2020-02-28 | 广州地理研究所 | Land water reserve prediction method, device and equipment based on decision tree algorithm |
| CN110852473A (en) * | 2019-09-24 | 2020-02-28 | 广州地理研究所 | Neural network algorithm-based land water reserve prediction method and equipment |
| CN112907441A (en) * | 2020-12-29 | 2021-06-04 | 中央财经大学 | Space downscaling method based on super-resolution of underground water satellite image |
| CN114218853A (en) * | 2021-11-18 | 2022-03-22 | 中国空间技术研究院 | Method for improving water reserve resolution and precision based on machine learning space downscaling |
| CN114491967A (en) * | 2021-12-30 | 2022-05-13 | 中国科学院地理科学与资源研究所 | Land water reserve prediction method, device, equipment and storage medium |
Non-Patent Citations (3)
| Title |
|---|
| 利用GRACE卫星重力数据监测关中地区地下水储量变化;李婉秋;王伟;章传银;杨强;冯伟;刘阳;;地球物理学报(第06期);全文 * |
| 基于多层感知机重建GRACE与GRACE-FO之间数据缺失的黄河流域陆地水储量变化;石通等;地球物理学报;第65卷(第07期);全文 * |
| 联合层次聚类与非线性回归模型的GRACE地下水储量变化降尺度反演方法;李名语;中国优秀硕士学位论文全文数据库 基础科学辑(第05期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116628442A (en) | 2023-08-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Feng et al. | A multi-layer perceptron approach for accelerated wave forecasting in Lake Michigan | |
| Maroufpoor et al. | Reference evapotranspiration estimating based on optimal input combination and hybrid artificial intelligent model: Hybridization of artificial neural network with grey wolf optimizer algorithm | |
| Vandal et al. | Intercomparison of machine learning methods for statistical downscaling: the case of daily and extreme precipitation | |
| Zhang et al. | Prediction of sea surface temperature using long short-term memory | |
| Pathiraja et al. | Hydrologic modeling in dynamic catchments: A data assimilation approach | |
| Samadianfard et al. | Wavelet neural networks and gene expression programming models to predict short-term soil temperature at different depths | |
| Lu et al. | Streamflow simulation in data-scarce basins using Bayesian and physics-informed machine learning models | |
| Campos et al. | Improving NCEP’s global-scale wave ensemble averages using neural networks | |
| Nair et al. | Prediction of monthly summer monsoon rainfall using global climate models through artificial neural network technique | |
| Bai et al. | A forecasting method of forest pests based on the rough set and PSO-BP neural network | |
| CN116628442B (en) | Groundwater reserve change space downscaling method based on artificial neural network | |
| Bai et al. | Accurate estimation of tidal level using bidirectional long short-term memory recurrent neural network | |
| Choi et al. | Deep-learning model for sea surface temperature prediction near the Korean Peninsula | |
| Watson-Parris et al. | Model calibration using ESEm v1. 1.0–an open, scalable Earth system emulator | |
| Parasuraman et al. | Spiking modular neural networks: A neural network modeling approach for hydrological processes | |
| Pachev et al. | A framework for flexible peak storm surge prediction | |
| Saulter et al. | An application of NEMOVAR for regional wave model data assimilation | |
| Darmawan et al. | GRU and XGBoost Performance with Hyperparameter Tuning Using GridSearchCV and Bayesian Optimization on an IoT-Based Weather Prediction System. | |
| CN114676621A (en) | Method for improving accuracy of land water reserve abnormity based on deep learning weight load | |
| Krasnopolsky et al. | Using machine learning for model physics: An overview | |
| Krasnopolsky et al. | Adjusting neural network to a particular problem: Neural network-based empirical biological model for chlorophyll concentration in the upper ocean | |
| Dayal et al. | Application of hybrid artificial neural network algorithm for the prediction of standardized precipitation index | |
| Wei et al. | Probabilistic optimal interpolation for data assimilation between machine learning model predictions and real time observations | |
| Khan et al. | Forecasting renewable energy for environmental resilience through computational intelligence | |
| Juan et al. | Data-driven and physics-based approach for wave downscaling: A comparative study |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |