CN114970302B - Regional groundwater condition prediction method based on groundwater monitoring system - Google Patents

Abstract

The invention provides a regional groundwater condition prediction method based on a groundwater monitoring system, which comprises the following steps: by combining observation data in a groundwater observation system with real-time monitoring data, based on SVR, a water pumping quantity prediction function and a water pumping quantity-groundwater water level prediction function of each observation well are established, and the groundwater level is predicted in real time, so that the problems that the conventional groundwater condition estimation modeling is long in time consumption and needs a large amount of hydrogeological parameters, and the conventional groundwater condition estimation modeling is limited by complicated geological conditions, uncertainty of water condition prediction parameters, mathematical modeling burden and other factors are solved, a relatively accurate regional groundwater condition quantification prediction tool is not formed, accurate prediction of each part of the regional groundwater condition cannot be refined, and urban, town regional groundwater management cannot be well serviced.

Description

A regional groundwater situation prediction method based on groundwater monitoring system

Technical field

The invention belongs to the technical field of computer machine learning, and specifically relates to a regional groundwater regime prediction method based on a groundwater monitoring system.

Background technique

In most areas of my country, under the influence of climate change and uneven distribution of water resources, the use of water resources has become increasingly severe. Due to complex hydrogeological conditions, relevant groundwater hydrological data is scarce. Because groundwater has the characteristics of stable water quantity and good water quality, it is one of the important water sources for agricultural irrigation, enterprise production and urban residents' lives. Especially in arid and semi-arid areas where surface water is scarce, groundwater is often used as the main local water source. Groundwater resources play a vital role in regional water supply problems. If water resources are not properly managed, excessive use of groundwater resources can easily cause problems such as stratum subsidence and collapse, resulting in water resource attenuation in local areas and even groundwater pollution problems. , causing serious environmental problems. In 2020, the entire North China Plain has seen a downward trend in groundwater levels due to generally low rainfall in recent years. Therefore, it is necessary to strengthen the rational use of groundwater resources, implement scientific and effective supervision, protect the safety of groundwater, and promote the sustainable development and utilization of groundwater resources.

Traditional estimation of groundwater regimes relies on various hydraulic tests and numerical models to estimate the amount of water available in the groundwater system. Modeling is time-consuming and requires a large number of hydrogeological parameters. It is limited by complex geological conditions, the uncertainty of water regime prediction parameters, and Due to factors such as the burden of mathematical modeling, a more accurate quantitative prediction tool for regional groundwater regimes has not yet been formed. It cannot be refined to accurately predict groundwater regimes in various parts of the region. It cannot serve regional water supply management well, especially based on a single model. The early warning parameter model makes arrangements for water outage, water supply, limited water supply volume, and water supply time for the entire area, which will cause inconvenience to residents' lives and even affect enterprise production or project schedules.

Contents of the invention

Therefore, the technical problem to be solved by the present invention is to provide a regional groundwater regime prediction method based on a groundwater monitoring system, which can solve the problem that the traditional estimation and modeling of groundwater regimes is time-consuming and requires a large number of hydrogeological parameters, which is limited by complex geology. Factors such as conditions, uncertainty of water regime prediction parameters, and mathematical modeling burden have not yet formed a more accurate quantitative prediction tool for regional groundwater regimes. It cannot be refined into accurate predictions of groundwater regimes in various parts of the region, and cannot serve well. Issues related to groundwater management in cities, towns and rural areas.

In order to solve the above problems, the present invention provides a regional groundwater regime prediction method based on a groundwater monitoring system. The method includes the following steps:

S1: Based on the observation data in the groundwater observation system, combined with real-time monitoring data, and based on support vector regression SVR, establish the pumping volume prediction function and the pumping volume-groundwater level prediction function of each observation well to predict the groundwater level in real time;

S2: Use the Markov-Monte Carlo method to calculate the pumping-water level release relationship between pumping wells and their corresponding observation wells, and use the observation data monitored in real time by the observation network to simultaneously calculate multiple groundwater regime parameters, based on Bayesian theory The established probability distribution is obtained by iteratively screening samples to obtain particles to approximate the posterior probability distribution. At the same time, the water level prediction results of each observation well and the water pumping volume prediction results of the observation well are calculated to avoid coupling between various prediction parameters;

S3: Based on the coordinates and distribution of the observation wells, define a prediction area covering all observation wells. The prediction area covers the entire monitoring network area. Grid the entire prediction area to generate n×n grid points;

S4: The Thiessen polygon in the structure monitoring network area is combined with the corresponding coordinate set U _e of the Euclidean distance transformation into the observation well coordinate set U _m to form a two-dimensional array U _m+e and then merged with the Thiessen polygon vertex coordinate set U _V , forming a two-dimensional array representing the coordinate point set U _m+e+V that collects observation well coordinates, Euclidean distance transformation coordinates and Thiessen polygon vertex coordinates;

Using the observation well water level prediction results, the prediction area covering the entire monitoring network is simulated according to the Thiessen polygon constructed in the observation area, assigning values to the generated n×n grid points, and setting the Thiessen value of observation well i in the Thiessen polygon. The polygon V _i area grid point is assigned a value, which is the groundwater level prediction result of the observation well i. At the same time, the observation well pumping volume prediction result is assigned to the V _i area grid point to generate a preliminary prediction result of the water pumping volume in the area, and at the same time, the area Preliminary prediction results of groundwater levels;

S5: Construct a convex polygon point set U _m+e+V containing all scattered points through the observation well coordinates, Euclidean distance transformation coordinates and Thiessen polygon vertex coordinates, and divide the convex polygon into mutually non-intersecting triangles. The adjacent triangles are adjusted to obtain a weighted triangulation. Through triangle interpolation calculation, corrections are made based on the preliminary prediction results corresponding to each point in the coordinate point set, and the generated preliminary prediction results of the regional distribution of groundwater level-pumping volume are corrected in real time;

S6: Use the Gaussian smoothing function to smooth the corrected groundwater level-pumping volume distribution area water regime prediction results to obtain the final prediction results. If the gradient double threshold detection is used to further correct the prediction results, the groundwater water regime using the support vector regression algorithm is Statistically analyze the prediction results and the groundwater regime prediction results using the Markov-Monte Carlo method algorithm. Compare and analyze the accuracy of the prediction results based on the statistical distribution of the comparison residuals. Create an array to save the results and determine the optimal prediction model to save. As a stable groundwater level prediction system model.

Optional, S1 specifically includes the following steps:

S11: Establish a training sample set in time series corresponding to the coordinates of different observation wells. Historical observation data representing water pumping volume, corresponding to the coordinates of different observation wells to establish a training sample set in time series/> Represents historical observation data of groundwater levels;

S12: Train the model function in the sample space, under the following constraints y _i ∈{-1,+1}, slack variable ξ _i ≥0, slack variable, η _i ≥0, i=0,1,2...t, i represents the i-th time point; t represents the total number of observation time points ;The loss function is:

Among them, C is a constant >0; ε gives the position of the decision surface; μ represents the normal vector of the decision surface; ξ _i and η _i are slack variables; to meet the optimization objectives, the precipitation prediction model function f ₁ (x ^a , t _i ), where t _i represents time and x ^a represents the amount of water pumped;

Under the following constraints Obtain the groundwater level prediction model function f ₂ (x ^a , x ^b ), where x ^a represents the pumping amount; x ^b represents the groundwater level; that is, real-time prediction of the pumping amount and groundwater level in the future;

S13: Observation data of water pumping volume and historical observation data of groundwater levels/> Perform linear regression to construct various pumping volume-groundwater level linear regression models, calculate the residuals, conduct comparative analysis, compare the prediction results of the linear regression model function with the groundwater level results predicted by support vector regression, and analyze the accuracy of the prediction. , get the optimized model.

Optionally, in step S2, the pumping-water level release relationship between the pumping well and its corresponding observation well is calculated based on the Markov-Monte Carlo method to avoid coupling between various prediction parameters, which specifically includes the following steps:

S21: Establish a probability model for continuous calculation of water regime parameters based on Bayesian theory. Historical water regime observation data can simultaneously calculate the probability distribution of multiple water regime parameters including pumping volume and groundwater level;

Set each observation well and the collected water regime observation data at time t ⁰ and before time t ⁰ as X _t , and the posterior probability P (x ^a , x ^b , t|X _t ) can be expressed as:

Among them, x ^a represents the pumping amount; x ^b represents the groundwater level; t ⁰ prediction starting time; t

is the prediction time;

S22: Take the sample simplified model φ _t (x ^a , x ^b , t|X _t ) to conform to the q distribution. q(·) is the transfer matrix of the prior distribution P (φ _t |φ _t-1 ), i means that it conforms to q The i-th particle of the distribution is denoted as t represents the time, iterated by i=1,...,N loop;

The posterior probability distribution can be approximately summarized as:

Among them, the total number of the first weight coefficient ⅴ _t ⁱ of the i-th particle at time t can be normalized to 1, N is the total number of particles, and δ represents the equal probability;

Based on the updated observation data X _t , participating in the calculation of water regime parameters, all particle weights are re-normalized to:

Among them, L(X _t |φ _t )=L(X _t |x ^a , x ^b , t) is used as the second weight coefficient. Through several weighted particles, the pumping-water level of the pumping well and its corresponding observation well is calculated. The relationship between discharge and drop is used to construct a regional groundwater regime prediction model.

Optional, S3 specifically includes the following steps:

S31: Delimit the prediction area covering all observation wells according to the coordinates and distribution of the observation wells. The prediction area covers the entire monitoring network area. Based on the real-time monitoring data provided by the monitoring network, the prediction area is divided into grids and the observation well coordinate array the establishment of;

S32: Based on the grid divided by the prediction area, create a two-dimensional array of grid point area centers and an n×n area U _n ={(x ₁ , y 1 ) centered on the horizontal and vertical coordinates (x _n , y _n ) _. ,(x ₂ ,y ₂ ),…,(x _n-1 ,y _n-1 ),(x _n ,y _n )}, initialize all points in the array to the identity matrix function, where l and h are x respectively The spacing between adjacent grid points in the y direction, x _n =l×n, y _n =h×n;

S33: Construct a two-dimensional array U _m ={(x ₁ , y ₁ ), (x ₂ , y ₂ ),..., (x _m-1 , y _m-1 ), (x _m ,y _m )}, (x _m , y _m ) is the water regime prediction result corresponding to the m-th observation well in U _m .

Optional, S4 specifically includes the following steps:

S41: Structural Thiessen polygons in the monitoring network area: Construct Thiessen polygons in the prediction area based on the coordinates of N observation wells, and use adjacent observation wells as vertical lines to generate the Thiessen polygons, that is, V _N diagrams;

Using the observation well water level prediction results, simulate the prediction area covering the entire monitoring network based on the V _N diagram of the observation area structure, assign values to the generated n×n grid points, and set the Tyson value of observation well i in the Tyson polygon. The grid points in the polygon V _i area are assigned to the groundwater level prediction results of the observation well i, and the prediction results of the water pumping volume of the observation well are assigned to the grid points in the V _i area;

S42: Use D _f (x _i , y _i ) to represent the Euclidean distance transformation calculation of coordinates (x _i , y _i ), and use f (x _j , y _j ) to represent the consumption of coordinates (x _j , y _j ) Function, (x _i ,y _i )∈U _n , (x _j ,y _j )∈U _n , D _f (x _i ,y _i )=min(((x _i ,y _i )-(x _j ,y _j )) ² +f(x _j ,y _j )) traverse n×n area U _n ={(x ₁ ,y ₁ ),(x ₂ ,y ₂ ),…,(x _n-1 ,y _{n- 1} ), (x _n ,y _n )}, the calculated coordinate set U _e corresponding to the Euclidean distance transformation is merged into the observation well coordinate set U _m to form a two-dimensional array U _m+e , and then the Thiessen polygons are merged The vertex coordinate set U _V forms a two-dimensional array representing the set observation well coordinates, the coordinate set U _e coordinates corresponding to the Euclidean distance transformation, and the coordinate point set U _m+e+V of the Thiessen polygon vertex coordinates.

Optional, S5 specifically includes the following steps:

S51: The generated two-dimensional array corresponds to the observation well position coordinates, the coordinate set U _e coordinates corresponding to the Euclidean distance transformation, and the two-dimensional array U _m+e+V of the Thiessen polygon vertex coordinates. Each coordinate point corresponds to U The corresponding water regime prediction results of the observation wells in _m+e+V . Each coordinate point in U _m+e+V corresponds to a time series of groundwater water regime prediction results. The points in U _m+e+V are used as The endpoints form closed line segments of the triangulation network;

S52: The points in U _m+e+V serve as endpoints to form a convex polygon of the triangulation network. The point set enveloped in the grid point U _n is U _k ={(x ₁ , y ₁ ), (x ₂ , y ₂ ),…,(x _k-1 ,y _k-1 ),(x _k ,y _k )}, the weight of any two points is ω _ij , and the convex polygon is divided into several non-intersecting triangles, Generate weighted triangulation triangulation network;

S53: Construct a triangulation containing all the coordinate point areas of U _m+e+V , recursively obtain the triangulation of the triangular network, perform triangle interpolation calculation on the triangular network, and calculate the corresponding coordinate point set U _k in the triangular network. The time series pumping volume prediction results and groundwater level prediction results are further revised, and the generated preliminary prediction results are optimized and updated.

Optional, S6 specifically includes the following steps:

S61: Grid points in the area U _n ={(x ₁ ,y ₁ ),(x ₂ ,y ₂ ),…,(x _n-1 ,y _n-1 ),(x _n ,y _n )} Each center (x _i , y _i ) constructs a Gaussian kernel function as follows:

Among them, σ is the standard deviation of the Gaussian distribution in the Gaussian kernel generator, which represents the degree of discreteness of the prediction result data;

S62: Perform Gaussian smoothing on the corrected pumping volume prediction results and groundwater level prediction results, use the sobel function to calculate the horizontal and vertical gradients of the prediction results, use the canny function to set double thresholds for edge detection, and use weak gradient values for double threshold detection , the prediction results of the entire region are further revised to obtain the final prediction results;

S63: Based on the measured data, the regional groundwater level distribution simulation results are generated. The generated prediction results are smoothed using the Gaussian smoothing function, and the final prediction results are obtained after further correction. The groundwater water regime prediction results using the regression algorithm of SVR are compared with the prediction results based on Marl The groundwater regime prediction results calculated by the Kov-Monte Carlo method are simulated for regional distribution;

S64: Compare the results predicted by different algorithms, statistically analyze the accuracy and calculation efficiency of the prediction, and add information on temperature, rainfall or/and snowfall, and surface subsidence during the monitoring period to analyze the impact of different conditions on the final prediction result. According to the accuracy and computational efficiency analysis to preserve a robust regional groundwater regime prediction model.

beneficial effects

The regional groundwater regime prediction method based on the groundwater monitoring system provided in the embodiment of the present invention can realize the prediction of the overall groundwater regime in the entire monitoring area, and is not affected by the location of the observation wells and the distribution of the observation wells, regardless of the observation Whether the data is sufficient, prediction can be made as long as the observation data is obtained from a limited number of observation wells and pumping wells (groups) discretely distributed within the monitoring area, and multiple different prediction methods can be carried out simultaneously using the existing observation data The comparative analysis of accuracy and calculation efficiency is refined to the distribution of groundwater levels in each part of the prediction area, reflecting the groundwater conditions in each part of the prediction area in real time. In addition, the present invention can predict groundwater conditions by constructing a pumping amount prediction function to establish a pumping amount-groundwater level release relationship function, and can also reduce the coupling between pumping amount and groundwater level, based on the Markov-Monte Carlo method. Establish a prediction function for the relationship between pumping volume and groundwater level release. Based on the accuracy analysis of actual case prediction results, a robust regional groundwater regime prediction model is saved. It solves the problem that the traditional estimation and modeling of groundwater regime is time-consuming and requires a large number of hydrogeological parameters. Due to factors such as complex geological conditions, uncertainty of water regime prediction parameters and mathematical modeling burden, a more accurate regional groundwater has not yet been formed. Quantitative water regime prediction tools cannot accurately predict groundwater regimes in various parts of the region, and cannot serve regional water supply management well, especially when the entire region is stopped, supplied, or restricted based on a single early warning parameter model. Determining the quantity and time of water supply will cause inconvenience to residents’ lives and may even affect enterprise production or project schedules.

Description of the drawings

Figure 1 is a flow chart of regional groundwater regime prediction based on a groundwater monitoring system according to an embodiment of the present invention;

Figure 2 is a schematic diagram of the regional simulation distribution of groundwater level observation data according to the embodiment of the present invention;

Detailed ways

According to an embodiment of the present invention, a regional groundwater regime prediction method based on a groundwater monitoring system, please refer to Figure 1, includes the following steps:

S1: Collect groundwater level observation information through water level observation equipment of each observation well (group) according to time series, and water pumping observation information near the observation well. In addition, it is also necessary to collect temperature and rainfall (snowfall) information, and based on each observation The coordinates of the wells are used to collect and organize the extracted observation data.

Using existing observation data, the pumping amount prediction function and the pumping amount-groundwater level prediction function are constructed through the Support Vector Regression (SVR) model to predict future groundwater levels.

S1 specifically includes the following steps:

S11: Establish a training sample set in time series corresponding to the coordinates of different observation wells. Historical observation data representing water pumping volume, corresponding to the coordinates of different observation wells to establish a training sample set in time series/> Represents historical observations of groundwater levels.

S12: Train the model function in the sample space, set a width ε to give the position of the decision surface, μ is the normal vector of the decision surface, which determines the direction of the decision surface, introduce slack variables ξ _i and η _i , under the following constraints y _i ∈{-1,+1} slack variable ξ _i ≥ 0, slack variable η _i ≥ 0, i=0,1,2...t, i represents the i-th time point, t represents the total number of observation time points, loss The function is as follows:

Among them, C is a constant > 0, ε gives the position of the decision surface, μ represents the normal vector of the decision surface, ξ _i and η _i are slack variables, which meet the optimization objectives and obtain the precipitation prediction model function f ₁ (x ^a , t _i ), where t _i represents time, x ^a represents the pumping amount, also under the following constraints The groundwater level prediction model function f ₂ (x ^a , x ^b ) is obtained, x ^a represents the pumping amount, x ^b represents the groundwater level, and the pumping amount and groundwater level in the future can be predicted in real time.

S13: Observation data of water pumping volume Historical observation data of groundwater levels/> Perform linear regression, construct various pumping volume-groundwater level linear regression models (such as least squares, polynomial regression, logarithmic linear regression), and calculate the residuals for comparative analysis. Compare the prediction results of the linear regression model function with the groundwater level results predicted by SVR in S12 to analyze the accuracy of the prediction.

S2: Based on the Markov-Monte Carlo method to estimate the pumping-water level release relationship between pumping wells (groups) and their corresponding observation wells, build a regional groundwater regime prediction model.

S2 specifically includes the following steps:

S21: Establish a probability model for continuous calculation of water regime parameters based on Bayesian method. Historical water regime observation data can simultaneously estimate the probability distribution of multiple water regime parameters including water pumping volume and groundwater level. Let the water regime observation data collected at and near each observation well at time t ⁰ and before time ^t0 be X _t , and the posterior probability P (x ^a , x ^b , t|X _t ) can be expressed as follows:

Among them, x ^a represents the pumping amount, x ^b represents the groundwater level, t ⁰ prediction starting time, and t is the prediction time.

S22: With a large number of weighted particles to approximate the posterior distribution, the estimation result can be calculated through iteration:

Taking a sample to simplify the model φ _t (x ^a , x ^b , t|X _t ) conforms to the q distribution, q (·) is usually simplified and defined as the transfer matrix of the prior distribution P (φ _t |φ _t-1 ), i represents The i-th particle that conforms to the q distribution is denoted as t represents the time, and the loop iteration starts from i=1,...,N. The posterior probability distribution can be approximately summarized as follows:

Among them, the weight coefficient V _t ⁱ of the i-th particle at time t, its total number can be normalized to 1, N is the total number of particles, and δ represents the equal probability.

As the updated observation data X _t at each step participate in the estimation of water regime parameters, all particle weights are re-normalized to

_{Among them} ^{, L} ₍ -The relationship between water level release and drop, and the construction of a regional groundwater regime prediction model.

S3: Delimit the prediction area covering all observation wells (groups) based on the coordinates and distribution of the observation wells. The prediction area covers the entire monitoring network area. Grid the area to provide subsequent Thiessen polygons combined with Euclidean The preliminary prediction of distance transformation, interpolation calculation and simulation smoothing of this area provide initial conditions, divide it into n×n grids, set the horizontal and vertical spacing l and h of the grid, and uniformly initialize it with the unit matrix function.

S3 specifically includes the following steps:

S31: Delimit the prediction area covering all observation wells (groups) based on the coordinates and distribution of the observation wells. The prediction area covers the entire monitoring network area. Make full use of the real-time monitoring data provided by the monitoring network as much as possible to grid the prediction area. Division and establishment of observation well coordinate array;

S32: Based on the grid divided by the prediction area, create a two-dimensional array of grid point area centers and an n×n area U _n ={(x ₁ , y 1 ) centered on the horizontal and vertical coordinates (x _n , y _n ) _. ,(x ₂ ,y ₂ ),…,(x _n-1 ,y _n-1 ),(x _n ,y _n )}, initialize all points in the array to the identity matrix function, where l and h are x respectively The spacing between adjacent grid points in the y direction, x _n =l×n, y _n =h×n;

S33: Construct a two-dimensional array U _m ={(x ₁ , y ₁ ), (x ₂ , y ₂ ),..., (x _m-1 , y _m-1 ), (x _m ,y _m )}, (x _m , y _m ) is the water regime prediction result corresponding to the m-th observation well in U _m ;

After this step, the division of the prediction area grid and the establishment of the array of observation well water regime predictions are completed, which provides conditions for subsequent preliminary predictions, correction calculations and smoothing calculations, ensuring that even if the monitoring network is sparsely distributed and lacks sufficient observations, Without data, precipitation and groundwater levels can still be predicted at any location within the prediction area.

S4: Simulate the preliminary prediction results of the prediction area: Construct the Voronoi polygon with the vertical line of adjacent observation wells, N observation wells, that is, the V _N diagram, divide it into n Thiessen polygons with the observation well coordinates as the center, and add one observation well, generate N+1 Thiessen polygons, that is, the V _N+1 graph, and combine the Euclidean distance transformation corresponding coordinate set U _e into the observation well coordinate set U _m to form a new two-dimensional array U _{m+ e} then merges the Thiessen polygon vertex coordinate set U _V to form a two-dimensional array representing the set of observation well coordinates, Euclidean distance transformed coordinates and the coordinate point set U _m+e+V of the Thiessen polygon vertex coordinates, thus the entire Simulate the prediction area to obtain preliminary prediction results for the entire area;

Simulate the entire prediction area to obtain preliminary prediction results of groundwater regime distribution in the entire area.

S4 specifically includes the following steps:

S41: Structural Thiessen polygons in the monitoring network area: Construct Thiessen polygons in the prediction area based on the coordinates of N observation wells, and use adjacent observation wells as vertical lines to generate the Thiessen polygons, that is, V _N diagrams. If a new observation well is added, a new V _N diagram will be generated;

Using the observation well water level prediction results, simulate the prediction area covering the entire monitoring network according to the V _N diagram of the observation area structure, assign values to the n×n grid points generated in S3, and make the Thai value of the observation well i in the Tyson polygon The grid points in the Sen polygon V _i area are assigned to the groundwater level prediction results of the observation well i, and the prediction results of the pumping volume near the observation well are assigned to the grid points in the V _i area.

S42: Use D _f (x _i , y _i ) to represent the Euclidean distance transformation calculation of coordinates (x _i , y _i ), and use f (x _j , y _j ) to represent the consumption of coordinates (x _j , y _j ) Function, (x _i ,y _i )∈U _n , (x _j ,y _j )∈U _n , D _f (x _i ,y _i )=min(((x _i ,y _i )-(x _j ,y _j )) ² +f(x _j ,y _j )) traverse n×n area U _n ={(x ₁ ,y ₁ ),(x ₂ ,y ₂ ),…,(x _n-1 ,y _{n- 1} ), (x _n ,y _n )}, the coordinate set U _e corresponding to the Euclidean distance transformation is calculated and merged into the observation well coordinate set U _m to form a new two-dimensional array U _m+e , and then merged with Tyson The polygon vertex coordinate set U _V forms a two-dimensional array representing the coordinate point set U _m+e+V that collects observation well coordinates, Euclidean distance transformed coordinates and Thiessen polygon vertex coordinates.

The groundwater water regime prediction results in steps S1 and S2 are assigned according to different Thiessen polygons constructed in the observation area. The Thiessen polygon with the observation well coordinates as the center is assigned the observed water level or predicted water level of the observation well. Combined with the coordinate set U _e corresponding to the Euclidean distance transformation, a new two-dimensional array U _m+e is formed, and then the Thiessen polygon vertex coordinate set U _V is combined to form a two-dimensional array representing the set of observation well coordinates and the Euclidean distance transformation corresponding coordinates and the coordinate point set U _m+e+V of the Thiessen polygon vertex coordinates, so as to simulate the entire prediction area and obtain the preliminary prediction results of the entire area;

S5: Construct a weighted triangulation containing all coordinate points U _m+e+V through the observation well coordinates, Euclidean distance transformation corresponding coordinates and Thiessen polygon vertex coordinates, adjust the adjacent triangles, and use the maximum convex hull The weighted triangulation of the point set corrects the preliminary prediction results in real time. Two triangles with common sides are formed from the observation wells to form a polygon, which surrounds all coordinate points. The maximum convex polygon point set is calculated based on the prediction results of each observation well. The corresponding water regime prediction results are corrected point by point, and the preliminary prediction results generated in step S4 are corrected in real time;

S5 specifically includes the following steps:

S51: Convert the two-dimensional array generated in step S4 to a two-dimensional array U _m+e+V corresponding to the observation well position coordinates, Euclidean distance transformation corresponding coordinates and Thiessen polygon vertex coordinates, and each coordinate point corresponds to U _m The corresponding water regime prediction results of the observation wells in step S1 and step S2 in +e+V. Each coordinate point in U _{m +e+V} corresponds to a time series of groundwater water regime prediction results. U _m+e The points in _+V serve as endpoints to form closed line segments of the triangulation network;

S52: The points in U _m+e+V serve as endpoints to form a convex polygon of the triangulation network. The point set enveloped in the grid point U _n is U _k ={(x ₁ , y ₁ ), (x ₂ , y ₂ ),…,(x _k-1 ,y _k-1 ),(x _k ,y _k )}, the weight of any two point shapes is ω _ij , and the convex polygon is divided into several non-intersecting triangles , generate a weighted triangulation triangulation network.

S53: Construct a triangulation of the area containing all coordinate points of U _m+e+V , recursively obtain the weight function value corresponding to the triangular network, perform triangle interpolation calculation on the triangular network, and calculate each coordinate point set U _k in the triangular network The corresponding time series pumping volume prediction results and groundwater level prediction results are further revised, thereby optimizing and updating the preliminary prediction results generated in step S4.

S6: Use the Gaussian smoothing function to smooth the prediction results of step S5 to obtain the final prediction result; use different regression algorithms for the groundwater water regime prediction results of step S1 and the prediction results based on the Markov-Monte Carlo method in step S2. The groundwater regime prediction results were statistically analyzed to analyze the accuracy of the prediction results.

S61: Grid points in the area U _n ={(x ₁ ,y ₁ ),(x ₂ ,y ₂ ),…,(x _n-1 ,y _n-1 ),(x _n ,y _n )} Each center (x _i , y _i ) constructs a Gaussian kernel function σ is the standard deviation of the Gaussian distribution in the Gaussian kernel generator, which represents the degree of discreteness of the prediction result data;

S62: Perform Gaussian smoothing on the corrected pumping volume prediction results and groundwater level prediction results in step S5, use the sobel function to calculate the horizontal and vertical gradients of the prediction results, use the canny function to set double thresholds for edge detection, and use weak gradient values to perform edge detection. Double threshold detection further corrects the prediction results of the entire area to obtain the final prediction results.

S63: Based on the measured data, use the regional groundwater level distribution simulation results generated in steps S4 to S6, use the Gaussian smoothing function to smooth the prediction results generated in step S5, and obtain the final prediction result after further correction. Use support vector regression for step S1, etc. The groundwater water regime prediction results of various regression algorithms and the groundwater water regime prediction results based on the Markov-Monte Carlo method in step S2 are simulated using steps S4 to S6 to simulate regional distribution.

S64: Different groundwater regime prediction algorithms can be loaded in steps S1 and S2, and the results predicted by different algorithms can be compared, the accuracy and calculation efficiency of predictions can be statistically analyzed, and the temperature and rainfall during the monitoring period and/or Snowfall, surface subsidence and other information, analyze the impact of these detailed conditions on the final prediction results, and save a robust regional groundwater regime prediction model based on accuracy and calculation efficiency analysis.

As shown in Figure 2, using the groundwater level observation data of a certain city construction area and a certain day through steps S4 to S6 to realize the numerical simulation process of the groundwater level distribution in the area, the groundwater level distribution within the area of 150m×40m is obtained, represented by dots The location of the observation well. The color from light to dark indicates the water level of the observation well from low to high (-5~25m). After steps S4 and S5, the groundwater level distribution after triangulation interpolation is obtained, and finally the area smoothed by step S6 is obtained. Groundwater level distribution.

This invention is different from the traditional numerical prediction model or machine learning method of groundwater water regime in the past, which uses discretely distributed single observation wells for prediction, and estimates the groundwater water regime in the entire region based on empirical formulas. When certain areas lack observation data, this estimation There will be a large error in the water regime distribution of the entire region, making the prediction of groundwater regime inaccurate, especially when the entire region is deployed for water outage, water supply, limited water supply volume, and water supply time based on a single early warning parameter model. , will cause inconvenience to residents’ lives, and even affect corporate production or project schedules. The present invention can realize the prediction of the overall groundwater conditions in the entire monitoring area, and is not affected by the location of the observation wells and the distribution of the observation wells. Regardless of whether the observation data is sufficient, as long as there are a limited number of observation wells discretely distributed within the monitoring area and Predictions can be made by obtaining observation data from pumping wells (groups), and the existing observation data can be used to conduct comparative analysis of accuracy and calculation efficiency of multiple different prediction methods at the same time, down to the distribution of groundwater levels in various parts of the prediction area. , reflecting the groundwater conditions in various parts of the predicted area in real time. In addition, the present invention can predict groundwater conditions by constructing a pumping amount prediction function to establish a pumping amount-groundwater level release relationship function, and can also reduce the coupling between pumping amount and groundwater level, based on the Markov-Monte Carlo method. Establish a prediction function for the relationship between pumping volume and groundwater level release. Based on the accuracy analysis of actual case prediction results, a robust regional groundwater regime prediction model is saved.

Those skilled in the art can easily understand that, provided there is no conflict, the above-mentioned advantageous methods can be freely combined and superimposed.

Claims (7)

1. The regional groundwater condition prediction method based on the groundwater monitoring system is characterized by comprising the following steps:

s1: establishing a water extraction prediction function and a water extraction-groundwater level prediction function of each observation well based on support vector regression by combining observation data in a groundwater observation system with real-time monitoring data, and predicting the groundwater level in real time;

s2: calculating the pumping-water level drainage relation between a pumping well and a corresponding observation well by using a Markov-Monte Carlo method, simultaneously calculating a plurality of groundwater water condition parameters by using observation data monitored in real time by using the observation network, obtaining particles to approximate posterior probability distribution by iterative screening on samples based on probability distribution established by Bayesian theory, and simultaneously calculating a water level prediction result of each observation well and a water pumping prediction result of each observation well so as to avoid coupling among each prediction parameter;

s3: according to the coordinates and distribution conditions of the observation wells, defining a prediction area covering all the observation wells, covering the whole monitoring network area by the prediction area, and performing grid division on the whole prediction area to generate n multiplied by n grid points;

s4: constructing Thiessen polygons of a monitoring network area, and combining Euclidean distance transformation to correspond to a coordinate set U _e Is incorporated into the observation well coordinate set U _m Wherein, two-dimensional array U is formed _m+e Reconsolidating Thiessen polygon vertex coordinate set U _V Coordinate point set U for forming two-dimensional array representation set observation well coordinates, euclidean distance transformation coordinates and Thiessen polygon vertex coordinates _m+e+V ；

Simulating a prediction area covering the whole monitoring network according to a Thiessen polygon constructed by an observation area by utilizing the observation well water level prediction result, assigning values to the generated n multiplied by n grid points, and setting a Thiessen polygon V of an observation well i in the Thiessen polygon _i Assigning values to regional grid points to obtain the ground water level prediction result of the observation well i, and simultaneouslyAssigning the prediction result of the water pumping quantity of the observation well to V _i Generating a preliminary prediction result of the water pumping quantity of the area and a preliminary prediction result of the groundwater level of the area at the grid points of the area;

s5: constructing a convex polygon point set U containing all scattered points by observing well coordinates, euclidean distance transformation intersection point coordinates and Thiessen polygon vertex coordinates _m+e+V Dividing the convex polygon into mutually disjoint triangles, adjusting adjacent triangles to obtain weighted triangulation, calculating through triangle interpolation values, correcting according to the preliminary prediction results corresponding to each point in the coordinate point set, and correcting the generated preliminary prediction results of groundwater level-water pumping area distribution in real time;

s6: smoothing the corrected groundwater level-water pumping distribution area water regime prediction result by using a Gaussian smoothing function to obtain a final prediction result, further correcting the prediction result by using a gradient double-threshold detection, carrying out statistical analysis on the groundwater regime prediction result by using a support vector regression algorithm and the groundwater regime prediction result by using a Markov-Monte Carlo algorithm, creating an array of storage results according to the accuracy of comparing the analysis prediction results with the residual error statistical distribution, and storing the optimized prediction model as a stable groundwater level prediction system model.

2. The regional groundwater condition prediction method based on a groundwater monitoring system according to claim 1, wherein S1 specifically comprises the following steps:

s11: establishing a training sample set according to time sequence corresponding to different observation well coordinatesHistorical observation data representing the pumping capacity, and establishing a training sample set according to time series corresponding to different observation well coordinates>Historical observation data representing groundwater level;

s12: training a model function in sample space under the following constraintsy _i Epsilon { -1, +1}, relaxation variable ζ _i Not less than 0, relaxation variable eta _i Gtoreq 0, i=0, 1,2 … t, i representing the i-th time point; t represents the total number of observation time points; the loss function is:

wherein C is>A constant of 0; epsilon gives the position of the decision surface; μ represents a normal vector of the decision plane; zeta type toy _i And eta _i Is a relaxation variable; meets the optimization target to obtain the precipitation prediction model function f ₁ (x ^a ，t _i ) Wherein t is _i Time x is represented ^a Representing the water extraction;

under the following constraint conditionsObtaining a groundwater level prediction model function f ₂ (x ^a ，x ^b ) Wherein x is ^a Representing the water extraction; x is x ^b Representing groundwater level; the water pumping quantity and the groundwater level at the future moment are predicted in real time;

s13: data on water extraction observationsAnd historical observations of groundwater level +.>Performing linear regression, constructing various water pumping quantity-groundwater level linear regression models, calculating to obtain residual errors, performing comparative analysis, comparing the prediction result of the linear regression model function with the groundwater level result of support vector regression prediction, analyzing the prediction precision, and optimizing the prediction model.

3. The regional groundwater condition prediction method based on a groundwater monitoring system according to claim 1, wherein in step S2, a pumping-water level drainage relationship between a pumping well and its corresponding observation well is calculated based on a markov-monte carlo method, so as to avoid coupling between each prediction parameter, and specifically comprising the following steps:

s21: establishing a probability model for continuously calculating water regime parameters based on a Bayesian theory, wherein the water regime observation historical data can calculate probability distribution of a plurality of water regime parameters including water pumping quantity and groundwater level at the same time;

setting t ⁰ Time t ⁰ Each observation well and the collected water condition observation data before the moment are X _t Posterior probability P (x ^a ,x ^b ,t|X _t ) Can be expressed as:

wherein x is ^a Representing the water extraction; x is x ^b Representing groundwater level; t is t ⁰ Predicting a starting time; t is the predicted time;

s22: sample taking simplified model phi _t (x ^a ，x ^b ，t|X _t ) Q (·) is a priori distributed P (phi) in conformity with the q distribution _t |φ _t-1 ) I represents the ith particle conforming to the q distribution, denoted ast represents the moment, iterated by i=1, …, N loops;

the posterior probability distribution can be approximated as:

wherein the first weight coefficient of the ith particle at time tThe total number can be normalized to 1,/or->N is the total number of particles; delta represents an equal proportion probability;

based on updated observed data X _t In the calculation of the water regime parameters, all particle weights are re-normalized to:

wherein L (X) _t |φ _t )＝L(X _t |x ^a ，x ^b T) as a second weight coefficient, calculating the water pumping-water level drainage relation of the water pumping well and the observation well corresponding to the water pumping well through a plurality of particles with weights, and constructing a regional groundwater condition prediction model.

4. The regional groundwater condition prediction method based on a groundwater monitoring system according to claim 1, wherein S3 specifically comprises the following steps:

s31: defining a prediction area covering all observation wells according to the coordinates and distribution conditions of the observation wells, wherein the prediction area covers the whole monitoring network area, and carrying out grid division on the prediction area and establishment of an observation well coordinate array based on real-time monitoring data provided by the monitoring network;

s32: based on the grid divided by the prediction area, a two-dimensional array of the grid point area center is created with the abscissa (x _n ，y _n ) An n×n region U as the center _n ＝{(x ₁ ,y ₁ ),(x ₂ ,y ₂ ),…,(x _n-1 ,y _n-1 ),(x _n ,y _n ) Initializing all points in the array as an identity matrix function, wherein l and h are the distances between adjacent grid points in the x direction and the y direction respectively, and x _n ＝l×n，y _n ＝h×n；

S33: constructing a two-dimensional array U corresponding to the position coordinates of the observation well _m ＝{(x ₁ ,y ₁ ),(x ₂ ,y ₂ ),…,(x _m-1 ,y _m-1 ),(x _m ,y _m )}，(x _m ，y _m ) Is U (U) _m The water regime prediction result corresponding to the m observation well.

5. The regional groundwater condition prediction method based on a groundwater monitoring system according to claim 1, wherein S4 specifically includes the following steps:

s41: construction of Thiessen polygons for monitoring network areas: constructing Thiessen polygons for the predicted area according to the N observation well coordinates, and generating Thiessen polygons by taking the neighboring observation wells as perpendicular bisectors, namely V _N A figure;

v constructed according to the observation area by using the observation well water level prediction result _N The graph simulates a prediction area covering the whole monitoring network, assigns values to the generated n multiplied by n grid points, and sets a Thiessen polygon V of an observation well i in the Thiessen polygons _i Assigning the regional grid points to obtain the ground water level prediction result of the observation well i, and assigning the water pumping prediction result of the observation well to V at the same time _i Regional grid points;

s42: in D _f (x _i ，y _i ) Representing coordinates (x) _i ，y _i ) Is calculated as the Euclidean distance transform of f (x _j ，y _j ) Representing coordinates (x) _j ，y _j ) The consumption function of (x) _i ,y _i )∈U _n ，(x _j ,y _j )∈U _n ，D _f (x _i ,y _i )＝min(((x _i ,y _i )-(x _j ,y _j )) ² +f(x _j ,y _j ) Traversing n x n region U) _n ＝{(x ₁ ,y ₁ ),(x ₂ ,y ₂ ),…,(x _n-1 ,y _n-1 ),(x _n ,y _n ) Calculating to obtain a coordinate set U corresponding to Euclidean distance transformation _e Is incorporated into the observation well coordinate set U _m Wherein, two-dimensional array U is formed _m+e Combining Thiessen polygon vertex coordinate set U _V Coordinate point set U for forming two-dimensional array representation set observation well coordinates, euclidean distance transformation coordinates and Thiessen polygon vertex coordinates _m+e+V 。

6. The regional groundwater condition prediction method based on a groundwater monitoring system according to claim 1, wherein S5 specifically includes the following steps:

s51: the generated two-dimensional array U corresponds to the two-dimensional array U of the position coordinates of the observation well, the Euclidean distance transformation coordinates and the Thiessen polygon vertex coordinates _m+e+V Each coordinate point corresponds to U _m+e+V Corresponding water condition prediction result of observation well, U _m+e+V Each coordinate point corresponds to a time sequence groundwater condition prediction result, which is formed by U _m+e+V The points in (a) are used as end points to form a closed line segment of the triangulation network;

S52：U _m+e+V the points in (a) are used as end points to form a triangulation network convex polygon, and the points in (U) are used as grid points _n The point set of the medium envelope is U _k ＝{(x ₁ ,y ₁ ),(x ₂ ,y ₂ ),…,(x _k-1 ,y _k-1 ),(x _k ,y _k ) Any two points of which weight is omega _ij Dividing the convex polygon into a plurality of mutually disjoint triangles to generate a triangle net with weight;

s53: the structure comprises U _m+e+V Triangle subdivision of all coordinate point areas is obtained by recursion, triangle subdivision of a triangular net is obtained, triangle interpolation calculation is carried out on the triangular net, and each coordinate point set U in the triangular net is obtained _k And further correcting the corresponding time sequence water extraction prediction result and the groundwater level prediction result, and optimizing and updating the generated preliminary prediction result.

7. The regional groundwater condition prediction method based on a groundwater monitoring system according to claim 1, wherein S6 specifically includes the following steps:

S61：U _n ＝{(x ₁ ,y ₁ ),(x ₂ ,y ₂ ),…,(x _n-1 ,y _n-1 ),(x _n ,y _n ) Each center of grid points in the } region (x _i ,y _i ) The gaussian kernel is constructed as follows:

wherein sigma is the standard deviation of Gaussian distribution in a Gaussian kernel generator, and represents the discrete degree of prediction result data;

s62: carrying out Gaussian smoothing on the corrected water extraction prediction result and the groundwater level prediction result, calculating transverse gradients and longitudinal gradients of the prediction result by adopting a sobel function, carrying out edge detection by adopting a canny function to set double thresholds, carrying out double threshold detection by utilizing a weak gradient value, and further correcting the prediction result of the whole area to obtain a final prediction result;

s63: smoothing the generated prediction result by adopting a Gaussian smoothing function based on the actually measured data, further correcting to obtain a final prediction result, and simulating regional distribution of the groundwater condition prediction result by adopting a regression algorithm of SVR and the groundwater condition prediction result calculated based on a Markov-Monte Carlo method;

s64: based on the comparison of the predicted results of different algorithms, the precision and the calculation efficiency of prediction are statistically analyzed, the temperature and rainfall or/and snowfall and surface subsidence information during monitoring are added, the influence of different conditions on the final predicted result is analyzed, and a regional groundwater water regime prediction model with robustness is stored according to the precision and the calculation efficiency analysis.