Archaeological excavation area underground water dynamic monitoring and ground surface settlement early warning method based on deep learning

Abstract

The invention relates to an archaeological excavation area underground water dynamic monitoring and ground surface settlement early warning method based on deep learning, and the method comprises the steps: obtaining to-be-detected field survey data and to-be-detected geological survey data in an archaeological excavation area, and inputting the to-be-detected field survey data and to-be-detected geological survey data into a rock-soil layer classification model and a specific yield classification model respectively, obtaining a rock-soil layer category result and a specific yield classification result, and inputting geological survey data to be measured into the boundary model to obtain a boundary result; the method comprises the following steps: covering an archaeological excavation area based on coordinates of drilling points in survey data of a to-be-measured site in combination with boundary conditions to perform grid division, and constructing a Thiessen polygon by taking adjacent drilling points as perpendicular bisectors to perform filling, thereby performing assignment on initial simulation conditions of an underground water seepage field of the archaeological excavation area; initial prediction conditions of the underground water seepage field at the future moment are obtained, an underground water seepage field prediction model is further obtained, the rock-soil layer category result, the specific yield classification result and the boundary result are input into the underground water seepage field prediction model, and an underground water seepage field prediction result is obtained.

Description

Technical Field

[0001] The present invention relates to the field of computer machine learning technology, and in particular to a method for dynamic monitoring of groundwater and early warning of surface subsidence in archaeological excavation areas based on deep learning. Background Art

[0002] Archaeological excavation areas in my country's vast monsoon climate zone are constantly threatened by rainwater, groundwater and even air moisture to varying degrees. Groundwater erodes the archaeological excavation area to a great extent, which can easily make the protection and survey of ancient sites difficult, increase the difficulty of in-situ protection of archaeological excavation areas, and even affect the water supply and water use of residential areas and villages around the archaeological excavation area, causing inconvenience to residents living near the archaeological excavation area. Therefore, it is necessary to provide real-time information such as groundwater seepage field and surface settlement distribution during the drainage process in the core area of ​​the site as a basis for evaluating the site's hydrogeological conditions and conducting surveys, avoiding environmental geological pollution, and facilitating the reasonable resettlement of nearby related demolitions, and strengthening the construction goal of integrating ancient heritage protection with urban and rural construction.

[0003] The prediction of groundwater seepage field in the process of drainage in traditional archaeological excavation areas is to estimate the groundwater seepage field based on various seepage tests and gradient field parameters. Regression modeling is often limited by complex geological conditions and a large number of uncertain hydrogeological parameters. It is impossible to update the prediction model in real time as the amount of data increases, and the increase in data volume easily causes a computational burden. In addition, information such as borehole water level and permeability coefficient collected in real time is usually used for simulation research of groundwater seepage field in archaeological excavation areas. Although the simulation results can be used as a basis for the design of drainage in archaeological excavation areas, a more accurate real-time prediction and update model for quantitative regional seepage field has not yet been formed. It is impossible to refine the accurate prediction of the distribution of groundwater seepage field coupled with surface settlement in each part of the drainage process in the archaeological excavation area, and it is impossible to provide real-time update services for important information in on-site exploration and ancient site protection in the archaeological excavation area. It will cause inconvenience for real-time adjustment and arrangement of drainage in the exploration and ancient site protection process, and may even affect the archaeological excavation in situ protection or project schedule. Summary of the invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to propose a method for dynamic monitoring of groundwater and early warning of surface subsidence in archaeological excavation areas based on deep learning. According to the statistical sampling data of groundwater drilling water level observations and field test data of permeability coefficient in the process of drainage in the archaeological excavation area, the groundwater seepage field and surface settlement distribution are predicted in real time based on a neural network algorithm. The method is suitable for the prediction of groundwater seepage field and surface settlement distribution under various hydrogeological conditions in archaeological excavation areas. It solves the problem that the traditional estimation modeling of groundwater seepage field is time-consuming and cannot be updated in real time with the increase of data volume, and cannot reflect the update of drainage effect in archaeological excavation areas in real time. It can serve the problem of groundwater management in the process of drainage in archaeological excavation areas in real time, and can be used for the prediction and analysis of the development and change trend of groundwater seepage field and drainage effect in archaeological excavation areas.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] The method of groundwater dynamic monitoring and surface subsidence early warning in archaeological excavation areas based on deep learning includes:

[0007] Obtaining on-site survey data and geological survey data to be tested within the archaeological excavation area;

[0008] Input the rock and soil mass information and water supply degree information in the field survey data to be tested into the rock and soil layer classification model and the water supply degree classification model respectively to obtain the rock and soil layer classification result and the water supply degree classification result, and input the geological survey data to be tested into the boundary model to obtain the boundary result;

[0009] Based on the coordinates of the drilling points in the field survey data to be tested, combined with the boundary conditions, the archaeological excavation area is covered for grid division, and Thiessen polygons are constructed with the perpendicular medians of the adjacent drilling points for filling, and the initial simulation conditions of the groundwater seepage field in the archaeological excavation area are assigned to obtain the initial prediction conditions of the groundwater seepage field at the future time;

[0010] The groundwater seepage field prediction model is obtained through the initial prediction conditions of the groundwater seepage field at the future moment, and the rock and soil layer classification results, the water supply degree classification results and the boundary results are input into the groundwater seepage field prediction model to obtain the groundwater seepage field prediction results.

[0011] Optionally, obtaining the rock and soil layer classification result, the water supply degree classification result and the boundary result includes:

[0012] The rock and soil body information in the field survey data to be tested is input into the rock and soil layer classification model to obtain the rock and soil layer classification result; the rock and soil layer classification model is trained using the first training set, and is obtained by reversely adjusting the layer by layer by minimizing the loss function in combination with the gradient descent method; wherein the training set includes: the rock and soil body information and the rock and soil layer permeability coefficient information after data processing;

[0013] Inputting the water supply degree information in the field survey data to be tested into a water supply degree classification model to obtain a water supply degree classification result, wherein the water supply degree classification model is trained using a second training set; wherein the second training set includes: water supply degree information after data processing;

[0014] The geological survey data to be tested is input into the boundary model to obtain the boundary result; the boundary model is trained using the third training set and is obtained by adjusting the loss function in combination with the three-component loss calculation; wherein the third training set includes: geological survey data after data processing.

[0015] Optionally, obtaining the rock and soil mass information after the data processing includes:

[0016] Extracting features of the rock and soil mass information using a directional gradient histogram to obtain rock and soil mass feature information;

[0017] A digital label is established, the rock and soil body characteristic information is matched with the digital label, the digital label with the rock and soil body characteristic information is normalized, and the rock and soil body information after the data processing is obtained.

[0018] Optionally, obtaining the geological survey data after the data processing includes:

[0019] The upper boundary, the lower boundary and the lateral boundary in the geological survey data are divided into a plurality of candidate frames, the boundary features are extracted using the plurality of candidate frames, and the frames are stored in an array. Save vertical boundary information, save the box by array Save lateral boundary information;

[0020] Among them, Z u,d Indicates vertical boundary information, N indicates that there are N vertical boundary information data in total, Represents the Nth upper boundary information, Indicates the Nth lower boundary information, Z s Indicates lateral boundary information, Represents the Mth lateral boundary information data.

[0021] Optionally, obtaining the initial prediction conditions of the groundwater seepage field at a future time includes:

[0022] Constructing a triangulated mesh of all grid points in the archaeological excavation area based on the boundary vertex coordinates and each grid point, and segmenting the triangulated mesh using convex polygons to obtain a plurality of mutually non-intersecting triangles;

[0023] A triangle interpolation calculation is performed on the triangle, and the initial simulation conditions of the assigned groundwater seepage field are corrected by triangulation of the maximum convex hull point set. The seepage field prediction results are smoothed using a cubic spline interpolation function to obtain the initial prediction conditions of the groundwater seepage field at future times.

[0024] Optionally, obtaining the groundwater seepage field prediction model includes:

[0025] The initial prediction condition of the groundwater seepage field at the future moment is used as the initial prediction condition of the groundwater seepage field at the first time node, and the groundwater seepage field at the t-1th time node is used as the initial prediction condition of the groundwater seepage field at the first time node. Predict the groundwater seepage field at the tth time node

[0026]

[0027] Among them, A is the dynamic system parameter matrix, u t-1 represents the increment of groundwater level observation value at the t-th time node, Γ represents the Kalman filter algorithm;

[0028] The groundwater seepage field at the t-th time node is used to perform repeated iterative calculations, and n time nodes that need to be predicted after the current moment are traversed to establish the groundwater seepage field prediction model.

[0029] Optionally, the method of performing iterative calculation using the groundwater seepage field at the t-th time node is:

[0030]

[0031] in, represents the prediction result at time t+1, Γ represents the Kalman filter algorithm, for, for.

[0032] Optionally, the method further comprises:

[0033] The surface settlement survey information to be measured is obtained, and the surface settlement in the surface settlement survey information and the groundwater seepage field prediction result are input into the long short-term memory network model to predict the surface settlement change, so as to obtain the surface settlement change caused by the drainage process of the archaeological excavation area; wherein the long short-term memory network model is trained with the fourth training set and adjusted with the multi-classification logarithmic loss function; wherein the fourth training set includes the surface settlement after data processing.

[0034] Optionally, obtaining the surface settlement amount after the data processing includes:

[0035] The surface settlement amount is sorted according to a time series and standardized to obtain the surface settlement amount after the data processing.

[0036] The beneficial effects of the present invention are:

[0037] The present invention discloses a method for dynamic monitoring of groundwater and early warning of surface subsidence in an archaeological excavation area based on deep learning. The method can realize the prediction of groundwater seepage field and surface subsidence distribution in the process of drainage in the entire archaeological excavation area, and is not affected by the amount of geological survey data and information. Even in the case of a large amount of observation data and relatively complex chart information, as long as the archaeological excavation area has a number of drilling point coordinates and drilling water levels as well as rock and soil layer parameters and permeability coefficient survey data, real-time prediction can be performed. The prediction model automatically extracts features without the need for a large amount of specific measurement data. The neural network model can be trained and updated in real time through the existing observation data, and the accuracy and computational efficiency are analyzed in real time, and the training and updating of the groundwater seepage field prediction model in each part of the prediction area are refined to reflect the groundwater drainage effect in each part of the prediction area in real time. The present invention solves the problem that traditional groundwater seepage field estimation modeling is time-consuming and cannot be updated in real time with the increase of data volume, and cannot reflect the drainage effect of archaeological excavation areas in real time. It can serve the problem of groundwater management during the drainage of archaeological excavation areas in real time, and can be used for predicting and analyzing the development and change trend of groundwater seepage fields in archaeological excavation areas and the drainage effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0039] Figure 1 A flow chart of a method for dynamic monitoring of groundwater in an archaeological excavation area according to an embodiment of the present invention;

[0040] Figure 2 This is a flow chart of the training of a model for dynamic monitoring of groundwater in an archaeological excavation area and real-time prediction of surface settlement distribution according to an embodiment of the present invention;

[0041] Figure 3 This is a flow chart of extracting boundary features of a groundwater dynamic monitoring model according to an embodiment of the present invention;

[0042] Figure 4Schematic diagram of the three-component loss training process of an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] like Figure 1 As shown, this embodiment discloses a method for dynamic monitoring of groundwater and early warning of surface subsidence in an archaeological excavation area based on deep learning, including: obtaining on-site survey data to be tested and geological survey data to be tested within the scope of the archaeological excavation area; inputting the rock and soil body information and water supply degree information in the on-site survey data to be tested into the rock and soil layer classification model and the water supply degree classification model respectively, obtaining the rock and soil layer category results and the water supply degree classification results, and inputting the geological survey data to be tested into the boundary model to obtain the boundary results; based on the coordinates of the drilling points in the on-site survey data to be tested combined with the boundary conditions, the archaeological excavation area is gridded, and Thiessen polygons are constructed with the medians of adjacent drilling points as perpendicular lines for filling, and the initial simulation conditions of the groundwater seepage field in the archaeological excavation area are assigned to obtain the initial prediction conditions of the groundwater seepage field at the future moment; through the initial prediction conditions of the groundwater seepage field at the future moment, a groundwater seepage field prediction model is obtained, and the rock and soil layer category results, the water supply degree classification results and the boundary results are input into the groundwater seepage field prediction model to obtain the groundwater seepage field prediction results.

[0046] Furthermore, obtaining rock and soil layer classification results, water supply degree classification results and boundary results includes: inputting rock and soil body information in the field survey data to be tested into the rock and soil layer classification model to obtain rock and soil layer classification results; the rock and soil layer classification model is trained using the first training set, and is obtained by reversely adjusting the layer by layer by minimizing the loss function in combination with the gradient descent method; wherein the training set includes: rock and soil body information and rock and soil layer permeability coefficient information after data processing; the water supply degree type in the field survey data to be tested is obtained to obtain boundary results; the boundary model is trained using the third training set, and is obtained by adjusting the loss function in combination with the three-component loss calculation; wherein the third training set includes: inputting geological information after data processing into the water supply degree classification model to obtain water supply degree classification results, and the water supply degree classification model is obtained by training with the second training set; wherein the second training set includes: water supply degree information after data processing; the geological survey data to be tested is input into the boundary model survey data.

[0047] Furthermore, obtaining the rock and soil information after data processing includes: using the directional gradient histogram to extract the features of the rock and soil information to obtain the rock and soil characteristic information; establishing digital labels, corresponding the rock and soil characteristic information to the digital labels, normalizing the digital labels with the rock and soil characteristic information, and obtaining the rock and soil information after data processing.

[0048] Further, obtaining the geological survey data after data processing includes: dividing the upper boundary, the lower boundary and the lateral boundary in the geological survey data into a plurality of candidate frames, extracting boundary features using the plurality of candidate frames, and storing the frames in an array. Save vertical boundary information, save the box by array Save lateral boundary information; where Z u,d Indicates vertical boundary information, N indicates that there are N vertical boundary information data in total, Represents the Nth upper boundary information, Indicates the Nth lower boundary information, Z s Indicates lateral boundary information, Represents the Mth lateral boundary information data.

[0049] Specifically, S1: through the field survey data within the scope of the archaeological excavation area, including the coordinates of several drilling points and the drilling water level as well as the rock and soil parameters, permeability coefficient and other information, the feature extraction is completed through the directional gradient histogram calculation, and a set of digital labels with feature data are established, the data features are corresponded to the digital labels, and the digital labels are uniformly normalized; according to the geological structure profile of the current archaeological excavation area, the original groundwater system survey data and the archaeological excavation area topographic map, hydrogeological conditions and stratigraphic structure and other geological survey data, the convolutional neural network algorithm is used to extract the frame features, the seed filling algorithm is used to fill the internal area of ​​the predicted boundary, and the stack is used to realize the polygon filling of the internal area of ​​the boundary, so as to generate the boundary conditions of the groundwater seepage field in the archaeological excavation area and generate the initial prediction conditions of the groundwater seepage field.

[0050] S1 specifically includes the following steps:

[0051] S11: Based on the on-site survey data within the archaeological excavation area, including the coordinates of several drilling points and the drilling water level as well as rock and soil parameters, permeability coefficient, water supply degree and other information, feature extraction is completed through directional gradient histogram calculation, and a set of digital labels with feature data is established. The data features are corresponded to the digital labels, and the digital labels are uniformly normalized.

[0052] S12: Based on the geological structure profile of the current archaeological excavation area, the original groundwater system survey data, the topographic map of the archaeological excavation area, the hydrogeological conditions and the stratigraphic structure and other geological survey data, the upper boundary, lower boundary and lateral boundary in the vertical geological information data and the lateral geological information data are divided into several small frames, and the candidate frames are extracted and directly applied to feature establishment. The position of the frame is marked in an array, and the frame features are extracted by the convolutional neural network algorithm. The process of extracting the position feature is as follows: Figure 3 As shown in the figure, the convolution layer and the sampling layer are alternately set to extract features, the internal area of ​​the predicted boundary is filled by the seed filling algorithm, and the polygon filling of the internal area of ​​the boundary is realized by using a stack.

[0053] S2: Utilize the existing rock and soil layer test and field survey data, establish a classification model through feature data, and use the gradient descent method to minimize the loss function to reversely adjust the weight parameters in the network layer by layer, thereby accelerating convergence and providing rock and soil layer category information for future groundwater seepage field predictions; collect, organize and save the water supply degree information involved in the rock and soil materials, extract the unit water supply degree characteristics based on the water supply degree information, establish unit water supply degree characteristics and digital labels, and complete the nonlinear classification of unit water supply degree through the constructed data set fitted by the support vector machine classification model. The data set is divided into a training set and a test set, and the unit water supply degree results of the training set are predicted. The predicted results are compared with the actual results to facilitate the analysis of the prediction accuracy and build a water supply degree prediction model.

[0054] S2 specifically includes the following steps:

[0055] S21: Using the existing rock and soil layer test and field survey data, the permeability coefficient information is saved in the order of the rock and soil layers to form a permeability coefficient training sample set, and the set K = {K1, K2, ..., K n} represents the permeability information corresponding to n different rock and soil layers;

[0056] For rock and soil materials, the rock layer features such as silt soil, silt soil, silt soil, weathered rock mass, bedrock, etc. are extracted, the feature extraction is completed through the directional gradient histogram calculation, the classification model is established, and the classifier training of each rock and soil layer is completed. By comparing the extracted features with the classification rules contained in the classification model, the category of the rock and soil layer can be obtained;

[0057] Establish a set of digital labels with characteristic data, correspond the characteristics representing the rock and soil layer categories to the digital labels, normalize the digital labels uniformly, and keep the data range consistent with the data characteristics provided by training when testing the model to avoid errors caused by the inability to identify a certain rock and soil layer category.

[0058] The rock and soil layer classification model function is trained in the sample space. The weight parameters in the network are reversely adjusted layer by layer by minimizing the loss function using the gradient descent method, and the deep model optimization algorithm is used to accelerate convergence.

[0059] Δω i =-κη i ,

[0060] ω i+1 =ω i +Δω i =ω i -κη i

[0061] Where к is the learning rate, ω i is the weight value during the i-th training, ω i+1 is the weight value during the i+1th training, η i is the gradient of the i-th training, Δω i is the update amount of the i-th weight, where ω0=0, which is fed back to the network for optimization after cyclic iteration. The accuracy of the network is improved through frequent iterative training.

[0062] S22: Collect, organize and save the water supply degree information involved in the rock and soil materials. The water supply degree information is represented by a set u = {u1, u2, ..., u n} means that the unit water supply data of n different rock formations are saved to construct a unit water supply training sample set;

[0063] The unit water content characteristics of rock and soil materials are extracted, and the feature extraction is completed through directional gradient histogram calculation. A classification model is established based on the obtained feature data to complete the classifier training of the water content of each rock and soil layer. The water content classification model is trained by the extracted features to obtain the category prediction function of the water content.

[0064] A set of unit water supply degree digital labels with feature data is established, the features representing the unit water supply degree categories are matched with the digital labels, the digital labels are uniformly normalized, and the unit water supply degree category features and digital labels are passed to the computer for training.

[0065] S23: Complete the construction of features and labels through S22, train the unit water supply degree classification model in the sample space, fit the constructed data set through the support vector machine classification model, train and separate positive samples and negative samples, complete the nonlinear classification of unit water supply degree, divide the data set into training set and test set, predict the unit water supply degree result of the training set, compare the predicted result with the actual result, and analyze the prediction accuracy.

[0066] S3: Establish an aquifer information set collected and organized through geological data, divide the upper boundary, lower boundary and lateral boundary into several small frames, perform target detection on the groundwater aquifer boundary, fill the internal area of ​​the predicted boundary through the seed filling algorithm, generate the boundary conditions of the groundwater seepage field in the archaeological excavation area, and establish a numerical model for predicting the groundwater seepage field in the archaeological excavation area.

[0067] S3 specifically includes the following steps:

[0068] S31: Establish a collection of aquifer information collected and collated through geological data, including geological structure profiles of archaeological excavation areas, original groundwater system survey data and topographic maps of archaeological excavation areas, hydrogeological conditions and stratigraphic structures, etc. Save vertical boundary information, Z u,d Indicates vertical boundary information, N indicates that there are N vertical boundary information data in total, Represents the Nth upper boundary information, Represents the Nth lower boundary information, Save lateral boundary information, Z s Indicates lateral boundary information, Represents the Mth lateral boundary information data.

[0069] S32: Divide the upper boundary, lower boundary and lateral boundary of the vertical geological information data and the lateral geological information data into a number of small frames, search for bounded frames one by one, extract candidate frames and directly apply them to feature establishment, with array z={z1,z2,…,z M+N+N}Save the data of the position annotation of the box, and extract the box features using the convolutional neural network algorithm. The process of extracting the position features is as follows Figure 3 As shown in the figure, the convolution layer and the sampling layer are alternately set to extract features, and the nonlinear support vector machine algorithm is used as the detection algorithm to classify the boundaries that appear. If it is determined that there is a boundary in a box, the position of this small box is output, and the training is carried out through the detection data set. The loss function is calculated by the three-element loss

[0070] L triplet =max(d(a,p)-d(a,n)+m)

[0071] Where L triplet Represents the three-element loss function, such as Figure 4As shown in the figure, a represents anchor, p represents positive and is a sample of the same category as a, n represents negative and is a sample of a different category from a, m takes a constant greater than 0, d(a,p) represents the Euclidean distance between the anchor and the positive, d(a,n) represents the Euclidean distance between the anchor and the negative, max represents the maximum value function, and inputs a triplet including anchor, positive, and negative. By optimizing the distance between the anchor and the positive to be less than the distance between the anchor and the negative, the similarity calculation between samples is realized, and the test is performed in the test set to establish the boundary prediction model function.

[0072] S33: Fill the internal area of ​​the predicted boundary in S32 through the seed filling algorithm, use the stack to realize the polygon filling of the internal area of ​​the boundary, generate the boundary conditions of the groundwater seepage field in the archaeological excavation area, establish the boundary conditions of the numerical model for predicting the groundwater seepage field in the archaeological excavation area, and generate the initial prediction conditions of the groundwater seepage field.

[0073] Furthermore, obtaining the initial prediction conditions of the groundwater seepage field in the future includes: constructing a triangulated mesh of all grid points in the archaeological excavation area based on the boundary vertex coordinates and each grid point, and dividing the triangulated mesh using convex polygons to obtain a number of non-intersecting triangles; performing triangle interpolation calculations on the triangles, and correcting the initial simulation conditions of the assigned groundwater seepage field through the triangulation of the maximum convex shell point set, and using the cubic spline interpolation function to smooth the seepage field distribution to obtain the initial prediction conditions of the groundwater seepage field in the future.

[0074] Furthermore, obtaining the groundwater seepage field prediction model includes: using the initial prediction condition of the groundwater seepage field at the future moment as the initial prediction condition of the groundwater seepage field at the first time node, and using the groundwater seepage field at the t-1th time node as the initial prediction condition of the groundwater seepage field Predict the groundwater seepage field at the tth time node

[0075]

[0076] Among them, A is the dynamic system parameter matrix, u t-1 represents the increment of the groundwater level observation value at the t-th time node, and Γ represents the Kalman filter algorithm. The groundwater seepage field at the t-th time node is used for repeated iterative calculations, and the n time nodes that need to be predicted after the current moment are traversed to establish a groundwater seepage field prediction model.

[0077] Furthermore, the method of iterative calculation using the groundwater seepage field at the tth time node is:

[0078]

[0079] in, represents the prediction result at time t+1, Γ represents the Kalman filter algorithm, is the prediction result at time t+n, is the groundwater seepage field distribution at time t.

[0080] Specifically, S4: collects observation information of the water level of each borehole corresponding to the coordinates of different borehole points, extracts the characteristics of the borehole water level and the coordinates of the borehole point, and in addition, it is necessary to collect rock and soil layer information and water supply degree information, and extract the characteristics of the rock and soil layer according to the category of each rock and soil layer, collects and organizes the observation data in time series, establishes a numerical model for simulating the groundwater seepage field in the archaeological excavation area, and generates the initial prediction conditions for the groundwater seepage field.

[0081] S4 specifically includes the following steps:

[0082] S41: Establish a training sample set according to the measured water level corresponding to the coordinates of different drilling points Represents the historical observation data of the borehole water level corresponding to the 1st to tth time nodes, where k represents the borehole point number, i represents the i-th time node, and t represents the total number of time nodes in the time series. According to the boundary conditions, coordinates and distribution of the borehole points assigned by the seed filling color algorithm in step S3, the prediction area covering the groundwater range of the entire archaeological excavation area is delineated. The prediction area covers the entire filling area. All the information and data provided by the field survey and test are used as much as possible to grid the prediction area and establish the observation well coordinate array;

[0083] S42: Based on the grids divided into the prediction area, Thiessen polygons are constructed with adjacent drilling points as perpendicular medians, and the Thiessen polygons are filled with a seed filling algorithm to assign initial simulation conditions for the groundwater seepage field in the archaeological excavation area;

[0084] S43: Construct a triangulated network containing all grid points in the prediction area through the boundary vertex coordinates and each internal grid point, divide the convex polygon into a number of non-intersecting triangles, perform triangle interpolation calculation on the triangulated network, and perform real-time correction on the initial simulation conditions in S42 through triangulation of the maximum convex shell point set. Use the cubic spline interpolation function to smooth the seepage field prediction results to obtain the initial prediction conditions of the groundwater seepage field at future times.

[0085] S5: The groundwater level distribution in the boundary area simulated in step S4 is used as the initial prediction condition of the groundwater seepage field at the first time node, and the groundwater seepage field in the area at the current time is used Real-time correction and update of the groundwater seepage field prediction results within the future time range, which can be specifically recorded as in It represents the prediction result at time t+1, Γ represents the Kalman filter. Through repeated iterative calculations, the prediction results of the groundwater seepage field in the area after n time nodes can be calculated and tested in the test set. The drainage effect of the archaeological excavation area is analyzed based on the accuracy prediction.

[0086] Furthermore, the method also includes: obtaining survey information of surface settlement to be measured, inputting the surface settlement and groundwater seepage field prediction results in the surface settlement survey information into the long short-term memory network model to predict surface settlement changes, and obtaining surface settlement changes caused by drainage in the archaeological excavation area; wherein, the long short-term memory network model is trained using the fourth training set and is obtained by adjusting the multi-classification logarithmic loss function; wherein, the fourth training set includes the surface settlement after data processing.

[0087] Furthermore, obtaining the surface settlement amount after data processing includes: arranging the surface settlement amount according to a time series, and performing standardization processing to obtain the surface settlement amount after data processing.

[0088] Specifically, S5 includes the following steps:

[0089] S51: Collect and sort the groundwater level distribution in the boundary area simulated by the borehole water level in step S4 as the initial prediction condition of the groundwater seepage field at the first time node, and the groundwater seepage field at the t-1th time node Predict the groundwater seepage field at the tth time node The error covariance matrix P at time t-1 is t-1 And the correction value matrix Q t , using matrix A to represent the dynamic system parameter matrix, predict the error covariance matrix P at time t t :

[0090] P t =A(P t-1 )A Τ +Q t

[0091] ut-1 represents the increment of groundwater level observation value at the t-th time node, Γ represents the Kalman filter algorithm,

[0092]

[0093] This can predict the groundwater seepage field at the tth time node

[0094] S52: Using the groundwater seepage field in the area at the current time Real-time correction and update of the groundwater seepage field prediction results within the future time range, which can be specifically recorded as in It represents the prediction result at time t+1. Through repeated iterative calculation, the prediction result of groundwater seepage field in the area after n time nodes can be calculated, as shown in the following formula:

[0095]

[0096] By traversing the n time nodes that need to be predicted after the current moment, a groundwater seepage field prediction model can be established and tested in the test set. The drainage effect of the archaeological excavation area can be analyzed based on the accuracy prediction.

[0097] S6: Create an array to save the current prediction results of groundwater seepage field and surface settlement distribution, select the latest survey data, update the prediction results in real time through the trained prediction model, sort them according to time order, put them into the result array, and finally get an array that saves all the prediction results at the current moment, such as Figure 1 As shown, with the increase of on-site survey information, the groundwater seepage field and surface settlement distribution prediction results can be updated in real time; according to the on-site survey information of surface settlement and the prediction results of the groundwater seepage field distributed in the archaeological excavation area in step S5, a real-time prediction model of surface settlement based on long-short-term memory network is established, and the observation data is used to construct a neural network model. Through the loop unit, a neural network of the long-distance time series dependency relationship between surface settlement and groundwater seepage field can be established, and the surface settlement changes caused by the drainage process in the ancient excavation area can be predicted in real time, which can assist in reflecting the drainage effect and avoiding the hazards caused by surface settlement, and timely repair and reinforcement can be carried out to reasonably arrange the field survey and project progress.

[0098] S6 specifically includes the following steps:

[0099] S61: Collect survey information on surface subsidence, organize it according to time series, establish training sets and test sets for surface subsidence, standardize the collected surface subsidence survey data, introduce a long short-term memory network (LSTM) to use the surface subsidence and the groundwater seepage field prediction information in step S5, and use the three thresholds of input, output, and forgetting to train the surface subsidence distribution prediction model during the drainage process of the archaeological excavation area. By constructing a multi-hidden layer neural network, the forward propagation process at each moment in the deep recurrent neural network is realized, and the output of the current long short-term memory network structure is passed into a fully connected layer to obtain the final output, and the multi-classification logarithmic loss function is calculated iteratively and cyclically.

[0100]

[0101] Where LML represents the loss function, nML represents the total number of predicted samples, mLL represents the number of classes, ηij represents the classification, if the i-th sample belongs to the j-th class, then ηij = 1, Pij represents the probability that the i-th sample prediction result belongs to the j-th class, and log takes the natural logarithm.

[0102] S62: Through the cyclic unit, a neural network of the long-distance temporal dependency between surface settlement and groundwater seepage field can be established. The long short-term memory network is used to test the surface settlement data test set of the archaeological excavation area, and the surface settlement changes caused by the drainage process in the ancient excavation area are predicted in real time, which helps to reflect the drainage effect and avoid the hazards caused by surface settlement, and timely repair and reinforcement, so as to reasonably arrange the field survey and engineering progress, such as Figure 2 The figure shows the training process of the real-time prediction model for the groundwater seepage field and surface settlement distribution in the archaeological excavation area.

[0103] S63: Create an array to save the current prediction results of groundwater seepage field and surface settlement distribution, select the latest survey data, update the prediction results in real time through the trained prediction model, sort them according to time sequence, put them into the result array, and finally obtain an array that saves all the prediction results at the current moment, such as Figure 1 As shown in the figure, with the increase of on-site survey information, the prediction results of groundwater seepage field and surface settlement distribution can be updated in real time.

[0104] The present invention is different from the prediction of groundwater seepage field in the process of drainage of traditional archaeological excavation areas. The groundwater seepage field is estimated by various seepage tests and parameters such as gradient field. Modeling is often limited by complex geological conditions and a large number of uncertain hydrogeological parameters. The prediction model cannot be updated in real time as the amount of data increases, and the increase in data volume easily causes a computational burden. In addition, data such as borehole water level and permeability coefficient collected in real time are usually used for simulation research of groundwater seepage field in archaeological excavation areas. Although the simulation results can be used as a basis for the design of drainage in ancient excavation areas, a more accurate real-time prediction and update model for quantitative regional seepage field has not yet been formed. It is impossible to refine the accurate prediction of the distribution of groundwater levels in various parts during the drainage process of archaeological excavation areas, and it is impossible to provide real-time update services for important information in on-site exploration and ancient site protection in archaeological excavation areas. It will cause inconvenience to the real-time adjustment and arrangement of exploration, drainage and other work processes of ancient site protection, and may even affect the in-situ protection of archaeological excavation areas or the progress of the project.

[0105] The deep learning-based groundwater dynamic monitoring and surface subsidence early warning method for archaeological excavation areas provided in the embodiments of the present invention can realize the prediction of groundwater seepage field and surface subsidence distribution in the drainage process in the entire archaeological excavation area, and is not affected by the amount of seismic survey data and information. Even in the case of a large amount of observation data and more complex chart information, as long as there are several drilling point coordinates and drilling water levels as well as rock and soil layer parameters and permeability coefficient survey data within the archaeological excavation area, real-time prediction can be performed. The prediction model automatically extracts features without the need for a large amount of specific measurement data, and the neural network model can be trained and updated in real time through existing observation data. In addition, the accuracy and computational efficiency are analyzed in real time, and the training and update of the groundwater seepage field prediction model in each part of the prediction area are refined to reflect the groundwater drainage effect in each part of the prediction area in real time. It solves the problem that traditional groundwater seepage field estimation modeling is time-consuming and cannot be updated in real time with the increase in data volume, and cannot reflect the drainage effect of archaeological excavation areas in real time. It can serve the problem of groundwater management during the drainage of archaeological excavation areas in real time, and can be used for predicting and analyzing the development and change trend of groundwater seepage fields in ancient excavation areas and the drainage effect.

[0106] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for dynamic monitoring of groundwater and early warning of surface subsidence in archaeological excavation areas based on deep learning, characterized in that: include: Obtaining on-site survey data and geological survey data to be tested within the archaeological excavation area; Input the rock and soil mass information and water supply degree information in the field survey data to be tested into the rock and soil layer classification model and the water supply degree classification model respectively to obtain the rock and soil layer classification result and the water supply degree classification result, and input the geological survey data to be tested into the boundary model to obtain the boundary result; Based on the coordinates of the drilling points in the field survey data to be tested, combined with the boundary conditions, the archaeological excavation area is covered for grid division, and Thiessen polygons are constructed with the perpendicular medians of the adjacent drilling points for filling, and the initial simulation conditions of the groundwater seepage field in the archaeological excavation area are assigned to obtain the initial prediction conditions of the groundwater seepage field at the future time; The groundwater seepage field prediction model is obtained through the initial prediction conditions of the groundwater seepage field at the future moment, and the rock and soil layer category and the water supply degree classification results are input into the groundwater seepage field prediction model to obtain the groundwater seepage field prediction results.

2. The method for dynamic monitoring of groundwater and early warning of surface subsidence in archaeological excavation areas based on deep learning according to claim 1 is characterized in that: Obtaining the rock and soil layer classification result, the water supply degree classification result and the boundary result includes: The rock and soil body information in the field survey data to be tested is input into the rock and soil layer classification model to obtain the rock and soil layer classification result; the rock and soil layer classification model is trained using the first training set, and is obtained by reversely adjusting the layer by layer by minimizing the loss function in combination with the gradient descent method; wherein the training set includes: the rock and soil body information and the rock and soil layer permeability coefficient information after data processing; Inputting the water supply degree information in the field survey data to be tested into a water supply degree classification model to obtain a water supply degree classification result, wherein the water supply degree classification model is trained using a second training set; wherein the second training set includes: water supply degree information after data processing; The geological survey data to be tested is input into the boundary model to obtain the boundary result; the boundary model is trained using the third training set and is obtained by adjusting the loss function in combination with the three-component loss calculation; wherein the third training set includes: geological survey data after data processing.

3. The method for dynamic monitoring of groundwater and early warning of surface subsidence in archaeological excavation areas based on deep learning according to claim 2 is characterized in that: The rock and soil mass information obtained after the data processing includes: Extracting features of the rock and soil mass information using a directional gradient histogram to obtain rock and soil mass feature information; A digital label is established, the rock and soil body characteristic information is matched with the digital label, the digital label with the rock and soil body characteristic information is normalized, and the rock and soil body information after the data processing is obtained.

4. The method for dynamic monitoring of groundwater and early warning of surface subsidence in archaeological excavation areas based on deep learning according to claim 2 is characterized in that: The geological survey data obtained after the data processing includes: The upper boundary, the lower boundary and the lateral boundary in the geological survey data are divided into a plurality of candidate frames, the boundary features are extracted using the plurality of candidate frames, and the frames are stored in an array. Save vertical boundary information, save the box by array Save lateral boundary information; Among them, Z u,d Indicates vertical boundary information, N indicates that there are N vertical boundary information data in total, Represents the Nth upper boundary information, Indicates the Nth lower boundary information, Z s Indicates lateral boundary information, Represents the Mth lateral boundary information data.

5. The method for dynamic monitoring of groundwater and early warning of surface subsidence in archaeological excavation areas based on deep learning according to claim 1 is characterized in that: The initial prediction conditions for obtaining the groundwater seepage field at future moments include: Constructing a triangulated mesh of all grid points in the archaeological excavation area based on the boundary vertex coordinates and each grid point, and segmenting the triangulated mesh using convex polygons to obtain a plurality of mutually non-intersecting triangles; A triangle interpolation calculation is performed on the triangle, and the initial simulation conditions of the assigned groundwater seepage field are corrected by triangulation of the maximum convex hull point set. The seepage field distribution is smoothed using a cubic spline interpolation function to obtain the initial prediction conditions of the groundwater seepage field at future times.

6. The method for dynamic monitoring of groundwater and early warning of surface subsidence in archaeological excavation areas based on deep learning according to claim 1 is characterized in that: Obtaining the groundwater seepage field prediction model includes: The initial prediction condition of the groundwater seepage field at the future moment is used as the initial prediction condition of the groundwater seepage field at the first time node, and the groundwater seepage field at the t-1th time node is used as the initial prediction condition of the groundwater seepage field at the first time node. Predict the groundwater seepage field at the tth time node Among them, A is the dynamic system parameter matrix, u t-1 represents the increment of groundwater level observation value at the t-th time node, Γ represents the Kalman filter algorithm; The groundwater seepage field at the t-th time node is used to perform repeated iterative calculations, and n time nodes that need to be predicted after the current moment are traversed to establish the groundwater seepage field prediction model.

7. The method for dynamic monitoring of groundwater and early warning of surface subsidence in archaeological excavation areas based on deep learning according to claim 6 is characterized in that: The method of iterative calculation using the groundwater seepage field at the tth time node is: in, represents the prediction result at time t+1, Γ represents the Kalman filter algorithm, is the prediction result at time t+n, is the groundwater seepage field distribution at time t.

8. The method for dynamic monitoring of groundwater and early warning of surface subsidence in archaeological excavation areas based on deep learning according to claim 1 is characterized in that: The method further comprises: The surface settlement survey information to be measured is obtained, and the surface settlement in the surface settlement survey information and the groundwater seepage field prediction result are input into the long short-term memory network model to predict the surface settlement change, so as to obtain the surface settlement change caused by the drainage process of the archaeological excavation area; wherein the long short-term memory network model is trained with the fourth training set and adjusted with the multi-classification logarithmic loss function; wherein the fourth training set includes the surface settlement after data processing.

9. The method for dynamic monitoring of groundwater and early warning of surface subsidence in archaeological excavation areas based on deep learning according to claim 8 is characterized in that: The surface settlement amount after obtaining the data processing includes: The surface settlement amount is sorted according to a time series and standardized to obtain the surface settlement amount after the data processing.

Images