LU502946B1 - Real-time prediction method of groundwater seepage field in cultural site areas - Google Patents

Real-time prediction method of groundwater seepage field in cultural site areas Download PDF

Info

Publication number
LU502946B1
LU502946B1 LU502946A LU502946A LU502946B1 LU 502946 B1 LU502946 B1 LU 502946B1 LU 502946 A LU502946 A LU 502946A LU 502946 A LU502946 A LU 502946A LU 502946 B1 LU502946 B1 LU 502946B1
Authority
LU
Luxembourg
Prior art keywords
prediction
seepage field
data
training
time
Prior art date
Application number
LU502946A
Other languages
German (de)
Inventor
Lu Yang
Original Assignee
Univ Shenyang Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Univ Shenyang Technology filed Critical Univ Shenyang Technology
Priority to LU502946A priority Critical patent/LU502946B1/en
Application granted granted Critical
Publication of LU502946B1 publication Critical patent/LU502946B1/en

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V99/00—Subject matter not provided for in other groups of this subclass
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Geophysics (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a real-time prediction method of groundwater seepage field in a cultural site area, comprising the following steps: using the data of rock and soil layer test and the field survey to extract features, establishing the classification model through historical data to build rock and soil layer category prediction function, building specific yield prediction model, training the boundary conditions of groundwater seepage field in cultural sites by target detection algorithm, building real-time prediction model of groundwater seepage field by Kalman filter algorithm, reflecting the real-time drainage effect of the site area, establishing a real-time prediction model of surface subsidence distribution based on long-term and short-term memory networks, making real-time prediction of the surface subsidence changes caused by the drainage process of the site area, so as to help reflect the drainage effect and avoid the harm caused by the surface subsidence, and timely repairing and reinforcing it, so as to reasonably arrange the field survey and project progress.

Description

REAL-TIME PREDICTION METHOD OF GROUNDWATER SEEPAGE HU502948
FIELD IN CULTURAL SITE AREAS
TECHNICAL FIELD
The invention belongs to the technical field of computer machine learning, which specifically relates to deep learning, pattern recognition, numerical simulation, and other algorithms. In particular, it involves the real-time prediction of groundwater seepage field and surface settlement distribution based on a neural network algorithm based on statistical sampling data of groundwater borehole the water level observation, and field test data of permeability coefficient in the process of drainage in cultural site areas, which is suitable for prediction of groundwater seepage field and surface settlement distribution under various hydrogeological conditions in cultural site areas.
BACKGROUND
Cultural sites in China's vast monsoon climate areas are constantly threatened by rain, groundwater, and even air moisture in different degrees. Among them, groundwater erodes cultural sites to a great extent, which easily makes it difficult to protect and survey cultural sites, increases the difficulty of in-situ protection of cultural sites, and even affects the water supply and water use in residential areas and villages around cultural sites, resulting in inconvenience to residents living near cultural sites. Therefore, it is necessary to provide real-time information on groundwater seepage field and surface subsidence distribution in the process of drainage in the core area of the site as the basis for evaluating the hydrogeological conditions of the site and carrying out investigation work, to avoid environmental geological pollution, facilitate reasonable resettlement of nearby related demolition, and strengthen the construction goal of integrating cultural site protection with urban and rural construction.
The prediction of groundwater seepage field in the process of drainage in traditional cultural sites is to estimate the groundwater seepage field with various parameters such as the infiltration test and gradient field. Regression modeling is often limited by complex geological conditions and a large number of uncertain 502946 hydrogeological parameters, so it is impossible to update the prediction model in real-time with the increase of data, and the increase of data is easy to cause a computational burden. In addition, the information on borehole water level and permeability coefficient collected in real-time is usually used for the simulation of groundwater seepage fields in cultural site areas. Although the simulation results can be used as the basis for the drainage design in the site area, a more accurate real-time prediction and update model for quantifying the regional seepage field has not been formed, which cannot be refined to the precise prediction of the distribution of ground subsidence coupled with groundwater seepage field in the drainage process of the cultural site area and cannot provide real-time update services for important information in the field exploration and cultural site protection in the cultural site area.
It will cause inconvenience to the real-time adjustment and arrangement of the drainage system for exploration and ancient site protection, and even affect the in-situ protection of cultural sites or the project schedule.
SUMMARY
The technical problem to be solved by the invention is to provide a real-time prediction method for the groundwater seepage field in the cultural site areas, which solves the problem that the estimation and modeling of the traditional groundwater seepage field takes a long time and cannot be updated in real-time with the increase of the data amount, and cannot reflect the drainage effect of the updated cultural site area in real-time, and can serve the groundwater management problem in the drainage process of the cultural site area in real-time, and can be used to predict the development trend of groundwater seepage field and drainage effect in the site area.
S1: surveying the data in the cultural site area, including the coordinates and the water level of several drilling points, parameters of rock and soil layers, permeability coefficient, specific yield, etc., establishing a group of digital labels with feature data, corresponding the feature data to the digital labels, and transmitting the category features and digital labels to the computer for prediction; according to the current geological profile map of the cultural site, the original groundwater system survey 502946 data, the topographic map of the cultural site, the hydrogeological conditions and the stratigraphic structure, etc, extracting the box feature by convolution neural network algorithm, filling the inner area of the predicted boundary through the seed filling algorithm, and using the stack to realize the polygon filling of the inner area of the boundary, so as to generate the boundary conditions of groundwater seepage field in cultural sites, and generating the initial prediction conditions of groundwater seepage field;
S2, using the existing rock and soil layer test and field survey data, establishing a classification model through feature data to construct a rock and soil layer category prediction function, use gradient descent method to minimize the loss function and reverse adjust the weight parameters in the network layer by layer, so as to speed up the convergence and provide geotechnical layer category information for future groundwater seepage field prediction; collect, sort out and save specific yield information involved in rock and soil materials; according to the features of specific yield information extraction unit specific yield, establish the features of unit specific yield and digital labels, and fit the constructed data set through the classification model of support vector machine to complete the nonlinear classification of unit specific yield; divide the data set into training set and test set, predict the unit specific yield result of the training set, and compare the predicted result with the real result, which is convenient to analyze the prediction accuracy and build the specific yield prediction model;
S3: establishing the aquifer information set collected and sorted by geological data, dividing the upper boundary, the lower boundary, and the lateral boundary into several small frames, detecting the target of the groundwater aquifer boundary, filling the inner area of the predicted boundary by the seed filling algorithm, generating the boundary conditions of groundwater seepage field in cultural sites, and establishing the numerical model for predicting groundwater seepage field in cultural sites;
S4: collect the observation information of each borehole water level corresponding to different coordinates of each borehole point, extract the features of borehole water level and borehole point coordinates, in addition, collect the 7506968 information of rock and soil layers and specific yield, extract the features of rock and soil layers according to the categories of each rock and soil layer, collect and sort out the observation data according to the time series, establish a numerical model of simulated groundwater seepage field in cultural sites, and generate the initial prediction conditions of groundwater seepage field;
SS: taking the groundwater level distribution in the boundary area simulated in
S4 as the initial prediction condition of the groundwater seepage field at the first time node, using the current regional groundwater seepage field U/, updating the prediction of future regional groundwater seepage field in real-time, recording as
UT, =T(U7) by Kalman filtering algorithm, where U}, represents the prediction result at +1, I" represents Kalman filtering algorithm, through repeated iterative calculation, calculating the prediction results of groundwater seepage field in this area after n time nodes, testing in the test set, and analyzing the drainage effect of cultural sites according to the precision prediction;
S6: create an array that stores the current prediction results of groundwater seepage field and surface subsidence distribution, screens the latest survey data, update the prediction results in real-time through the trained prediction model, and finally get an array that stores all the current prediction results according to the time, as shown in Figure 1; with the increase of on-site survey information, update the prediction results of groundwater seepage field and surface subsidence distribution in real time; through the circulation unit, establish the neural network of the long-distance time-series dependence relationship between the surface subsidence and the groundwater seepage field, which can help reflect the drainage effect and avoid the harm caused by the surface subsidence, and repair and reinforce it in time, to reasonably arrange the field survey and project progress.
Effects of Invention
A real-time prediction method of groundwater seepage fields in cultural site areas can realize the prediction of groundwater seepage field and surface subsidence distribution in the process of drainage in the entire cultural site area and is not 17008968 affected by the amount of geological survey data and information. Even in the case of a large number of observation data and complicated chart information, as long as there are some borehole coordinates and borehole water levels, geotechnical 5 parameters, and permeability coefficient survey data in the cultural site area, real-time prediction can be made, and the prediction model can automatically extract features.
There is no need for a large number of specific measurement data, and the existing observation data can be used to train and update the neural network model in real time, and the accuracy and calculation efficiency can be analyzed in real-time, to refine the training and update of the groundwater seepage field prediction model in each part of the prediction area, and reflect the groundwater drainage effect in each part of the prediction area in real-time. In addition, the prediction method solves the problems that the traditional estimation and modeling of groundwater seepage fields takes a long time and can't be updated in real-time with the increase of data, and can't reflect the drainage effect of updated cultural sites in real-time. It can serve the problem of groundwater management in the drainage process of cultural sites in real-time and can be used to predict and analyze the development trend and drainage effect of groundwater seepage fields in sites.
BRIEF DESCRIPTION OF THE FIGURES
Fig. 1 is a flow chart of a real-time prediction system of groundwater seepage field and surface settlement distribution in cultural site area;
Fig. 2 is a training flow chart of a real-time prediction model of groundwater seepage field and surface subsidence distribution in cultural site area;
Fig. 3 is a low chart of extracting boundary features of the groundwater seepage field prediction model;
Fig. 4 is a schematic diagram of the three-component loss training process.
DESCRIPTION OF THE INVENTION
The invention provides a real-time prediction method of groundwater seepage 502946 fields in cultural site areas. According to the embodiment of the present invention, referring to the flow chart of the real-time prediction system of groundwater seepage field and surface settlement distribution in the base of cultural heritage area as shown in Figure 1, including the following steps:
S1 comprises the following steps:
S11: surveying the data in the cultural site area, including the coordinates and the water level of several drilling points, parameters of rock and soil layers, permeability coefficient, specific yield, etc, completing the feature extraction through the calculation of directional gradient histogram, establishing a group of digital labels with feature data, corresponding the feature data to the digital labels, unifying the digital labels, and transmitting the category features and digital labels to the computer for prediction;
S12: dividing the upper boundary, the lower boundary and the lateral boundary of the vertical geological information data and the lateral geological information data into a plurality of small frames according to the current geological profile map of the cultural heritage area, the original groundwater system survey data, the topographic map of the cultural site, the hydrogeological conditions and the stratigraphic structure, etc, extracting the candidate box and directly applying to the feature building, marking the position of the array to save the box, extracting the box feature by convolution neural network algorithm, extracting the position feature flow, and alternately setting the convolution layer and sampling layer to extract the feature; filling the inner area of the predicted boundary through the seed filling algorithm, and using the stack to realize the polygon filling of the inner area of the boundary, so as to generate the boundary conditions of groundwater seepage field in cultural sites, establishing the boundary conditions of the numerical model of groundwater seepage field prediction in cultural sites, and generating the initial prediction conditions of groundwater seepage field;
S2 comprises the following steps:
S21: using the existing rock and soil layer test and field survey data, preserving 502946 the permeability coefficient information according to the rock and soil layer order, forming a permeability coefficient training sample set, and establishing a set
KK Ki} to represent the permeability coefficient information corresponding to n different rock and soil layers; for rock and soil materials, extracting the features of silt, weathered rock mass, bedrock, etc., of each rock and soil layer, completing the feature extraction by calculating the directional gradient histogram, establishing the classification model according to the obtained feature data, completing the classifier training of each rock and soil layer, and comparing the extracted features with the classification rules contained in the classification model to obtain the category of rock and soil layer; setting up a group of digital labels with feature data, corresponding the features representing rock and soil layer categories to the digital labels, unifying the digital labels, and transferring the category features and digital labels to the computer for training, when testing the model, keeping the data range consistent with the data features provided by training, and avoiding errors caused by the inability to identify a certain rock and soil layer category; training the geotechnical layer classification prediction model function in the sample space, adjusting the weight parameters in the network layer by layer by minimizing the loss function with the gradient descent method, and accelerating the convergence with the depth model optimization algorithm:
Arg, = KR, @ Tate = @ - KR, wherein x is the learning rate, w; is the weight of the i-th training, w+ is the weight of the /+1-th training, #; is the gradient of the i-th training, Aw; is the update of the weight of the i-th training, wo= 0, to optimize the network through cyclic iteration and feedback, the accuracy of the network is improved through frequent iterative training;
S22: collecting, sorting out, and saving specific yield information involved in 502946 rock and soil materials, expressing specific yield information as a set de bio} oo 1502 ** saving specific yield data information of m different rock formations, and building a training sample set of unit-specific yield; extracting unit-specific yield features for rock and soil materials, completing feature extraction through directional gradient histogram calculation, establishing a classification model according to the obtained feature data, completing the classifier training of specific yield of each rock and soil layer, and training specific yield classification model through the extracted features to obtain the category prediction function of specific yield; setting up a group of digital labels of unit-specific yield with feature data, corresponding the features representing the specific yield category of the unit to the digital labels, unifying the digital labels, and transferring the features of the specific yield category of the unit and the digital labels to the computer for training;
S23: completing the construction of features and labels through S22, training the unit-specific yield classification prediction model function in the sample space, fitting the constructed data set through the support vector machine classification model, training and separating positive samples and negative samples, completing the nonlinear classification of unit specific yield, dividing the data set into the training set and test set, predicting the unit specific yield result of the training set, comparing the predicted result with the real result, and analyzing the prediction accuracy;
S3 comprises the following steps:
S31: establishing an aquifer information collection collected and sorted through geological data, including geological section map of cultural sites, survey data of original groundwater system, topographic map of cultural sites, hydrogeological conditions, stratum structure, and other geological survey data; saving the vertical 10 779 mn 78 m8 SAS Lui boundary information with Wada SEE Ss SÉRE | wherein 4 represents the vertical boundary information, N represents a total of N vertical boundary information materials, Zu represents the N-th upper boundary information
, LU502946 materials, a represents the N-th lower boundary information materials, and
Z = WB zi holds the lateral boundary information materials, # represents the lateral boundary information, and Ze represents the M-th lateral boundary information data;
S32: dividing the upper boundary, the lower boundary, and the lateral boundary in the vertical geological information data and the lateral geological information data into a plurality of small frames, checking and finding the frames with boundaries one by one, extracting the candidate frames and directly applying them to the featuring = fe > = 3 establishment, using the array = 7 =i: »Ss+84%5 to save the data marked by the position of the box, using the convolution neural network algorithm to extract the box features and the position features, alternately setting the convolution layer and the sampling layer to extract features, using the nonlinear support vector machine algorithm as the detection algorithm to classify the boundary; if judging that there is a boundary in a box, outputting the position of this small box, training through the detection data set, and calculating the loss function from the three-component loss:
Lyn = axiale, gy d{a,r) + pi wherein Liripiet represents a three-component loss function, a represents anchor, p represents positive samples in the same category as a, n represents negative samples, and a samples in different categories, m takes a constant greater than 0, d(a,p) represents Euclidean distance between the anchor and positive, d(a,n) represents
Euclidean distance between the anchor and negative samples, max means maximum function; inputting triple, including anchor, positive and negative; by optimizing the distance between the anchor and positive to be less than that between the anchor and negative, calculating the similarity between samples, and establishing the boundary prediction model function by testing in the test set;
S4 comprises the following steps:
S41: establishing a training sample set
Z's WEE 6 => dl, EC corresponding to coordinates of different drilling points according to measured water levels, and representing 502946 historical observation data of the drilling water levels corresponding to the 1 to / time nodes, wherein Æ represents the serial number of drilling points, / represents the i-th time node, and 7 represents the total number of time nodes in the time series; according to the boundary conditions, coordinates and distribution of drilling points assigned by the seed filling color algorithm in S3, covering all the predicted areas of the groundwater range in the cultural site area, wherein the predicted areas cover the whole filled area; doing all the information and data provided by field investigation and test as much as possible, and dividing the forecast area into grids and establishing the coordinate array of observation wells;
S42: according to the grid divided by the predicted area range, constructing the
Tai Sen polygon with the adjacent drilling points as the vertical lines, and filling the
Tai Sen polygon with the seed filling algorithm, to evaluate the initial simulation conditions of the groundwater seepage field in the cultural site area;
S43: constructing a triangular mesh containing all the grid points in the prediction area through the boundary vertex coordinates and the internal grid points, dividing the convex polygon into a plurality of mutually disjoint triangles, performing triangle interpolation calculation on the triangular mesh, correcting the initial simulation conditions in S42 in real-time by triangulation of the maximum convex hull point set, and using cubic spline interpolation function to smooth the prediction results of seepage field, to obtain the initial prediction conditions of groundwater seepage field in the future;
SS comprises the following steps:
S51: collecting and sorting out 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, predicting the groundwater seepage field Ur at the / time node from the groundwater seepage field
A at the 7-1 time node, and expressing the dynamic system parameter matrix by the error covariance matrix Pri and the correction matrix Ü; at the 7-1 time, and predicting 17008968 the error matrix P; at the / time node by the matrix A:
B= AP IA +0 wherein 2-1 represents the increment of groundwater level observation value to the t-th time node, I" represents the Kalman filtering algorithm,
UP = AUX +T 64) and therefore, predicting the groundwater seepage field uy at the t time node;
S52: using the current regional groundwater seepage field vr updating the prediction result of future regional groundwater seepage field in real-time, recording as Ur, = I) , where Ul represents the prediction result at #+1, through repeated iterative calculation, calculating the prediction results of groundwater seepage field in this area after n time nodes, as shown in the following formula:
UL = FU) = THU) = DE PO traversing the n time nodes that need to be predicted after the current time, establishing the prediction model of groundwater seepage field, testing in the test set, and analyzing the drainage effect of cultural sites according to the precise prediction;
S6 comprises the following steps:
S61: creating an array to save the current prediction results of groundwater seepage field and surface subsidence distribution, saving the array of all the current prediction results, with the increase of field survey information, and updating the prediction results of groundwater seepage field and surface subsidence distribution in real-time;
S62: standardizing the collected survey data of surface subsidence, introducing long-term and short-term memory networks, and using the prediction information of surface subsidence and groundwater seepage field in S5 to train the prediction model of surface subsidence distribution in the process of drainage in cultural sites, and iterative loop calculating the multi-classification logarithmic loss function:
wherein Lu. represents the loss function, mv represents the total number of predicted samples, mm. represents the number of classes, and 7; represents the classification; if the i-th sample belongs to the j class, #;=1, Pj; represents the probability that the predicted result of the i-th sample belongs to the j class, and log takes the natural logarithm.
Through the circulation unit, establishing the neural network of the long-distance time-series dependence relationship between the surface subsidence and the groundwater seepage field, assisting in reflecting the drainage effect and avoiding the harm caused by the surface subsidence, making timely repair and reinforcement, reasonably arranging the field survey and the project progress. Fig.2 is training the process for the real-time prediction model of the groundwater seepage field and the surface subsidence distribution in the cultural site base.
The method is different from the previous prediction of groundwater seepage field in the process of dewatering and drainage in traditional cultural sites, in that the groundwater seepage field is estimated by various parameters such as infiltration test and gradient field, and the modeling is often limited by complex geological conditions and a large number of uncertain hydrogeological parameters so that the prediction model cannot be updated in real-time with the increase of data volume, and the increase of data volume is easy to cause the calculation burden, and data such as borehole water level and permeability coefficient collected in real time are usually used for simulation research of groundwater seepage field in cultural sites. Although the simulation results can be used as the basis for the design of drainage in cultural sites, a more accurate real-time prediction and update model of regional seepage field has not been formed, which can't refine the accurate prediction of groundwater level distribution in each part of the drainage process in cultural sites, and can't provide real-time update services of important information for on-site exploration and protection of cultural sites. It will cause inconvenience to real-time adjustment and deployment in the process of exploration, drainage of ancient sites protection, and 502946 even affect the in-situ protection of cultural sites or the project schedule.
The real-time prediction method of groundwater seepage field in the cultural site area is provided by the embodiment of the invention can realize the prediction of groundwater seepage field and the prediction of surface subsidence distribution in the whole cultural site area in the process of drainage, and is not affected by the amount of seismic survey data and information. Even in the case of a large number of observation data and complicated chart information, real-time prediction can be carried out as long as there are several borehole coordinates and borehole water levels, geotechnical parameters, and permeability coefficient survey data in the cultural site area. The prediction model automatically extracts features without a large number of specific measurement data, and the existing observation data can be used to train and update the neural network model in real time, the accuracy and calculation efficiency can be analyzed in real-time, to refine the training and update of the groundwater seepage field prediction model in each part of the prediction area, and reflect the groundwater drainage effect of each part of the prediction area in real-time. The method solves the problems that the traditional estimation and modeling of groundwater seepage fields take a long time, cannot be updated in real-time with the increase of data volume, and cannot reflect the drainage effect of updated cultural sites in real-time, can serve the problem of groundwater management in the drainage process of cultural sites in real-time, and can be used for forecasting and analyzing the development trend and drainage effect of groundwater seepage field in sites.
Those skilled in art can easily understand that the above advantages can be freely combined and superimposed without conflict.

Claims (6)

CLAIMS LU502946
1. A real-time prediction method of groundwater seepage field in cultural site areas, characterized in that S1 comprises: S11: surveying the data in the cultural site areas, including the coordinates and the water level of several drilling points, parameters of rock and soil layers, permeability coefficient and specific yield, completing the feature extraction through calculation of directional gradient histogram, establishing a group of digital labels with feature data, corresponding the feature data to the digital labels, unifying the digital labels, and transmitting the category features and digital labels to the computer for prediction; S12: dividing the upper boundary, the lower boundary and the lateral boundary of the vertical geological information data and the lateral geological information data into a plurality of small frames according to the current geological profile map of the cultural site, the original groundwater system survey data, the topographic map of the cultural site, the hydrogeological conditions and the stratigraphic structure; extracting the candidate box and directly applying to feature building, marking the position of the array to save the box, extracting the box feature by convolution neural network algorithm, extracting the position feature flow, and alternately setting the convolution layer and sampling layer to extract the feature; filling the inner area of the predicted boundary through the seed filling algorithm, and using the stack to realize polygon filling of the inner area of the boundary to generate the boundary conditions of groundwater seepage field in cultural sites, establishing the boundary conditions of the numerical model of groundwater seepage field prediction in cultural sites, and generating the initial prediction conditions of groundwater seepage field.
2. The real-time prediction method of groundwater seepage field in cultural site areas according to claim 1, characterized in that S2 comprises: S21: using the existing rock and soil layer test and field survey data, preserving the permeability coefficient information according to the rock and soil layer order, forming a permeability coefficient training sample set, and establishing a set
K= {K,K,,,K,} to represent the permeability coefficient information corresponding to n different rock and soil layers; for rock and soil materials, extracting the features of silt, weathered rock, bedrock of each rock and soil layer, completing the feature extraction by calculating the directional gradient histogram, establishing the classification model according to the obtained feature data, completing the classifier training of each rock and soil layer, and comparing the extracted features with the classification rules contained in the classification model to obtain the category of rock and soil layer; setting up a group of digital labels with feature data, corresponding the features representing rock and soil layer categories to the digital labels, unifying the digital labels, and transferring the category features and digital labels to the computer for training; when testing the model, making the data range the same as the data characteristics of training, and avoiding errors caused by the inability to identify a certain rock and soil layer category;
training the geotechnical layer category prediction model function in the sample space, adjusting the weight parameters in the network layer by layer by minimizing the loss function with the gradient descent method, and accelerating the convergence with the depth model optimization algorithm:
Ao, =—Kn,, oO, = 0, +Ao, =o, —kn,, wherein x is the learning rate, w; is the weight of the i-th training, 1 is the weight of the i+1-th training, 7; is the gradient of the i-th training, Aw; is the update of the weight of the i-th training, wo= 0, to optimize the network through cyclic iteration and feedback, the accuracy of the network is improved through frequent iterative training;
S22: collecting, sorting out, and saving specific yield information of rock and soil materials, expressing specific yield information as a set u = fu uy, U, }, saving specific yield information of mn different rock formations, and building a training 17008968 sample set of specific yield in a unit, extracting specific yield in a unit for rock and soil materials, completing feature extraction through directional gradient histogram calculation, establishing a classification model based on the obtained feature data, completing the classifier training of specific yield of each rock and soil layer, and training specific yield classification model through the extracted features to obtain the category prediction function of specific yield, setting up a group of digital labels of unit-specific yield with feature data, corresponding the features representing the specific yield category of the unit to the digital labels, unifying the digital labels, and transferring the features of the specific yield category of the unit and the digital labels to the computer for training; S23: completing the construction of features and labels through S22, training the specific yield classification prediction model function in the sample space, fitting the constructed data set through the support vector machine classification model, training and separating positive samples and negative samples, completing the nonlinear classification of specific yield in the unit, dividing the data set into the training set and test set, predicting the specific yield result of the training set, comparing the predicted result with the real result, and analyzing the prediction accuracy.
3. The real-time prediction method of groundwater seepage field in cultural site areas according to claim 1, characterized in that S3 comprises: S31: establishing an aquifer information collection collected and sorted through geological data, including geological profile map of cultural sites, survey data of original groundwater system, topographic map of cultural sites, hydrogeological conditions, stratum structure, and other geological survey data; saving the vertical boundary information with Z"* = ZZ ZU. 20,78 ZU}, wherein Z"* represents the vertical boundary information, N represents a total of N vertical boundary information materials, Z, represents the N-th upper boundary information materials, Z§ represents the N-th lower boundary information materials, and
Z° = {2:75.72} holds the lateral boundary information materials, 75 represents the lateral boundary information, and Z}, represents the M-th lateral boundary information data; S32: dividing the upper boundary, the lower boundary, and the lateral boundary in the vertical geological information data and the lateral geological information data into a plurality of small frames, checking and finding the frames with boundaries one by one, extracting the candidate frames and directly applying to the establishment of features, using the array z= 2.20, Zune to save the data marked by the position of the box, using the convolution neural network algorithm to extract the box features and the position features, alternately setting the convolution layer and the sampling layer to extract features, using the nonlinear support vector machine algorithm as the detection algorithm to classify the boundary; if judging that there is a boundary in a box, outputting the position of the box, training through the detection data set, and calculating the loss function from the three-component loss: Lip = Max(d(a, p)-d(a,n)+m), wherein Liripiet represents a three-component loss function, a represents anchor, p represents positive samples in the same category as a, n represents negative samples and is a sample that is in a different category from a, m takes a constant greater than 0, d(a,p) represents Euclidean distance between the anchor and positive samples, d(a,n) represents Euclidean distance between the anchor and negative samples, max means maximum function; inputting triple, including anchor, positive and negative; by optimizing the distance between the anchor and positive, so that the distance between the anchor and positive is less than the distance between the anchor and negative, calculating the similarity between samples, and establishing the boundary prediction model function by testing in the test set.
4. The real-time prediction method of groundwater seepage field in cultural site areas according to claim 1, characterized in that S4 comprises:
S41: establishing a training sample set 502946 Zh = {a, z5),2, 25) (i, 25) (t= 1,25), (¢, 2 )} corresponding to coordinates of different drilling points according to measured water levels and representing historical observation data of the drilling water levels corresponding to the 1 to 7 time nodes, wherein Æ represents the serial number of drilling points, i represents the i-th time node, and 7 represents the total number of time nodes in the time series; according to the boundary conditions, coordinates and distribution of drilling points assigned by the seed filling color algorithm in S3, covering all the predicted areas of the groundwater range in the cultural site area, wherein the predicted areas cover the whole filled area; based on all the information and data provided by field investigation and test, dividing the forecast area into grids and establishing the coordinate array of observation wells; S42: according to the grid divided by the predicted area, constructing the Tai Sen polygon with the adjacent drilling points as the vertical lines, and filling the Tai Sen polygon with the seed filling algorithm to evaluate the initial simulation conditions of the groundwater seepage field in the cultural site area; S43: constructing a triangular mesh containing all the grid points in the prediction area through the boundary vertex coordinates and the internal grid points, dividing the convex polygon into a plurality of mutually disjoint triangles, performing triangle interpolation calculation on the triangular mesh, correcting the initial simulation conditions in S42 in real-time by triangulation of the maximum convex hull point set, and using cubic spline interpolation function to smooth the prediction results of seepage field, to obtain the initial prediction conditions of groundwater seepage field in the future.
5. The real-time prediction method of groundwater seepage field in cultural site areas according to claim 1, characterized in that S5 comprises: S51: collecting and sorting out the groundwater level distribution in the boundary area simulated by the drill hole water level in S4 as the initial prediction condition of the groundwater seepage field at the first time node, predicting the groundwater seepage field U; at the / time node from the groundwater seepage field U; atthe 7-1 time node, and expressing the dynamic system parameter matrix by the error covariance matrix P,-1 and the correction matrix (J; at the 7-1 time, and predicting the error matrix P; at the 7 time node by the matrix A: T F=AF)4" +0, wherein 2-1 represents the increment of groundwater level observation value to the t-th time node, I" represents the Kalman filtering algorithm, Ur = AUF +T'(u, |), and therefore, the groundwater seepage field U/ at the / time node is predicted; S52: using the current regional groundwater seepage field U/, updating the prediction of future regional groundwater seepage field in real-time, recording as UT, =T(U}), wherein U}, represents the prediction result at #+1 time node, through repeated iterative calculation, calculating the prediction results of groundwater seepage field in this area after n time nodes, as shown in the following formula: Up, TUE) = Up) ==" (U)=T"U)), traversing the n time nodes predicted after the current time, establishing the prediction model of the groundwater seepage field, testing in the test set, and analyzing the drainage effect of cultural sites according to the precise prediction.
6. The real-time prediction method of groundwater seepage field in cultural site areas according to claim 1, characterized in that S6 comprises: S61: establishing an array to save the current prediction results of groundwater seepage field and surface subsidence distribution, saving the array of all the current prediction results, with the increase of field survey information, and updating the prediction results of groundwater seepage field and surface subsidence distribution in real-time; S62: standardizing the collected survey data of surface subsidence, introducing long-term and short-term memory networks, and using the prediction information of surface subsidence and groundwater seepage field in SS to train the prediction model 502946 of surface subsidence distribution in the process of drainage in cultural sites, and iterative loop and calculating the multi-classification logarithmic loss function: Ly LSS > My, i=1 j= wherein Lm represents the loss function, nm. represents the total number of predicted samples, mm represents the number of classes, and #; represents the classification; if the i-th sample belongs to the j class, #;=1; Pj; represents the probability that the predicted result of the i-th sample belongs to the j class, and log is the natural logarithm;
through the circulation unit, establishing the neural network of the long-distance and time-series relationship between the surface subsidence and the groundwater seepage field, assisting in reflecting the drainage effect and avoiding the harm caused by the surface subsidence, timely repairing and reinforcing, reasonably arranging the field survey and the project progress, and training the process for the real-time prediction model of the groundwater seepage field and the surface subsidence distribution in the cultural site.
LU502946A 2022-10-23 2022-10-23 Real-time prediction method of groundwater seepage field in cultural site areas LU502946B1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
LU502946A LU502946B1 (en) 2022-10-23 2022-10-23 Real-time prediction method of groundwater seepage field in cultural site areas

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
LU502946A LU502946B1 (en) 2022-10-23 2022-10-23 Real-time prediction method of groundwater seepage field in cultural site areas

Publications (1)

Publication Number Publication Date
LU502946B1 true LU502946B1 (en) 2023-04-24

Family

ID=86097378

Family Applications (1)

Application Number Title Priority Date Filing Date
LU502946A LU502946B1 (en) 2022-10-23 2022-10-23 Real-time prediction method of groundwater seepage field in cultural site areas

Country Status (1)

Country Link
LU (1) LU502946B1 (en)

Similar Documents

Publication Publication Date Title
CN109584510B (en) Road high slope landslide hazard early warning method based on evaluation function training
Greve et al. Quantifying the ability of environmental parameters to predict soil texture fractions using regression-tree model with GIS and LIDAR data: The case study of Denmark
CN110490100A (en) Ground automatic identification based on deep learning names method and system
Hajihassani et al. 3D prediction of tunneling-induced ground movements based on a hybrid ANN and empirical methods
CN112966722A (en) Regional landslide susceptibility prediction method based on semi-supervised random forest model
CN107315862A (en) A kind of method for setting up open-cut foundation ditch engineering investigation and analog parameter relation
CN105321204B (en) A kind of three-dimensional geological grid body Model construction method
CN114036841A (en) Landslide incidence prediction method and system based on semi-supervised support vector machine model
CN102253423B (en) Proper water supply position intelligent recognition technology based on multi-source hydrogeology survey information
CN115730684A (en) Air quality detection system based on LSTM-CNN model
CN113128106B (en) Method for determining surface subsidence caused by shield construction of karst stratum
Shi et al. Data-driven sequential development of geological cross-sections along tunnel trajectory
CN112396312B (en) Open caisson sinking construction automatic decision-making method and system
LU502946B1 (en) Real-time prediction method of groundwater seepage field in cultural site areas
CN109542995B (en) Method for analyzing landslide of reservoir area
CN116628886A (en) Real-time optimization method for shield tunneling machine crossing construction parameters based on limited engineering data
KR20120004113A (en) A development of prediction model for groundwater productivity potential based on probability and statistic
CN115200540A (en) Mine roadway deformation monitoring and early warning method and system
CN114066271A (en) Tunnel water inrush disaster monitoring and management system
CN115182398A (en) Underground water level and surface subsidence prediction method for earthquake early warning area
KR102408840B1 (en) Detecting method of water saturated weak area of forest soil for improving landslide danger map using forest water map and the system
KR102408827B1 (en) Preparation method of forest water map using computer program and the system
CN114357678A (en) Novel optimization design method for regional underground water level monitoring network
Firoozi et al. Application of Machine Learning in Geotechnical Engineering for Risk Assessment
CN113449362A (en) Road reconnaissance design method and system based on 5G mobile network

Legal Events

Date Code Title Description
FG Patent granted

Effective date: 20230424