Surveying and mapping deformation monitoring method and system based on spatiotemporal deep learning and application

Abstract

This invention discloses a method and system for monitoring surveying deformation based on spatiotemporal deep learning, relating to the technical field of computer systems for surveying deformation monitoring. It obtains periodic and sudden factors affecting surveying deformation. Based on a prediction model, it outputs time-series information of deformation prediction vectors. It obtains predicted values ​​for periodic factors; when the fluctuation vector is greater than or equal to a threshold, it retrieves first spatiotemporal sample data, statistically analyzes the time-series information of the first deformation deviation vector, and corrects the time-series information of the deformation prediction vector to obtain first-order deformation correction vector time-series information. It then statistically analyzes the time-series information of the second deformation deviation vector and corrects the first-order deformation correction vector time-series information. Finally, it adds the deformation correction vector time-series information to the surveying deformation monitoring data within the target time window.

Description

Technical Field

[0001] This invention relates to the technical field of computer systems for surveying deformation monitoring, specifically to a surveying deformation monitoring method, system, and application based on spatiotemporal deep learning. Background Technology

[0002] With the increasing demands for accuracy and timeliness in surface deformation monitoring from infrastructure construction, geological disaster monitoring, and urban planning, deformation monitoring is gradually shifting from post-hoc summarization to intelligent prediction. Current technologies typically predict future deformation data based on historical deformation time-series data, providing a basis for early prevention. While this method yields accurate results under ideal conditions where deformation factors change periodically without fluctuations, the actual cycles and fluctuations of these factors are uncertain, leading to unreliable predictions. Therefore, current deformation monitoring methods suffer from low accuracy in complex environments. Improving the accuracy of deformation prediction has become an urgent technical challenge. Summary of the Invention

[0003] To address the problems existing in current technologies, this invention provides a surveying deformation monitoring method, system, and application based on spatiotemporal deep learning. This solves the technical problem of low accuracy in deformation prediction under complex environments in existing surveying deformation monitoring methods. It achieves year-on-year and month-on-month fusion prediction of periodic factors, fault-tolerant fluctuation judgment, and multi-round deviation correction based on similar samples, thereby effectively improving the accuracy of surveying deformation prediction and its response to abnormal disturbances, and enhancing the accuracy of surface deformation monitoring in complex environments.

[0004] This invention is implemented as follows: a surveying deformation monitoring method based on spatiotemporal deep learning. The method includes: obtaining periodic factors and sudden factors affecting surveying deformation in a target area, wherein the periodic factors have a fault-tolerant fluctuation threshold; processing historical surveying deformation time-series data through a surveying deformation prediction model, and outputting deformation prediction vector time-series information, wherein the historical surveying deformation time-series data has a periodic factor baseline value, and the deformation prediction vector time-series information belongs to a target time window; performing month-on-month and year-on-year fusion prediction on the periodic factors according to the target time window to obtain predicted values ​​for the periodic factors; when the magnitude of the fluctuation vector between the predicted value of the periodic factor and the baseline value of the periodic factor is greater than... Alternatively, if the value is equal to the fault tolerance fluctuation threshold, retrieve first spatiotemporal sample data that satisfies the fluctuation vector and the spatiotemporal data of the target area, statistically analyze the first deformation deviation vector time series information, and correct the deformation prediction vector time series information to obtain first deformation correction vector time series information; retrieve second spatiotemporal sample data that satisfies the sudden factor, the first deformation correction vector time series information, and the target area spatiotemporal data, statistically analyze the second deformation deviation vector time series information, and correct the first deformation correction vector time series information to obtain second deformation correction vector time series information; add the first deformation correction vector time series information and the second deformation correction vector time series information into the target time window mapping deformation monitoring data.

[0005] In the implementation, a deformation prediction model is used to process historical deformation time-series data and output deformation prediction vector time-series information. This includes: receiving target area distribution information from a target area surveying device up to the Nth period; obtaining a surveying device size accuracy threshold as the grid edge length; traversing the target area distribution information from the first period to the Nth period for grid segmentation to obtain the target area grid distribution information from the first period to the Nth period, where the surveying device size accuracy threshold represents the maximum identifiable accuracy scale of the surveying device; comparing the target area grid distribution information from the first period to the second period to obtain a historical deformation vector for the first time zone; continuing this process until comparing the target area grid distribution information from the (N-1)th period and the Nth period to obtain a historical deformation vector for the (N-1)th time zone; and sequentially concatenating the historical deformation vectors from the first and (N-1)th time zones to generate the historical deformation time-series data.

[0006] In the implementation, a deformation prediction model is used to process historical deformation time-series data and output deformation prediction vector time-series information. This includes: configuring the periodic factor baseline value via a user terminal; collecting selected deformation time-series data of the target area using the periodic factor baseline value as a constraint, wherein the periodic factor deviation of the comparison period of the selected deformation vector in any time zone of the selected deformation time-series data is less than or equal to the periodic factor baseline value; randomly deploying the target time window from the fourth time zone to the last time zone from the selected deformation time-series data, selecting the deformation prediction vector time-series information to supervise the ground truth, using data before the target time window as training input deformation time-series data to train the deformation prediction model, and binding and storing it with the target area and the periodic factor baseline value.

[0007] In the implementation, based on the target time window, a month-on-month and year-on-year fusion prediction is performed on the cyclical factors to obtain predicted values ​​for the cyclical factors. This includes: extracting a first cyclical factor attribute from the cyclical factors; retrieving the first cyclical factor record value of the first cyclical factor attribute; retrieving the month-on-month record value sequence of the first cyclical factor based on the first cyclical factor record value, wherein the last record value of the month-on-month record value sequence of the first cyclical factor is the previous month-on-month period; retrieving the year-on-year record value sequence of the first cyclical factor based on the first cyclical factor record value, wherein the last record value of the year-on-year record value sequence of the first cyclical factor is the previous year-on-year period; and statistically analyzing the first cyclical factor month-on-month growth rate feature vector of the first cyclical factor month-on-month record value sequence as supervisory data, using the first cyclical factor... Using the month-on-month record value sequence as input data, a fully connected neural network is trained to generate a first feature vector extraction branch. The first year-on-year growth rate feature vector of the first periodic factor record value sequence is used as supervised data. Using the first year-on-year record value sequence of the first periodic factor as input data, a fully connected neural network is trained to generate a second feature vector extraction branch. Using the first and second feature vector extraction branches as input, and the first periodic factor record value as supervision, a fully connected neural network is trained to generate a month-on-month and year-on-year fusion backbone. The first feature vector extraction branch, the second feature vector extraction branch, and the month-on-month and year-on-year fusion backbone are merged to generate a periodic factor prediction model. Based on the target time window, a month-on-month and year-on-year fusion prediction is performed on the periodic factor to obtain the predicted value of the periodic factor.

[0008] In the implementation method, the feature vector of the first periodic factor month-on-month growth rate of the first periodic factor in the first periodic factor month-on-month record value sequence is used as supervised data, including: statistically analyzing the first periodic factor month-on-month growth rate vector sequence of the first periodic factor month-on-month record value sequence; performing outlier vector deletion on the first periodic factor month-on-month growth rate vector sequence to obtain a representative growth rate vector set; and performing mean analysis on the representative growth rate vector set to obtain the feature vector of the first periodic factor month-on-month growth rate.

[0009] In the implementation method, obtaining the periodic factors and sudden factors affecting the mapping deformation of the target area includes: configuring the periodic factors and initial sudden factors through the user terminal; traversing the initial sudden factors, retrieving the historical sudden frequency in the target area, and adding the initial sudden factors whose historical sudden frequency is greater than or equal to the sudden frequency threshold to the sudden factors; collecting several periodic factor record values ​​for several periods in the target area, performing same-attribute central tendency analysis, and obtaining representative values ​​of the periodic factors; calculating the set of fluctuation vector magnitude values ​​of the representative values ​​of the periodic factors and the baseline values ​​of the periodic factors; performing outlier analysis on the set of fluctuation vector magnitude values ​​to obtain a set of fluctuation vector magnitude outlier factors; based on the set of fluctuation vector magnitude outlier factors, extracting the maximum value of the selected fluctuation vector magnitude values ​​from the set of fluctuation vector magnitude values ​​where the fluctuation vector magnitude outlier factors are less than or equal to the outlier factor threshold, and setting it as the fault-tolerant fluctuation threshold.

[0010] In the implementation method, first spatiotemporal sample data satisfying the fluctuation vector and target area spatiotemporal data are retrieved, and the time series information of the first deformation deviation vector is statistically analyzed, including: extracting target area temporal feature data and target area spatial feature data from the target area spatiotemporal data, wherein the temporal feature represents at least five years of surveying deformation data and periodic factor benchmark values, and the spatial feature represents geological structure data and climate data; obtaining sample historical surveying deformation data, wherein the sample historical surveying deformation data has preset periodic factor benchmark values ​​and preset climate data; randomly dividing the sample historical surveying deformation data into two parts to obtain previous sample historical surveying deformation data and subsequent sample historical surveying deformation data; extracting the geological structure data of the last sample of the previous sample historical surveying deformation data. The structural data involves extracting at least five years of previous sample mapping deformation data from the previous sample historical mapping deformation data, truncated from the tail to the head. When the previous sample mapping deformation data matches the current mapping deformation data, and the periodic factor benchmark value matches the preset periodic factor benchmark value, and the geological structure data matches the geological structure data of the last sample, and the climate data matches the preset climate data, the subsequent sample historical mapping deformation data is added to the first spatiotemporal sample data. The previous sample historical mapping deformation data is processed using the mapping deformation prediction model to obtain the temporal information of the sample deformation prediction vector. The first spatiotemporal sample data is compared with the temporal information of the sample deformation prediction vector to obtain the temporal information of the first deformation deviation vector.

[0011] This invention also provides a mapping deformation monitoring system based on spatiotemporal deep learning, comprising: a deformation factor acquisition module, used to acquire periodic factors and sudden factors affecting mapping deformation in a target area, wherein the periodic factors have a fault-tolerant fluctuation threshold; a deformation prediction module, used to process historical mapping deformation time-series data through a mapping deformation prediction model and output deformation prediction vector time-series information, wherein the historical mapping deformation time-series data has a periodic factor baseline value, and the deformation prediction vector time-series information belongs to a target time window; a periodic factor prediction module, used to perform month-on-month and year-on-year fusion prediction on the periodic factors according to the target time window to obtain the periodic factor prediction value; and a periodic factor deviation acquisition module, used to acquire the deviation when the fluctuation of the periodic factor prediction value and the periodic factor baseline value is... If the magnitude of the quantity is greater than or equal to the fault tolerance fluctuation threshold, first spatiotemporal sample data that satisfies the fluctuation vector and the spatiotemporal data of the target area are retrieved, the time series information of the first deformation deviation vector is statistically analyzed, and the time series information of the deformation prediction vector is corrected to obtain the time series information of the first deformation correction vector; the sudden factor deviation acquisition module is used to retrieve second spatiotemporal sample data that satisfies the sudden factor, the time series information of the first deformation correction vector, and the spatiotemporal data of the target area, statistically analyze the time series information of the second deformation deviation vector, and correct the time series information of the first deformation correction vector to obtain the time series information of the second deformation correction vector; the deformation correction module is used to add the time series information of the first deformation correction vector and the time series information of the second deformation correction vector into the deformation monitoring data of the target time window.

[0012] A computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the above-described method.

[0013] A computer device includes a memory, a processor, and a program stored in the memory and executable thereon, the program being executed by the processor to implement the steps of the method described above.

[0014] The advantages and technical effects of this invention are as follows: The surveying deformation monitoring method and system based on spatiotemporal deep learning proposed in this invention obtains the periodic and sudden factors affecting surveying deformation. Based on the prediction model, it outputs the time-series information of the deformation prediction vector. It obtains the predicted value of the periodic factor; when the fluctuation vector is greater than or equal to a threshold, it retrieves the first spatiotemporal sample data, statistically analyzes the time-series information of the first deformation deviation vector, and corrects the time-series information of the deformation prediction vector to obtain the time-series information of the first deformation correction vector. It then statistically analyzes the time-series information of the second deformation deviation vector and corrects the time-series information of the first deformation correction vector. Finally, it adds the time-series information of the deformation correction vector to the surveying deformation monitoring data within the target time window. This solves the technical problem of low deformation prediction accuracy in existing surveying deformation monitoring methods under complex environments. It achieves year-on-year and month-on-month fusion prediction of periodic factors, fault-tolerant fluctuation judgment, and multi-round deviation correction based on similar samples, thereby effectively improving the accuracy of surveying deformation prediction and the response capability to abnormal disturbances, and enhancing the accuracy of surface deformation monitoring in complex environments. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this invention to illustrate the operations performed by the system according to the embodiments of the present invention. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0016] Figure 1 A schematic diagram of the mapping deformation monitoring method based on spatiotemporal deep learning provided in an embodiment of the present invention;

[0017] Figure 2 This is a schematic diagram of the structure of a mapping deformation monitoring system based on spatiotemporal deep learning, provided in an embodiment of the present invention.

[0018] In the diagram: 11 Deformation factor acquisition module, 12 Deformation prediction module, 13 Periodic factor prediction module, 14 Periodic factor deviation acquisition module, 15 Sudden factor deviation acquisition module, 16 Deformation correction module. Detailed Implementation

[0019] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below.

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" can be the same or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only.

[0022] This invention provides a mapping deformation monitoring method and system based on spatiotemporal deep learning, such as... Figure 1 As shown, the method includes:

[0023] The system obtains periodic and sudden factors affecting the deformation of the target area, wherein the periodic factors have a fault-tolerant fluctuation threshold; through a deformation prediction model, it processes historical deformation time-series data and outputs deformation prediction vector time-series information, wherein the historical deformation time-series data has a baseline value for the periodic factors, and the deformation prediction vector time-series information belongs to the target time window; based on the target time window, it performs month-on-month and year-on-year fusion prediction on the periodic factors to obtain the predicted values ​​of the periodic factors.

[0024] The system identifies periodic and sudden factors influencing the deformation of the target area. The periodic factors, which are influencing factors exhibiting periodic variation patterns (such as seasonal temperature changes, river water level changes, and seasonal precipitation changes), and sudden factors (such as earthquakes, rainstorms, and construction blasting), each have a tolerance threshold. This tolerance threshold represents the allowable deviation range between the predicted and baseline values ​​of the periodic factors; exceeding this threshold indicates abnormal fluctuations in the predicted values. Furthermore, the system processes historical time-series data of the deformation, acquired through long-term surveying and observation at fixed surveying cycles. This data records the deformation data obtained over consecutive measurement cycles and arranged chronologically. The historical time-series data is input into the deformation prediction model, outputting a deformation prediction vector time-series information. The historical time-series data includes a baseline value for the periodic factors, and the deformation prediction vector time-series information belongs to the target time window. Furthermore, based on the target time window, a month-on-month and year-on-year fusion forecast is performed on the cyclical factors to obtain the predicted values ​​of the cyclical factors.

[0025] The method provided in this embodiment of the invention further includes: receiving target area distribution information from a target area mapping device for a first period up to the Nth period; obtaining a mapping device size accuracy threshold as the grid edge length; traversing the target area distribution information for the first period up to the Nth period to perform grid segmentation to obtain target area grid distribution information for the first period up to the Nth period, wherein the mapping device size accuracy threshold characterizes the maximum accuracy scale that the mapping device can identify; comparing the target area grid distribution information for the first period and the target area grid distribution information for the second period to obtain a historical mapping deformation vector for the first time zone; comparing the target area grid distribution information for the (N-1)th period and the target area grid distribution information for the Nth period to obtain a historical mapping deformation vector for the (N-1)th time zone; sequentially concatenating the historical mapping deformation vector for the first time zone and the historical mapping deformation vector for the (N-1)th time zone to generate the historical mapping deformation time series data.

[0026] By using a deformation prediction model, historical deformation time-series data is processed to output deformation prediction vector time-series information. This process includes: data acquisition at fixed monitoring cycles from surveying equipment (such as a terrestrial laser scanner or synthetic aperture radar) positioned in the target area; and receiving target area distribution information from the first cycle up to the Nth cycle, where N is a positive integer greater than 1. For example, from January 2020 to December 2024, regular observations are conducted quarterly to obtain the target area spatial distribution information from the first quarter of 2020 (the first cycle) to the fourth quarter of 2024 (the Nth cycle). Further, based on the minimum size accuracy limit of the surveying equipment itself, a surveying equipment size accuracy threshold is obtained. Using this threshold as the grid edge length, the target area distribution information from the first cycle up to the Nth cycle is divided into grids to obtain the target area grid distribution information from the first cycle up to the Nth cycle. The surveying equipment size accuracy threshold represents the maximum accuracy scale that the surveying equipment can identify, i.e., the highest accuracy scale.

[0027] Furthermore, by comparing the grid distribution information of the target area in the first period and the grid distribution information of the target area in the second period, the spatial positional difference between the two periods is calculated. Specifically, a reference point can be obtained as the benchmark point for calculating the spatial positional difference. The first and second positional coordinates of the reference point in the grid distribution information of the target area in the first and second periods are obtained. The difference is calculated based on the positional coordinates to obtain the historical mapping deformation vector of the first time zone. The same method is used to continue comparing the second and third periods until the grid distribution information of the target area in the (N-1)th period and the grid distribution information of the target area in the Nth period are compared to obtain the historical mapping deformation vector of the (N-1)th time zone. Finally, in chronological order, the historical mapping deformation vector of the first time zone is used as the starting point, and the historical mapping deformation vectors of the second and third time zones are sequentially concatenated until the historical mapping deformation vector of the (N-1)th time zone is generated to produce the historical mapping deformation time series data. The stitched historical mapping deformation time series data completely records the spatial changes of the target area from the first cycle to the Nth cycle, and can reflect the continuous trend of deformation of the mapping area in each time interval.

[0028] The method provided in this embodiment of the invention further includes: configuring the periodic factor benchmark value through a user terminal; collecting selected mapping deformation time series data of the target area under the constraint of the periodic factor benchmark value, wherein the periodic factor deviation of the comparison period of the selected mapping deformation vector in any time zone of the selected mapping deformation time series data is less than or equal to the periodic factor benchmark value; randomly deploying the target time window from the fourth time zone to the last time zone from the selected mapping deformation time series data, selecting the deformation prediction vector time series information supervision ground value, using the data before the target time window as the training input mapping deformation time series data, training the mapping deformation prediction model, and binding and storing it with the target area and the periodic factor benchmark value.

[0029] The deformation prediction model processes historical deformation time-series data and outputs deformation prediction vector time-series information. This includes configuring the periodic factor baseline value via a user terminal. Subsequently, using the periodic factor baseline value as a constraint, each periodic data point is compared against the historical deformation time-series data. The deformation data for each time zone is compared with the periodic factor baseline value, and deformation data less than or equal to the periodic factor baseline value is obtained, thus acquiring the selected deformation time-series data.

[0030] Finally, since the input data for the mapping deformation prediction model must include data from at least three consecutive time zones, a target time window is randomly determined from the selected mapping deformation time series data, starting from the fourth time zone and continuing until the end of the time series data. The output of the prediction model is corrected using the supervised ground truth of the data deformation prediction vector time series information within the target time window. Data prior to the target time window is used as the training input mapping deformation time series data to obtain the training input for the mapping deformation prediction model, which is then trained. During training, the training input mapping deformation time series data is input into the prediction model. The prediction error is calculated by comparing the supervised ground truth with the model's output prediction results. The model parameters are then optimized and adjusted based on the prediction error, and multiple rounds of iterative training are performed until the model reaches the ideal accuracy. After training, the trained mapping deformation prediction model is bound and stored with the target area and periodic factor benchmark values, facilitating subsequent calls to the corresponding target area model and improving monitoring efficiency and prediction reliability.

[0031] The method provided in this embodiment of the invention further includes: extracting a first periodic factor attribute from the periodic factor; retrieving the first periodic factor record value of the first periodic factor attribute; retrieving a first periodic factor month-on-month record value sequence based on the first periodic factor record value, wherein the last record value of the first periodic factor month-on-month record value sequence is the previous month-on-month period of the first periodic factor record value; retrieving a first periodic factor year-on-year record value sequence based on the first periodic factor record value, wherein the last record value of the first periodic factor year-on-year record value sequence is the previous year-on-year period of the first periodic factor record value; and using the first periodic factor month-on-month record value sequence as input data, training a fully connected circuit. A neural network is connected to generate a first feature vector extraction branch. The first year-on-year growth rate feature vector of the first periodic factor is statistically analyzed as the first periodic factor year-on-year record value sequence as supervised data. Using the first periodic factor year-on-year record value sequence as input data, a fully connected neural network is trained to generate a second feature vector extraction branch. Using the first and second feature vector extraction branches as inputs and the first periodic factor record values ​​as supervision, a fully connected neural network is trained to generate a month-on-month and year-on-year fusion backbone. The first feature vector extraction branch, the second feature vector extraction branch, and the month-on-month and year-on-year fusion backbone are merged to generate a periodic factor prediction model. Based on the target time window, month-on-month and year-on-year fusion prediction is performed on the periodic factor to obtain the predicted value of the periodic factor.

[0032] Based on the target time window, a month-on-month and year-on-year fusion prediction is performed on the cyclical factors to obtain predicted values ​​for the cyclical factors. This includes: acquiring the cyclical factors in the prediction analysis, such as seasonal temperature changes, river water level changes, and seasonal precipitation changes. From the cyclical factors, a first cyclical factor attribute is extracted. The first cyclical factor attribute is a specific influencing indicator of the cyclical factor. Taking seasonal precipitation changes as an example, the corresponding first cyclical factor attribute can be the rainfall amount within a fixed period. Further, the first cyclical factor recorded value of the first cyclical factor attribute is retrieved. The first cyclical factor recorded value is the historical time window cyclical factor recorded value for which the prediction needs to be performed.

[0033] Based on the first periodic factor recorded value, the month-on-month recorded value sequence of the first periodic factor is retrieved. This sequence comprises multiple consecutive recorded periods preceding the first periodic factor recorded value, with the last record in the sequence representing the preceding month-on-month period. For example, if the first periodic factor recorded value is 80 mm of rainfall in April 2025, its month-on-month sequence could include rainfall data from several consecutive months, such as March, February, and January 2025. Based on the first periodic factor recorded value, the year-on-year recorded value sequence of the first periodic factor is retrieved. This sequence comprises periodic factor data at the same or similar time points as the first periodic factor recorded value, with the last record in the year-on-year sequence representing the preceding year-on-year period. For example, if the first periodic factor recorded value is 80 mm of rainfall in April 2025, the year-on-year recorded value sequence would include rainfall data from May 2024, May 2023, and May 2022.

[0034] The month-on-month growth rate sequence of the first-period factor's month-on-month record value sequence is statistically analyzed across consecutive year-on-year periods. Then, outlier filtering and mean analysis are performed on the month-on-month growth rate sequence to obtain a representative feature vector of the first-period factor's month-on-month growth rate. Using this feature vector as supervised data and the first-period factor's month-on-month record value sequence as input data, a fully connected neural network is trained to generate the first feature vector extraction branch.

[0035] The year-on-year growth rate sequence of the first-period factor's year-on-year record value sequence is statistically analyzed across consecutive year-on-year periods. Outlier filtering and mean analysis are performed on the year-on-year growth rate sequence to obtain a representative feature vector of the first-period factor's year-on-year growth rate. Using the feature vector of the first-period factor's year-on-year growth rate as supervised data and the first-period factor's year-on-year record value sequence as input data, a fully connected neural network is trained to generate a second feature vector extraction branch.

[0036] Further, using the first feature vector extraction branch and the second feature vector extraction branch as inputs, and the first periodic factor record value as supervision, a fully connected neural network is trained under supervision to generate a month-on-month and year-on-year fusion backbone. Finally, the first feature vector extraction branch, the second feature vector extraction branch, and the month-on-month and year-on-year fusion backbone are merged to generate a periodic factor prediction model. Through the target time window, month-on-month and year-on-year fusion prediction is performed on the periodic factors, obtaining the month-on-month growth rate feature vector and the year-on-year growth rate feature vector output by the first feature vector extraction branch and the second feature vector extraction branch, respectively. The month-on-month product of the month-on-month growth rate feature vector and the previous periodic factor value in the target time window is obtained. The month-on-month product calculation result is summed and compared with the previous month-on-month periodic factor value to obtain the month-on-month periodic factor prediction value. The year-on-year product of the year-on-year growth rate feature vector and the previous year-on-year periodic factor value in the target time window is obtained. The year-on-year product calculation result is summed and compared with the previous year-on-year periodic factor value to obtain the year-on-year periodic factor prediction value. The mean of the month-on-month periodic factor prediction value and the year-on-year periodic factor prediction value is calculated to obtain the periodic factor prediction value.

[0037] The method provided in this embodiment of the invention further includes: statistically analyzing the first periodic factor month-on-month growth rate vector sequence of the first periodic factor month-on-month record value sequence; performing outlier vector deletion on the first periodic factor month-on-month growth rate vector sequence to obtain a representative growth rate vector set; and performing mean analysis on the representative growth rate vector set to obtain the first periodic factor month-on-month growth rate feature vector.

[0038] The first-period factor month-on-month growth rate feature vector is used as supervised data in the first-period factor month-on-month record value sequence. This includes: calculating the first-period factor month-on-month growth rate vector sequence in the first-period factor month-on-month record value sequence, where the first-period factor month-on-month growth rate vector sequence is a sequence formed by vector transformation of the year-on-year growth rates between consecutive year-on-year periods. Outlier vector deletion is performed on the first-period factor month-on-month growth rate vector sequence to remove data points that are significantly far removed from other data, obtaining a representative growth rate vector set. Finally, the mean is obtained from the representative growth rate vector set to obtain the first-period factor month-on-month growth rate feature vector.

[0039] The method provided in this embodiment of the invention further includes: configuring the periodic factors and initial sudden factors through a user terminal; traversing the initial sudden factors, retrieving historical sudden frequencies in the target area, and adding the initial sudden factors whose historical sudden frequencies are greater than or equal to a sudden frequency threshold to the sudden factors; collecting several periodic factor record values ​​for several periods in the target area, performing same-attribute central tendency analysis, and obtaining representative values ​​of the periodic factors; calculating the set of fluctuation vector magnitude values ​​of the representative values ​​of the periodic factors and the baseline values ​​of the periodic factors; performing outlier analysis on the set of fluctuation vector magnitude values ​​to obtain a set of fluctuation vector magnitude outlier factors; based on the set of fluctuation vector magnitude outlier factors, extracting the maximum value of the selected fluctuation vector magnitude values ​​from the set of fluctuation vector magnitude values ​​where the fluctuation vector magnitude outlier factors are less than or equal to an outlier factor threshold, and setting it as the fault-tolerant fluctuation threshold.

[0040] The user inputs and configures the periodic factors and initial outbreak factors through the user terminal. The initial outbreak factors are traversed, and their historical outbreak frequencies in the target area are retrieved. If the historical trigger frequency of an initial outbreak factor is lower than a outbreak frequency threshold, its probability of occurrence is low. Initial outbreak factors with historical outbreak frequencies greater than or equal to the outbreak frequency threshold are added to the outbreak factors. The outbreak frequency threshold is a preset minimum frequency threshold; if the trigger frequency is greater than or equal to this threshold, it is added to the outbreak factors. Factor record values ​​for multiple periods in the target area are collected to form a time series. Subsequently, a common-attribute central tendency analysis is performed, and data from several periodic factor record values ​​are obtained as representative values ​​of the periodic factors based on the 3σ principle of normal distribution. The set of fluctuation vector magnitudes between the representative values ​​of the periodic factors and the baseline values ​​of the periodic factors is calculated. The set of fluctuation vector magnitudes represents the absolute value of the deviation between the periodic factor record values ​​and the baseline values. Outlier analysis is performed on the set of fluctuation vector magnitudes to obtain outlier parameters, resulting in a set of fluctuation vector magnitude outlier factors. Based on the set of outlier factors for the volatility vector magnitude, the maximum value of the selected volatility vector magnitude that is less than or equal to the outlier factor threshold is extracted from the set of volatility vector magnitudes. That is, the maximum value of the selected volatility vector magnitude that is less than or equal to the outlier factor threshold in the set of volatility vector magnitudes is obtained and set as the fault-tolerant volatility threshold. The fault-tolerant volatility threshold is a positive and negative parameter range. The outlier factor threshold is a preset maximum selected volatility vector magnitude. If the value is greater than this threshold, the corresponding data has poor reliability and may be due to data anomalies caused by other reasons.

[0041] When the magnitude of the fluctuation vector between the predicted value of the periodic factor and the baseline value of the periodic factor is greater than or equal to the fault-tolerant fluctuation threshold, first spatiotemporal sample data that satisfies the fluctuation vector and the spatiotemporal data of the target area are retrieved, the time series information of the first deformation deviation vector is statistically analyzed, and the time series information of the deformation prediction vector is corrected to obtain the time series information of the first deformation correction vector; second spatiotemporal sample data that satisfies the sudden factor, the time series information of the first deformation correction vector, and the spatiotemporal data of the target area are retrieved, the time series information of the second deformation deviation vector is statistically analyzed, and the time series information of the first deformation correction vector is corrected to obtain the time series information of the second deformation correction vector; the time series information of the first deformation correction vector and the time series information of the second deformation correction vector are added to the deformation monitoring data of the target time window.

[0042] When the magnitude of the fluctuation vector of any attribute between the predicted value of the periodic factor and the baseline value of the periodic factor is greater than or equal to the fault-tolerant fluctuation threshold, the deviation between the two is large, and the data may be abnormal. In this case, first spatiotemporal sample data that satisfies the fluctuation vector and the spatiotemporal data of the target region is retrieved, and the time series information of the first deformation deviation vector is statistically analyzed. The time series information of the deformation prediction vector is then compensated and corrected based on the time series information of the first deformation deviation vector to obtain a first-order deformation correction vector time series information, thereby obtaining more accurate prediction data.

[0043] The acquired primary deformation correction vector time series information already considers the influence of periodic factors. Therefore, it is necessary to introduce the influence of sudden factors. By retrieving historical data similar to the current sudden factor situation, the primary deformation correction vector time series information, and the spatiotemporal data of the target area, this data is used as the second spatiotemporal sample data, i.e., the actually collected data. The second deformation deviation vector time series information generated under the influence of the same sudden factors is statistically analyzed. This second deformation deviation vector time series information is used to correct the deviation influence of the primary deformation correction vector time series information; that is, the second deformation deviation vector time series information is simultaneously compensated within the primary deformation correction vector time series information to obtain the secondary deformation correction vector time series information. Finally, the primary and secondary deformation correction vector time series information are added to the deformation monitoring data within the target time window. This solves the technical problem of low deformation prediction accuracy in existing deformation monitoring methods under complex environments. It achieves month-on-month and year-on-year fusion prediction of cyclical factors, fault-tolerant fluctuation judgment, and multi-round deviation correction based on similar samples, thereby effectively improving the accuracy of surveying and mapping deformation prediction and the ability to respond to abnormal disturbances, and enhancing the accuracy of surface deformation monitoring in complex environments.

[0044] The method provided in this embodiment of the invention further includes: extracting temporal feature data and spatial feature data of the target area from the spatiotemporal data of the target area, wherein the temporal features represent at least five years of surveying and mapping deformation data and periodic factor benchmark values, and the spatial features represent geological structure data and climate data; obtaining sample historical surveying and mapping deformation data, wherein the sample historical surveying and mapping deformation data has preset periodic factor benchmark values ​​and preset climate data; randomly dividing the sample historical surveying and mapping deformation data into two parts to obtain previous sample historical surveying and mapping deformation data and subsequent sample historical surveying and mapping deformation data; extracting the geological structure data of the last sample of the previous sample historical surveying and mapping deformation data, and processing the previous sample historical surveying and mapping deformation data... At least five years of preceding sample mapping deformation data are extracted from the tail to the head. When the preceding sample mapping deformation data is consistent with the mapping deformation data, and the periodic factor benchmark value is consistent with the periodic factor preset benchmark value, and the geological structure data is consistent with the geological structure data of the last sample, and the climate data is consistent with the climate preset data, the historical mapping deformation data of the subsequent sample is added to the first spatiotemporal sample data. The preceding sample historical mapping deformation data is processed through the mapping deformation prediction model to obtain the sample deformation prediction vector time series information. The first spatiotemporal sample data is compared with the sample deformation prediction vector time series information to obtain the first deformation deviation vector time series information.

[0045] Retrieve first spatiotemporal sample data that satisfies the fluctuation vector and the spatiotemporal data of the target area, and statistically analyze the time-series information of the first deformation deviation vector, including: extracting the temporal feature data and spatial feature data of the target area from the spatiotemporal data of the target area, wherein the temporal feature represents at least five years of surveying deformation data and periodic factor benchmark values, and the spatial feature represents geological structure data and climate data. Further, obtain sample historical surveying deformation data, which is recorded data ordered chronologically and includes preset benchmark values ​​for periodic factors and preset climate data. Perform random binary splitting on the sample historical surveying deformation data to divide it into two parts, obtaining preceding sample historical surveying deformation data and subsequent sample historical surveying deformation data. Further, extract the geological structure data of the last sample from the preceding sample historical surveying deformation data, and extract at least five years of preceding sample surveying deformation data from the preceding sample historical surveying deformation data from tail to head. When the deformation data of the preceding sample is consistent with the deformation data of the target area spatiotemporal data, the periodic factor benchmark value is consistent with the preset benchmark value of the periodic factor, the geological structure data is consistent with the geological structure data of the last sample, and the climate data is consistent with the preset climate data, that is, when the deformation data, the preset benchmark value of the periodic factor, and the geological structure data are all consistent, the historical deformation data of the subsequent sample is added to the first spatiotemporal sample data.

[0046] At this point, based on the above judgment that the historical mapping deformation data of the preceding and subsequent samples have a high similarity to the spatiotemporal data of the target area, the mapping deformation prediction model is used to process the historical mapping deformation data of the preceding samples to obtain the temporal information of the sample deformation prediction vector, thereby realizing the prediction of sample deformation. Finally, by comparing the differences between the first spatiotemporal sample data and the mapping deformation data in the temporal information of the sample deformation prediction vector, the temporal information of the first deformation deviation vector is obtained. The temporal information of the first deformation deviation vector is used to reflect the deformation deviation parameters generated under the influence of corresponding periodic factors.

[0047] In the above text, refer to Figure 1 A mapping deformation monitoring method based on spatiotemporal deep learning according to embodiments of the present invention is described in detail. Next, reference will be made to... Figure 2 A mapping deformation monitoring system based on spatiotemporal deep learning according to an embodiment of the present invention is described.

[0048] The mapping deformation monitoring system based on spatiotemporal deep learning according to embodiments of the present invention solves the technical problem of low deformation prediction accuracy in existing mapping deformation monitoring methods under complex environments. It achieves year-on-year and month-on-month fusion prediction of periodic factors, fault-tolerant fluctuation judgment, and multi-round deviation correction based on similar samples, thereby effectively improving the accuracy of mapping deformation prediction and its response to abnormal disturbances, and enhancing the accuracy of surface deformation monitoring in complex environments. The mapping deformation monitoring system based on spatiotemporal deep learning includes: a deformation factor acquisition module 11, a deformation prediction module 12, a periodic factor prediction module 13, a periodic factor deviation acquisition module 14, a sudden factor deviation acquisition module 15, and a deformation correction module 16.

[0049] The deformation factor acquisition module 11 is used to acquire periodic factors and sudden factors that affect the mapping deformation of the target area, wherein the periodic factors have a fault-tolerant fluctuation threshold.

[0050] The deformation prediction module 12 is used to process historical surveyed deformation time series data through a surveyed deformation prediction model and output deformation prediction vector time series information. The historical surveyed deformation time series data has a periodic factor reference value, and the deformation prediction vector time series information belongs to the target time window.

[0051] The cyclical factor prediction module 13 is used to perform month-on-month and year-on-year fusion prediction on the cyclical factors according to the target time window to obtain the predicted value of the cyclical factors;

[0052] The periodic factor deviation acquisition module 14 is used to retrieve first spatiotemporal sample data that satisfies the fluctuation vector and the target area spatiotemporal data when the magnitude of the fluctuation vector of the predicted periodic factor value and the baseline periodic factor value is greater than or equal to the fault tolerance fluctuation threshold, statistically analyze the first deformation deviation vector time series information, correct the deformation prediction vector time series information, and obtain a first deformation correction vector time series information.

[0053] The sudden factor deviation acquisition module 15 is used to retrieve second spatiotemporal sample data that satisfies the sudden factor, the first deformation correction vector time series information and the target area spatiotemporal data, statistically analyze the second deformation deviation vector time series information, correct the first deformation correction vector time series information, and obtain the second deformation correction vector time series information.

[0054] The deformation correction module 16 is used to add the timing information of the primary deformation correction vector and the timing information of the secondary deformation correction vector into the deformation monitoring data of the target time window.

[0055] The specific configuration of the deformation prediction module 12 will be described in detail below. The deformation prediction module 12 may further include: processing historical surveying deformation time-series data through a surveying deformation prediction model, and outputting deformation prediction vector time-series information, which includes: receiving target area distribution information from a target area surveying device up to the Nth period; obtaining a surveying device size accuracy threshold as the grid edge length, traversing the target area distribution information from the first period to the Nth period to perform grid segmentation, obtaining the target area grid distribution information from the first period to the Nth period, where the surveying device size accuracy threshold characterizes the maximum accuracy scale that the surveying device can identify; comparing the target area grid distribution information from the first period to the second period to obtain a historical surveying deformation vector for the first time zone; comparing the target area grid distribution information from the (N-1)th period and the target area grid distribution information from the Nth period to obtain a historical surveying deformation vector for the (N-1)th time zone; and sequentially concatenating the historical surveying deformation vector from the first time zone and the historical surveying deformation vector from the (N-1)th time zone to generate the historical surveying deformation time-series data.

[0056] The specific configuration of the deformation prediction module 12 will be described in detail below. The deformation prediction module 12 may further include: processing historical surveying deformation time-series data through a surveying deformation prediction model, and outputting deformation prediction vector time-series information, including: configuring the periodic factor benchmark value through a user terminal; collecting selected surveying deformation time-series data of the target area under the constraint of the periodic factor benchmark value, wherein the periodic factor deviation of the comparison period of the selected surveying deformation vector in any time zone of the selected surveying deformation time-series data is less than or equal to the periodic factor benchmark value; randomly deploying the target time window from the fourth time zone to the last time zone from the selected surveying deformation time-series data, selecting the deformation prediction vector time-series information supervision ground truth, using the data before the target time window as the training input surveying deformation time-series data, training the surveying deformation prediction model, and binding and storing it with the target area and the periodic factor benchmark value.

[0057] The specific configuration of the cyclical factor prediction module 13 will be described in detail below. The cyclical factor prediction module 13 may further include: performing month-on-month and year-on-year fusion prediction on the cyclical factors according to the target time window to obtain predicted values ​​for the cyclical factors, including: extracting a first cyclical factor attribute from the cyclical factors; retrieving the first cyclical factor record value of the first cyclical factor attribute; retrieving the first cyclical factor month-on-month record value sequence based on the first cyclical factor record value, wherein the last record value of the first cyclical factor month-on-month record value sequence is the previous month-on-month period of the first cyclical factor record value; retrieving the first cyclical factor year-on-year record value sequence based on the first cyclical factor record value, wherein the last record value of the first cyclical factor year-on-year record value sequence is the previous year-on-year period of the first cyclical factor record value; and statistically analyzing the first cyclical factor month-on-month growth rate feature vector of the first cyclical factor month-on-month record value sequence as supervisory data, using the... Using the first cyclical factor month-on-month record value sequence as input data, a fully connected neural network is trained to generate a first feature vector extraction branch. The first cyclical factor year-on-year growth rate feature vector of the first cyclical factor year-on-year record value sequence is used as supervised data. Using the first cyclical factor year-on-year record value sequence as input data, a fully connected neural network is trained to generate a second feature vector extraction branch. Using the first and second feature vector extraction branches as input, and the first cyclical factor record value as supervision, a fully connected neural network is trained to generate a month-on-month and year-on-year fusion backbone. The first feature vector extraction branch, the second feature vector extraction branch, and the month-on-month and year-on-year fusion backbone are merged to generate a cyclical factor prediction model. Based on the target time window, a month-on-month and year-on-year fusion prediction is performed on the cyclical factor to obtain the predicted value of the cyclical factor.

[0058] The specific configuration of the cyclical factor prediction module 13 will be described in detail below. The cyclical factor prediction module 13 further includes: collecting the first cyclical factor month-on-month growth rate feature vector as supervised data from the first cyclical factor month-on-month record value sequence, including: collecting the first cyclical factor month-on-month growth rate vector sequence from the first cyclical factor month-on-month record value sequence; performing outlier vector deletion on the first cyclical factor month-on-month growth rate vector sequence to obtain a representative growth rate vector set; and performing mean analysis on the representative growth rate vector set to obtain the first cyclical factor month-on-month growth rate feature vector.

[0059] The specific configuration of the deformation factor acquisition module 11 will be described in detail below. The deformation factor acquisition module 11 further includes: obtaining periodic factors and sudden factors affecting the mapping deformation of the target area, including: configuring the periodic factors and initial sudden factors through a user terminal; traversing the initial sudden factors, retrieving historical sudden frequencies in the target area, and adding the initial sudden factors whose historical sudden frequencies are greater than or equal to a sudden frequency threshold to the sudden factors; collecting several periodic factor record values ​​for several periods in the target area, performing same-attribute central tendency analysis to obtain representative values ​​of the periodic factors; calculating the set of fluctuation vector magnitude values ​​of the representative values ​​of the periodic factors and the baseline values ​​of the periodic factors; performing outlier analysis on the set of fluctuation vector magnitude values ​​to obtain a set of fluctuation vector magnitude outlier factors; based on the set of fluctuation vector magnitude outlier factors, extracting the maximum value of selected fluctuation vector magnitude values ​​whose fluctuation vector magnitude outlier factors are less than or equal to an outlier factor threshold from the set of fluctuation vector magnitude values, and setting it as the fault-tolerant fluctuation threshold.

[0060] The specific configuration of the periodic factor deviation acquisition module 14 will be described in detail below. The periodic factor deviation acquisition module 14 further includes: retrieving first spatiotemporal sample data that satisfies the fluctuation vector and the spatiotemporal data of the target area; and statistically analyzing the time-series information of the first deformation deviation vector, including: extracting target area temporal feature data and target area spatial feature data from the target area spatiotemporal data, wherein the temporal features represent at least five years of surveying deformation data and periodic factor benchmark values, and the spatial features represent geological structure data and climate data; obtaining sample historical surveying deformation data, wherein the sample historical surveying deformation data has preset periodic factor benchmark values ​​and preset climate data; randomly dividing the sample historical surveying deformation data into two parts to obtain previous sample historical surveying deformation data and subsequent sample historical surveying deformation data; and extracting the previous sample historical surveying deformation data. The geological structure data of the final sample is obtained by extracting at least five years of previous sample mapping deformation data from the previous sample historical mapping deformation data from tail to head. When the previous sample mapping deformation data is consistent with the mapping deformation data, and the periodic factor benchmark value is consistent with the periodic factor preset benchmark value, and the geological structure data is consistent with the final sample geological structure data, and the climate data is consistent with the climate preset data, the subsequent sample historical mapping deformation data is added to the first spatiotemporal sample data. The previous sample historical mapping deformation data is processed through the mapping deformation prediction model to obtain the sample deformation prediction vector time series information. The first spatiotemporal sample data is compared with the sample deformation prediction vector time series information to obtain the first deformation deviation vector time series information.

[0061] The mapping deformation monitoring system based on spatiotemporal deep learning provided in this embodiment of the invention can execute the mapping deformation monitoring method based on spatiotemporal deep learning provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0062] Although this invention makes various references to certain modules in the system according to embodiments of the present invention, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of the present invention.

[0063] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented, in whole or in part, as a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape) or an optical medium.

[0064] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A spatiotemporal deep learning based mapping deformation monitoring method, characterized in that, include: (1) Obtain the periodic factors and sudden factors that affect the mapping deformation of the target area, wherein the periodic factors have a fault tolerance fluctuation threshold; (2) By using the deformation prediction model, the historical deformation time series data is processed and the deformation prediction vector time series information is output. The historical deformation time series data has a periodic factor reference value and the deformation prediction vector time series information belongs to the target time window. (3) Based on the target time window, perform month-on-month and year-on-year fusion forecasting on the cyclical factors to obtain the predicted values ​​of the cyclical factors, including: Extract the attributes of the first periodic factor from the aforementioned periodic factors; Retrieve the first period factor record value of the first period factor attribute; Based on the first periodic factor record value, retrieve the first periodic factor month-on-month record value sequence, wherein the last record value of the first periodic factor month-on-month record value sequence is the month-on-month period preceding the first periodic factor record value; Based on the first period factor record value, retrieve the first period factor year-on-year record value sequence, wherein the last record value of the first period factor year-on-year record value sequence is the previous year-on-year period before the first period factor record value; The first-period factor month-on-month growth rate feature vector of the first-period factor month-on-month record value sequence is used as supervision data. The first-period factor month-on-month record value sequence is used as input data to train a fully connected neural network and generate the first feature vector extraction branch. The first-period factor year-on-year growth rate feature vector of the first-period factor year-on-year record value sequence is used as supervision data. The first-period factor year-on-year record value sequence is used as input data to train a fully connected neural network and generate a second feature vector extraction branch. Using the first feature vector extraction branch and the second feature vector extraction branch as inputs, and the first periodic factor record value as supervision, a fully connected neural network is trained to generate a month-on-month and year-on-year fusion backbone. Merge the first feature vector extraction branch, the second feature vector extraction branch, and the month-on-month and year-on-year fusion backbone to generate a cyclical factor prediction model. Perform month-on-month and year-on-year fusion prediction on the cyclical factor according to the target time window to obtain the predicted value of the cyclical factor. (4) When the magnitude of the fluctuation vector of the predicted value of the periodic factor and the benchmark value of the periodic factor is greater than or equal to the fault tolerance fluctuation threshold, retrieve the first spatiotemporal sample data that satisfies the fluctuation vector and the spatiotemporal data of the target area, statistically analyze the time series information of the first deformation deviation vector, correct the time series information of the deformation prediction vector, and obtain the time series information of the first deformation correction vector. (5) Retrieve second spatiotemporal sample data that satisfies the sudden factor, the time series information of the first deformation correction vector and the spatiotemporal data of the target area, statistically analyze the time series information of the second deformation deviation vector, correct the time series information of the first deformation correction vector, and obtain the time series information of the second deformation correction vector. (6) Add the timing information of the first deformation correction vector and the timing information of the second deformation correction vector into the target time window to map the deformation monitoring data.

2. The mapping deformation monitoring method based on spatiotemporal deep learning according to claim 1, characterized in that, Step (1) precedes the following: Receive target area distribution information from the target area mapping equipment for the first period up to the target area distribution information for the Nth period; The size accuracy threshold of the surveying equipment is obtained and used as the grid side length. The distribution information of the target area in the first period is traversed until the distribution information of the target area in the Nth period is divided into grids to obtain the grid distribution information of the target area in the first period until the distribution information of the target area in the Nth period. The size accuracy threshold of the surveying equipment represents the maximum accuracy scale that the surveying equipment can identify. By comparing the grid distribution information of the target area in the first period with the grid distribution information of the target area in the second period, the historical mapping deformation vector of the first time zone is obtained; Until the grid distribution information of the target area in the N-1th period is compared with the grid distribution information of the target area in the Nth period, the historical mapping deformation vector of the N-1th time zone is obtained; The historical mapping deformation vectors of the first time zone and the N-1th time zone are sequentially concatenated to generate the historical mapping deformation time series data.

3. The mapping deformation monitoring method based on spatiotemporal deep learning according to claim 2, characterized in that, By using a deformation prediction model, historical deformation time-series data is processed to output deformation prediction vector time-series information, including: Configure the baseline value of the periodic factor through the user terminal; Using the periodic factor reference value as a constraint, selected mapping deformation time series data of the target area are collected, wherein the periodic factor deviation of the comparison period of the selected mapping deformation vector in any time zone of the selected mapping deformation time series data is less than or equal to the periodic factor reference value. From the selected time series data of deformation mapping, from the fourth time zone to the last time zone, the target time window is randomly deployed, the deformation prediction vector time series information supervision ground value is selected, and the data before the target time window is used as the training input time series data of deformation mapping to train the deformation mapping prediction model, and the model is bound and stored with the target area and the periodic factor benchmark value.

4. The mapping deformation monitoring method based on spatiotemporal deep learning according to claim 1, characterized in that, The feature vector of the first-period factor month-on-month growth rate of the first-period factor in the first-period factor month-on-month record value sequence is used as supervised data, including: Statistically analyze the vector sequence of the first-period factor month-on-month growth rate of the first-period factor month-on-month record value sequence; Outlier vectors are removed from the first periodic factor month-on-month growth rate vector sequence to obtain a representative growth rate vector set. Perform mean analysis on the representative growth rate vector set to obtain the characteristic vector of the month-on-month growth rate of the first periodic factor.

5. The mapping deformation monitoring method based on spatiotemporal deep learning according to claim 1, characterized in that, Step (1) includes: Configure the periodic factors and initial sudden factors through the user terminal; Traverse the initial burst factors, retrieve the historical burst frequency in the target area, and add the initial burst factors whose historical burst frequency is greater than or equal to the burst frequency threshold into the burst factors; Collect periodic factor records for several periods in the target area, perform central tendency analysis of the same attribute, and obtain representative values ​​of periodic factors; Calculate the set of fluctuation vector magnitudes of the representative value of the periodic factor and the benchmark value of the periodic factor; Outlier analysis is performed on the set of fluctuation vector magnitudes to obtain the set of outlier factors for fluctuation vector magnitudes; Based on the set of outlier factors of the volatility vector magnitude, the maximum value of the selected volatility vector magnitude that is less than or equal to the outlier factor threshold is extracted from the set of volatility vector magnitudes and set as the fault-tolerant volatility threshold.

6. The mapping deformation monitoring method based on spatiotemporal deep learning according to claim 1, characterized in that, Retrieve first spatiotemporal sample data that satisfies the fluctuation vector and the spatiotemporal data of the target region, and statistically analyze the time series information of the first deformation deviation vector, including: From the spatiotemporal data of the target area, extract the temporal feature data and the spatial feature data of the target area. The temporal feature data represents at least five years of surveying and mapping deformation data and periodic factor benchmark values, while the spatial feature data represents geological structure data and climate data. Obtain historical deformation data of the sample, wherein the historical deformation data of the sample includes preset baseline values ​​for periodic factors and preset climate data; The historical deformation data of the sample is randomly divided into two parts to obtain the historical deformation data of the previous sample and the historical deformation data of the subsequent sample. Extract the geological structure data of the last sample from the historical deformation data of the preceding samples, and extract at least five years of the historical deformation data of the preceding samples from the tail to the head. When the deformation data of the preceding sample is consistent with the deformation data, the baseline value of the periodic factor is consistent with the preset baseline value of the periodic factor, the geological structure data is consistent with the geological structure data of the last sample, and the climate data is consistent with the preset climate data, the historical deformation data of the subsequent sample is added to the first spatiotemporal sample data. The deformation prediction model is used to process historical deformation data of previous samples to obtain temporal information of sample deformation prediction vectors. By comparing the first spatiotemporal sample data with the temporal information of the sample deformation prediction vector, the temporal information of the first deformation deviation vector is obtained.

7. A mapping deformation monitoring system based on spatiotemporal deep learning, characterized in that, The system is used to perform the method according to any one of claims 1-6, the system comprising: The deformation factor acquisition module is used to obtain periodic factors and sudden factors that affect the mapping deformation of the target area, wherein the periodic factors have a fault-tolerant fluctuation threshold; The deformation prediction module is used to process historical deformation time series data through a deformation prediction model and output deformation prediction vector time series information. The historical deformation time series data has a periodic factor baseline value, and the deformation prediction vector time series information belongs to the target time window. The cyclical factor prediction module is used to perform month-on-month and year-on-year fusion prediction on the cyclical factors according to the target time window to obtain the predicted value of the cyclical factors; The periodic factor deviation acquisition module is used to retrieve first spatiotemporal sample data that satisfies the fluctuation vector and the target area spatiotemporal data when the magnitude of the fluctuation vector between the predicted value of the periodic factor and the benchmark value of the periodic factor is greater than or equal to the fault tolerance fluctuation threshold, to statistically analyze the first deformation deviation vector time series information, to correct the deformation prediction vector time series information, and to obtain a first deformation correction vector time series information. The sudden factor deviation acquisition module is used to retrieve second spatiotemporal sample data that satisfies the sudden factor, the first deformation correction vector time series information and the target area spatiotemporal data, statistically analyze the second deformation deviation vector time series information, correct the first deformation correction vector time series information, and obtain the second deformation correction vector time series information. The deformation correction module is used to add the timing information of the primary deformation correction vector and the timing information of the secondary deformation correction vector into the deformation monitoring data of the target time window.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the method as described in any one of claims 1-6.

9. A computer device, characterized in that, The computer device includes a memory, a processor, and a program stored in and executable on the memory, the program being executed by the processor to implement the steps of the method as described in any one of claims 1-6.

Images