Method and apparatus for analyzing land subsidence

Abstract

The application belongs to the technical field of geological analysis, and discloses a ground subsidence analysis method and an analysis device. The analysis method comprises the following steps: matching a sensor arrangement strategy corresponding to the meteorological data of a basic subsidence risk and a monitoring area; collecting ground subsidence data of the monitoring area according to the sensor; performing time-frequency analysis on the ground subsidence data to obtain a ground sound spectrum corresponding to the ground subsidence data; inputting the ground sound spectrum into a preset neural network to output a subsidence semantic vector corresponding to the ground sound spectrum; acquiring and analyzing historical ground subsidence data to construct a subsidence rule corresponding to the historical ground subsidence data; and inputting the subsidence semantic vector and the subsidence rule into a preset space-time analysis model to predict a future subsidence trend of the monitoring area. The method disclosed by the application can solve the problems that all risks in a certain area cannot be investigated when geological data is monitored based on local monitoring points, and the analysis efficiency of geological risks is low.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological analysis, in particular to an analysis method and an analysis device for land subsidence. BACKGROUND

[0002] With the acceleration of urbanization and the continuous expansion of infrastructure construction, land subsidence has become an important factor affecting urban safety and sustainable development. Land subsidence not only can cause damage to buildings, roads, bridges and other infrastructure, but also can affect the safety of subways, underground pipe networks and other urban underground facilities, and even can trigger landslides, floods and other secondary disasters.

[0003] Traditional land subsidence monitoring methods usually rely on local monitoring points, measure the subsidence amount by precise instruments, and analyze the monitoring data periodically. Although this land subsidence monitoring technology provides monitoring data and early warning mechanism to a certain extent, this monitoring method is usually limited to a single monitoring point, the monitoring range is small, and cannot cover large-scale urban areas, and the data of each monitoring point is independent, which cannot be effectively spatially coordinated and data fused, which makes it difficult to obtain comprehensive subsidence data in real time in complex geological environment or irregular land subsidence process, and potential risks are difficult to discover in time.

[0004] The traditional monitoring system lacks real-time data acquisition and processing, and often relies on manual data acquisition and periodic inspection, which makes it difficult to respond to rapid land subsidence events in time, affecting the speed of emergency response and decision-making efficiency. SUMMARY

[0005] The present application provides an analysis method and an analysis device for land subsidence, which solves the problem that all risks in a certain area cannot be investigated when monitoring geological data based on local monitoring points, and the problem of low analysis efficiency of geological risks.

[0006] In order to solve the above technical problems, in a first aspect, the present application provides an analysis method for land subsidence, which comprises:

[0007] Obtaining geological data and meteorological data of a monitoring area;

[0008] According to the geological data, evaluating the basic subsidence risk of the monitoring area;

[0009] Matching a sensor deployment strategy corresponding to the basic subsidence risk and the meteorological data;

[0010] According to the sensor deployment strategy, deploying sensors to collect land subsidence data of the monitoring area;

[0011] perform time-frequency analysis on the land subsidence data to obtain a ground sound spectrogram corresponding to the land subsidence data;

[0012] input the ground sound spectrogram into a preset neural network to output a subsidence semantic vector corresponding to the ground sound spectrogram;

[0013] obtain and analyze historical land subsidence data to construct a subsidence rule corresponding to the historical land subsidence data;

[0014] input the subsidence semantic vector and the subsidence rule into a preset space-time analysis model to predict a future subsidence trend of the monitoring area.

[0015] In a second aspect, the present application provides an analysis device for land subsidence, comprising:

[0016] a sensor module arranged in a monitoring area and configured to obtain land subsidence data;

[0017] an optical fiber transmission module, which comprises a plurality of optical fibers, the outer layer of the optical fiber is an optical fiber outer layer modified by a nano material or a flexible composite material, the optical fiber is electrically connected with the sensor module, configured to receive the land subsidence data and transmit the land subsidence data;

[0018] a control module, which is electrically connected with the optical fiber transmission module, configured to receive the transmitted land subsidence data, analyze the land subsidence data and output a data analysis result.

[0019] Compared with the prior art, the analysis method for land subsidence provided by the present application has the following beneficial effects:

[0020] The present application obtains geological data and meteorological data of the monitoring area, and then evaluates the basic subsidence risk according to the geological data, thereby realizing the risk global scanning of the whole monitoring area. Instead of focusing on a single monitoring point, the present application collects geological data and meteorological data of the whole area, divides the risk level of the whole area through multi-index quantitative evaluation, and clearly defines the risk distribution boundary. The further scheme breaks the blindness of uniform or local point distribution by matching the sensor distribution strategy corresponding to the basic subsidence risk and meteorological data. The sensors are densely distributed in high-risk areas and sparsely distributed in low-risk areas to ensure that all high-risk points are not missed and low-risk areas are not wasted, thereby realizing full-area dead-angle-free coverage. Moreover, the present application dynamically adjusts in combination with meteorological data to cope with the dynamic risk of irregular subsidence, thereby solving the problem of missed detection caused by the change of risk with the environment.

[0021] The sensors of the scheme work in coordination according to the risk level, and the data is self-labeled with space labels (such as position and risk level), instead of independent and scattered data; the subsequent time-frequency analysis and semantic vector conversion are based on the full-area data, which can capture the settlement correlation of different areas, realize spatial collaborative analysis, and solve the defect that the independent data cannot be linked.

[0022] In addition, in the scheme, the sensor data acquisition does not require manual intervention, solving the low efficiency problem of traditional manual recording and periodic export. Moreover, the scheme realizes efficient data processing through time-frequency analysis and further through a neural network to generate a settlement semantic vector: time-frequency analysis quickly strips environmental noise and extracts core settlement features, and the processing time is compressed from the traditional hour level to the second level; the neural network automatically converts the ground sound spectrum into a settlement semantic vector, replacing the traditional cumbersome process of manually interpreting data and labeling features, so that the analysis efficiency is improved. Further, the scheme breaks the lag of post-analysis by constructing a historical settlement rule and further predicting future trends through a time-space model. The historical settlement rule library (time sequence, space, and causal rule) is constructed in advance to avoid starting from scratch every time, shorten the analysis period, and the time-space analysis model combines real-time semantic vectors (current settlement features) with historical rules to quickly output future settlement trends, with the whole process time controlled within minutes from data collection to trend prediction; the scheme can respond to rapid settlement events in real time, output warnings in a timely manner, and solve the efficiency pain points of traditional periodic analysis that cannot respond to sudden risks. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only represent a part of the embodiments of the present application, and not all the embodiments. For those skilled in the art, other drawings obtained according to these drawings without creative labor belong to the scope of protection of the present application.

[0024] Figure 1 is a flowchart of a ground settlement analysis method provided by an embodiment of the present application.

[0025] Figure 2 is Figure 1 is a flowchart of step S130 in the embodiment.

[0026] Figure 3 is Figure 1 is another flowchart of step S130 in the embodiment.

[0027] Figure 4 is Figure 1 is a flowchart of step S170 in the embodiment.

[0028] Figure 5 is Figure 1A flowchart of step S180.

[0029] Figure 6 is a structural schematic diagram of an analysis device for ground settlement provided by an embodiment of the present application.

[0030] Figure 7 is a structural schematic diagram of another analysis device for ground settlement provided by an embodiment of the present application.

[0031] Figure 8 is a structural schematic diagram of still another analysis device for ground settlement provided by an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to make the objects, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application in combination with the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0033] In order to make the description of the present disclosure more detailed and complete, the following describes the embodiments of the present application and specific embodiments; but this is not the only form of implementation or use of the specific embodiments of the present application. The embodiments include the features of the specific embodiments and the method steps and their order used to construct and operate these specific embodiments. However, other specific embodiments can also be used to achieve the same or equivalent functions and step sequences. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present application.

[0034] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0035] In the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in the text only describes the relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone. In addition, in the description of the embodiments of the present application, "multiple" means two or more than two, and other quantifiers similar thereto should be understood. The preferred embodiments described herein are only used to illustrate and explain the present application, and do not limit the present application, and the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0036] like Figure 1 The flowchart described above is a method for analyzing ground subsidence provided in an embodiment of the present invention, and includes the following steps.

[0037] Step S110: Obtain geological and meteorological data for the monitoring area.

[0038] Specifically, in this step, for scenarios with stringent data accuracy requirements, such as urban core areas, large bridges, or subways, an integrated acquisition mode combining fiber optic integrated sensing and multi-source cross-verification can be used to acquire geological and meteorological data for the monitoring area. When collecting meteorological data, a microclimate sensing module can be integrated at key nodes of the fiber optic network (e.g., one module every 500 meters, prioritizing soil-rock contact surfaces, historical subsidence sensitive areas, and areas surrounding infrastructure). This module contains miniature temperature and humidity sensors, rainfall sensors, and air pressure sensors. Geological data can include soil type data, rock stability data, groundwater level depth data, and historical subsidence data. During geological data acquisition, soil pressure sensors and strain sensors can be deployed within a 1-meter radius of each fiber optic node. Soil pressure sensors can be used to infer soil type by monitoring changes in soil pressure; strain sensors can capture strain fluctuations at the contact surface between rock and soil in real time, and when the daily strain fluctuation exceeds a certain value, it can be determined that the stability of the rock layer has decreased; groundwater level depth data can be estimated through a correlation model between soil moisture and pressure values; historical settlement within a preset historical interval needs to be retrieved from the blockchain data integrity verification module.

[0039] Optionally, the collected raw data can first be processed by the edge computing nodes at the fiber optic aggregation point, and then processed through 3D processing. An outlier detection algorithm (i.e., removing invalid data exceeding three standard deviations above or below the data mean, such as blank values ​​when the sensor is offline or out-of-range data caused by transient electromagnetic interference) is used for cross-comparison with borehole data from the regional geological survey institute (e.g., at least one borehole per square kilometer, 3-5 meters deep, with laboratory analysis to obtain soil compressibility coefficient and rock layer distribution thickness) and data from national meteorological observation stations (e.g., less than or equal to 10 kilometers from the monitoring area, with data accuracy higher than that of on-site sensors). If the deviation of a certain type of data exceeds a preset deviation threshold, redundant node sensors (e.g., one spare sensor of the same type for each key node) can be activated to re-collect data until the data deviation is less than or equal to the preset deviation threshold, ultimately ensuring that the overall data accuracy meets the standard.

[0040] Specifically, in this step, for the wide-area rapid survey scene (such as urban group subsidence monitoring, the area is greater than 50 square kilometers), satellite remote sensing technology can be superimposed to reduce the ground sensing cost. When collecting geological data, synthetic aperture radar interferometry technology can be used to obtain monthly ground surface subsidence rate, covering the subsidence trend of a large area. At the same time, one ground calibration point can be arranged every 10 square kilometers, and a fiber optic sensor is used to obtain high-precision data of a single point to calibrate the deviation of satellite data; meteorological data uses satellite cloud image inversion rainfall and temperature data, combined with a small amount of real-time data correction of ground micro weather stations, reduces the layout density of ground sensors, the core is to ensure the spatial coverage and trend accuracy of the data, without pursuing real-time high-frequency monitoring of each point.

[0041] In step S120, according to the geological data, the foundation subsidence risk of the monitoring area is evaluated.

[0042] Specifically, in this step, for areas with diverse geological conditions and dense infrastructure, a hierarchical index quantification and expert collaborative calibration evaluation logic can be used to ensure accurate risk classification and actual scene adaptation. For example, 6 geological indexes that are strongly related to ground subsidence can be selected, and the definition, data source, weight, and classification standard of each index are as follows. The first one can be the soil compressibility coefficient, which refers to the compression deformation ability of soil under external pressure. The data can come from laboratory analysis of artificial borehole samples, and the weight can account for 0.25 in the evaluation system. The second one can be the groundwater level decline rate, which refers to the average decline of monthly groundwater level. The data can come from real-time monitoring by fiber optic sensors, and the weight can account for 0.2. The third one can be the rock stability, which can be determined by the single-day strain fluctuation value of the rock-soil contact surface. The data can come from strain sensors, and the weight can be set to 0.18. The fourth one can be the historical subsidence amount, which can be the cumulative subsidence value of the monitoring area in the past 3 years. The data can come from historical records stored in the blockchain, and the weight can be set to 0.15. The fifth one can be the building load strength, which can be the average load value of infrastructure foundations in the monitoring area. The data can come from building archives of the city planning department, and the weight can be set to 0.12. The sixth one can be the long-term average soil moisture, which can be the average value of annual soil moisture content. The data can come from historical statistics of soil moisture content detected by sensors in each year, and the weight can be set to 0.10. When quantifying each index, the indexes can be standardized first. The min-max normalization method can be used to convert index values of different dimensions to scores in the range of 0-1. Further, the comprehensive risk score can be calculated according to the weight of each index. Finally, the risk level can be divided according to the comprehensive score, such as greater than 0.7 for high risk, 0.4-0.7 for medium risk, and less than 0.4 for low risk. To avoid algorithm bias, feedback from geologists and hydrologists can be collected (such as reducing the weight of soil moisture average in clay soil area to 0.08 and increasing the weight of groundwater level decline rate to 0.22).

[0043] Step S130, matching the sensor deployment strategy corresponding to the foundation subsidence risk and the meteorological data.

[0044] Specifically, in this step, the principle of higher risk, more significant meteorological influence, better sensor configuration, and higher density can be followed to develop a refined deployment strategy in combination with the characteristics of distributed fiber optic sensing technology.

[0045] For example, in high-risk areas, if there is a significant meteorological impact (such as rainfall greater than 150 mm / month, daily temperature difference greater than 15 degrees Celsius) at the same time, a combination of fiber optic acoustic sensors, strain sensors, and pressure sensors can be used. The sensor layout spacing can be controlled at 5-8 meters, and the spacing can be further encrypted to 3-5 meters at key points such as building foundation periphery, underground pipe network interface, and rock fracture zone; the installation depth can be adjusted according to the soil type, such as 1.5-2 meters in sandy soil area (to avoid the influence of surface soil disturbance), 1-1.5 meters in clay soil area (clay soil has higher stability in deep layer); the communication network can use a redundant design of fiber optic main path and LoRa backup path, the main path transmission rate remains fast, and the backup path automatically switches when the fiber optic fails, ensuring uninterrupted data transmission.

[0046] For example, in medium-risk areas, if the meteorological impact is moderate (such as rainfall of 50-150 mm / month, daily temperature difference of 5-15 degrees Celsius), a combination of fiber optic acoustic sensors and micro-meteorological sensors can be used. The sensor spacing can be relaxed to 10-15 meters, and pressure sensors are not required; the installation depth can be uniform at 1.2-1.5 meters (taking into account different soil types); communication can only retain the fiber optic main path without the need for a backup path (reducing costs), and the outer layer of the fiber optic is still adapted to the environment to ensure signal stability.

[0047] For example, in low-risk areas, if the meteorological impact is weak (rainfall less than 50 mm / month, daily temperature difference less than 5 degrees Celsius), only fiber optic acoustic sensors can be used. The spacing can be further relaxed to 20-30 meters, and the installation depth can be 1-1.2 meters; low-cost fiber optics can be used for communication, and the sampling frequency can be reduced to maximize cost reduction while meeting monitoring requirements.

[0048] In an optional implementation, as shown in Figure 2 The step S130 includes the following steps.

[0049] Step S131, analyzing the influence weight of the meteorological data on the foundation settlement risk.

[0050] Specifically, in this step, the dynamic influence degree of each meteorological factor on the settlement risk can be determined through multi-factor analysis and expert calibration, providing a quantitative basis for density adjustment.

[0051] For example, from the meteorological data obtained in S110, 4 factors that have the most significant impact on the settlement risk are selected, which are rainfall, daily temperature difference, monthly average humidity, and instantaneous wind speed. Among them, rainfall directly affects the soil saturation degree (the greater the rainfall, the stronger the soil compressibility, and the higher the risk), daily temperature difference affects soil frost heaving and thawing settlement (the greater the temperature difference, the more soil freeze-thawing, and the greater the settlement risk), monthly average humidity affects soil cohesion (the higher the humidity, the lower the cohesion, and the easier the soil deformation), and instantaneous wind speed affects surface soil erosion (the greater the wind speed, the more surface soil loss, and the indirectly higher the settlement risk).

[0052] Further, a random forest regression algorithm can be used to calculate the initial weight. The algorithm constructs 100 decision trees (the maximum depth is set to 10 to avoid overfitting), takes the basic settlement risk score of S120 as the dependent variable, and the standardized data (min-max normalized to 0-1) of the four meteorological factors as the independent variable. After training, the weight of each factor is output through the feature importance. For example, rainfall has the most direct impact on soil compressibility, with the highest weight (0.4); daily temperature difference is second (0.25); monthly average humidity (0.2) and instantaneous wind speed (0.15) have lower weights. To avoid algorithm bias, feedback from 3-5 hydrology and geology experts can be combined for calibration. For example, experts determine that the rainfall weight in clay soil areas should be increased to 0.5, and the wind speed weight should be reduced to 0.1. Then the initial weight is adjusted to ensure that the weight fits the regional soil characteristics.

[0053] Further, the weight can be recalculated every month based on newly collected meteorological data and settlement data (preliminary calculation is completed locally by the group collaboration edge computing node, and then uploaded to the center system for calibration) to adapt to seasonal changes. For example, the daily temperature difference has a greater impact on settlement risk in winter (December-February), and the weight can be increased from 0.25 to 0.3. In the rainy season (June-August), the rainfall weight increases from 0.4 to 0.45, and the wind speed weight decreases to 0.1 due to smaller wind force in the rainy season, ensuring that the weight always matches the actual meteorological impact.

[0054] Optionally, in this step, when calculating the meteorological impact weight, the contact surface state data of the geology-fiber coupling interface monitoring module (such as soil and rock contact surface strain fluctuation) can be combined. When the contact surface state data is poor, the weight can be modified. For example, in clay soil areas and unstable contact surfaces, the rainfall weight can be additionally increased by 8%, as rainwater is more likely to penetrate and exacerbate settlement in this scenario.

[0055] In step S132, the sensor deployment density in the sensor deployment strategy is determined based on the basic settlement risk and the impact weight.

[0056] The sensor deployment density is directly proportional to the basic settlement risk and the impact weight.

[0057] Specifically, in this step, a density coefficient model of the basic risk and the meteorological weight can be constructed to ensure that the density is positively correlated with the risk and the meteorological influence, while the key points and the complex terrain are taken into account.

[0058] For example, the density coefficient calculation model can define the laying density coefficient K as the core calculation index, and the formula is K = basic settlement risk score The meteorological influence weight comprehensive value is calculated by normalizing (min-max normalization to 0-1) the actual monitoring value of each meteorological factor, multiplying the corresponding weight, and summing up.

[0059] Further, five density levels can be divided according to the K value, corresponding to different sensor spacings. When K is greater than 0.8, it is a special dense level, the sensor spacing is 3-5 meters, and it is suitable for high-risk and strong meteorological influence scenarios (such as basic risk score 0.9, meteorological comprehensive value 0.9, K = 0.81, corresponding to high-risk area in rainy season); when K is greater than 0.6 and less than or equal to 0.8, it is a dense level, the spacing is 5-8 meters, and it is suitable for high-risk and weak meteorological influence or medium-risk and strong meteorological influence; when K is greater than 0.4 and less than or equal to 0.6, it is a regular level, the spacing is 8-15 meters, and it is suitable for medium-risk and weak meteorological influence or low-risk and strong meteorological influence; when K is greater than 0.2 and less than or equal to 0.4, it is a sparse level, the spacing is 15-25 meters, and it is suitable for low-risk and weak meteorological influence; when K is less than or equal to 0.2, it is a special sparse level, the spacing is 25-35 meters, and it is suitable for extremely low-risk area.

[0060] It can be understood that the above embodiment of the application quantifies the dynamic weight of meteorological data on settlement risk, so that the sensor density is positively correlated with the risk level and the meteorological influence degree, and is adaptively adjusted. This not only avoids the detection of small settlement in high-influence areas due to insufficient density, but also eliminates the cost waste caused by excessive layout in low-influence areas, and realizes the dynamic balance between precision and cost.

[0061] In an optional implementation manner, as shown in Figure 3 After step S132, the following steps are further included.

[0062] Step S133, analyzing the influence degree of the meteorological data on the sensitivity of the sensor.

[0063] Specifically, in this step, the sensitivity error rate can be selected as the core index (sensitivity error rate = (actual measured value after interference - standard value) / standard value), and the influence of specific meteorological factors is analyzed for different sensor types.

[0064] For example, for the optical fiber acoustic sensor (monitoring settlement vibration signals), the analysis shows that when the soil humidity increases from 30% to 90% relative depth, the optical signal of the outer layer of the optical fiber attenuates due to water vapor penetration, and the sensitivity error rate increases from 5% to 18%; if the outer layer of the optical fiber is modified by a nano material (carbon nanotube), the error rate increases from 5% to 10%, and the interference amplitude decreases by 44%; and the analysis shows that when the ambient temperature decreases from 25 degrees Celsius to -5 degrees Celsius, the optical path changes due to the shrinkage of the optical fiber material, and the sensitivity error rate increases from 5% to 12%; when the temperature rises to 40 degrees Celsius, the error rate increases to 10% due to the drift of electronic components; and the analysis shows that for every 10 hPa change in air pressure, the error rate fluctuates by plus or minus 1% (weak influence, can be ignored).

[0065] For example, for the strain sensor (monitoring soil deformation), the analysis shows that when the humidity is greater than 85% relative depth, the sensor probe is affected by moisture, causing strain measurement deviation, and the error rate increases from 3% to 15%; and for every 10 degrees Celsius change in temperature, the error rate changes by plus or minus 2% (for example, from 25 degrees Celsius to 35 degrees Celsius, the error rate increases from 3% to 5%).

[0066] Optionally, in this step, when analyzing the degree of influence on sensitivity, if the monitoring shows that the degree of peeling between the soil and the optical fiber is greater than the preset peeling threshold, regardless of the weather conditions, it is determined as high degree of influence, and the installation depth of the sensor is adjusted (increased by 0.3 meters) to reattach to the soil.

[0067] Step S134, based on the degree of influence of the meteorological data on the sensitivity of the sensor, adjust the sensitivity or working mode of the sensor.

[0068] Specifically, in this step, when adjusting the sensitivity, for high-influence scenarios (error rate > 15%), i.e. when the humidity is greater than 85%RH (relative humidity), the optical signal gain of the optical fiber acoustic sensor can be increased from 20dB (decibels) to 25dB by the sensor controller, and a signal compensation algorithm can be started (based on historical humidity-error correlation data, automatically correcting the measured value, such as multiplying the measured value by 1.15 compensation factor when the humidity is 90%); for high-influence scenarios where the temperature is less than 0 degrees Celsius, a built-in heating sheet can be started to maintain the internal temperature of the sensor between 8 degrees Celsius and 12 degrees Celsius to avoid material shrinkage; for high-influence scenarios where the temperature is greater than 40 degrees Celsius, a cooling fan (power 3 watts) can be started to maintain the internal temperature less than or equal to 35 degrees Celsius;

[0069] For the middle influence scene where the humidity is located at 50%-85%RH, the gain can be increased from 20dB to 22dB without compensation algorithm; for the middle influence scene where the temperature is located at 0-10 degrees Celsius or the temperature is located at 30-40 degrees Celsius, only temperature monitoring can be started, without active heating or heat dissipation, and if the temperature fluctuation is greater than 5 degrees Celsius or less than 5 degrees Celsius, adjustment is triggered again.

[0070] Specifically, when adjusting the working mode, for the scene of heavy rain (rainfall greater than 50mm / day, humidity greater than 90%RH, high sensitivity to sensor), the sensor sampling can be switched from normal sampling (100Hz) to high-frequency sampling (200Hz), so as to reduce the influence of single noise on data by increasing the number of sampling points. For the scene of low temperature in winter (temperature less than-5 degrees Celsius, high sensitivity to sensor), low-power sleep can be started at night (22:00-6:00), the sampling frequency is reduced to 50Hz (Hz), the transmission interval is extended, and the heating sheet is preferentially powered; the normal mode is restored during the day.

[0071] It can be understood that the above embodiment of the application is different from the passive collection mode of the conventional sensor fixed parameters, and the sensitivity parameters are dynamically adjusted or the working mode is switched by analyzing the comprehensive influence of the meteorological and geological interface state on the sensitivity, thereby reducing the collection error of the sensor on the data.

[0072] Step S140, the sensor arranged according to the sensor arrangement strategy collects the land subsidence data of the monitoring area.

[0073] Specifically, in this step, the high sensitivity characteristics of the optical fiber sensor can be relied on to realize uninterrupted and high-precision data collection and transmission. When setting the collection parameters, the sampling frequency can be set differently according to the risk level, such as setting 500Hz for high-risk areas (which can capture the small soil vibration caused by subsidence, with a frequency range of 0.1-10Hz), 200Hz for medium-risk areas, and 100Hz for low-risk areas, and all sensors can be set to collect continuously for 24 hours without interruption (to avoid missing sudden subsidence events). Specifically, the signal types collected by the sensor can include optical fiber acoustic signals (reflecting soil particle vibration, indirectly reflecting subsidence rate), soil strain signals (directly reflecting soil deformation), and soil pressure signals (reflecting pressure changes caused by subsidence), and real-time environmental data (temperature, humidity, rainfall) of the micro-meteorological sensor can be collected simultaneously for subsequent signal noise reduction to distinguish between subsidence signals and environmental interference.

[0074] Optionally, the ground subsidence original data collected in this step can be transmitted to the group collaboration type edge computing node of the optical fiber convergence point first, and the data is preliminarily screened (blank data when the sensor is offline, invalid data with signal strength less than the threshold value) at the same time; the screened data can be uploaded to the center processing system through the optical fiber network.

[0075] Optionally, the sensor collected data can be uploaded to the edge node, and in addition to the preliminary screening, the data can also be subjected to hash check before being chained through the block chain data integrity verification module, a unique hash value is generated for each batch of data, which is bound with the node identifier and the time stamp and temporarily stored, so that the subsequent transmission to the center system can be traced and tamper-proof.

[0076] Step S150, time-frequency analysis is performed on the ground subsidence data to obtain the corresponding ground sound spectrogram of the ground subsidence data.

[0077] Specifically, in this step, the time domain original data can be converted into a visual feature map by signal preprocessing and time-frequency analysis, highlighting the frequency and amplitude characteristics of the subsidence signal.

[0078] For example, when performing signal preprocessing, Kalman filtering algorithm can be used to smooth the original data first. The algorithm dynamically corrects data deviation through the prediction-to-update cycle, and reduces the influence of environmental noise (such as vehicle vibration and wind interference). Further, the effective signal after preprocessing can be subjected to short-time Fourier transform for time-frequency analysis. The time window length can be set to 0.5 seconds (taking into account the time resolution and frequency resolution), and the window overlap rate can be set to 50% (to avoid signal discontinuity). The time domain signal is converted into a time-frequency-amplitude three-dimensional matrix (time dimension in minutes, frequency dimension in Hz, and amplitude dimension in voltage value). Then, the three-dimensional matrix is converted into a two-dimensional ground sound spectrogram, in which the horizontal axis represents time, the vertical axis represents frequency, and the color depth represents amplitude (the larger the amplitude, the darker the color, corresponding to more significant subsidence). At the same time, key feature points (such as the stable frequency band of 2-5 Hz corresponding to uniform subsidence, and the instantaneous frequency peak of 8-10 Hz corresponding to sudden subsidence) are marked on the spectrogram, providing a clear feature carrier for subsequent semantic conversion.

[0079] In an optional implementation, step S150 includes the following steps.

[0080] Step S151, wavelet transform is performed on the ground subsidence data to separate the low-frequency subsidence signal and the high-frequency noise of the ground subsidence data.

[0081] Specifically, in this step, the input ground subsidence data can be a time-domain signal (such as an optical fiber acoustic signal) collected by a sensor, which can be first processed by mean removal (subtracting the data mean to eliminate the interference of the direct current component) and then normalized.

[0082] Further, db4 wavelet basis can be selected, and 3-layer decomposition is performed, wherein 1-layer decomposition separates signals less than 250Hz, 2-layer separation separates signals less than 125Hz, and 3-layer separation separates signals less than 62.5Hz; in combination with the actual frequency range (0.1-10Hz) of the subsidence signal, 3-layer decomposition is sufficient to separate high-frequency noise and low-frequency signals; further, the 3-layer decomposition is realized by convolution-downsampling iteration using the standard algorithm of discrete wavelet transform, and the approximation coefficients (low frequency, corresponding to effective signals) and detail coefficients (high frequency, corresponding to noise) of each layer are output.

[0083] The low-frequency subsidence signal can be the approximation coefficient extracted after 3-layer decomposition, which is reconstructed into a time-domain signal by inverse wavelet transform, and the frequency range is concentrated in 0.1-10Hz; the high-frequency noise can be the detail coefficient extracted after 3-layer decomposition, which is combined into a high-frequency noise signal, and the frequency range is greater than 10Hz (mainly environmental interference, such as vehicle vibration (20-50Hz), wind noise (15-30Hz), and sensor electronic noise (greater than 100Hz)).

[0084] In step S152, the wavelet coefficient features of the high-frequency noise are clustered in real time to generate the inverse noise signal of the high-frequency noise.

[0085] Specifically, in this step, the detail coefficients of 3-layer decomposition in S151 can be extracted, and 4 types of features (such as the amplitude mean of 1-layer detail coefficient 0.2, the standard deviation 0.05, the mean of 2-layer 0.15, and the standard deviation 0.03) of each group of coefficients can be calculated to form a noise feature vector (dimension 3 4=12 dimensions).

[0086] Further specifically, a real-time clustering algorithm (K-means clustering) can be used to cluster the wavelet coefficient features of the high-frequency noise in real time, the clustering parameters of the clustering algorithm can be set as K=3 (such as vehicle vibration noise, wind noise, and electronic noise, corresponding to 3 cluster centers), the iteration number can be set to 50 times, and the convergence threshold can be set to 0.001; during clustering, the 12-dimensional noise feature vector can be input into the K-means algorithm, and the Euclidean distance between the vector and the cluster center can be calculated, so that the high-frequency noise at each time can be classified into the corresponding noise type (such as the feature vector at a certain time is closest to the vehicle vibration cluster center, which is determined as vehicle noise); finally, the coefficient distribution models of 3 types of noise can be obtained (such as the detail coefficient amplitude of vehicle noise is concentrated in 0.15-0.25, and the wind noise is concentrated in 0.1-0.15).

[0087] Furthermore, the amplitude of the detail coefficients for each type of noise can be kept unchanged but the sign can be reversed (e.g., if the original coefficient is 0.2, it becomes -0.2 after reversal; if the original coefficient is -0.1, it becomes 0.1 after reversal). Then, the three layers of detail coefficients after reversal can be reconstructed into a time-domain signal through Mallat inverse wavelet transform to obtain the inverse noise signal.

[0088] Step S153: Superimpose the reverse noise signal and the ground settlement data to obtain noise-free ground settlement data;

[0089] Step S154: Perform time-frequency analysis on the noise-free ground subsidence data to obtain the ground acoustic spectrum corresponding to the ground subsidence data.

[0090] Specifically, short-time Fourier transform can be used to perform time-frequency analysis on ground subsidence data. The time window of this short-time Fourier transform can be the Hanning window (window length 0.5 seconds, corresponding to 250 data points, because 0.5 seconds can take into account both time resolution (capturing subsidence abrupt changes within 1 second) and frequency resolution (distinguishing frequency differences of 1 Hz)). The window overlap rate can be 50% (that is, adjacent windows overlap by 125 data points to avoid signal discontinuity and ensure the continuity of the time-frequency spectrum). The frequency resolution is calculated by dividing the sampling frequency of 500 Hz by the window length of 250 points, which equals 2 Hz (that is, the frequency interval of the vertical axis of the spectrum is 2 Hz, covering 0-250 Hz, but in practice only the subsidence signal frequency band of 0-10 Hz is considered).

[0091] In this way, a short-time Fourier transform is performed on the noise-free data to output a three-dimensional matrix of time, frequency, and amplitude (time dimension: 1440 points, corresponding to a step size of 1 hour / 0.25 seconds; frequency dimension: 126 points, corresponding to 0-250Hz; amplitude dimension: matrix element values, representing the signal strength at the corresponding time-frequency point). Further, the amplitude of the three-dimensional matrix is ​​mapped to color intensity through dimensional compression (using Jet color mapping, where a larger amplitude corresponds to a darker color, e.g., amplitude 0.5 corresponds to red, 0.3 to yellow, and 0.1 to blue). This is further achieved by making the horizontal axis represent time, the vertical axis represent frequency, and the color represent amplitude, generating a two-dimensional geosonic spectrogram.

[0092] Optionally, in this step, in addition to the Hanning window (0.5 seconds), when the settlement signal is abrupt (such as a daily settlement greater than 5 mm), a rectangular window (0.2 seconds) can be switched to improve the time resolution and accurately capture the moment of abrupt change; when the settlement is uniform, a Hamming window (0.8 seconds) can be used to improve the frequency resolution and realize the window's adaptation to the settlement signal.

[0093] It can be understood that, on the one hand, the reverse noise generation of the above-mentioned embodiments of the application is not simply filtering high-frequency components, but constructing a reverse signal based on noise clustering features to realize noise amplitude cancellation and completely remove vehicle vibration, wind and other interference; on the other hand, the above-mentioned embodiments of the application dynamically adjust the wavelet decomposition layer and the Short-Time Fourier Transform Window (STFT) window to ensure that the subtle features of the sedimentation signal are not damaged. The final output of the ground sound spectrogram can accurately represent the sedimentation mode and provide high-quality data support for subsequent semantic conversion and trend prediction, which is the precision that traditional mean filtering and single wavelet denoising technology cannot achieve.

[0094] In step S160, the ground sound spectrogram is input into a preset neural network, and a sedimentation semantic vector corresponding to the ground sound spectrogram is output.

[0095] Specifically, in this step, the spatial features of the ground sound spectrogram can be extracted by a convolutional neural network, and the image features are mapped to semantic vectors by combining a fully connected layer, so that the machine can understand the sedimentation characteristics.

[0096] For example, the input layer of the convolutional neural network can receive a 256 pixel normalized ground sound spectrogram (normalizing the amplitude value to the 0-1 interval and converting it to a grayscale image); the convolutional layer can include 3 convolutional blocks, each of which can be composed of 1 convolutional layer (convolution kernel size 3 3, step 1, activation function ReLU) and 1 max pooling layer (pooling kernel size 2 2, step 2), which is used to extract local features of the ground sound spectrogram (such as frequency peak position, amplitude change trend, frequency band distribution); the fully connected layer can include 2 hidden layers and 1 output layer, wherein the first hidden layer can be set to 1024 neurons, and the second hidden layer can be set to 512 neurons (the activation function can be set to ReLU), which is used to fuse local features into global features, and the output layer can be set to 128 neurons (without activation function), outputting a 128-dimensional sedimentation semantic vector. Each dimension of the vector corresponds to a specific sedimentation feature, for example, dimensions 1-20 correspond to the sedimentation rate (the larger the value, the faster the rate), dimensions 21-40 correspond to the sedimentation amplitude (the larger the value, the greater the sedimentation amplitude), dimensions 41-60 correspond to the sedimentation stability (the smaller the value, the more stable the sedimentation), and dimensions 61-128 correspond to other auxiliary features (such as the degree of association with environmental factors);

[0097] In the training of the model, the training data set of the model can select more than 1000 groups of different subsidence patterns of ground sound spectrograms, which can cover three typical patterns of uniform subsidence (rate 5-10 mm / month), sudden subsidence (single day subsidence greater than 5 mm) and periodic subsidence (seasonal fluctuation), each group of spectrogram is labeled with the corresponding subsidence characteristic label; the cross entropy loss function can be used to calculate the error of the model prediction value and the label, the Adam optimizer (initial learning rate 0.001, decay 10% every 20 rounds) is used for training, a total of 100 rounds, during which 5-fold cross validation can be used to avoid overfitting; after training, the semantic vector matching accuracy of the model on the test set can be greater than or equal to the preset accuracy threshold (i.e. the cosine similarity between the predicted vector and the true label vector is greater than or equal to the preset accuracy threshold), to ensure accurate representation of the subsidence characteristics.

[0098] Step S170, obtaining and analyzing historical ground subsidence data, and constructing the subsidence law corresponding to the historical ground subsidence data.

[0099] Specifically, in this step, the time-space-cause three-layer law can be extracted from the reliable historical ground subsidence data to provide historical basis for subsequent prediction, and ensure that the prediction fits the regional subsidence characteristics.

[0100] For example, the historical data can preferentially select the monitoring data of more than 5 years stored by the blockchain data integrity verification module, supplemented by artificial measurement records (such as level measurement data in the past 10 years) in the regional geological disaster database; in the preprocessing stage, outliers and missing values can be removed, and finally a certain amount of valid data is retained to ensure the reliability of the law mining;

[0101] Further, the monthly subsidence rate and the monthly rainfall can be input into the traditional time series model (ARIMA model).

[0102] Further, the autoregressive term of the autoregressive integrated moving average model (ARIMA) model can be used to capture the periodic law (such as the rate in June-August of the past two years is higher than that in other months, which is determined as the ground subsidence rate is faster in the rainy season), and the trend term can be used to extract the short-term trend (such as the rate in the past year decreases from 5 mm / month to 4.5 mm / month, which is determined as the rate decreases slowly).

[0103] Optionally, in this step, when removing abnormal historical data, the clustering-based anomaly detection algorithm (isolated forest) can be used to identify isolated points (such as abnormal data with single-day subsidence greater than 3 times the historical mean) that are difficult to combine with other data points by randomly splitting the historical subsidence data points multiple times, to further improve the purity of the effective data.

[0104] In an alternative implementation, step S170 comprises the following steps.

[0105] Step S1711, obtaining the historical ground subsidence data.

[0106] Specifically, in this step, the main data source in the historical ground subsidence data can be the 5-year monitoring data stored by the blockchain data integrity verification module, which stores the full amount of fiber node data at 1 hour / time, including the subsidence amount (mm) of each node, the collection time (accurate to seconds), the node position (longitude and latitude, error less than or equal to 1 meter), and the synchronous environmental data (temperature, humidity, from the microclimate perception module). The data is tamper-proof and the hash value is updated every hour to ensure authenticity.

[0107] The supplementary data source in the historical ground subsidence data can be the historical abnormal records of regional geological disaster database (such as sudden subsidence event report, containing event time, impact range, subsidence amount measured value), artificial leveling historical data (full area measurement every 2 years in the past 10 years, used for calibrating fiber data).

[0108] Further specifically, when selecting the historical ground subsidence data, the fiber nodes uniformly distributed in the monitoring area can be selected (at least 3 nodes per 1 square kilometer to avoid sparse data), and the time span covers a complete annual cycle (at least 12 months, including different climate stages such as rainy season and winter, to ensure that the data can reflect the spatio-temporal changes); and the sensor offline data (blank value) and extreme interference data (such as earthquake caused out-of-range data, subsidence amount greater than 100 mm / day) can be excluded, and normal monitoring data (subsidence amount -50 mm to 50 mm / year, consistent with regional geological characteristics) can be retained to ensure that the effective data ratio meets the needs; then the collection time of different nodes can be unified to Beijing time whole point (such as 14:00:00, data with deviation more than 1 minute is corrected by linear interpolation), and the position coordinates can be unified to WGS84 coordinate system to avoid spatio-temporal misplacement; and the subsidence amount can be standardized by Z-score (mean = 0, standard deviation = 1) to eliminate the data magnitude deviation caused by different node monitoring accuracy differences.

[0109] Step S1712, in the historical ground subsidence data, analyzing the historical ground subsidence data of the monitoring area in the same time period or adjacent time period between the preset nodes to obtain the similarity of the historical ground subsidence data between the preset nodes.

[0110] Specifically, in this step, when selecting the preset nodes, the node pairs can be selected according to geological correlation and spatial distribution to avoid irrelevant node interference, for example, adjacent nodes with a distance of 500 meters to 1 kilometer can be selected first; nodes with the same soil type and groundwater depth (for example, both are clay and the groundwater depth is 5 meters) are selected to exclude low mutual information caused by geological differences (non-exception factors); node pairs can also be additionally added in the settlement sensitive area (such as the periphery of building foundation and underground pipe network interface) to ensure that the anomaly can be captured, such as node 5 (north of bridge foundation) - node 6 (south of bridge foundation).

[0111] Specifically, in this step, when determining the historical ground subsidence data in the same time period or adjacent time period, the time window can adopt a sliding time window, the window length can be 1 month, the window sliding step is 15 days, and the window covers the entire historical period (such as 5 years, a total of 118 windows), and 1 month window is selected because subsidence is a slow process, short-term (such as 1 day) data fluctuation is large, and long-term (such as 3 months) is easy to cover sudden anomalies;

[0112] When calculating the similarity of the historical ground subsidence data between the preset nodes, assuming that two preset nodes A and B, the standardized subsidence data of preset node A is X, the standardized subsidence data of preset node B is Y, and the similarity of the two is The calculation formula is:

[0113]

[0114] Among them , is the marginal probability density of X and Y, is the joint probability density, which can be calculated by kernel density estimation method;

[0115] In this way, in each time window, the mutual information value of a group of node pairs, that is, the similarity value (range 0-1, 0 represents complete irrelevance, and 1 represents complete correlation) is output.

[0116] Step S1713, when the similarity of the historical ground subsidence data is lower than the preset threshold value, the historical ground subsidence data corresponding to the preset node is determined as the historical abnormal subsidence data.

[0117] Specifically, in this step, the threshold value can be set by using statistical method and expert calibration strategy to avoid deviation of single empirical value. Specifically, the mutual information value of all node pairs in the historical normal period (a period without known subsidence anomaly, such as the whole year of 2021) can be calculated, and 90% of the minimum value is taken as the initial threshold value, for example, the minimum value of the mutual information of all node pairs in the normal period is 0.7, and the initial threshold value is 0.7 0.9=0.63; further, geologists can adjust the threshold to 0.6 by combining regional historical abnormal cases (e.g., in 2022, a node caused abnormal settlement due to underground pipe network leakage, and at that time, the mutual information of the node and the surrounding nodes was 0.58).

[0118] Further, abnormal data can be labeled as abnormal, which can include abnormal node identification information, abnormal time window, mutual information value, abnormal type (such as local sudden settlement), and be stored separately in a historical abnormal database (separated from normal data, facilitating subsequent rule mining), while being associated with geological data (such as soil type, underground water level change) of the node to provide context for rule construction.

[0119] Step S1714, constructing a settlement rule corresponding to the historical abnormal settlement data.

[0120] Specifically, in this step, the node positions of all historical abnormal settlement data can be counted, and an abnormal settlement space heat map can be drawn to represent the abnormal frequency with color depth and identify abnormal hot areas; the months and seasons of abnormal occurrence can also be counted, and an abnormal time distribution graph can be drawn to find that the abnormal proportion in the rainy season (June-August) is 55%, and the abnormal proportion in winter (December-February) is 15%, forming a time rule that abnormal settlement is more likely to occur in the rainy season;

[0121] Further, the correlation between abnormal data and environmental or geological factors can also be calculated, such as the proportion of abnormal occurrence when the monthly decrease of underground water level is greater than 0.3m is 80%, and the proportion of soil humidity greater than 35% RH is 75%, and it is concluded that rapid decrease of underground water level and high soil humidity are the main inducing factors;

[0122] In this way, through cross analysis, it is found that when the monthly decrease of underground water level is greater than 0.3m and the soil humidity is greater than 35% RH, the probability of abnormal occurrence is 3 times that of other conditions, forming a combination rule of double factors superimposed to easily cause abnormalities;

[0123] Further, multiple sets of abnormal data in the same abnormal hot area can also be tracked, such as node 4 first abnormal (settlement amount 8mm) in August 2023, abnormal intensification (12mm) in September, and recovery (3mm) in October, and the period from abnormal occurrence to recovery is about 2-3 months;

[0124] The subsequent data of the surrounding nodes of the abnormal node can also be analyzed, such as node 4 (abnormal in August 2023)-node 5 (abnormal in September, distance from node 4 is 0.8 kilometers)-node 6 (abnormal in October, distance from node 5 is 0.5 kilometers), and it is concluded that the abnormality spreads from the center to the periphery, and the spread speed is about 0.5-1 kilometer / month.

[0125] It can be understood that the above embodiments of the application associate the historical data of the surrounding monitoring nodes with the target area data through the spatio-temporal autoregressive model, capture the influence of the time period, and can also strip the interference caused by the area association. Compared with the traditional ARIMA model which can only identify a single time trend, the above embodiments of the application can simultaneously extract the periodicity, long-term trend and the precise law after the spatial correlation correction, so that the time dimension error of subsequent prediction is reduced.

[0126] In an optional implementation, as shown in FIG. 17, Figure 4 step S170 includes the following steps.

[0127] Step S1721, obtaining historical land subsidence data;

[0128] Specifically, in this step, the historical land subsidence data can include fiber monitoring data, satellite auxiliary data, and environmental or geological data, etc., wherein the fiber monitoring data can come from a distributed fiber acoustic sensing system, containing daily subsidence amount, subsidence rate and acoustic signal characteristics of each node; the satellite auxiliary data can come from a multi-scale fiber-satellite fusion engine module, containing monthly ground elevation data of an interferometric synthetic aperture radar (InSAR) satellite, quarterly subsidence data of a global navigation satellite system (GNSS), used to supplement data of areas not covered by the fiber; the environmental or geological data can come from a microclimate perception module (5-year daily temperature and humidity, rainfall), a geological survey database (soil compressibility coefficient, monthly data of underground water level depth), human activity data (building load in city planning archives, annual data of underground water extraction amount).

[0129] Step S1722, inputting the historical land subsidence data into a spatio-temporal autoregressive model, and outputting time sequence correlation results of the historical land subsidence data and spatial correlation results of the historical land subsidence data.

[0130] Specifically, in this step, the input of the model can be 50m (meter) monthly subsidence amount of a 50m spatial grid (5 years, a total of 60 time steps, set as Yt,i,j, t=1-60, i,j is the grid coordinate), while inputting lag 1 period subsidence amount Yt-1,i,j (time correlation), adjacent grid subsidence amount Yt,i±1,j, Yt,i,j±1 (spatial correlation), monthly rainfall X1,t and underground water level X2,t (external influencing factors);

[0131] Further specifically, the model can use maximum likelihood estimation method to train the model parameters when training the parameters, the iteration number is 100, the convergence threshold is 1e-5, the training data set accounts for 80% of the total data (48 time steps), and the verification set accounts for 20% (12 time steps).

[0132] Further specifically, the time series correlation result output by the model is the output time autoregression coefficient p1, p1 is close to 1, indicating that the monthly subsidence amount is related to the height of the previous month (strong time series continuity); the model also outputs the fitting value and residual of each time step of the time series correlation result, and the time step with small residual corresponds to the stable subsidence period.

[0133] Further specifically, the spatial correlation result output by the model includes a spatial autoregression coefficient and a spatial correlation matrix, wherein the spatial autoregression coefficient p2 is close to 1, indicating that the subsidence amount of a certain grid is related to the height of the adjacent grid (strong spatial continuity); the larger the matrix element value of the spatial correlation matrix, the stronger the spatial correlation of the corresponding two grids.

[0134] Optionally, in this step, the spatio-temporal autoregression model can introduce a spatial weight matrix, i.e., taking the distance between each monitoring node as the weight (the closer the distance, the higher the weight), and fuse the historical data of the surrounding nodes to assist in analysis, for example, the time series law of the city center node needs to superimpose the historical data of 3 nodes within 1 km to improve the reliability of the law.

[0135] Step S1723, inputting the time series correlation result into a preset time series analysis model to obtain a long-term change trend of the historical subsidence rate.

[0136] Specifically, in this step, the preset time series analysis model inputs the monthly subsidence amount time series correlation result output by the spatio-temporal autoregression model and the annual groundwater exploitation amount. In the preset time series analysis model, when processing data, the layers reached by the data are input layer (dimension 1, corresponding to monthly subsidence amount)-LSTM layer (64 neurons, capturing long-term dependence)-full connection layer (32 neurons)-output layer (dimension 1, corresponding to annual subsidence rate), the activation function is tanh, and the optimizer is Adam (learning rate 0.001).

[0137] Further specifically, the monthly subsidence amount output by the long short-term memory network (LSTM) can be aggregated by year to calculate the annual average subsidence rate; further, the annual rate sequence is fitted by linear regression to obtain the trend line equation: annual rate v=-0.5 The year + 1019.5, wherein the slope is -0.5, indicating that the average annual subsidence rate decreases by 0.5 mm / year in the last 5 years, and the long-term trend is slowing down; further, the inflection point in the trend can be identified, such as the inflection point in 2021, the rate decreases rapidly (0.5 mm / year on average) from 2019 to 2021, and the rate decreases slowly (0.25 mm / year on average) from 2021 to 2023, corresponding to the influence of the policy of groundwater level control since 2021.

[0138] Step S1724, based on the spatial correlation analysis technology, the spatial correlation result is analyzed to obtain the diffusion path of the historical subsidence range.

[0139] Specifically, in this step, the expansion direction and speed of the subsidence hot spot area are extracted from the spatial correlation result, and the spatial diffusion law is determined.

[0140] Further specifically, the Gi statistical value of each grid can be calculated based on the hot spot analysis in the spatial correlation analysis technology on the monthly subsidence amount spatial data output by the spatiotemporal autoregressive model, when the statistical value is greater than 1.96 (95% confidence), it can be determined as a hot spot grid (large and concentrated subsidence amount), and when the statistical value is less than -1.96, it can be determined as a cold spot grid (small and concentrated subsidence amount); based on the spatial interpolation in the spatial correlation analysis technology, the hot spot grid of each month is generated by using the inverse distance weighted interpolation method to generate a hot spot area planar graph (connecting discrete hot spot grids into continuous areas), and the area of the region is marked.

[0141] In this way, further specifically, the geometric center (longitude and latitude coordinates) of the hot spot area of each month is calculated; and the direction of the connecting line of the hot spot centers of adjacent two years is calculated, such as the connecting line direction from 2019 to 2020 is northeast, and from 2020 to 2021 is east by south 10 degrees, and the overall diffusion direction is determined as northeast-southeast; further, the distance between the hot spot center of each year and the initial center (January 2019) can be calculated, such as the distance to the initial center in December 2023 is 2 kilometers, the diffusion time is 5 years, and the average annual diffusion speed is 0.4 kilometers per year; further, the preferential direction in the diffusion can be identified, such as the diffusion to the northeast suburb with large groundwater exploitation is fast (0.5 kilometers / year), and the diffusion to the southeast direction of the ecological protection zone is slow (0.3 kilometers / year). Finally, the diffusion path and the subsidence path of the subsidence area can be compared, if the coincidence degree is greater than or equal to 80% (such as both diffusing to the northeast suburb), it is determined that the diffusion path rule is effective; if the coincidence degree is less than 80%, the soil type distribution analysis (such as the slow diffusion area is sandy soil with low compressibility) is supplemented to correct the path rule.

[0142] Step S1725, based on the association rule mining algorithm, the time series correlation result and the spatial correlation result are mined to obtain the causal correlation data of the environmental conditions and the subsidence mode in the historical land subsidence data.

[0143] Specifically, in this step, the association rule mining algorithm can be the Apriori algorithm. When the association rule mining algorithm is used to mine the time sequence association result and the spatial association result, the time sequence association result (settling rate), the spatial association result (settling type: hot spot / non-hot spot) and the environmental condition (such as rainfall, underground water level drop and soil humidity) can be converted into transaction data, such as "transaction 1: high rate, hot spot, high rainfall, fast water level drop and high humidity". When the algorithm parameters are set, the minimum support of the Apriori algorithm can be set to 20% (the probability of the rule appearing in the transaction is greater than or equal to 20%), the minimum confidence is 80% (the probability of the rule being true is greater than or equal to 80%), and the maximum item set length is 4 (a maximum of 4 condition combinations). When mining, the strong association rules that meet the parameters can be filtered out as the causal association data through the two steps of the frequent item set generation and rule generation of the Apriori algorithm.

[0144] In step S1726, the long-term change trend of the historical settlement rate, the diffusion path of the historical settlement range and the causal association data are associated to the historical land subsidence data as the settlement rule.

[0145] Specifically, in this step, a rule label can be added to each time step-space grid of the historical data, such as that in June 2023, the grid (i=10, j=15) is labeled as "consistent with rule 1 (high rainfall + fast water level drop - high rate), located in a hot spot area, corresponding to the northeast section of the diffusion path, and the long-term trend is a rate decrease of 0.5 mm / year". Further, for each settlement mode (such as high rate), the corresponding causal association data (such as "high rate in June 2023 due to rainfall of 120 mm / month + water level drop of 0.4 m / month") and the long-term trend influence (such as "although the rate is high, it is 0.3 mm / month lower than the same period in 2022, consistent with the long-term slowing trend") are associated.

[0146] Further, a settlement rule knowledge base can be set up. The knowledge base structure can be divided into a time sequence rule base, a spatial rule base and a causal rule base. The three bases can be associated through time identifiers-space identifiers. For example, the time sequence rule base stores annual rate trend lines, inflection points and policy factors; the spatial rule base stores monthly hot spot area maps, diffusion path coordinates and diffusion speed; and the causal rule base stores strong association rules, rule confidence or support.

[0147] It can be understood that the above-mentioned embodiments of the application focus on the combination of multiple factors through the association rule mining algorithm, and clearly show the coupling effect of the superposition of multiple factors on settlement. Compared with the traditional regression analysis which can only identify linear single causality, this technology can accurately locate the core driving combination of settlement, improve the credibility of the causal relationship, and provide a targeted intervention direction for subsequent risk management and control, rather than a generalized single measure.

[0148] Step S180, input the settlement semantic vector and the settlement rule into a preset spatio-temporal analysis model to predict a future settlement trend of the monitoring area.

[0149] Specifically, in this step, the settlement semantic vector of three core dimensions of settlement rate, amplitude stability and recent settlement trend can be extracted, such as dimensions 1-5 in the semantic vector corresponding to the rate (taking the average value as the current rate), dimensions 21-25 corresponding to the amplitude stability (taking the minimum value to determine the stability), and dimensions 61-65 corresponding to the trend in the past month (taking the difference value to determine the increase or decrease); the historical rule can select the time sequence period characteristics (such as a 30% increase in rate during the rainy season and a 15% decrease in winter).

[0150] Specifically, in this step, ARIMA can be used as a preset spatio-temporal analysis model, time sequence data is processed through ARIMA, the core semantic dimension data is sorted by time (such as the daily average rate in the past 30 days), the historical time sequence rule (rainy season or winter adjustment coefficient) is combined, and the model parameters (such as p=3, d=1, q=2, which can be determined by AIC criterion) are trained. Further, the model can output the values of two core indicators of the cumulative settlement and the average rate in the future 1-3 months.

[0151] Optionally, after predicting the future settlement trend, the result can be input into a digital twin ground surface engine module for dynamic evolution simulation, that is, based on a virtual ground surface model, the stress changes of key facilities such as underground pipe network and bridge support caused by settlement in the future 1-6 months are deduced according to the predicted trend (such as the stress of the pipe network interface increases to 150 kilopascals when the settlement is 25 mm), and the simulation result is used to correct the prediction model in the opposite direction (such as when the stress exceeds the threshold, the predicted value of the future settlement is reduced by 5%).

[0152] In an optional implementation manner, as shown in Figure 5 Step S180 includes the following steps.

[0153] Step S181, input the settlement semantic vector into a spatio-temporal convolutional neural network in the preset spatio-temporal analysis model to extract the spatial evolution features in the settlement semantic vector.

[0154] Specifically, in this step, the spatial correlation information (such as settlement hot area expansion and different area settlement correlation) contained in the settlement semantic vector can be captured by a spatio-temporal convolutional neural network (STCN) to provide feature support for subsequent trend prediction in the spatial dimension.

[0155] The STCN layer architecture can include a network level, a convolution kernel parameter, and a neuron number, wherein the network level contains 2 spatio-temporal convolution blocks (each block is composed of a spatio-temporal convolution layer, a batch normalization layer, and a ReLU activation layer), no pooling layer (to avoid loss of spatial information); the convolution kernel parameter can adopt a 3D spatio-temporal convolution kernel of 3 3 3 (the first two dimensions correspond to the spatial feature dimension, and the third dimension corresponds to the time step dimension), the convolution step is set to 1 (to ensure that the feature map size is consistent with the input), and the padding mode is set to the same (to avoid loss of edge features); the neuron number: the output channel number of the first spatio-temporal convolution block can be 64, and the output channel number of the second spatio-temporal convolution block can be 128.

[0156] Further specifically, the input of the settling semantic vector can be a 128-dimensional single-sample semantic vector (corresponding to the full-area settling feature of a certain monitoring time), which needs to be reshaped into a 3D tensor of “spatial grid dimension semantic feature dimension time step” ; assuming that the monitoring area is divided into 32 32 spatial grids (a total of 1024 grid units, corresponding to the fiber node layout density), the time step is set to 1 (single-time prediction), and the tensor dimension is “32 32 128 1” (spatial high spatial width semantic feature number time step”); further, the spatial correlation features in the settling semantic vector (such as the inter-regional settling similarity corresponding to dimensions 61-128) can be subjected to min-max normalization to avoid gradient imbalance caused by feature amplitude differences; further, the first layer spatio-temporal convolution block can perform convolution on the tensor of 32 32 128 1 to capture local spatial correlation features, such as the settling semantic similarity of a certain grid unit and the surrounding 8 units (output a local correlation feature map of 64 channels, each channel corresponding to a local correlation mode, such as center-edge collaborative settling and local sudden settling); the second layer spatio-temporal convolution block can further convolve the 64-channel feature map to integrate local features into global spatial evolution features, such as the current distribution pattern of the full-area settling hot spot, the boundary profile of the hot spot and the non-hot spot, and the potential expansion direction (such as the semantic feature of the grid unit in the northeast direction of the city center hot spot showing a high risk trend); finally, an output of 128-channel 32 32 spatial feature maps is obtained, and each feature map pixel value represents the spatial evolution potential of the corresponding grid unit (the larger the value, the higher the probability of significant settling of the unit in the future).

[0157] Optionally, in this step, when extracting spatial evolution features, the 3D convolution kernel of STCN can be based on the spatial database to construct a spatial weight matrix, such as assigning a weight of 0.8 to adjacent grid cells in a 32x32 grid and a weight of 0.2 to non-adjacent cells, to ensure that the convolution process preferentially captures local spatial correlations (such as the coordinated settlement of hot spot areas and surrounding areas).

[0158] Step S182, input the settlement semantic vector into the LSTM layer in the preset spatiotemporal analysis model to obtain the evolution law of the settlement semantic vector in the time dimension.

[0159] Specifically, in this step, the network structure of the LSTM layer can include 1 layer of hidden layer and 1 layer of output layer, the number of hidden layer neurons is set to 256, and the activation function uses tanh (adapted to the nonlinear change of time series data); the forgetting gate threshold of the LSTM layer can be set to 0.7 (preferentially retaining long-term effective time series features, such as annual rainy season settlement rules), the input gate threshold can be set to 0.3 (controlling the introduction intensity of new time series features, avoiding short-term noise interference), and the output gate threshold can be set to 0.5 (balancing the output weight of historical and current features); further, Dropout (dropout rate set to 0.2) can be used to prevent overfitting, and 20% of the neuron connections are randomly discarded during training.

[0160] Further specifically, in this step, the settling semantic vector input by the LSTM layer can select the settling semantic vector group of the recent 30 days to form a time sequence (i.e., the sequence length input by the LSTM is 30, and each time step is a 128-dimensional semantic vector), covering short-term fluctuations (such as weekly rate changes) and medium-term trends (such as monthly seasonal effects); further, all semantic vectors can be collected at the same time of day (such as 12:00), ensuring uniform time step intervals (24 hours / step), and if data is missing for a day, linear interpolation can be used to supplement the missing values (based on the feature trends of the semantic vectors of the previous and next two days); further, time feature encoding (such as month encoding: 1-12, date encoding: 1-31) can be added to the time sequence, which is converted into a 10-dimensional vector and concatenated with the 128-dimensional semantic vector to form a 138-dimensional / step input sequence (enhancing the model's sensitivity to seasonality and date). Further, the time sequence of the recent 30 days can be traversed, and the forget gate automatically filters out invalid short-term features (such as abnormal fluctuations in semantic vectors caused by temporary sensor failure on a single day), retaining valid long-term features (such as the rule that the rate feature rises by 10%-15% during rainy days (when the humidity feature in the semantic vector is greater than 0.8) in the past 30 days); further, the input gate can receive the semantic vector features of the current time step, combined with the historical cell state (such as the rate change trend of the previous 29 days), to update the cell state to a fusion state of the current feature and the historical trend (such as the current rate of 8 mm / month, combined with the trend of a 30% rate increase during the previous 29-day rainy season, the cell state records that the rate may rise to 10.4 mm / month in the next month); further, the output gate can filter the cell state and output a time dimension evolution rule vector (64-dimensional), which can include short-term rules, medium-term rules, and long-term rules. The short-term rules (dimensions 1-20) can show the rate change trend in the next 7 days (such as stable rate in the first 3 days, and rising to 9 mm / month due to rainfall in the fourth to seventh days); the medium-term rules (dimensions 21-40) can show the amplitude fluctuation range in the next 30 days; and the long-term rules (dimensions 41-64) can show the overall rate trend in the next 3 months (such as rising rate due to the rainy season in the previous 2 months, and falling to 7 mm / month after the end of the rainy season in the third month).

[0161] In step S183, the settling rule is input into the preset space-time analysis model, and a target settling rule corresponding to the spatial evolution feature and the time dimension evolution rule is matched to infer the future settling trend according to the target settling rule.

[0162] Specifically, in this step, the settlement law constructed in S170 can be converted into a law feature vector (128 dimensions), which can be split into time sequence law features, spatial law features, and causal law features according to the law type, wherein the time sequence law features (dimensions 1-40) can be, for example, the rain season rate increase, the winter rate decrease, and the average annual rate decrease in the past 5 years; the spatial law features (dimensions 41-80) can be, for example, the average expansion speed of the settlement hot spot area, the difference between the hot spot area and the suburban settlement amount, and the regional correlation coefficient; the causal law features (dimensions 81-128) can be, for example, the confidence (0.8) that the ground subsidence rate increases by 50% when the rainfall is greater than 100 mm / month and the groundwater level decreases by more than 0.3 m / month. All features can be normalized by min-max to ensure the same amplitude as the spatial and time features.

[0163] In matching the target settlement law corresponding to the spatial evolution feature and the time dimension evolution law, the spatial evolution feature of S181, the time evolution law vector of S182, and the historical law feature vector can be spliced into a 320-dimensional comprehensive feature vector. Then, the cosine similarity between the current comprehensive feature vector and each group of historical vectors can be calculated from the historical law database (storing more than 1000 groups of historical law feature vectors in the past 5 years). Further, the top 3 groups of historical laws with the highest similarity can be selected, and their average value can be taken as the target settlement law.

[0164] Further specifically, in predicting the future settlement trend based on the target settlement law, spatial trend prediction, time-settlement amount prediction, and risk level and impact prediction can be performed.

[0165] The spatial trend prediction can combine the spatial expansion speed of the target law with the potential expansion direction of the hot spot area in the current spatial evolution feature to predict the settlement impact range in the future 1 month, 3 months, or 6 months, such as the target law showing an average annual expansion of 500 meters in the hot spot area, and the current spatial feature showing the highest potential expansion in the northeast direction. Then, it is predicted that the hot spot area will expand 125 meters northeast in the future 3 months (0.25 years), covering 2 main roads.

[0166] The time-settlement amount prediction can combine the time sequence increase of the target law with the rate trend of the current time evolution law to calculate the future cumulative settlement amount, such as the current rate being 8 mm / month, and the target law showing a 30% increase in the rain season (1 month in the future). Then, the future 1-month cumulative settlement amount = 8 1.3 30 days / 30 days = 10.4 mm; future 3 months (1 month of rain season + 2 months of non-rain season, non-rain season rate 6 mm / month) cumulative = 10.4 + 6 2 = 22.4 mm;

[0167] Wherein, the risk level and the influence speculation can combine the cause-effect correlation of the target law (such as the ground subsidence rate greater than 10mm / month means high risk), so that the risk level of future trend can be determined based on the cause-effect correlation, such as the subsidence rate in the future 1 month is 10.4mm / month, which means high risk, and the influence on the infrastructure is speculated (such as the trunk road A section accumulates subsidence of 22.4mm in the future 3 months, which exceeds the safety threshold of 20mm, and the roadbed needs to be reinforced).

[0168] It can be understood that the above embodiment of the application extracts the spatial evolution characteristics such as the expansion direction and the boundary contour of the hot area by the STCN, and the LSTM captures the short-term rate fluctuation, the medium and long-term trend and the like, and then the cosine similarity matching is performed with the historical subsidence law library to form the bidirectional calibration of the real-time features and the historical benchmark. Compared with the traditional model which can only output a single subsidence value, the above embodiment of the application can simultaneously and accurately predict the spatial influence range and the time-subside curve, and the adaptability to new scenes (without direct historical data matching) is improved (reliable prediction is realized by similar historical law migration), and the core problems of the traditional prediction, such as spatial ambiguity, time inaccuracy and poor scene adaptation, are completely solved.

[0169] As shown in Figure 6 It is a structural schematic diagram of an analysis device for ground subsidence provided by an embodiment of the application, which comprises the following modules.

[0170] The sensor module 610 is arranged in the monitoring area and is used to acquire ground subsidence data;

[0171] The optical fiber transmission module 620 comprises a plurality of optical fibers, the outer layer of the optical fiber is an optical fiber outer layer modified by a nano material or a flexible composite material, the optical fiber is electrically connected with the sensor module 610, is used to receive the ground subsidence data, and transmits the ground subsidence data;

[0172] The control module 630 is electrically connected with the optical fiber transmission module 620, is used to receive the transmitted ground subsidence data, and outputs data analysis results after analyzing the ground subsidence data.

[0173] Specifically, the sensor module 610 serves as a data acquisition terminal and is responsible for capturing physical signals related to ground subsidence to provide an original data source for subsequent analysis.

[0174] Specifically, the layout position and density of the sensor can be different according to the risk level of the monitoring area. For example, one sensor can be laid every 5-8 meters in a high-risk area (such as a comprehensive risk score greater than 0.7), and one sensor can be laid every 3-5 meters in a key point (such as a building foundation, an underground pipe network interface, and a rock fracture zone). One sensor can be laid every 10-15 meters in a medium-risk area (0.4-0.7), and one sensor can be laid every 20-30 meters in a low-risk area (less than 0.4), to ensure the comprehensiveness and accuracy of signal collection.

[0175] Specifically, the sensor module 610 can use a combination of multiple types of sensors to adapt to different subsidence signal monitoring needs. Specifically, it can include a fiber optic acoustic sensor, a soil strain sensor, and a microclimate sensor. The fiber optic acoustic sensor is used as the core sensor to collect soil vibration acoustic signals (frequency range 0.1-10 Hz) caused by ground subsidence. The soil strain sensor can be laid in conjunction with the soil deformation strain signal to directly reflect the soil compression caused by subsidence. The microclimate sensor can be integrated into the fiber node to synchronously collect environmental signals such as temperature and humidity, providing a reference for subsequent data noise reduction.

[0176] Specifically, the sensor module 610 can output an analog signal (such as an acoustic signal voltage value of 0-5V) for real-time uploading through the fiber transmission module 620.

[0177] Specifically, the fiber transmission module 620 can serve as a signal transmission channel to ensure stable and low-loss transmission of sensor data. The outer layer of the fiber of the fiber transmission module 620 can be modified with nanomaterials (such as carbon nanotubes) or flexible composite materials (such as polyimide), with a thickness of 0.5-1 mm. It can automatically adjust the optical transmission characteristics, such as absorbing moisture and expanding to form a protective film to reduce light signal attenuation in the face of high soil humidity, and the material elastic deformation offsets the effects of thermal expansion and cold contraction to avoid light fiber breakage in the face of extreme temperatures (-10 degrees Celsius to 40 degrees Celsius). In addition, the fiber in the fiber transmission module 620 can use a single-mode fiber to connect each sensor module 610 by fusion splicing. When laying, it can be laid along the soil or rock contact surface, avoiding large underground pipelines and sharp rocks, and can use an S-shaped path to reduce tensile stress to ensure transmission stability.

[0178] Specifically, the control module 630 serves as the central brain of the device, responsible for receiving and processing signals from the transmission module and outputting data analysis results.

[0179] Specifically, the control module 630 can use an industrial-grade single-chip microcomputer as a core processor to support high-speed data operation; and be matched with an AD converter to convert sensor analog signals into digital signals. The control module 630 can receive digital signals of the optical fiber transmission module 620 in real time through an optical fiber interface, support multi-channel parallel reception; the control module 630 can be internally provided with Kalman filtering, short-time Fourier transform and other algorithms to perform noise reduction, time-frequency analysis on received data, and extract subsidence characteristics (such as frequency, amplitude, subsidence rate); the control module 630 can output data analysis results (such as subsidence amount, risk level) through an Ethernet interface or a wireless module, and support a connected host computer system (such as a city subsidence monitoring platform).

[0180] Optionally, in addition to receiving and processing data, the control module 630 can also integrate local pre-analysis functions, that is, perform simple subsidence rate calculation (such as near-1-hour average rate) and abnormal threshold judgment (directly trigger a local alarm when exceeding a preset threshold) on received sensor data, without waiting for central system instructions, to improve emergency response speed.

[0181] It can be understood that the control module 630 and the sensor module 610 establish bidirectional communication through the optical fiber transmission module 620, the control module 630 can send parameter adjustment instructions (such as sampling frequency, sensitivity) to the sensor module 610, and the sensor module 610 feeds back working states (such as whether offline or signal strength) to realize closed-loop control.

[0182] It can be understood that the above-mentioned embodiment of the application links the nanometer material cladding design (moisture and temperature resistance) of the optical fiber transmission module 620 and the edge pre-analysis function of the control module 630, so that the device is upgraded from a single data acquisition to a real-time processing-dynamic control intelligent unit, the control module 630 can send sampling frequency and sensitivity adjustment instructions to the sensor, the sensor feeds back working states to form a closed loop, and the core pain points of data distortion, control lag and low reliability of traditional devices are solved.

[0183] In an optional implementation manner, as shown in Figure 7 The analysis device of ground subsidence further includes a signal amplification module 640, the signal amplification module 640 includes a plurality of micro acoustic cavities, each micro acoustic cavity is located at each first preset key node around the optical fiber transmission module 620, the signal amplification module 640 is electrically connected with the control module 630, and is used for receiving the ground subsidence data and outputting the ground subsidence data to the control module 630 after amplification.

[0184] Specifically, the signal amplification module 640 can be composed of a micro acoustic cavity, a piezoelectric amplifier, and a signal conditioning circuit. Each micro acoustic cavity can correspond to one first preset key node around the optical fiber transmission module 620. The first preset key node can be specifically the contact surface of soil or rock, the geological structure change point, the settlement sensitive area, and the optical fiber node within a 5-meter range around the key infrastructure (building foundation, bridge support point), and correspond one-to-one with the sensor module 610 (one sensor module 610 is matched with one signal amplification module 640). The micro acoustic cavity design can adopt a cylindrical ceramic acoustic cavity, the inner wall of which can be plated with an aluminum film (to enhance sound wave reflection). The acoustic cavity axis is consistent with the optical fiber transmission direction, and the optical fiber passes through the center of the acoustic cavity. By designing a specific size of the acoustic cavity resonance frequency (0.1-10 Hz, matching the settlement signal frequency), the acoustic signal transmitted by the optical fiber and the environmental sound wave form a resonance coupling, and the signal amplitude is amplified by 3-5 times.

[0185] It can be understood that the weak settlement acoustic signal collected by the sensor module 610 is transmitted to the micro acoustic cavity through the optical fiber, causing the acoustic cavity to resonate, and the signal amplitude is preliminarily amplified. The resonated signal is input into the piezoelectric amplifier (with adjustable gain, range 20-40 dB) in the signal amplification module 640, which further amplifies the signal amplitude to 0.5V-2V, meeting the input requirements of the AD converter. The signal conditioning circuit (including a low-pass filter with a cutoff frequency of 10 Hz) filters the high-frequency noise (greater than 10 Hz) after amplification to ensure signal purity, and then outputs to the control module 630.

[0186] Specifically, when the signal amplification module 640 is physically coupled with the optical fiber transmission module 620, the micro acoustic cavity can be fixed at the optical fiber node by epoxy resin to ensure close contact between the acoustic cavity and the optical fiber. When the signal amplification module 640 is electrically connected with the control module 630, the amplified signal can be transmitted to the Analog-to-Digital (AD) converter input end of the control module 630 through a shielded cable. The control module 630 can adjust the gain of the piezoelectric amplifier through instructions (e.g., adjusting to 40 dB for weak signals and adjusting to 20 dB for strong signals), realizing adaptive amplification.

[0187] Specifically, in order to realize resonance coupling between the micro acoustic cavity and the optical fiber, a micro groove consistent with the optical fiber transmission direction can be formed in the inner wall of the acoustic cavity, so that the contact area is increased to 95% after the optical fiber is embedded, and the resonance coupling efficiency is increased from 90% to 98%.

[0188] It can be understood that the above-mentioned embodiment of the application is different from the defect of the traditional fixed gain amplifier that blindly amplifies all signals (including noise). Through the customized coupling design of the micro acoustic cavity and the optical fiber, the amplitude of the weak signal is increased after resonance coupling amplification, solving the missed detection problem caused by the simultaneous amplification of noise and signal in the traditional amplification technology.

[0189] In an alternative implementation, as shown in Figure 8 The analysis device of land subsidence further comprises:

[0190] An environmental parameter acquisition module 650 is arranged at each second preset key node around the optical fiber transmission module 620, and is electrically connected with the control module 630, for acquiring the value of the environmental parameter of the second preset key node, and outputting the value of the environmental parameter to the control module 630, so that the control module 630 matches the optical fiber energy collection strategy corresponding to the value of the environmental parameter.

[0191] An energy collection module 660 is electrically connected with the control module 630, for receiving the optical fiber energy collection strategy, and collecting and storing energy based on the optical fiber energy collection strategy.

[0192] Specifically, the environmental parameter acquisition module 650 can be composed of a micro temperature and humidity sensor, an air pressure sensor, and a soil moisture sensor, and is arranged at the second preset key node around the optical fiber transmission module 620. The second preset key node can be set to be within 1 meter around the optical fiber node, and partially overlaps with the first preset key node (the arrangement point of the signal amplification module 640). At least one environmental parameter acquisition module 650 can be arranged for every 5 optical fiber nodes, to ensure the spatial coverage of environmental data.

[0193] The temperature and humidity sensor is used to collect air temperature and air humidity, the air pressure sensor is used to collect atmospheric pressure, and the soil moisture sensor is used to collect soil volume water content, directly reflecting the influence of soil humidity on subsidence.

[0194] Specifically, the environmental parameter acquisition module 650 can be electrically connected with the control module 630 through an I2C communication protocol, and real-time upload the environmental parameter value to the control module 630. Further specifically, after receiving the environmental parameter, the control module 630 can construct an environment-subside correlation model in combination with subsidence data, and match the corresponding optical fiber energy collection strategy. For example, when the soil humidity is greater than 35%, the piezoelectric energy collection efficiency of soil is improved, and the control module 630 instructs the energy collection module 660 to preferentially start the piezoelectric collection mode. When the air temperature is greater than 35 degrees Celsius, the ground heat gradient power generation efficiency is improved, and the instruction preferentially starts the heat gradient collection mode.

[0195] The energy collection module 660 can be composed of a ground thermal gradient power generation unit, a soil piezoelectric power generation unit, a vibration energy power generation unit, an energy storage unit (such as a lithium battery), and an energy management unit, integrated at a fiber convergence point (one energy collection module 660 for every 10 fiber nodes) to supply power to the sensor module 610, the signal amplification module 640, and the control module 630. Specifically, the ground thermal gradient power generation unit can utilize the temperature difference between soil and air to output voltage using a bismuth telluride thermoelectric generator; the soil piezoelectric power generation unit can utilize the pressure deformation of soil settlement to generate electricity using a piezoelectric ceramic sheet to output voltage; and the vibration energy power generation unit can utilize environmental vibrations (such as vehicle travel and wind) to generate electricity using an electromagnetic induction type vibration generator to output voltage.

[0196] It can be understood that the energy management unit monitors the output power of each power generation unit and the lithium battery capacity in real time, and optimizes the energy collection efficiency through a maximum power point tracking algorithm, such as stopping charging when the lithium battery capacity is greater than or equal to 80%, starting all power generation units when the capacity is less than 20%, and matching the collection strategy according to the environmental parameters when the capacity is between 20% and 80%.

[0197] Specifically, the values of the environmental parameters and the fiber energy collection strategy can be set based on specific application requirements, such as when the soil humidity is greater than 35% and the soil settlement rate is greater than 5 mm / month, the soil piezoelectric power generation efficiency is high, and the soil piezoelectric power generation unit is preferentially enabled, and the vibration energy power generation unit is auxiliary enabled; when the air temperature is greater than 35 degrees Celsius or less than 5 degrees Celsius, the ground thermal gradient temperature difference is large, and the ground thermal gradient power generation unit can be preferentially enabled; when the wind speed is greater than 3 m / s, the vibration energy power generation efficiency is high, and the vibration energy power generation unit can be preferentially enabled, and other units can be auxiliary enabled; in extreme environments (such as heavy rain and low temperature), all power generation units can be controlled to be enabled at the same time to ensure stable power supply.

[0198] Specifically, the energy collection module 660 can automatically adjust various energy collection and storage logic, such as when the lithium battery capacity is between 20% and 50%, preferentially allocating 80% of the power to the sensor module 610 and 20% to the transmission module; when the capacity is between 50% and 80%, the power is evenly distributed.

[0199] It can be understood that the above-mentioned embodiment of the application collects environmental parameter data through the environmental parameter module and dynamically matches the power generation strategy through the linkage control module 630. Compared with traditional devices that rely on solar energy or simple lithium battery power supply devices, this technology can adapt to various environments and achieve full-environment self-power supply.

[0200] The technical solutions provided by the present application are described in detail above, the principles and implementation manners of the present application are described by using specific examples, the above examples are only used to help understand the method and core idea of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges can be changed, and the above description should not be understood as a limitation of the present application.

[0201] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Accordingly, the present application can be embodied in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can be embodied in the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk memory, CD-ROMs, optical memory, etc.) having computer usable program code embodied thereon.

[0202] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The functions specified in a flow or multiple flows and / or blocks

[0203] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The functions specified in a flow or multiple flows and / or blocks

[0204] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The functions specified in a flow or multiple flows and / or blocks

[0205] Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A method of analyzing ground subsidence, characterized by, include: Obtain geological and meteorological data for the monitoring area; Based on the geological data, assess the risk of foundation subsidence in the monitored area; Match sensor deployment strategies with the aforementioned basic settlement risk and the aforementioned meteorological data; Sensors deployed according to the aforementioned sensor deployment strategy collect ground subsidence data for the monitored area. Time-frequency analysis was performed on the ground subsidence data to obtain the ground acoustic spectrum corresponding to the ground subsidence data; The ground acoustic spectrum is input into a preset neural network, and the corresponding subsidence semantic vector is output. Acquire and analyze historical ground subsidence data to construct the subsidence patterns corresponding to the historical ground subsidence data; The settlement semantic vector and the settlement law are input into a preset spatiotemporal analysis model to predict the future settlement trend of the monitored area.

2. The method for analyzing ground subsidence according to claim 1, characterized in that, The sensor deployment strategy that matches the basic settlement risk and the meteorological data includes: Analyze the weighting of the impact of the meteorological data on the foundation settlement risk; Based on the foundation settlement risk and the influence weight, the sensor deployment density in the sensor deployment strategy is determined, wherein the sensor deployment density is proportional to both the foundation settlement risk and the influence weight.

3. The method for analyzing ground subsidence according to claim 2, characterized in that, After determining the sensor deployment density in the sensor deployment strategy based on the foundation settlement risk and the impact weight, the method further includes: Analyze the extent to which the meteorological data affects the sensitivity of the sensor; Based on the degree of influence of the meteorological data on the sensor's sensitivity, the sensor's sensitivity or operating mode is adjusted.

4. The method for analyzing ground subsidence according to claim 1, characterized in that, The step of performing time-frequency analysis on the ground subsidence data to obtain the corresponding geosonic spectrum includes: Wavelet transform is performed on the ground subsidence data to separate the low-frequency subsidence signal and high-frequency noise. The wavelet coefficient features of the high-frequency noise are clustered in real time to generate the inverse noise signal of the high-frequency noise; By superimposing the reverse noise signal and the ground subsidence data, noise-free ground subsidence data is obtained; Time-frequency analysis was performed on the noise-free ground subsidence data to obtain the corresponding ground acoustic spectrum.

5. The method for analyzing ground subsidence according to claim 1, characterized in that, The acquisition and analysis of historical ground subsidence data, and the construction of subsidence patterns corresponding to the historical ground subsidence data, include: Obtain the historical ground subsidence data; In the historical ground subsidence data, the similarity of the historical ground subsidence data between the preset nodes is obtained by analyzing the historical ground subsidence data of the monitoring area in the same time period or adjacent time periods between preset nodes. When the similarity of the historical ground subsidence data is lower than a preset threshold, the historical ground subsidence data corresponding to the preset node is identified as historical abnormal subsidence data. Construct the settlement patterns corresponding to the historical abnormal settlement data.

6. The method for analyzing ground subsidence according to claim 1, characterized in that, The acquisition and analysis of historical ground subsidence data, and the construction of subsidence patterns corresponding to the historical ground subsidence data, include: Obtain historical ground subsidence data; The historical ground subsidence data is input into the spatiotemporal autoregressive model, and the temporal correlation results and spatial correlation results of the historical ground subsidence data are output. The time-series correlation results are input into a preset time-series analysis model to obtain the long-term trend of historical settlement rate. Based on spatial correlation analysis technology, the diffusion path of the historical subsidence range is obtained by analyzing the spatial correlation results. Based on the association rule mining algorithm, the temporal association results and the spatial association results are mined to obtain the causal association data between environmental conditions and subsidence patterns in the historical ground subsidence data. The long-term trend of the historical subsidence rate, the diffusion path of the historical subsidence range, and the causal correlation data are used as the subsidence law and associated with the historical ground subsidence data.

7. The method for analyzing ground subsidence according to claim 1, characterized in that, The step of inputting the settlement semantic vector and the settlement law into a preset spatiotemporal analysis model to predict the future settlement trend of the monitored area includes: The settlement semantic vector is input into the spatiotemporal convolutional neural network in the preset spatiotemporal analysis model to extract the spatial evolution features in the settlement semantic vector; The settlement semantic vector is input into the LSTM layer of a preset spatiotemporal analysis model to obtain the evolution law of the settlement semantic vector in the time dimension. The settlement pattern is input into the preset spatiotemporal analysis model, and the target settlement pattern corresponding to the spatial evolution characteristics and the evolution pattern of the time dimension is matched, so as to predict the future settlement trend based on the target settlement pattern.

8. A ground settlement analysis apparatus, said apparatus being used to implement the steps of the ground settlement analysis method as described in any one of claims 1-7, characterized in that, include: The sensor module, located in the monitoring area, is used to acquire ground subsidence data; The optical fiber transmission module includes several optical fibers. The outer layer of each optical fiber is modified by nanomaterials or flexible composite materials. The optical fibers are electrically connected to the sensor module and are used to receive and transmit the ground subsidence data. The control module, electrically connected to the optical fiber transmission module, is used to receive the transmitted ground subsidence data, analyze the ground subsidence data, and output the data analysis results.

9. The ground subsidence analysis device according to claim 8, characterized in that, It also includes a signal amplification module, which includes several miniature acoustic cavities. Each miniature acoustic cavity is located at a first preset key node around the optical fiber transmission module. The signal amplification module is electrically connected to the control module and is used to receive the ground subsidence data and amplify the ground subsidence data before outputting it to the control module.

10. The ground subsidence analysis apparatus according to claim 8, characterized in that, Also includes: An environmental parameter acquisition module is provided, which is set at each of the second preset key nodes around the optical fiber transmission module. The environmental parameter acquisition module is electrically connected to the control module and is used to acquire the values ​​of the environmental parameters of the second preset key nodes and output the values ​​of the environmental parameters to the control module so that the control module matches the optical fiber energy harvesting strategy corresponding to the values ​​of the environmental parameters. An energy harvesting module, electrically connected to the control module, is used to receive the fiber optic energy harvesting strategy, and to harvest and store energy based on the fiber optic energy harvesting strategy.

Images