Surface collapse dynamic early warning method based on probability data assimilation edge calculation

Abstract

The invention discloses a surface collapse dynamic early warning method based on probability data assimilation edge calculation. The method comprises the steps of sensor arrangement, data acquisition, probability data assimilation, dynamic threshold calculation and anomaly recognition and early warning. The method comprises the following steps of: horizontally burying an HDPE (High-Density Polyethylene) pipeline provided with an FCF (Fiber Filter Filter) sensor in a surface collapse waiting area to form a quasi-distributed flexible sensor array; acquiring and processing resistance change rate signals of each channel in the distributed flexible sensor array, and calculating the resistance change rate of each channel in each time step; carrying out dynamic weight calculation and data assimilation on the resistance change rate signal; performing dynamic calculation on the assimilation resistance change rate signal in the sliding window to obtain a local mean value and a standard deviation of the assimilation resistance change rate signal; constructing a dynamic confidence interval; and when the assimilation resistance change rate continuously exceeds the upper and lower boundaries of the dynamic confidence interval and reaches a preset time step number, early warning is carried out. According to the invention, the response efficiency and early warning precision of the surface collapse event are improved.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, and in particular to a dynamic early warning method for ground subsidence based on probabilistic data assimilation edge calculation. Background Technology

[0002] Ground subsidence is a geological hazard process in which the surface soil and rock layers suddenly sink or break down, forming pits or depressions. It is characterized by complex mechanisms, sudden onset, and rapid evolution, and is widely found in karst areas, mining subsidence areas, areas with dense underground engineering, and soft foundations. In recent years, with the accelerated development of urban underground space and the increasing frequency of extreme weather events, the risk of ground subsidence has significantly increased, becoming one of the major hidden dangers affecting urban safety and green development. Therefore, developing automated and intelligent ground subsidence monitoring and early warning methods is crucial.

[0003] Current ground subsidence monitoring methods mostly rely on point sensors such as resistance strain gauges, levels, and crack gauges, which often suffer from inflexible deployment, significant blind spots, and poor coupling with the soil. Distributed fiber optic sensors offer advantages in long-distance sensing, but are prone to physical damage under conditions of strong disturbance or large deformation. Therefore, under complex geological conditions, the above-mentioned sensing methods are insufficient to fully perceive the entire process of ground subsidence, especially during the rapid deformation process in the imminent stage of subsidence, making it difficult to guarantee monitoring effectiveness.

[0004] Furthermore, traditional geological disaster monitoring systems generally employ a centralized data processing architecture, where data collected on-site is uploaded to remote servers or cloud platforms for analysis. This model relies on stable data links and high computing resources, resulting in problems such as large communication latency, slow response speed, and high system latency, making it difficult to meet the urgent needs of ground subsidence disasters for rapid response and edge processing capabilities. Once communication is interrupted or computing resources are strained, the system will face risks such as delayed perception, interrupted early warnings, and false alarms or missed alarms, lacking practicality and robustness.

[0005] Regarding early warning algorithms, existing methods mostly rely on empirical rules to set static thresholds for single indicators such as displacement increment, settlement rate, or groundwater level changes. These methods struggle to address the nonlinear evolution and dynamic trend changes during the collapse process. Especially during periods of significant noise interference or abrupt trend changes, static threshold early warning algorithms are prone to false alarms or missed alarms, exhibiting insufficient robustness and generalization ability. Therefore, current technologies have not yet provided a systematic solution integrating high-sensitivity awareness, edge computing, and dynamic threshold early warning. Summary of the Invention

[0006] Objective of the Invention: To address the shortcomings of the existing technologies, this invention proposes a dynamic early warning method for ground subsidence based on probabilistic data assimilation edge computing. This method solves the problems of traditional rigid or point sensors in complex geological conditions, such as small range, difficult deployment, easy damage, and blind spots. It improves the response efficiency and early warning accuracy of ground subsidence events, reduces the dependence on remote transmission and cloud computing, and enhances the real-time performance and accuracy of monitoring and early warning. This provides support for the early identification and intelligent early warning of ground subsidence disasters in complex geological environments.

[0007] Technical solution: The early warning system adopted in this invention includes hardware and software components:

[0008] The hardware components include a flexible conductive thin-film sensor, a multi-channel analog-to-digital converter (ADC), a Zynq-7000 chip, and a wireless communication module. The Zynq-7000 chip serves as an edge computing platform, integrating a programmable gate array (FPGA) architecture and an advanced reduced instruction set machine (ARM) architecture. It is responsible for controlling the operation of modules such as data acquisition, probabilistic data assimilation, dynamic threshold calculation, and anomaly identification and early warning.

[0009] The software component includes high-frequency synchronous acquisition and preliminary processing of multi-channel signals, a trend-compensated exponentially weighted moving average (TC-EWMA) model, a recursive Bayesian weighted fusion (RBWF) algorithm, and a continuous anomaly time step judgment mechanism.

[0010] This invention relates to a dynamic early warning method for ground subsidence based on probabilistic data assimilation edge computation, comprising the following steps:

[0011] 1) Set several equidistant cross sections on the surface of the HDPE pipe, and install FCF sensors on each cross section. Horizontally bury the HDPE pipe with FCF sensors in the area where ground subsidence is expected to occur to form a quasi-distributed flexible sensor array, which enhances the coupling between the sensor and the soil and reduces the occurrence of blind spots.

[0012] 2) The resistance change rate signals of each channel in the distributed flexible sensor array are synchronously acquired and processed by a multi-channel analog-to-digital converter, and the resistance change rate of each channel at each time step t is calculated.

[0013] 3) Based on the ARM architecture of the Zynq-7000 chip, a recursive Bayesian weighted fusion (RBWF) algorithm is used to dynamically calculate weights and assimilate data from multi-channel resistance change rate signals from a quasi-distributed flexible sensor array. The process is as follows: at each sampling time, the residual of the resistance change rate signal for each channel is calculated, and the noise variance of the corresponding channel is calculated based on this residual; the weight of each channel's resistance change rate signal in the assimilated resistance change rate signal is determined according to the magnitude of the noise variance (the smaller the noise variance, the larger the weight); then, all channel resistance change rate signals are weighted and summed to obtain the assimilated resistance change rate signal. The assimilated resistance change rate signal is used to characterize the macroscopic deformation trend of the observation area as a whole, thereby reducing the impact of single-channel noise on the stability of the early warning system.

[0014] 4) Based on the ARM architecture of the Zynq-7000 chip, the trend-compensated exponentially weighted moving average (TC-EWMA) method is used to dynamically calculate the assimilation resistance change rate signal within a sliding window, obtaining the local mean and standard deviation of the assimilation resistance change rate signal. The assimilation resistance change rate signal within the sliding window is fitted and trend-compensated based on local least squares regression. The dynamic confidence interval is constructed by combining the trend compensation results with the local mean and standard deviation of the assimilation resistance change rate signal, and the upper and lower bounds of the dynamic confidence interval are obtained as reference thresholds for subsequent anomaly identification and early warning judgment.

[0015] 5) At each time step, the assimilation resistance change rate signal is compared with the constructed dynamic confidence interval in real time. When the assimilation resistance change rate signal value continuously exceeds the upper and lower boundaries of the dynamic confidence interval for a preset number of time steps, it is determined to be a ground collapse early warning event and an early warning is issued.

[0016] In step 3), the assimilation resistance change rate signal The calculation process is as follows:

[0017] 3.1) First, initialize the assimilation resistance change rate signal of the multi-channel assimilation resistance change rate in the sensor array at the previous moment. Assimilation covariance P t-1 and noise variance Calculate the residual for each channel at each time step t:

[0018] 3.2) Real-time calculation of the noise variance of each channel signal: Where λ is the attenuation factor;

[0019] 3.3) Calculate the weight w of each channel's resistance change rate signal in the assimilated resistance change rate signal at time t. i,t :

[0020]

[0021] Where N is the total number of channels in the sensor array, and all weights w i,t The sum of these is 1; This represents the current noise variance; The current assimilation posterior covariance;

[0022] 3.4) Obtain the assimilation resistance change rate signal by weighted averaging of the signals from each channel.

[0023]

[0024] Step 4) involves first calculating the local weighted mean and standard deviation of the assimilated resistance change rate signal using the exponentially weighted moving average method:

[0025]

[0026] Where, μ t σ is the local mean of the assimilation resistance change rate signal within the sliding time window up to the current time step. t μ is the standard deviation of the assimilation resistance rate signal within the sliding time window up to the current time step. t-1 This is the local mean of the assimilation resistance change rate signal within the sliding time window up to the previous time step. The standard deviation of the assimilation resistance rate signal within the sliding time window up to the previous time step is given by α = 2 / (s+1), which is the smoothing coefficient.

[0027] At each time step t, calculate the local slope of the assimilation resistance rate of change signal curve within the current sliding window:

[0028]

[0029] Where, β t denoted as the local slope of the assimilation signal curve, W is the sliding window size used in the least squares regression, and c is the constant term in the least squares regression.

[0030] Construct dynamic confidence intervals and dynamically adjust the upper and lower boundaries of the early warning control threshold:

[0031]

[0032] Among them, U t L is the upper bound of the dynamic confidence interval. t μ is the lower bound of the dynamic confidence interval. t This represents the trend compensation result for the assimilation resistance change rate signal within the sliding time window up to the current time step. K is the statistical confidence factor, corresponding to a confidence level of 95%–99%, γ>0 is the trend compensation coefficient, and γ·βt This is the result of trend compensation.

[0033] In step 4), at each time step t, the local slope of the assimilation resistance rate of change signal curve within a current sliding window is calculated using least squares regression over a sliding window of length W.

[0034] In step 4), the TC-EWMA algorithm is used to construct a dynamic confidence interval and dynamically adjust the upper and lower boundaries of the early warning control threshold.

[0035] In step 4), the early warning conditions for ground subsidence events are derived: or Where ε is the noise tolerance coefficient.

[0036] In step 2), the rate of change of resistance x of channel i at time step t is obtained. i,t =(r i,t -r i,0 ) / t, where r i,t Let r be the resistance value of channel i at time step t. i,0 Let be the initial resistance value of channel i.

[0037] In step 3.2), the noise variance of each channel signal is calculated in real time using a recursive method in the form of a first-order low-pass filter.

[0038] In step 1), the sensor arrangement uses HDPE pipe as a carrier, and sets several equally spaced monitoring sections on its surface. Each section is arranged with a set of FCF sensors, which are arranged along the axis of the HDPE pipe to form a quasi-distributed flexible sensor array.

[0039] In step 1), the conductive material of the FCF sensor is liquid metal gallium alloy, and the substrate is modified polyurethane elastomer. It is encapsulated on the top and bottom surfaces of the HDPE pipe with epoxy resin. The FCF sensors are arranged along the pipe axis, and each FCF sensor transmits resistance change signals through wires.

[0040] In step 5), an anomaly identification and early warning module is employed. Based on a continuous anomaly time-step judgment mechanism, the assimilation resistance change rate signal is compared in real time with the constructed confidence interval at each time step: when the assimilation resistance change rate signal is between the upper and lower boundaries of the confidence interval, it is considered normal, and the system does not alarm; when the assimilation resistance change rate signal value is greater than the upper boundary of the confidence interval or less than the lower boundary, it is considered abnormal; when the assimilation resistance change rate signal exceeds the upper or lower boundary of the confidence interval for multiple consecutive time steps, i.e., the abnormal state occurs in three consecutive time steps, the system determines that a ground collapse event will occur and triggers an early warning signal. The early warning is triggered by the ARM processor in the edge computing platform, issuing a local alarm through LED indicators and a buzzer, or remotely sending early warning information through a wireless communication module, realizing local and remote coordinated response.

[0041] Working Principle: This invention, during early warning, encapsulates multiple flexible conductive film (FCF) sensors onto the surface of a high-density polyethylene (HDPE) pipe. The encapsulated HDPE pipe is then horizontally buried in the soil of an area prone to ground subsidence, forming a quasi-distributed flexible sensor array. A programmable gate array (FPGA) architecture based on the Zynq-7000 chip performs high-frequency synchronous acquisition and subsequent processing of the resistance change rate signals from multiple channels in the flexible sensor array. Leveraging the edge computing power of the Zynq-7000 chip, a recursive Bayesian weighted fusion (RBWF) algorithm is used to process the resistance change rate signals from multiple channels. The system dynamically weights and integrates real-time acquired signals to generate an assimilated resistance change rate signal to comprehensively represent the overall deformation trend of the observation area. It calculates the trend of the assimilated resistance change rate signal based on the exponentially weighted moving average (EWMA) model. Simultaneously, it dynamically constructs confidence intervals based on the TC-EWMA model with trend compensation using local least squares regression. A continuous anomaly identification criterion is adopted: if the value of the assimilated resistance change rate signal exceeds the confidence interval boundary for multiple consecutive time steps, a ground subsidence warning signal is triggered and transmitted locally and remotely via LED indicators, buzzers, or wireless modules.

[0042] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0043] (1) This invention constructs a quasi-distributed flexible sensor array with a stable structure by coupling a flexible conductive thin film sensor with a high-density polyethylene pipe. It has excellent ability to sense large deformations in the strata and effectively solves the problems of small range, difficult deployment, easy damage and blind spots of traditional rigid or point sensors under complex geological conditions. It improves the spatial coverage and measurement accuracy of monitoring and early warning.

[0044] (2) This invention relies on the edge computing power of the Zynq-7000 system-on-a-chip to realize closed-loop processing of multi-channel signal synchronous acquisition, dynamic probability data assimilation and early warning discrimination at the edge, reducing the dependence on remote transmission and cloud computing, and has the advantages of low latency and fast response.

[0045] (3) This invention employs a trend-compensated exponentially weighted moving average algorithm and a continuous anomaly identification mechanism to construct a dynamic threshold early warning strategy, effectively suppressing noise interference and improving the accuracy of collapse event identification. The early warning method of this invention has high integration, high robustness, and strong real-time performance, possessing practical value and engineering application prospects.

[0046] (4) This invention has the characteristics of strong real-time performance, lightweight calculation and dynamic adjustment. It is suitable for intelligent early warning deployment of geological disasters in environments with limited data processing and transmission conditions, and can effectively improve the response efficiency and early warning accuracy of ground subsidence events. Attached Figure Description

[0047] Figure 1 This is a flowchart of the dynamic early warning method for ground subsidence based on probabilistic data assimilation edge calculation according to the present invention;

[0048] Figure 2 This is a scene diagram showing the sensor arrangement of the present invention;

[0049] Figure 3 This is a schematic diagram of the overall structure of the quasi-distributed flexible sensor array of the present invention;

[0050] Figure 4 This is a schematic diagram of the cross-sectional structure of the quasi-distributed flexible sensor array of the present invention;

[0051] Figure 5 This is a flowchart of the RBWF probability data assimilation process of the present invention. Detailed Implementation

[0052] like Figures 1 to 5 As shown, the ground subsidence dynamic early warning method based on probabilistic data assimilation edge calculation of the present invention is as follows: Figure 1 As shown, the process includes the following steps: sensor deployment, data acquisition, probability data assimilation, dynamic threshold calculation, anomaly identification and early warning.

[0053] In the sensor arrangement stage of step 1), several flexible conductive film FCF sensors are encapsulated on the surface of high-density polyethylene (HDPE) pipes, and the pipes are horizontally buried in areas prone to ground subsidence to form a quasi-distributed flexible sensor array.

[0054] In step 2), during the data acquisition stage, the Zynq-7000 chip's programmable gate array (FPGA) architecture is used to achieve high-frequency synchronous acquisition and preliminary processing of the resistance change rate signals of multiple channels in the sensor array, thereby obtaining the resistance change rate signal of the FCF sensor array caused by formation deformation.

[0055] In step 3), the probability data assimilation stage, the recursive Bayesian weighted fusion RBWF algorithm is run on the ARM architecture based on the Zynq-7000 chip to perform dynamic weight calculation and Bayesian weighted assimilation on the multi-channel signals, generating an assimilation resistance change rate signal to characterize the overall deformation state of the soil in the observation area.

[0056] In step 4), during the dynamic threshold calculation stage, the trend-compensated exponentially weighted moving average model TC-EWMA is used to dynamically weight and assimilate the changing trend of the resistance change rate signal, thereby constructing a confidence interval.

[0057] In step 5), anomaly identification and early warning: a continuous anomaly time step judgment mechanism is adopted to improve the robustness of the early warning. When the number of time steps in which the signal continuously exceeds the dynamic threshold range reaches a preset value, a ground collapse early warning is triggered.

[0058] In this embodiment, by comparing the signal assimilation value with the constructed dynamic threshold in real time, if the assimilation value exceeds the dynamic threshold for three consecutive time steps, a ground collapse early warning signal is triggered.

[0059] Figure 2 This is a typical deployment scenario for an FCF sensor array. First, a subsidence-prone area is selected, and the encapsulated FCF sensor array is horizontally deployed above the critical depth of the underground cavities within the area. The FCF sensors are tightly attached to the HDPE pipe carrier. When the carrier deforms under the influence of the soil, the FCF sensors deform synchronously, and their resistance values ​​change accordingly. Therefore, the calculated resistance change rate signal reflects the stress and deformation of the HDPE pipe. Because the HDPE pipe carrier is tightly coupled with the surrounding soil, they can deform synchronously and in coordination. Therefore, the deformation state of the HDPE pipe can be deduced using the obtained FCF sensor resistance change rate signal, and the deformation of the soil at the corresponding location can be further inferred. When geological conditions such as groundwater levels change and underground cavities develop upwards, the instability and deformation of the soil in local areas will exert stress on the HDPE pipe carrier buried in the strata, causing the resistance change rate signal output by the FCF sensor array to continuously change over time. By continuously acquiring and analyzing the resistance change rate signal, the formation and expansion process of underground cavities can be detected, enabling real-time monitoring of underground cavity development.

[0060] In step 1), the sensor array is arranged as follows: Figure 3 As shown, its cross-sectional perspective is as follows Figure 4As shown, an HDPE pipe 1 with an outer diameter greater than 20mm and a length greater than 3m is selected as the carrier. Several monitoring sections are evenly divided on the HDPE pipe according to the actual situation. Each section is equipped with a bottom FCF sensor 2 and a top FCF sensor 3. The FCF sensor is a flexible film 40mm long and 10mm wide, with liquid gallium alloy as the conductive material and modified polyurethane elastomer as the substrate material. Therefore, the FCF sensor has good conductivity, flexibility, and durability. The FCF sensors are all encapsulated on the surface of the HDPE pipe with epoxy resin. The sensor axis is arranged parallel to the pipe direction, and the distance between adjacent sensors is greater than 10cm, forming a stable, blind-spot-free quasi-distributed sensing array. All sensors are connected to the Zynq-7000 SoC edge computing platform 5 via wires 4. Subsequent data acquisition, probability data assimilation, dynamic threshold calculation, and anomaly identification and early warning steps are all implemented on this platform, which integrates FPGA and ARM architectures.

[0061] In step 2), data acquisition is implemented based on the FPGA architecture inside the Zynq-7000 SoC. The FPGA controls the multi-channel analog-to-digital converter (ADC) to perform high-frequency synchronous acquisition of the multi-channel resistance signals of the FCF sensor, with the sampling frequency set to above 10Hz. During the acquisition process, the FPGA architecture completes the conditioning of the acquired signals of each channel, the calculation of the resistance change rate, and the data buffering, which are then used for subsequent probability data assimilation and early warning algorithms.

[0062] The acquired raw signal first enters the FPGA's internal processing unit for signal conditioning (such as filtering and amplification), timing calibration (to ensure alignment of different channels), and resistance change rate calculation, obtaining the resistance change rate signal x for each channel at each time step t. i,t Among them, x i,t =(r i,t -r i,0 ) / t, r i,t Let r be the resistance value of each channel at time step t. i,0 These are the initial resistance values ​​for each channel. These calculations are cached and transmitted to the ARM architecture within the Zynq-7000 SoC for subsequent probabilistic data assimilation calculations.

[0063] In step 3), the probability data assimilation is implemented based on the ARM architecture inside the Zynq-7000 SoC, and the recursive Bayesian weighted fusion (RBWF) algorithm is used to process the multi-channel resistance change rate signal x. i,t (i represents the channel number, t is the time step), integrated into a global data assimilation signal. This represents the overall deformation trend of the observation area. The calculated values ​​of each channel are weighted to obtain the data assimilation result, where the weight is proportional to the reciprocal of the variance of the channel measurement noise, thus ensuring that the channel with less noise has a higher proportion in the assimilation result.

[0064] To achieve real-time dynamic updates of signal weights for each channel, a dynamic calculation of noise variance based on residuals and covariance is adopted to suppress biased performance channels and enhance the reliability and anti-interference capability of signal assimilation results.

[0065] like Figure 5 As shown, the assimilation resistance change rate signal The calculation process is as follows:

[0066] Step 3.1) First, initialize the assimilation resistance change rate signal of the multi-channel assimilation resistance change rate in the sensor array at the previous moment (or initial). Assimilation covariance P t-1 and noise variance Calculate the residual for each channel at each time step t:

[0067] Step 3.2): Calculate the noise variance of each channel signal in real time using a recursive method in the form of a first-order low-pass filter.

[0068] Where λ is the attenuation factor, which can be taken from 0.5 to 0.99. The larger the value, the more attention is paid to historical noise.

[0069] 3.3) Further, the observation information of each channel is dynamically assigned a confidence level based on its uncertainty. The assimilation result of each channel's data is equivalent to the posterior mean, balancing the contribution of each channel's signal to the assimilated signal. Based on the latest noise calculation results, the algorithm further assigns weights to each channel, with the weights proportional to their reliability, which is characterized by the reciprocal of the channel noise (information content). The weight w of each channel's resistance change rate signal in the assimilated resistance change rate signal at time t is calculated. i,t :

[0070]

[0071] Where N is the total number of channels in the sensor array, and all weights w i,t The sum of these is 1; This represents the current noise variance; This represents the current posterior covariance after assimilation. After data assimilation is complete... and P t This serves as the input for the next time step, thus forming a recursive structure.

[0072] 3.4) The assimilation resistance change rate signal is calculated by weighted averaging of the signals from each channel:

[0073]

[0074] Obtain the assimilation resistance change rate signal Then, proceed to step 4 of the dynamic threshold calculation.

[0075] In step 4), based on the ARM architecture, the TC-EWMA model is used to calculate the assimilation resistance change rate signal in real time. The signal within the sliding window is fitted and trend compensated based on local least squares regression. The local weighted mean and standard deviation of the assimilation resistance change rate signal are calculated. The dynamic confidence interval is constructed in combination with the trend compensation results as a reference threshold for anomaly identification and early warning judgment.

[0076] This step is also implemented based on the ARM architecture of the Zynq-7000 SoC. It uses the trend-compensated exponentially weighted moving average (TC-EWMA) algorithm to obtain the overall trend and fluctuation range of the current signal and form dynamic warning upper and lower boundaries.

[0077] Step 4) involves setting a fixed-length sliding time window s at each time step t, with a window size ranging from 20 to 100. First, the local mean and standard deviation of the assimilated signal are calculated using the exponentially weighted moving average method.

[0078]

[0079] Where, μ t σ is the mean of the assimilated signal within the sliding time window up to the current time step. t μ is the standard deviation of the assimilated signal within the sliding time window up to the current time step. t-1 This is the mean of the assimilated signal within the sliding time window up to the previous time step. Let α = 2 / (s+1) be the variance of the assimilated signal within the sliding time window up to the previous time step, and let α = 2 / (s+1) be the smoothing coefficient, which depends on the size of the sliding time window. To reduce false alarms in the presence of long-term trends, it is necessary to use the changes in the assimilated signal near the current time as trend compensation. Therefore, at each time step t, a local slope is calculated using least-squares regression over a sliding window of length W:

[0080]

[0081] Where, β tThe local slope of the assimilation signal curve quantifies the local linear trend of the signal, i.e., the slope of the assimilation resistance rate curve at that point. W is the sliding window size used in the least squares regression, and c is the constant term in the least squares regression. Next, using the slope calculation result, a dynamic confidence interval is constructed using the TC-EWMA algorithm to dynamically adjust the upper and lower boundaries of the early warning control threshold.

[0082]

[0083] Among them, U t L is the upper bound of the dynamic confidence interval. t μ is the lower bound of the dynamic confidence interval. t Let γ be the mean of the assimilated signal within the sliding time window up to the current time step, K be the statistical confidence factor, typically 2–3, corresponding to a 95%–99% confidence level, γ>0 be the trend compensation coefficient, typically 0.1–1, and γ·β t This is the result of trend compensation. Slope β t The upper and lower bounds of the confidence interval for the early warning threshold control were modified to adapt to the monotonic trend of the assimilation resistance change rate signal, maintaining sensitivity to true outliers without overreacting to trend changes. In summary, the early warning conditions for ground subsidence events are as follows:

[0084] or

[0085] Where ε is the noise tolerance coefficient, with a value range of [0,1]. It allows the assimilated resistance change rate signal to deviate slightly from the upper and lower limits without immediately triggering an alarm, thus dealing with small and high-frequency fluctuations in the signal and reducing false alarms.

[0086] Step 5) Anomaly identification and early warning are also implemented based on the ARM architecture of the Zynq-7000 SoC. Based on the dynamically calculated and updated confidence interval described above, the assimilation resistance change rate signal is compared with the constructed confidence interval in real time at each time step. If the above-mentioned ground subsidence event early warning conditions are met, it is considered an anomaly. To further improve the robustness of the early warning algorithm, a continuous anomaly judgment mechanism is adopted: when the assimilation resistance change rate signal continuously exceeds the upper and lower boundaries of the dynamic confidence interval for a preset number of time steps (e.g., 3 consecutive steps), it is determined that there is sudden instability or potential ground subsidence risk in the strata of the observation area, thereby triggering local alarm devices (e.g., buzzers, LED lights) and sending alarm information to the remote monitoring platform via a wireless module.

[0087] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A dynamic early warning method for ground subsidence based on probabilistic data assimilation edge computation, characterized in that: Includes the following steps: 1) Set several equidistant cross sections on the surface of the HDPE pipe, and install FCF sensors on each cross section. Horizontally bury the HDPE pipe with FCF sensors in the area where ground subsidence is expected to occur to form a quasi-distributed flexible sensor array. 2) The resistance change rate signals of each channel in the distributed flexible sensor array are synchronously acquired and conditioned by a multi-channel analog-to-digital converter, and the resistance change rate of each channel at each time step t is calculated. 3) The RBWF algorithm is used to dynamically weight and assimilate the multi-channel resistance change rate signals from the quasi-distributed flexible sensor array. The process is as follows: at each sampling time, the residual of the resistance change rate signal of each channel is calculated, and the noise variance of the corresponding channel is calculated based on the residual; the weight of the resistance change rate signal of each channel in the assimilated resistance change rate signal is determined according to the magnitude of the noise variance of each channel; then, the resistance change rate signals of all channels are weighted and summed to obtain the assimilated resistance change rate signal. 4) Using the exponentially weighted moving average method, the assimilation resistance change rate signal is calculated within a sliding window to obtain the local mean and standard deviation of the assimilation resistance change rate signal; the assimilation resistance change rate signal within the sliding window is fitted and trend compensated based on local least squares regression; a dynamic confidence interval is constructed based on the obtained trend compensation results and the local mean and standard deviation of the assimilation resistance change rate signal, and the upper and lower bounds of the dynamic confidence interval are obtained. 5) At each time step, the assimilation resistance change rate signal is compared with the constructed dynamic confidence interval in real time. When the assimilation resistance change rate signal value continuously exceeds the upper and lower boundaries of the dynamic confidence interval for a preset number of time steps, it is determined to be a ground collapse early warning event and an early warning is issued.

2. The method for dynamic early warning of ground subsidence based on probabilistic data assimilation edge calculation according to claim 1, characterized in that: In step 3), the assimilation resistance change rate signal The calculation process is as follows: 3.1) First, initialize the assimilation resistance change rate signal of the multi-channel assimilation resistance change rate in the sensor array at the previous moment. Assimilation covariance P t-1 and noise variance Calculate the residual for each channel at each time step t: 3.2) Real-time calculation of the noise variance of each channel signal: Where λ is the attenuation factor; 3.3) Calculate the weight w of each channel's resistance change rate signal in the assimilated resistance change rate signal at time t. i,t : Where N is the total number of channels in the sensor array, and all weights w i,t The sum of these is 1; This represents the current noise variance; The current assimilation posterior covariance; 3.4) Obtain the assimilation resistance change rate signal by weighted averaging of the signals from each channel.

3. The method for dynamic early warning of ground subsidence based on probabilistic data assimilation edge calculation according to claim 1, characterized in that: Step 4) involves first calculating the local weighted mean and standard deviation of the assimilated resistance change rate signal using the exponentially weighted moving average method: Where, μ t σ is the local mean of the assimilation resistance change rate signal within the sliding time window up to the current time step. t μ is the standard deviation of the assimilation resistance rate signal within the sliding time window up to the current time step. t-1 This is the local mean of the assimilation resistance change rate signal within the sliding time window up to the previous time step. The standard deviation of the assimilation resistance rate signal within the sliding time window up to the previous time step is given by α = 2 / (s+1), which is the smoothing coefficient. At each time step t, calculate the local slope of the assimilation resistance rate of change signal curve within the current sliding window: Where, β t denoted as the local slope of the assimilation signal curve, W is the sliding window size used in the least squares regression, and c is the constant term in the least squares regression. Construct dynamic confidence intervals and dynamically adjust the upper and lower boundaries of the early warning control threshold: Among them, U t L is the upper bound of the dynamic confidence interval. t μ is the lower bound of the dynamic confidence interval. t This represents the trend compensation result for the assimilation resistance change rate signal within the sliding time window up to the current time step. K is the statistical confidence factor, corresponding to a confidence level of 95%–99%, γ>0 is the trend compensation coefficient, and γ·β t This is the result of trend compensation.

4. The method for dynamic early warning of ground subsidence based on probabilistic data assimilation edge calculation according to claim 3, characterized in that: In step 4), at each time step t, the local slope of the assimilation resistance rate of change signal curve within a current sliding window is calculated using least squares regression over a sliding window of length W.

5. The method for dynamic early warning of ground subsidence based on probabilistic data assimilation edge calculation according to claim 3, characterized in that: In step 4), the TC-EWMA algorithm is used to construct a dynamic confidence interval and dynamically adjust the upper and lower boundaries of the early warning control threshold.

6. The method for dynamic early warning of ground subsidence based on probabilistic data assimilation edge calculation according to claim 3, characterized in that: In step 4), the early warning conditions for ground subsidence events are derived: or Where ε is the noise tolerance coefficient.

7. The method for dynamic early warning of ground subsidence based on probabilistic data assimilation edge calculation according to claim 1, characterized in that: In step 1), the FCF sensor is arranged along the axis of the HDPE pipe.

8. The method for dynamic early warning of ground subsidence based on probabilistic data assimilation edge calculation according to claim 1, characterized in that: In step 2), the rate of change of resistance x of channel i at time step t is obtained. i,t =(r i,t -r i,0 ) / t, where r i,t Let r be the resistance value of channel i at time step t. i,0 Let be the initial resistance value of channel i.

9. The method for dynamic early warning of ground subsidence based on probabilistic data assimilation edge calculation according to claim 2, characterized in that: In step 3.2), the noise variance of each channel signal is calculated in real time using a recursive method in the form of a first-order low-pass filter.

10. The method for dynamic early warning of ground subsidence based on probabilistic data assimilation edge calculation according to claim 1, characterized in that: In step 1), the conductive material of the FCF sensor is liquid metal gallium alloy, and the substrate is modified polyurethane elastomer.

Images