Open caisson construction soil gushing dynamic early warning system based on multi-parameter fusion

Abstract

The invention relates to the field of civil engineering construction safety monitoring, and discloses an open caisson construction soil gushing dynamic early warning system based on multi-parameter fusion, which comprises a sensor array module, a data processing module, a dynamic parameter correlation analysis module, a self-adaptive covariance optimization module, a risk index generation module and an early warning output module, collecting data in real time through a multi-source sensor and generating a parameter vector; the dynamic parameter correlation analysis module updates the dynamic matrix on line and represents real-time correlation among soil pressure, displacement and water level; the adaptive covariance optimization module dynamically adjusts a covariance matrix and suppresses noise interference; the risk index generation module fuses the dynamic matrix, the covariance inverse matrix and the parameter vector calculation weight, and corrects the risk index in combination with the vibration state of the construction machinery; and the early warning output module triggers grading alarm and feeds back to form a closed loop optimization link. According to the invention, real-time accurate early warning and self-adaptive prevention and control of the soil gushing risk are realized.

Description

Technical Field

[0001] The invention relates to the technical field of civil engineering construction safety monitoring, and in particular to a dynamic early warning system for soil gushing during caisson construction based on multi-parameter fusion. Background Art

[0002] Caisson construction is a core construction method for bridge foundations, underground pump stations, tunnel shafts and other projects. It achieves controlled sinking by overcoming soil friction through the deadweight of the wellbore or external loading. However, under complex geological conditions (such as soft soil, quicksand layer, water-rich strata) or construction disturbances (sudden load, biased excavation), the imbalance of pressure inside and outside the wellbore wall can easily cause soil inrush - the soil around the wellbore instantly rushes into the wellbore, causing the wellbore to tilt, the surface to subside, and even landslides. Traditional monitoring technology has two major limitations: Single parameter threshold alarm mechanism: The static threshold of a single parameter such as displacement or soil pressure is used to trigger the warning, which cannot capture the multi-parameter coupling effect in the cause of soil inrush (such as the synergistic effect of sudden drop in water level and sudden change in soil pressure), resulting in high false alarm and missed alarm rates. Fixed-weight empirical model: Some improved technologies attempt to integrate parameters such as displacement, soil pressure, and water level, but use a fixed-weight distribution model preset before construction. The weight value relies on manual experience and cannot dynamically respond to changes in geological conditions during construction (such as cohesion attenuation, water content fluctuations) and mechanical disturbances, resulting in significant early warning lags and difficulty in timely prevention and control of soil gushing risks.

[0003] The above defects make it difficult for existing technologies to meet the real-time and accurate early warning requirements for soil inrush risks in complex construction environments. Therefore, there is an urgent need for an intelligent early warning system that can dynamically perceive the correlation of multiple parameters, adaptively adjust weight distribution, and suppress interference, so as to improve the timeliness and reliability of soil inrush prevention and control. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a dynamic early warning system for soil gushing during caisson construction based on multi-parameter fusion, which solves the problems of traditional early warning methods for soil gushing during caisson construction, such as delayed warning, high misjudgment rate, and inability to dynamically adapt to geological conditions and construction disturbances due to reliance on a single parameter threshold alarm and a fixed weight empirical model.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a dynamic early warning system for soil inrush during caisson construction based on multi-parameter fusion, comprising: The sensor array module is used to collect the soil pressure change rate, displacement change rate, water level change rate and soil property parameters in real time during caisson construction; The data processing module normalizes the collected data and generates a normalized parameter vector; Dynamic parameter correlation analysis module, based on the normalized parameter vector and the rate of change of soil property parameters, updates the dynamic matrix online through recursive least squares method to characterize the real-time correlation of soil pressure, displacement and water level; Adaptive covariance optimization module dynamically adjusts the covariance matrix based on the risk gradient to suppress noise interference; The risk index generation module calculates the weights based on the dynamic matrix and the covariance matrix, fuses the normalized parameter vector to generate the soil inrush risk index, and performs risk correction based on the state of the construction machinery; The early warning output module triggers a graded alarm signal according to the corrected risk index, and feeds back data to the dynamic parameter association analysis module and the adaptive covariance optimization module for closed-loop optimization.

[0006] Preferably, the sensor array module comprises: Earth pressure sensors are evenly arranged along the circumference of the caisson wall; Displacement sensors installed at the top, middle and bottom nodes of the well wall; Groundwater level monitors arranged outside the caisson; The soil property detector buried in the construction area can obtain the friction angle, cohesion and their change rate in the soil in real time.

[0007] Preferably, the dynamic matrix updating method of the dynamic parameter association analysis module comprises the following steps: receiving a normalized parameter vector from a data preprocessing module; Based on recursive least squares method for dynamic matrix To perform an online update: ; in, For the moment The dynamic matrix represents the correlation between earth pressure, displacement and water level; is the learning rate; is the Kalman gain matrix; is the prediction error vector; is the transpose symbol; Combined with the cohesion change rate in the soil property parameter change rate and moisture content change rate Corrected dynamic matrix update direction: ; in, , are the coupling coefficients of cohesion and moisture content change rate, is the identity matrix, is an all-one matrix.

[0008] Preferably, the dynamic matrix is ​​initialized as follows: Constructing the initial dynamic matrix , Represents the initial dynamic matrix is a 3rd-order real matrix with all diagonal elements being 1 and all off-diagonal elements being , , , set according to the initial correlation of earth pressure, displacement and water level; The initial matrix form is: ; in, , and .

[0009] Preferably, the method for dynamically adjusting the covariance matrix of the adaptive covariance optimization module includes: receiving a risk index gradient from a risk index generation module; Adaptively adjust the forgetting factor based on the risk index gradient : ; in, is the initial forgetting factor, is the attenuation coefficient, is the risk index gradient; Update the covariance matrix : ; in, is the normalized parameter vector, express and The covariance matrix of represents the normalized parameter vector at the previous moment, represents the covariance matrix of the previous moment; The updated covariance matrix Output to the risk index generation module.

[0010] Preferably, the risk index gradient is calculated as follows: ; in, and are the risk indexes at the current moment and the previous moment respectively; when When , the enhanced update of the covariance matrix is ​​triggered: ; in, is the noise threshold, is the noise suppression coefficient, is the identity matrix.

[0011] Preferably, the soil bursting risk index calculation method of the risk index generation module includes: Receive dynamic matrix from dynamic parameter association analysis module , the covariance matrix of the adaptive covariance optimization module , and the normalized parameter vector of the data processing module ; Calculate the weight vector : ; in, is the inverse matrix of the covariance matrix, and the weight vector satisfies , Represents the weight vector No. Quantity; Generate soil inrush risk index : ; in, is the regularization coefficient, Represents the weight vector The transpose of Represents the Frobenius norm of a matrix.

[0012] Preferably, the risk index generation module performs risk correction in combination with the state of the construction machinery, including: Receive the construction machinery vibration status signal from the sensor array module, and when the vibration amplitude exceeds the threshold, determine it as mechanical operation interference; Modification of risk index: ; in, express The mechanical vibration acceleration measurement value at the time, To preset the vibration threshold, is the correction factor; The revised risk index Output to the early warning output module.

[0013] Preferably, the hierarchical alarm triggering method of the early warning output module includes: Receive the revised risk index from the risk index generation module ; Based on the preset risk threshold Triggering graded alarm signals: Level 1 Alarm: When the risk index is corrected Greater than or equal to the preset risk threshold and is less than the preset risk threshold Triggered when 1.2 times of Second level alarm: when the risk index is corrected Greater than or equal to the preset risk threshold Triggered when 1.2 times of No alarm: When the risk index is corrected Less than the preset risk threshold Maintain no alarm status; When a level 1 or level 2 alarm is triggered, the normalized parameter vector at the alarm time is , Modified Risk Index , Dynamic Matrix and the covariance matrix Synchronously feed back to the dynamic parameter association analysis module and the adaptive covariance optimization module to update the real-time parameters of the dynamic matrix and the covariance matrix.

[0014] Preferably, the closed-loop optimization method comprises: After receiving the feedback data, the dynamic parameter association analysis module updates the dynamic matrix through reinforcement learning: ; in, For the base Incremental matrix of differences from historical data; After receiving the feedback data, the adaptive covariance optimization module resets the forgetting factor of the covariance matrix: ; Based on Recalculate , suppress the impact of outdated data; Updated and It flows back to the risk index generation module to form a closed-loop optimization link.

[0015] The present invention provides a dynamic early warning system for soil inrush during caisson construction based on multi-parameter fusion. It has the following beneficial effects: 1. The present invention realizes the coordinated perception of soil mechanical behavior, hydrological conditions and mechanical disturbances during caisson construction through multi-dimensional real-time collection of soil pressure, displacement, water level and soil property parameters, combined with dynamic matrix correlation analysis and covariance optimization. Compared with the traditional single parameter monitoring method, it can more comprehensively capture the potential causes of soil inrush risk and significantly reduce the probability of missed reports and false reports.

[0016] 2. The dynamic parameter correlation analysis module of the present invention uses the recursive least squares method to update the dynamic matrix online, reflecting the changes in the correlation strength between soil pressure, displacement, and water level parameters in real time. This mechanism can automatically adapt to the soil response characteristics at different construction stages, avoid the prediction deviation of the traditional static model caused by changes in geological conditions, and improve the timeliness of risk warning.

[0017] 3. The adaptive covariance optimization module of the present invention distinguishes between normal data fluctuations and abnormal noise interference by dynamically adjusting the forgetting factor of the covariance matrix, and dynamically corrects the contribution of the risk gradient to the historical data. This design effectively suppresses the influence of interference such as sensor noise and mechanical vibration on the risk index calculation, ensuring the stability of weight allocation and risk quantification.

[0018] 4. By introducing the vibration state signal of the construction machinery, the system can distinguish between the real soil gushing risk and the mechanical operation interference, and dynamically scale the risk index using the correction coefficient. At the same time, the corrected risk index is fed back to the dynamic parameter association analysis module to suppress the abnormal matrix update caused by mechanical interference, form a closed-loop control logic, and enhance the system's anti-interference ability in complex construction environments.

[0019] 5. The early warning output module triggers a graded alarm based on the corrected risk index, and simultaneously feeds back the key parameters at the alarm time to the associated module. This mechanism enables the system to continuously optimize the dynamic matrix update rules and covariance adjustment strategies based on historical alarm data, realize the upgrade from passive early warning to active learning, and improve the adaptability of long-term construction risk prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of the system architecture of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Please see attached Figure 1 The present invention provides a dynamic early warning system for soil gushing during caisson construction based on multi-parameter fusion. The system collects soil pressure, displacement, water level and soil parameters in real time, combines dynamic correlation analysis with a closed-loop optimization mechanism, and realizes accurate early warning of soil gushing risks.

[0023] like Figure 1As shown in the figure, the system includes a sensor array module, a data processing module, a dynamic parameter correlation analysis module, an adaptive covariance optimization module, a risk index generation module and an early warning output module. The following are the specific implementation details of each module: Sensor array module This module uses multiple types of sensors to collaboratively collect soil pressure, displacement, water level and soil parameter data, establishes a mapping relationship between physical quantities and digital signals, and provides the original data basis for subsequent parameter correlation analysis and risk calculation.

[0024] The sensor array module includes an earth pressure change rate acquisition unit, a displacement change rate acquisition unit, a water level change rate acquisition unit, and a soil property parameter acquisition unit. The earth pressure change rate acquisition unit exemplarily uses a fiber Bragg grating pressure sensor, which is equidistantly arranged along the caisson blade foot. The installation spacing is preferably 2 meters, and the sampling frequency is set to 10Hz. By measuring the pressure fluctuations at the contact surface between the blade foot and the soil, the earth pressure change rate is obtained in real time. In this unit, the adjacent sensor data are processed by spatial weighted average, and the calculation formula is: ; in, Indicates The sensors in The measured value at the moment, is the sampling interval, is the total number of circumferential sensors. This calculation can eliminate the measurement deviation caused by local soil heterogeneity and provide a stable input for the dynamic parameter correlation analysis module.

[0025] The displacement change rate acquisition unit is realized by combining a laser rangefinder and an inclinometer. The laser rangefinder is vertically installed at the top center of the caisson, and the inclinometer is circumferentially arranged on the side wall of the caisson. The sampling frequency is set to 5Hz. The change of inclination angle , calculate the comprehensive displacement change rate: ; in, is the caisson height, It is calculated by the difference between adjacent samples of the laser rangefinder. is the change in the inclinometer measurement value. This formula quantifies vertical sinking and tilt displacement in a unified manner, which can accurately characterize the overall displacement trend of the caisson and provide multi-dimensional displacement parameters for the risk index generation module.

[0026] The water level change rate acquisition unit is realized by a piezoresistive level gauge. Two sets of sensors are arranged inside and outside the caisson, and the sampling period is set to 10 seconds. Calculated as: ; in, and They are The water level inside and outside the well at the moment. The water level change rate is further calculated by difference: ; This parameter can reflect the influence of groundwater flow velocity on soil stability. After normalization, it is input into the data processing module and participates in weight calculation together with soil pressure and displacement parameters.

[0027] The soil property parameter collection unit includes resistivity probes and moisture content sensors, which are arranged in a grid pattern 1 meter below the blade foot and collect soil cohesion every 30 minutes. and moisture content . Cohesion is calculated using the resistivity inversion model. Moisture content The dielectric constant is measured directly by a dielectric constant sensor and the data is corrected by a temperature compensation algorithm.

[0028] The above parameters affect the dynamic matrix update through the soil state equation in the dynamic parameter association analysis module, which is specifically manifested as follows: ; This design enables the change of soil parameters to adjust the correlation strength between parameters in real time, thus enhancing the system's adaptability to changes in geological conditions.

[0029] Preferably, all sensor data are transmitted to the data processing module through a hybrid networking of RS-485 bus and LoRa wireless communication. Among them, RS-485 bus is used for wired stable transmission of soil pressure, displacement, and water level sensors, and LoRa is used for wireless coverage blind area data return of soil parameter sensors and mobile monitoring points.

[0030] Data processing module This module receives the raw data stream from the sensor array module, generates a standardized parameter vector through multi-level signal processing and normalization conversion, and provides an input basis for dynamic parameter correlation analysis and risk calculation.

[0031] The data processing module first preprocesses the original signals of soil pressure, displacement, and water level. For soil pressure data, a sliding average filter is used to eliminate high-frequency noise. The filter window length is set to 5 seconds for example. The calculation formula is: ; in, is the number of sampling points in the window. This filtering process can suppress short-term mechanical vibration interference and retain the low-frequency trend component reflecting soil stability.

[0032] The displacement data includes vertical displacement and inclination measurement values, which need to be integrated to calculate the comprehensive displacement change rate. First, the laser rangefinder data Perform wavelet threshold denoising, select sym4 wavelet basis for 3-layer decomposition, and the hard threshold processing formula is: ; in, For the Layer wavelet coefficients, threshold , is the estimated value of the noise standard deviation. This processing can effectively separate the real displacement signal from the measurement noise, and is especially suitable for the denoising needs of non-stationary signals.

[0033] The change rate is calculated based on the preprocessed data. Earth pressure change rate Calculated by first-order differences: ; in is the sampling interval. Displacement change rate Calculated by fusion displacement difference: ; Water level change rate Calculated using the symmetric difference method: ; This method can reduce phase delay, accurately capture water level mutation events, and provide real-time guarantee for the risk index generation module.

[0034] The normalization processing unit converts the change rate of each parameter into a dimensionless quantity. The normalization formula of earth pressure is: ; in, and They are the historical mean and standard deviation of the earth pressure change rate, updated every hour. The mean and standard deviation are calculated by sliding window statistics, and the window length is set to 24 hours for example. The calculation formula is: ; Similarly, the normalized values ​​of displacement and water level are calculated as and This normalization method can eliminate the dimension differences of parameters and enable multi-source data to participate in matrix operations at a unified scale.

[0035] Finally, the normalized parameter vector is constructed as: ; This vector is transmitted to the dynamic parameter correlation analysis module through a standardized interface.

[0036] Soil property parameter processing includes cohesion and moisture content Fusion calculation.

[0037] Resistivity probe measurements Convert to cohesion through inversion model: ; in, and is the calibration coefficient, is the temperature compensation term.

[0038] Moisture content After being measured by the dielectric constant sensor, it is corrected using the Topp formula.

[0039] The above parameters affect the matrix update process through the state equations in the dynamic parameter association analysis module.

[0040] Dynamic parameter correlation analysis module This module updates the dynamic matrix online through the recursive least squares method, quantifies the real-time correlation strength of soil pressure, displacement, and water level, and provides a dynamic weight basis for the risk index calculation.

[0041] The core function of the dynamic parameter correlation analysis module is to capture the time-varying correlation characteristics of multiple parameters in the soil system. To achieve this goal, we first define the dynamic matrix As a mathematical representation of the parameter association relationship, its elements are set according to the initial association prior data of earth pressure (P), displacement (D), and water level (W). The specific rules are as follows: Diagonal elements: , characterizing the dominant role of the historical values ​​of the parameters themselves, which is consistent with the inertial characteristics of the soil parameters in the early stage of caisson construction.

[0042] Off-diagonal elements: When there is a priori correlation, it is set according to soil engineering survey data or historical construction statistics: , For parameters and The initial correlation coefficient ranges from [-1,1].

[0043] When there is no prior data, the default setting is: ,at this time degenerates to the identity matrix, reflecting the initial independence assumption of the parameters.

[0044] Among them, the initialization data sources include: Geological exploration data: The initial relationship between cohesion and moisture content is obtained through soil shear test, and the indirect derivation is ; Historical construction data: extract parameter covariance matrix from similar project database and normalize it into correlation coefficient; Expert experience value assignment: When there is a lack of quantitative data, it is qualitatively set by engineers based on the soil type.

[0045] The initialization strategy is based on the inertial characteristics of the soil system and assumes that each parameter in the initial state is dominated by its own historical value, which is consistent with the physical reality that the parameters are highly independent in the early stage of caisson construction.

[0046] In view of the time-varying characteristics of the soil parameter correlation, the recursive least squares (RLS) algorithm is used to realize the online update of the matrix. This algorithm avoids repeated storage of historical data through recursive calculation and meets the real-time requirements. The specific iterative process first calculates the prediction error vector: ; in is the normalized parameter vector output by the data processing module, is the sampling interval. The error vector Reflects the prediction deviation of the current parameter association model and is used for driving matrix correction.

[0047] Kalman Gain Through the inverse covariance matrix The calculation results in: ; The forgetting factor λ=0.99 is used to balance the influence weights of historical data and new data to prevent excessive fluctuations in the matrix. Adopting the rank-1 update rule: ; This update mechanism avoids matrix inversion operations through recursive calculations, significantly reducing computational complexity.

[0048] The iterative correction formula of the dynamic matrix is: ; Learning Rate Control the matrix update step size to prevent the model from becoming unstable due to a single error that is too large. To ensure the physical interpretability of the matrix, the following constraints are imposed: 1. Autocorrelation constraints: diagonal elements , ensuring the dominance of the parameter’s own historical values; 2. Directional consistency constraint: The symbols of non-diagonal elements maintain the initial set direction to avoid sudden changes in association relationships.

[0049] When the iteration result violates the constraint, the projection correction algorithm is started: ; in is the Kronecker delta function, which limits the parameters to a reasonable range and maintains the stability of the model.

[0050] Soil property parameters affect the matrix update through differential coupling terms. Cohesion change rate Introducing the fix: ; Where α is the coupling coefficient, is the unit matrix. Similarly, the moisture content change rate Through the all-ones matrix Generate correction amount: ; The expanded dynamic matrix update formula is: ; This design enables real-time adjustment of associated weights in response to soil parameter changes, thus enhancing the system’s adaptability to dynamic changes in geological conditions.

[0051] The early warning feedback mechanism achieves closed-loop optimization by adjusting the learning rate. When the signal is received, the learning rate is dynamically adjusted to: ; in This mechanism compresses the learning rate when the risk increases, suppressing the interference of abnormal data on the associated model while retaining the model's ability to track new trends.

[0052] Adaptive Covariance Optimization Module This module dynamically senses changes in the system's risk status and establishes a closed-loop adjustment mechanism for the parameter covariance matrix, providing noise suppression and statistical stability guarantees for risk weight calculation.

[0053] The input data stream of the adaptive covariance optimization module includes the real-time risk index from the risk index generation module , and the normalized parameter vector output by the data processing module This module dynamically adjusts the covariance matrix update strategy by integrating risk status and parameter statistical characteristics. The specific implementation process is as follows: Step 1: Risk gradient calculation and state perception Receive the risk index generation module After that, the risk gradient is first calculated to quantify the rate of change of the system state: ; in and are the risk indexes at the current moment and the previous moment respectively. The gradient value is transmitted to the dynamic parameter association analysis module through the feedback channel of the early warning output module, forming a cross-module state sharing mechanism. hour, is the noise threshold; it determines that the system is in a high-risk transient state, triggering the enhanced update of the covariance matrix.

[0054] Step 2: Dynamic adjustment of the forgetting factor Adaptively adjust historical data weight coefficients based on risk gradient : ; in is the initial forgetting factor, is the attenuation coefficient. When the risk gradient increases, Exponential decay, reducing the historical covariance matrix The weight of makes the system more sensitive to the current parameter changes. This adjustment process is related to the learning rate of the dynamic parameter association analysis module. Form a collaborative control to jointly suppress the interference of sudden disturbances on model updating.

[0055] Step 3: Iterative update of the covariance matrix Using the normalized parameter vector Update the covariance matrix: ; in, is the normalized parameter vector, express and The covariance matrix of represents the normalized parameter vector at the previous moment, represents the covariance matrix at the previous moment.

[0056] Real-time covariance term Compute via sliding window: ; in is the number of window samples, This calculation strategy not only retains the time series correlation of parameters, but also avoids the memory overhead caused by storing historical data, which matches the real-time requirements of the data processing module.

[0057] Step 4: Noise suppression and matrix stabilization When the risk gradient exceeds the limit, the covariance matrix diagonal hardening is performed: ; in is the noise suppression coefficient, is the unit matrix. This operation suppresses abnormal correlation calculation caused by burst noise by enhancing the parameter autocorrelation weight.

[0058] Step 5: Matrix verification and output Perform symmetry and positive semidefiniteness checks on the updated covariance matrix: Symmetry correction: for asymmetric elements , forced symmetry processing: ; Semi-positive definiteness correction: truncated projection of negative eigenvalues: ; in is the eigenvalue diagonal matrix, is the eigenvector matrix. The modified covariance matrix is ​​output to the risk index generation module through a standardized interface, and its data structure is aligned with the matrix output of the dynamic parameter association analysis module to ensure the dimensional consistency of the weight calculation.

[0059] Risk Index Generation Module This module constructs a closed-loop calculation model of the soil inrush risk index by integrating dynamic matrix correlation, parameter covariance characteristics and construction machinery status, thereby realizing a multi-dimensional quantitative assessment of soil instability risk.

[0060] The risk index generation module receives the dynamic matrix from the dynamic parameter association analysis module , the covariance matrix of the adaptive covariance optimization module , and the normalized parameter vector of the data processing module The module processing flow is as follows: Step 1: Weight vector calculation First, the parameter weight vector is calculated based on the dynamic matrix and the inverse covariance matrix: ; The inverse covariance matrix Obtained through the reversibility guarantee algorithm: ; is the micro-disturbance coefficient, is the identity matrix. The weight vector The normalization constraints need to be satisfied: ; This constraint is implemented via post-processing scaling: ; Normalized weight vector Used for subsequent risk calculations.

[0061] Step 2: Generate soil inrush risk index The weight vector is linearly combined with the parameter vector, and a regularization term for the dynamic matrix change is introduced: ; in, is the regularization coefficient, Represents the weight vector The transpose of Represents the Frobenius norm of a matrix.

[0062] Step 3: Construction machinery status correction Receive mechanical vibration acceleration signal from sensor array module ,when When the mechanical operation causes significant disturbance to the soil, the risk index correction formula is: ; is a correction factor, which is set exemplarily based on engineering experience.

[0063] Step 4: Data packaging and output Modified risk index Encapsulate into standardized data packets and output to the warning output module.

[0064] Warning output module This module builds a closed-loop control system for risk warning and model self-update through hierarchical risk threshold judgment and multi-module collaborative optimization mechanism, realizing real-time monitoring and adaptive adjustment of construction safety status.

[0065] The early warning output module receives the corrected risk index output by the risk index generation module , and obtain the dynamic parameter association analysis module , Adaptive Covariance Optimization Module and data processing modules The module processing flow and interaction mechanism are as follows: Step 1: Triggering of graded alarms Based on preset risk thresholds Triggering the three-level alarm strategy: 1. No alarm status: When When the system is running normally, only data is recorded; 2. Level 1 alarm: When When the yellow warning of the sound and light alarm is activated; 3. Second level alarm: When An orange alert is triggered.

[0066] Step 2: Alarm data packaging When an alarm is triggered, the four-dimensional data packet is encapsulated and fed back to the dynamic parameter association analysis module and the adaptive covariance optimization module: ; Step 3: Closed-loop optimization execution Dynamic parameter association analysis module receives Then, update the dynamic matrix: ; The increment matrix Calculated by historical data difference: ; in, is the sliding window length. The function maps the risk index to the interval [-1,1] to suppress excessive corrections.

[0067] After receiving the data, the adaptive covariance optimization module resets the forgetting factor: ; Based on Recompute the covariance matrix: ; This operation reduces the weight of historical data and enhances the representation of the current state.

[0068] Step 4: Optimize Parameters for Reflow Updated and It flows back to the risk index generation module through the standardized interface and participates in the weight calculation of the next cycle, forming a closed-loop feedback link from risk warning to model parameter update, thereby improving the system's adaptability to dynamic changes in the construction environment.

[0069] In general, the present invention collects the data of the rate of change of soil pressure, displacement, water level and soil parameters in real time through the sensor array, generates parameter vectors through normalization by the data processing module, and updates the dynamic matrix reflecting the real-time correlation between parameters online in combination with the recursive least squares method of the dynamic parameter association analysis module and the risk gradient driven noise suppression mechanism of the adaptive covariance optimization module, generates a soil gushing risk index integrating the dynamic weight and the inverse covariance matrix, and performs dynamic correction based on the vibration state of the construction machinery; triggers a graded alarm through the early warning output module, and feeds back the corrected risk index and key matrix to the dynamic parameter association module and the covariance optimization module at the same time, forming a full-process adaptive early warning system, realizing the coordinated perception of soil pressure mutations, displacement anomalies and water level fluctuations and the precise prevention and control of soil gushing risks.

[0070] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A dynamic early warning system for soil gushing during caisson construction based on multi-parameter fusion, characterized in that: include: The sensor array module is used to collect the soil pressure change rate, displacement change rate, water level change rate and soil property parameters in real time during caisson construction; The data processing module normalizes the collected data and generates a normalized parameter vector; Dynamic parameter correlation analysis module, based on the normalized parameter vector and the rate of change of soil property parameters, updates the dynamic matrix online through recursive least squares method to characterize the real-time correlation of soil pressure, displacement and water level; Adaptive covariance optimization module dynamically adjusts the covariance matrix based on the risk gradient to suppress noise interference; The risk index generation module calculates the weights based on the dynamic matrix and the covariance matrix, fuses the normalized parameter vector to generate the soil inrush risk index, and performs risk correction based on the state of the construction machinery; The early warning output module triggers a graded alarm signal according to the corrected risk index, and feeds back data to the dynamic parameter association analysis module and the adaptive covariance optimization module for closed-loop optimization.

2. The caisson construction soil gushing dynamic early warning system based on multi-parameter fusion according to claim 1 is characterized in that: The sensor array module comprises: Earth pressure sensors are evenly arranged along the circumference of the caisson wall; Displacement sensors installed at the top, middle and bottom nodes of the well wall; Groundwater level monitors arranged outside the caisson; The soil property detector buried in the construction area can obtain the friction angle, cohesion and their change rate in the soil in real time.

3. The caisson construction soil gushing dynamic early warning system based on multi-parameter fusion according to claim 1 is characterized in that: The dynamic matrix updating method of the dynamic parameter association analysis module comprises the following steps: receiving a normalized parameter vector from a data preprocessing module; Based on recursive least squares method for dynamic matrix To perform an online update: ； in, For the moment The dynamic matrix represents the correlation between earth pressure, displacement and water level; is the learning rate; is the Kalman gain matrix; is the prediction error vector; is the transpose symbol; Combined with the cohesion change rate in the soil property parameter change rate and moisture content change rate Corrected dynamic matrix update direction: ； in, , are the coupling coefficients of cohesion and moisture content change rate, is the identity matrix, is an all-one matrix.

4. The caisson construction soil gushing dynamic early warning system based on multi-parameter fusion according to claim 3 is characterized in that: The initialization method of the dynamic matrix is: Constructing the initial dynamic matrix , Represents the initial dynamic matrix It is a 3rd-order real matrix whose diagonal elements are 1 and the off-diagonal elements are set according to the initial correlation of earth pressure, displacement, and water level.

5. The dynamic early warning system for soil gushing during caisson construction based on multi-parameter fusion according to claim 4 is characterized in that: The covariance matrix dynamic adjustment method of the adaptive covariance optimization module includes: receiving a risk index gradient from a risk index generation module; Adaptively adjust the forgetting factor based on the risk index gradient : ； in, is the initial forgetting factor, is the attenuation coefficient, is the risk index gradient; Update the covariance matrix : ； in, is the normalized parameter vector, express and The covariance matrix of represents the normalized parameter vector at the previous moment, represents the covariance matrix of the previous moment; The updated covariance matrix Output to the risk index generation module.

6. The caisson construction soil gushing dynamic early warning system based on multi-parameter fusion according to claim 5 is characterized in that: The calculation method of the risk index gradient is: ； in, and are the risk indexes at the current moment and the previous moment respectively; when When , the enhanced update of the covariance matrix is ​​triggered: ； in, is the noise threshold, is the noise suppression coefficient, is the identity matrix.

7. The caisson construction soil gushing dynamic early warning system based on multi-parameter fusion according to claim 6 is characterized in that: The soil gushing risk index calculation method of the risk index generation module includes: Receive dynamic matrix from dynamic parameter association analysis module , the covariance matrix of the adaptive covariance optimization module , and the normalized parameter vector of the data processing module ; Calculate the weight vector : ； in, is the inverse matrix of the covariance matrix, and the weight vector satisfies , Represents the weight vector No. Quantity; Generate soil inrush risk index : ； in, is the regularization coefficient, Represents the weight vector The transpose of Represents the Frobenius norm of a matrix.

8. The caisson construction soil gushing dynamic early warning system based on multi-parameter fusion according to claim 7 is characterized in that: The risk index generation module performs risk correction in combination with the construction machinery status, including: Receive the construction machinery vibration status signal from the sensor array module, and when the vibration amplitude exceeds the threshold, determine it as mechanical operation interference; Modification of risk index: ； in, express The mechanical vibration acceleration measurement value at the time, To preset the vibration threshold, is the correction factor; The revised risk index Output to the early warning output module.

9. The caisson construction soil gushing dynamic early warning system based on multi-parameter fusion according to claim 8 is characterized in that: The hierarchical alarm triggering method of the early warning output module includes: Receive the revised risk index from the risk index generation module ; Based on the preset risk threshold Triggering graded alarm signals: Level 1 Alarm: When the risk index is corrected Greater than or equal to the preset risk threshold and is less than the preset risk threshold Triggered when 1.2 times of Second level alarm: when the risk index is corrected Greater than or equal to the preset risk threshold Triggered when 1.2 times of No alarm: When the risk index is corrected Less than the preset risk threshold Maintain no alarm status; When a level 1 or level 2 alarm is triggered, the normalized parameter vector at the alarm time is , Modified Risk Index , Dynamic Matrix and the covariance matrix Synchronously feed back to the dynamic parameter association analysis module and the adaptive covariance optimization module to update the real-time parameters of the dynamic matrix and the covariance matrix.

10. The caisson construction soil gushing dynamic early warning system based on multi-parameter fusion according to claim 9 is characterized in that: The closed-loop optimization method comprises: After receiving the feedback data, the dynamic parameter association analysis module updates the dynamic matrix through reinforcement learning: ； in, For the base Incremental matrix of differences from historical data; After receiving the feedback data, the adaptive covariance optimization module resets the forgetting factor of the covariance matrix: ； Based on Recalculate , suppress the impact of outdated data; Updated and It flows back to the risk index generation module to form a closed-loop optimization link.

Images