A dynamic prediction method, device and medium for bearing capacity of wet land construction

Abstract

The present application discloses a method, device, and medium for dynamic prediction of bearing capacity for wetland construction, which relates to the field of intelligent monitoring of civil engineering. The method includes: training a dynamic bearing capacity prediction model pre-deployed at the edge based on laboratory calibration data and historical engineering data; collecting multi-source sensor data of the wetland construction area in real time through a sensor network; obtaining a bearing capacity prediction value and confidence interval based on the multi-source sensor data through a dynamic bearing capacity prediction model; adjusting the bearing capacity warning threshold corresponding to different warning levels based on the confidence interval, and judging whether there is a construction risk based on the bearing capacity prediction value and the bearing capacity warning threshold. The dynamic bearing capacity prediction model updates soil parameters in real time, effectively overcoming the limitations of traditional static models that rely on pre-construction geological data, and realizing real-time data-driven dynamic bearing capacity assessment.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent monitoring of civil engineering projects, and in particular to a method, equipment, and medium for dynamically predicting the bearing capacity of wet land construction. Background Art

[0002] Wet land (such as soft soil, silt soil, fill areas, etc.) has characteristics such as high water content, low permeability and rheological properties. Its construction process is very likely to cause sudden changes in bearing capacity due to changes in external loads, environmental changes (such as rainfall, groundwater level changes) or construction disturbances, which in turn may lead to major safety accidents such as foundation settlement, slope sliding, pile foundation displacement and even machinery overturning.

[0003] Safety monitoring of construction sites is often accompanied by monitoring of bearing capacity. In wet land construction scenarios, there are mainly two monitoring methods. One is static bearing capacity assessment based on indoor geotechnical tests or empirical formulas. For example, the shear strength parameters are determined through triaxial shear tests on undisturbed soil samples, and the theoretical value is calculated in combination with the Terzaghi ultimate bearing capacity formula. The other is a threshold alarm system based on on-site monitoring. For example, displacement meters or pressure sensors are deployed at key points, and an alarm is triggered when the monitoring value exceeds the preset threshold.

[0004] Both of these traditional monitoring methods are static assessments, using a single monitoring tool. However, these methods rely on pre-construction geological survey data and are unable to reflect dynamic changes in bearing capacity during construction due to mechanical vibration, rainfall penetration, or soil creep. This results in inaccurate bearing capacity monitoring, and consequently, construction adjustment strategies based on bearing capacity are also inaccurate and subject to lag. Summary of the Invention

[0005] To solve the above problems, this application proposes a dynamic prediction method for bearing capacity of wet land construction, including:

[0006] The dynamic load-bearing capacity prediction model pre-deployed at the edge is trained based on laboratory calibration data and historical engineering data.

[0007] Real-time collection of multi-source sensor data in wetland construction areas through sensor networks;

[0008] Obtaining a bearing capacity prediction value and a confidence interval based on the multi-source sensor data using the dynamic bearing capacity prediction model;

[0009] The bearing capacity warning thresholds corresponding to different warning levels are adjusted according to the confidence intervals, and whether there is a construction risk is determined according to the bearing capacity prediction value and the bearing capacity warning thresholds.

[0010] On the other hand, the present application also proposes a dynamic prediction device for bearing capacity of wet land construction, comprising:

[0011] at least one processor; and,

[0012] a memory communicatively connected to the at least one processor; wherein,

[0013] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a dynamic prediction method for bearing capacity of wet land construction as described in the above example.

[0014] On the other hand, the present application also proposes a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured as: a dynamic prediction method for bearing capacity of wet land construction as described in the above example.

[0015] This application proposes a dynamic prediction method for bearing capacity of wet land construction, which can bring the following beneficial effects:

[0016] This application uses a dynamic bearing capacity prediction model to update soil parameters in real time, effectively overcoming the limitations of traditional static models that rely on pre-construction geological data. Real-time data-driven dynamic bearing capacity assessment, through intelligent linkage with construction machinery, upgrades risk prevention and control from passive response to active intervention, realizing a "perception-analysis-decision-control" closed loop, and upgrading from "monitoring-alarm-manual processing" to "prediction-intervention-automated adjustment." BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 Schematic diagram of a flow chart of a method for dynamically predicting bearing capacity of wet land construction in an embodiment of the present application;

[0019] Figure 2 This is a schematic diagram of a dynamic bearing capacity prediction device for wet land construction in an embodiment of the present application. DETAILED DESCRIPTION

[0020] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0021] The following describes in detail the technical solutions provided by various embodiments of the present application in conjunction with the accompanying drawings.

[0022] like Figure 1 As shown, the embodiment of the present application provides a method for dynamically predicting the bearing capacity of wet land construction, including:

[0023] S101: Based on laboratory calibration data and historical engineering data, the dynamic load-bearing capacity prediction model pre-deployed at the edge is trained.

[0024] Specifically, a CNN-Transformer hybrid architecture is adopted, combining the spatial feature extraction capabilities of convolutional neural networks (CNN) with the global time series modeling advantages of Transformer to construct a dynamic bearing capacity prediction model. The dynamic bearing capacity prediction model is pre-deployed on the edge and trained based on laboratory calibration data and historical engineering data to initialize the model parameters.

[0025] It should be noted that laboratory calibration data is obtained through controlled experiments to establish a precise mapping between multi-source sensor data and bearing capacity parameters. Specifically, professionals collected wet soil from various geological conditions (such as soft soil, silt soil, and fill soil), prepared samples with varying moisture contents according to standard methods, and filled test tanks with these samples in layers to simulate the soil structure experienced during actual construction. Based on this simulated soil structure, multi-source sensor data and bearing capacity values were collected. The experiments were repeated multiple times, and the laboratory calibration data was recorded.

[0026] Historical engineering data is derived from long-term monitoring records of actual construction projects, covering diverse geological conditions, construction scenarios, and risk events. Historical data from a multi-source sensor network is extracted from records of completed wetland construction projects. In this embodiment, the historical engineering data covers typical wetland types (such as soft soil, silt, and fill areas) and extreme working conditions (such as heavy rainfall and high loads).

[0027] By combining laboratory calibration data with historical engineering data, the dynamic bearing capacity prediction model is trained. This not only balances the sample data and avoids model overfitting due to data skew, but also corrects data deviations caused by environmental differences, thereby increasing the prediction accuracy and physical interpretability of the dynamic bearing capacity prediction model.

[0028] In addition, in the embodiment of the present application, the weighted mean square error is used to balance the heteroscedasticity of multi-source sensor data. Its core purpose is to solve the model bias problem caused by the inconsistent quality of multi-source sensor data by assigning differentiated weights to the error terms of different sensors. The execution formula is: ,in The weights are dynamically adjusted based on the sensor error distribution. A linear warmup and cosine annealing strategy are applied simultaneously to prevent gradient instability during the initial training phase. Spatiotemporal dropout is introduced to randomly mask local spatial regions or time step features to enhance robustness to sensor noise. Model capabilities can also be evaluated using other performance metrics, such as mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²).

[0029] S102: Collect multi-source sensor data of the wet land construction area in real time through the sensor network.

[0030] Specifically, the sensor network includes probes, soil pressure gauges, plate load testers, inclinometers and weather stations. GPS modules or network time protocol (NTP) are used to provide accurate UTC time sources for all sensor devices, ensuring that each sensor device in the sensor network is in a unified time base. Multi-source sensor data, including soil reflection waves, soil vibration string frequency, soil displacement data, weather forecast data, etc., are continuously collected in real time through the sensor network, and accurate timestamps are added to each piece of multi-source sensor data.

[0031] Furthermore, because different devices have varying acquisition frequencies, multi-source data alignment is necessary to address these differences. Specifically, based on acquisition timestamps, multi-source sensor data is aligned to obtain missing values. The corresponding sensor devices are identified and filled in using linear interpolation of adjacent data collected by the sensor devices at adjacent timestamps or historical averages. Once this filling is complete, each sensor device in the multi-source sensor data is confirmed to have its corresponding value, thus constructing a multidimensional feature matrix.

[0032] Furthermore, the multi-source sensor data is cached at the edge, and based on a preset time window, the dynamic carrying capacity model is trained using the cached multi-source sensor data to update the model parameters. After the training is completed, the multi-source sensor data cached at the edge is cleared.

[0033] It should be noted that the probe emits high-frequency electromagnetic waves and receives reflected signals, analyzing the changes in dielectric constant based on the difference in the propagation time of electromagnetic waves in the soil, thereby accurately inverting the volumetric water content at different depths. Through a preset program, TDR continuously monitors soil moisture dynamics at a sampling frequency of 1 to 10 minutes, and supports automatic triggering of high-frequency acquisition by abnormal data. After filtering and feature extraction, the output time-domain reflection waveform data can quantify the soil moisture content, and combined with historical moisture curves and pore water pressure data, the saturation state of the soil layer can be evaluated in real time. In addition, TDR's multi-layer synchronous monitoring capability can reveal the laws of water migration in the vertical direction, such as identifying the infiltration of shallow water accumulation or the rise of deep capillary water, providing high-resolution moisture input for dynamic bearing capacity models.

[0034] Soil pressure gauges are buried in key stress-bearing locations within the construction area, such as the bottom of the foundation, inside the slope, or beneath the loading area. These sensors utilize a high-strength stainless steel housing and a waterproof seal design, which can withstand long-term soil acid and alkali corrosion and mechanical extrusion. They are usually pre-buried or embedded in the soil layer. Their core sensitive elements sense the deformation caused by the vertical / horizontal pressure exerted on the soil and convert the mechanical signal into an electrical signal. The vibrating string type reflects the pressure magnitude through frequency changes, while the resistance strain type outputs the strain value based on the Wheatstone bridge. The equipment continuously collects soil stress data at intervals of 1 to 5 minutes, and automatically switches to high-frequency monitoring in seconds when construction machinery is operating at high intensity. By deploying multiple groups of sensors in a grid, a two-dimensional / three-dimensional soil stress distribution cloud map can be constructed to track the load transfer path in real time, such as identifying stress diffusion or local over-limit concentration caused by loading.

[0035] The flat plate load tester is installed at the pre-compacted test point, and vertical loads are applied in stages to the standard circular bearing plate through a hydraulic servo system and a reaction frame. The device uses a high-precision laser displacement meter and a dynamic pressure sensor to capture the settlement and load intensity of the bearing plate in real time, and constructs a dynamic model of the load-settlement (ps) curve with a high-frequency sampling of 5-20 seconds per time. This curve not only reveals the mechanical response of the soil throughout the entire process from the elastic deformation stage to the plastic development stage to the destruction, but also can accurately determine the ultimate bearing capacity threshold through inflection point identification. For complex layered foundations, the multi-stage pressure stabilization control during the loading process can separate the differences in the compression characteristics of each soil layer, and maintain test stability even under conditions of groundwater level fluctuations or soil crack development.

[0036] Inclinometers are deployed at key nodes of slopes, bridge pier foundations or building structures. They are designed with high-precision MEMS or dual-axis electrolyte sensors, have IP68 waterproof and shockproof performance, and can be embedded in concrete structures or fixed to steel brackets for long-term monitoring. By measuring changes in the X / Y dual-axis tilt angle, they can capture tiny posture deviations of structures or soil in real time. The sampling frequency is set to 10-30 minutes / time during the stable period, and automatically switches to 1-5 seconds / time for continuous monitoring in the event of heavy rainfall or vibration from nearby construction. The settlement plate is pre-buried in the surface layer of the foundation, the fill layer interface or the soft foundation treatment area. It consists of a rigid base plate with anchor piles and a high-sensitivity liquid level sensor or laser rangefinder to continuously record vertical settlement. Deep layered settlement plates can also achieve separate measurement of compression of different soil layers through a magnetic ring array and a measuring rod.

[0037] The weather station is deployed in an open area of the construction site (such as outside the material storage yard or in the center of the foundation treatment area). It features an integrated column structure and is equipped with an anemometer, a tipping bucket rain gauge, a temperature and humidity sensor, and an atmospheric pressure sensor. All components meet IP66 protection standards, making it suitable for extreme weather conditions such as heavy rain and dust storms. Powered by both solar and battery power, the station transmits real-time meteorological data wirelessly via LoRa / NB-IoT. The sampling frequency is set at 5-10 minutes on sunny days, automatically switching to a high-frequency recording rate of 1 minute when wind speeds exceed 10 m / s or rainfall exceeds 5 mm / h. The collected minute-by-minute rainfall intensity, sustained wind direction and speed, and temperature and humidity curves are synchronized with a dynamic bearing capacity prediction model to correct theoretically calculated soil moisture content (for example, pore water pressure surges caused by heavy rain infiltration). The data is then combined with satellite meteorological data to predict weather impacts for the next 6-12 hours. For example, when the system detects rainfall greater than 20 mm for one hour, it automatically triggers the high-frequency monitoring mode of the TDR soil moisture sensor and the groundwater level meter, and associates the three-dimensional geological model to simulate the rainwater infiltration path and predict the risk of slope slippage.

[0038] Furthermore, features corresponding to multi-source sensor data without missing values are extracted. Specifically, the reflection peak time delay difference in the soil reflection wave at different depths is extracted to determine the soil moisture content at different depths. The collected grid-distributed soil vibration string frequency is Fourier transformed to obtain soil pressure values corresponding to different grid areas. The corresponding soil displacement rate is determined based on the soil displacement data collected at different coordinate positions. The soil inclination direction and soil inclination amplitude are determined based on the soil displacement rate at the different coordinate positions. The soil moisture content, the soil pressure value, the soil inclination amplitude and the meteorological forecast data are constructed into a multidimensional feature matrix based on a preset time and space grid.

[0039] It should be noted that in the embodiments of this application, the original reflected waveform is decomposed using the Daubechies wavelet basis for soil reflection waves. A soft thresholding method is used to filter out high-frequency noise (such as electromagnetic interference or soil particle reflection clutter), retaining the effective signal components. Polynomial fitting is used to extract low-frequency baseline trends, eliminating signal offsets caused by changes in probe contact resistance or temperature drift. The rise time, reflection peak amplitude, and time delay are extracted from the denoised waveform, and the soil dielectric constant (ε) is calculated based on the probe's geometric parameters. A nonlinear regression model based on ε-water content is constructed based on laboratory calibration data (standard soil samples with different moisture contents) to dynamically compensate for the influence of soil texture (such as clay / sand) on dielectric properties.

[0040] Regarding the soil vibrating string frequency, an FFT (Fast Fourier Transform) is performed on the frequency signal output by the vibrating string sensor to extract the fundamental frequency and convert it into a pressure value. For the resistance strain data, a temperature sensor is used to synchronously collect data, and a linear regression model is used to eliminate the zero-point drift caused by ambient temperature changes. Based on the grid-based sensor nodes, the Kriging interpolation algorithm is used to construct a two-dimensional / three-dimensional soil stress distribution field to identify stress concentration areas. The high-frequency sampling data (second level) is subjected to a short-time Fourier transform (STFT) to extract the main frequency component of the construction machinery vibration load (such as the impact frequency of the pile driver) and evaluate its periodic impact on soil stability.

[0041] For soil displacement data, the GPS latitude and longitude data are converted into a local construction coordinate system (such as UTM projection) to ensure spatial consistency with the total station and inclinometer data. The original displacement series is windowed and averaged to suppress measurement noise (such as multipath effect or instrument jitter). The first-order derivative of the displacement time series is calculated, and a graded warning threshold is set. For the dual-axis inclinometer data, the total tilt direction and amplitude are calculated through vector synthesis. The potential slip surface direction is analyzed in combination with the displacement field. The displacement data is synchronized with the construction machinery operation log, and a time-lag correlation model between displacement increment and load application is constructed.

[0042] For meteorological forecast data, rainfall, temperature, wind speed data are aligned with soil moisture and displacement data using UTC timestamps to address the problem of different sampling frequencies of different devices. For short-term missing meteorological data, linear interpolation is performed using neighboring sensor data or historical average values for the same period. Based on minute-level rainfall intensity data, the Green-Ampt infiltration formula is combined to calculate the rainwater infiltration depth and rate, predict the rising trend of pore water pressure, and construct a multivariate regression model of groundwater level changes and TDR moisture content and soil pressure data to quantify the contribution of water level fluctuations to effective stress and bearing capacity. A sliding window is used to detect periods of continuous rainfall exceeding the threshold, marking them as high-risk intervals, triggering adaptive adjustment of model parameters.

[0043] It should be noted that construction site data is first cleaned by an on-site edge computer to remove problematic data and reduce the burden on cloud servers. The edge computer then processes the data collected by each sensor accordingly. Specifically, for TDR (time domain reflectometry) data, baseline calibration is performed to eliminate inherent device noise, and wavelet transforms or sliding average filters are used to suppress high-frequency interference. The time difference between reflection peaks is then extracted and combined with a dielectric constant model to invert soil moisture content. The timing of multiple sensors is also aligned to enable simultaneous analysis. Earth pressure gauge data undergoes temperature compensation and low-pass filtering to eliminate ambient temperature drift and mechanical vibration noise. The voltage signal is then converted to standard pressure units. A three-dimensional earth pressure distribution model is constructed through stress direction decomposition, providing a quantitative basis for studying soil mechanical behavior. In the processing of plate load test data, outlier removal and smoothing algorithms are used to optimize the load-settlement curve. A continuous function is interpolated to calculate key parameters such as the foundation reaction modulus. The time stamps of the loading phases are strictly aligned to ensure temporal consistency between the load rate and settlement deformation, providing a reliable data foundation for foundation bearing capacity assessment. Inclinometer data requires zero-point drift correction. Kalman filtering is used to fuse multi-axis signals to suppress vibration noise. Coordinate system transformation is used to analyze the structure's spatial posture and separate horizontal and vertical tilt components. Settlement plate data is used to calculate cumulative settlement through differential calculations. Sliding window filtering is used to eliminate random errors, and correlation with environmental parameters (such as rainfall and groundwater level) is used to distinguish settlement effects caused by loads from those caused by natural factors. Processing of meteorological station data focuses on multi-parameter fusion and feature engineering. By interpolating missing values and unifying the sampling frequency to generate standardized time series, statistical features of cumulative rainfall and wind speed and direction are further extracted. Z-score normalization is used to eliminate dimensional differences, supporting coupled analysis between meteorological conditions and engineering responses.

[0044] S103: Obtaining a bearing capacity prediction value and a confidence interval based on the multi-source sensor data using the dynamic bearing capacity prediction model.

[0045] Specifically, the multidimensional feature matrix is input into the dynamic bearing capacity prediction model. The spatial features of the multidimensional feature matrix are extracted through the convolutional neural network of the dynamic bearing capacity prediction model. The spatial features are input into the self-attention layer of the dynamic bearing capacity prediction model to capture the dependency relationship and temporal characteristics between the multidimensional features in different spatial positions. According to the dependency relationship and temporal characteristics, the bearing capacity prediction values and confidence intervals of different spatial positions are obtained.

[0046] Preprocessed multi-source sensor data is fed into a trained CNN-Transformer hybrid architecture. This architecture uses a convolutional neural network (CNN) to extract local spatial features, such as soil moisture and pressure distribution, while leveraging the Transformer's self-attention mechanism to capture global dependencies among soil mechanical parameters and their time-series dynamics. The CNN module utilizes a lightweight MobileNetV3 architecture, whose depthwise separable convolution significantly reduces computational redundancy in complex soil environments. The Transformer layer, through a multi-head attention mechanism, enables cross-temporal and spatial modeling of soil moisture gradients and the dynamic response of bearing capacity. The fused multi-scale features are coupled and optimized through a dynamic weight allocation network, ultimately outputting a time-series prediction of wetland bearing capacity and its 95% confidence interval. The entire inference process is deployed on the Jetson AGX Xavier edge computing platform. Combined with TensorRT's layer fusion optimization of the hybrid architecture, this achieves real-time inference latency below 20ms while maintaining millimeter-level prediction accuracy.

[0047] It should be noted that in the spatial feature extraction layer of the dynamic bearing capacity prediction model, a lightweight convolution module (such as the inverted residual structure in MobileNetV3) is used to process spatial distribution data (such as soil moisture heat map) and output a high-dimensional feature tensor. ,in, , for input spatial data , after convolution operation Get the features.

[0048] In the time series modeling layer of the dynamic bearing capacity prediction model, the feature graph is expanded into (T is the time step), input to the Transformer encoder, and use the multi-head self-attention mechanism to capture long-term dependencies. Self-attention calculation: Output timing characteristics .

[0049] In the multimodal fusion layer of the dynamic carrying capacity prediction model, environmental data (such as temperature and rainfall) are encoded into Fe RD, after splicing with spatiotemporal features, is weightedly fused through a gated fusion mechanism: ,in for function, are learnable parameters.

[0050] S104: adjusting the bearing capacity warning thresholds corresponding to different warning levels according to the confidence intervals, and determining whether there is a construction risk according to the bearing capacity prediction value and the bearing capacity warning thresholds.

[0051] Specifically, based on the benchmark threshold corresponding to each warning level, a benchmark width is obtained. Based on the width corresponding to the confidence interval and the benchmark width, a normalized interval width is obtained. Based on the normalized interval width and meteorological forecast data, the initial safety factor is adjusted. Based on the adjusted safety factor, the bearing capacity warning threshold corresponding to each warning level is determined. In the embodiment of the present application, the bearing capacity warning includes loadable, critical, and overload.

[0052] Furthermore, based on the bearing capacity warning thresholds corresponding to different warning levels, warning intervals corresponding to different warning levels are constructed, and the predicted bearing capacity value is sequentially determined to determine whether it falls within the warning interval. If a construction risk exists, the warning interval to which the predicted bearing capacity value belongs is determined. Based on the warning interval, the warning level is determined, and control instructions are generated based on the warning level and sent to the equipment controller to adjust the parameters of the construction equipment.

[0053] It's important to note that the final results are sent in a structured format to a central control system or other relevant applications for management review or automated decision-making. Furthermore, the system records each inference result and its corresponding metadata (timestamp, sensor ID) to facilitate subsequent auditing and optimization. If anomalies or recognition errors are detected, an alarm mechanism can be triggered, prompting relevant personnel to intervene promptly.

[0054] This application uses a dynamic bearing capacity prediction model to update soil parameters in real time, effectively overcoming the limitations of traditional static models that rely on pre-construction geological data. Real-time data-driven dynamic bearing capacity assessment, through intelligent linkage with construction machinery, upgrades risk prevention and control from passive response to active intervention, realizing a "perception-analysis-decision-control" closed loop, and upgrading from "monitoring-alarm-manual processing" to "prediction-intervention-automated adjustment."

[0055] like Figure 2 As shown, the embodiment of the present application also proposes a dynamic prediction device for bearing capacity of wet land construction, including:

[0056] at least one processor; and,

[0057] a memory communicatively connected to the at least one processor; wherein,

[0058] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a dynamic prediction method for bearing capacity of wet land construction as described in any of the above embodiments.

[0059] An embodiment of the present application further provides a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured to be: a method for dynamically predicting the bearing capacity of wet land construction as described in any of the above embodiments.

[0060] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the device and medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simple. For relevant portions, refer to the descriptions of the method embodiments.

[0061] The devices and media provided in the embodiments of the present application correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0062] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for dynamically predicting the bearing capacity of wet land construction, characterized in that: include: The dynamic load-bearing capacity prediction model pre-deployed at the edge is trained based on laboratory calibration data and historical engineering data. Real-time collection of multi-source sensor data in wetland construction areas through sensor networks; The multi-source sensor data includes soil reflection wave, soil vibration string frequency, soil displacement data, and weather forecast data; The method further comprises: Aligning the multi-source sensor data based on acquisition timestamps to obtain missing values; Determine the sensor device corresponding to the missing value, and fill in the missing value using adjacent data collected by the sensor device at adjacent timestamps; Extract features corresponding to multi-source sensor data without missing values and construct a multi-dimensional feature matrix; The extracting of features corresponding to multi-source sensor data without missing values and constructing a multi-dimensional feature matrix specifically includes: Extract the reflection peak time delay difference in the soil reflection wave at different depths to determine the soil moisture content at different depths; Perform Fourier transformation on the collected grid-distributed soil vibration string frequency to obtain the soil pressure values corresponding to different grid areas; Determining the corresponding soil displacement rate based on the soil displacement data collected at different coordinate positions, and determining the soil tilt direction and soil tilt amplitude based on the soil displacement rate at the different coordinate positions; Constructing a multidimensional feature matrix based on a preset time and space grid using the soil moisture content, the soil pressure value, the soil tilt amplitude and the weather forecast data; Obtaining a bearing capacity prediction value and a confidence interval based on the multi-source sensor data using the dynamic bearing capacity prediction model; The method of obtaining a bearing capacity prediction value and a confidence interval based on the multi-source sensor data using the dynamic bearing capacity prediction model specifically includes: Extracting spatial features of the multidimensional feature matrix through the convolutional neural network of the dynamic bearing capacity prediction model; Inputting the spatial features into the self-attention layer of the dynamic carrying capacity prediction model to capture the dependencies and temporal characteristics between the multidimensional features in different spatial positions; Obtaining predicted values and confidence intervals of carrying capacity at different spatial locations according to the dependency relationship and time series characteristics; The bearing capacity warning thresholds corresponding to different warning levels are adjusted according to the confidence intervals, and whether there is a construction risk is determined according to the bearing capacity prediction value and the bearing capacity warning thresholds.

2. The method for dynamic prediction of bearing capacity of wet land construction according to claim 1, characterized in that: After collecting multi-source sensor data of the wet land construction area in real time through the sensor network, the method further includes: caching the multi-source sensor data to the edge end; Based on a preset time window, the dynamic bearing capacity prediction model is trained by using cached multi-source sensor data to update model parameters; After the training is completed, the multi-source sensor data cached at the edge end is cleared.

3. The method for dynamic prediction of bearing capacity of wet land construction according to claim 1, characterized in that: The adjusting of the bearing capacity warning thresholds corresponding to different warning levels according to the confidence interval specifically includes: Based on the benchmark thresholds corresponding to each warning level, the benchmark width is obtained; Obtaining a normalized interval width according to the width corresponding to the confidence interval and the reference width; adjusting an initial safety factor according to the normalized interval width and the weather forecast data; Based on the adjusted safety factor, the bearing capacity warning threshold corresponding to each warning level is determined.

4. The method for dynamic prediction of bearing capacity of wet land construction according to claim 3, characterized in that: The determining whether there is a construction risk based on the bearing capacity prediction value and the bearing capacity warning threshold specifically includes: Constructing warning intervals corresponding to the different warning levels based on the carrying capacity warning thresholds corresponding to the different warning levels; It is determined in turn whether the predicted bearing capacity value is within the warning interval.

5. The method for dynamic prediction of bearing capacity of wet land construction according to claim 4, characterized in that: After determining whether there is a construction risk based on the bearing capacity prediction value and the bearing capacity warning threshold, the method further includes: When there is a construction risk, the warning interval to which the bearing capacity prediction value belongs is determined, and the warning level is determined based on the warning interval; A control instruction is generated based on the warning level and sent to the equipment controller to adjust the parameters of the construction equipment.

6. A dynamic prediction device for bearing capacity of wet land construction, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the dynamic prediction method for bearing capacity of wet land construction as described in any one of claims 1 to 5.

7. A non-volatile computer storage medium storing computer-executable instructions, characterized in that: The computer executable instructions are configured as a method for dynamically predicting bearing capacity of wet land construction as described in any one of claims 1 to 5.

Images