A method and system for intelligent monitoring of soil voiding at the bottom of a caisson

Abstract

The application discloses a kind of caisson bottom soil body void intelligent monitoring method and system, it is related to caisson construction monitoring technical field, the method includes in caisson bottom preset multiple monitoring positions, obtains the strain time series data of each monitoring position in construction process;Each strain time series data is labeled and length uniform, and the label includes void and non-void;Strain time series data and its label are made into two-class samples after length uniform, and two-class sample set is constructed;The one-dimensional strain time series data in the two-class sample is carried out short-time Fourier transform and generates two-dimensional time-frequency spectrogram;Model training is carried out to the void prediction model pre-set by two-class sample set after processing;The trained void prediction model is deployed and applied, and the intelligent monitoring of caisson bottom soil body void is realized.The application can realize the global, real-time, accurate discrimination of caisson bottom void state.

Description

Technical Field

[0001] This invention relates to the field of caisson construction monitoring technology, and in particular to an intelligent monitoring method and system for soil voiding at the bottom of a caisson. Background Technology

[0002] With the rapid advancement of infrastructure projects such as large bridges and port terminals, caisson structures are widely used in deep-water foundations and underground engineering due to their high load-bearing capacity and adaptability to complex geological conditions. Voiding of the soil at the bottom of the caisson is a typical hidden danger during construction. The presence of voided areas leads to uneven stress on the caisson, significantly reducing the foundation's bearing stability and even causing safety accidents such as structural displacement and cracking. Therefore, accurately determining the state of soil voiding at the bottom of the caisson is a crucial step in ensuring construction safety.

[0003] Currently, monitoring of voids at the bottom of caissons mainly relies on point sensors (such as earth pressure cells) or manual inspections, which has the following shortcomings:

[0004] 1. Limited monitoring range: Point sensors can only reflect local conditions and are difficult to monitor the entire area;

[0005] 2. Poor real-time performance: Manual inspections rely on experience and cannot achieve continuous, real-time monitoring;

[0006] 3. Insufficient feature extraction: Existing methods are mostly based on simple threshold judgments, which fail to fully explore the time-frequency features and spatial correlation information contained in time-series strain signals;

[0007] 4. Low level of intelligence: The lack of a systematic discrimination method that integrates multi-source data and intelligent algorithms leads to insufficient model generalization ability and accuracy.

[0008] To address the aforementioned shortcomings, this application proposes an intelligent monitoring method and system for soil voiding at the bottom of a caisson. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide an intelligent monitoring method and system for soil voiding at the bottom of a caisson, which can realize the full-area, real-time and accurate identification of the voiding state at the bottom of the caisson.

[0010] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0011] Firstly, a method for intelligent monitoring of soil voids at the bottom of a caisson includes:

[0012] Multiple monitoring locations are pre-set at the bottom of the caisson to obtain strain time-series data at each monitoring location during construction.

[0013] Each strain time series data is labeled and has a uniform length; the labels include those with and without empty spaces.

[0014] By combining strain time-series data of uniform length and their labels into binary classification samples, a binary classification sample set is constructed.

[0015] A short-time Fourier transform is performed on the one-dimensional strain time-series data in the binary classification samples to generate a two-dimensional time-spectrum graph.

[0016] The pre-set empty prediction model is trained using the processed binary sample set.

[0017] The trained voiding prediction model is deployed and applied to achieve intelligent monitoring of voiding in the soil at the bottom of the caisson.

[0018] Optionally, the monitoring locations include the footrest surface of the blade, the bottom surface of the partition wall, and the bottom surface of the well wall.

[0019] Optionally, acquiring strain time-series data at each monitoring location during construction includes:

[0020] Deformable mounting plates are pre-embedded at each of the aforementioned monitoring locations;

[0021] Fiber Bragg grating sensors are densely distributed on each of the deformable mounting plates;

[0022] The fiber optic sensor, the deformable mounting plate, and the caisson are coupled together by concrete pouring to form a combined structure.

[0023] During construction, strain timing data is collected and uploaded using the fiber optic grating sensor.

[0024] Optionally, the deformable mounting plate is made of angle steel.

[0025] Optionally, labeling each of the strain time series data includes:

[0026] Each strain time series data is divided into a front-end data and a back-end data according to the drainage sinking stage and the non-drainage sinking stage;

[0027] During the drainage and settling stage, if a void is found based on the on-site investigation, the data at the moment the void occurs and the preceding data are extracted and marked as void, and the marking of the current strain time series data is terminated.

[0028] If the data is not empty, then mark the preceding data as not empty;

[0029] During the non-drainage settling stage, a sliding window is used to segment the subsequent data, and the strain rate is calculated by taking the first data point of each segment. :

[0030]

[0031] In the formula, The first The first data in each segment The step size of the sliding window;

[0032] According to the variable speed Calculate the emptying coefficient :

[0033] In the formula, The first The rate of increase in strain within each segment For the first The emptying coefficient within each segment This represents the total number of units during non-empty periods;

[0034] Threshold determination steps:

[0035] like Then cut off the first The data is segmented and marked as empty, and the labeling of the current strain time series data is ended.

[0036] like and Then the subsequent data will be marked as non-empty;

[0037] like and ,but Recalculate the devotion coefficient And repeat the threshold determination step;

[0038] in, The threshold for determining emptiness. This represents the number of segments in the subsequent data.

[0039] Optionally, the uniform length includes:

[0040] Set a uniform length benchmark value, and delete strain time series data with lengths smaller than the specified length benchmark value after labeling;

[0041] The remaining strain time series data after labeling are retained forward from the last digit, taking the length reference value as the starting point.

[0042] Optionally, the step of generating a two-dimensional time-spectrum graph by performing a short-time Fourier transform on the one-dimensional strain time-series data in the binary classification samples includes:

[0043] Short-time Fourier transform of one-dimensional strain time series data Transformed into a two-dimensional time-frequency matrix And calculate the time-frequency matrix. energy amplitude ;

[0044] energy amplitude Perform a logarithmic transformation to generate a two-dimensional time-frequency spectrum. .

[0045] Optionally, the empty prediction model is constructed based on a convolutional neural network, coarse and fine-grained convolutional modules, a self-attention mechanism, and a classifier;

[0046] The convolutional neural network includes multiple convolutional layers for extracting spatial features from the two-dimensional time-spectrum map;

[0047] The coarse-grained and fine-grained convolution modules include a coarse-grained convolution module and a fine-grained convolution module, which are used to extract coarse-grained and fine-grained temporal features from the sequence of spatial features, respectively.

[0048] The self-attention mechanism is used to perform self-attention encoding and feature fusion on coarse-grained and fine-grained temporal features, capturing long-distance temporal dependencies to generate fused temporal features;

[0049] The spatial features and the fused temporal features are globally pooled and concatenated, and then input into the classifier for prediction of whether the features are empty or not.

[0050] Secondly, the present invention provides an intelligent monitoring system for soil voids at the bottom of a caisson, comprising:

[0051] The data acquisition module is configured to preset multiple monitoring locations at the bottom of the caisson and acquire strain time-series data at each monitoring location during construction.

[0052] The data processing module is configured to label and unify the length of each strain time series data, wherein the labels include those with and without empty spaces;

[0053] The sample preparation module is configured to combine strain time-series data of uniform length and their labels into binary samples to construct a binary sample set.

[0054] The time-frequency conversion module is configured to perform a short-time Fourier transform on the one-dimensional strain time-series data in the binary classification samples to generate a two-dimensional time-frequency spectrum.

[0055] The model training module is configured to train a pre-set empty prediction model using the processed binary classification sample set.

[0056] The model application module is configured to deploy and apply the trained voiding prediction model to achieve intelligent monitoring of voiding in the soil at the bottom of the caisson.

[0057] Thirdly, the present invention provides an electronic device, including a processor and a storage medium;

[0058] The storage medium is used to store instructions;

[0059] The processor is configured to operate according to the instructions to perform the steps according to the method described above.

[0060] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0061] This invention provides an intelligent monitoring method for soil voiding at the bottom of a caisson. It collects strain time-series data from monitoring locations at the bottom of the caisson as the basis for predicting and identifying voiding and non-voiding. Hidden features in the strain time-series data are mined using short-time Fourier transform, and analyzed and predicted using a deep learning-based voiding prediction model, providing technical support for caisson construction safety monitoring. Simultaneously, in the voiding prediction model, spatial correlation information is extracted through a convolutional neural network, and coarse- and fine-grained temporal features are fused with self-attention mechanisms to improve the accuracy and generalization ability of voiding detection. Attached Figure Description

[0062] Figure 1 This is a flowchart of the intelligent monitoring method for soil voids at the bottom of a caisson provided in an embodiment of the present invention;

[0063] Figure 2 This is a schematic diagram of the training process of the void prediction model provided in an embodiment of the present invention;

[0064] Figure 3 This is a schematic diagram of the training effect evaluation metrics for both empty and non-empty datasets provided in this embodiment of the invention.

[0065] Figure 4 This is a schematic diagram illustrating the application effect of the void prediction model provided in the embodiments of the present invention. Detailed Implementation

[0066] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0067] Example 1

[0068] like Figure 1 As shown in the figure, this invention provides an intelligent monitoring method for soil voids at the bottom of a caisson, comprising the following steps:

[0069] Step S1: Pre-set multiple monitoring locations at the bottom of the caisson to obtain strain time sequence data at each monitoring location during construction.

[0070] Specifically, in this embodiment, the monitoring locations include, but are not limited to, the cutting edge surface, the bottom surface of the partition wall, and the bottom surface of the well wall. The cutting edge surface, located at the outermost edge of the caisson, is the primary point for cutting into the soil layer, transmitting vertical loads, and resisting horizontal earth pressure. The soil at this location is most susceptible to erosion and seepage from surrounding water flow, making it prone to voiding. The bottom surface of the partition wall, located directly below the internal structure of the caisson, bears the loads of the equipment inside the caisson and the forces transmitted by the superstructure. Voiding may begin to develop from the central area inside the caisson, a more concealed and dangerous situation. The bottom surface of the well wall covers the entire bottom area of ​​the caisson and is the main foundation surface bearing all vertical loads. It is more effective in confirming whether voiding has occurred and in locating the extent, shape, and development trend of voiding. This embodiment preferably uses three monitoring locations: the cutting edge surface, the bottom surface of the partition wall, and the bottom surface of the well wall, forming a three-dimensional monitoring system from the outer edge (cutting edge) to the interior (partition wall), and from key points to the entire surface, achieving comprehensive data monitoring.

[0071] Specifically, in this embodiment, acquiring strain time-series data at each monitoring location during construction includes:

[0072] (1) A deformable mounting plate is pre-embedded at each monitoring location. The deformable mounting plate is preferably an angle steel.

[0073] Angle steel, as a mounting base material, is readily available during caisson construction. Its high rigidity provides a stable, deformation-free mounting surface for fiber Bragg grating (FBG) sensors, ensuring synchronized deformation between the sensor and the structure and reducing measurement errors. The L-shaped structure of angle steel allows for easy and secure installation at the monitoring location within the caisson using bolts, welding, or bonding, adapting to various installation scenarios. The flat surface of angle steel facilitates tight bonding or welding with the FBG sensor, ensuring effective transmission of structural strain to the sensor's sensitive grating. Simultaneously, angle steel provides a degree of physical protection for the fragile FBG sensor and its leads, preventing accidental collisions and scratches during construction or use, and reducing deformation of the FBG cable caused by subsequent concrete pouring or other construction processes, thus minimizing its effective measurement range.

[0074] (2) Fiber Bragg grating sensors are densely distributed on each deformable mounting plate.

[0075] The densely distributed setup can ensure full-area, real-time monitoring of the bottom of the caisson, collect comprehensive spatial strain distribution characteristics, and gain a more accurate and detailed understanding of the voiding situation at the bottom of the caisson.

[0076] (3) The fiber optic grating sensor, deformable mounting plate and caisson are coupled together by concrete pouring to form a combined structure.

[0077] By forming a combined structure, spatial strain parameters can be collected more accurately.

[0078] (4) During the construction process, strain timing data is collected and uploaded through fiber optic grating sensors.

[0079] Step S2: Label each strain time series data and unify its length. The labels include those with and without empty space.

[0080] Specifically, in this embodiment, labeling each strain time series data includes:

[0081] (1) Divide each strain time series data into front-end data and back-end data according to the drainage sinking stage and the non-drainage sinking stage;

[0082] (2) During the drainage and sinking stage, if there is a void, the data of the time when the void occurs and the preceding data are extracted and marked as void, and the marking of the current strain time series data is ended.

[0083] If the data is not empty, then mark the preceding data as not empty;

[0084] (3) During the non-drainage settling stage, a sliding window is used to segment the data of the later stage, and the strain rate is calculated by taking the first data of each segment. :

[0085]

[0086] In the formula, The first The first data in each segment The step size of the sliding window;

[0087] According to the variable growth rate Calculate the emptying coefficient :

[0088] In the formula, The first The rate of increase in strain within each segment For the first The emptying coefficient within each segment This represents the total number of units during non-empty periods;

[0089] Threshold determination steps:

[0090] like Then cut off the first The data is segmented and marked as empty, and the labeling of the current strain time series data is terminated.

[0091] like and Then the subsequent data will be marked as non-empty;

[0092] like and ,but Recalculate the devotion coefficient And repeat the threshold determination step;

[0093] in, The threshold for determining emptiness. This represents the number of segments in the subsequent data.

[0094] Specifically, in this embodiment, the uniform length includes:

[0095] Set a uniform length baseline value and delete strain time series data with lengths shorter than the length baseline value after labeling;

[0096] The remaining strain time series data, after labeling, retains the length baseline value starting from the last digit.

[0097] By retaining the end data of the strain time series data, it is possible to ensure that the key strain time series data that produce detachment or non-detachment are saved, while unifying the signal length to facilitate subsequent feature extraction and other processing.

[0098] Step S3: Combine strain time series data of uniform length and their labels into binary classification samples to construct a binary classification sample set.

[0099] Specifically, in this embodiment, the empty label is set to 1 in the binary classification samples, and the non-empty label is set to 0. To facilitate subsequent model training and validation, the binary classification sample set is usually divided into a training set and a validation set proportionally, and a stratified random sampling method is used to ensure a balanced distribution of categories, thereby improving the efficiency of model training and the performance of the final model.

[0100] Step S4: Perform a short-time Fourier transform on the one-dimensional strain time series data in the binary classification samples to generate a two-dimensional time spectrum.

[0101] Specifically, in this embodiment, generating a two-dimensional time-spectrum graph includes:

[0102] Short-time Fourier transform of one-dimensional strain time series data Transformed into a two-dimensional time-frequency matrix And calculate the time-frequency matrix. energy amplitude ;

[0103] energy amplitude Perform a logarithmic transformation to generate a two-dimensional time-frequency spectrum. .

[0104] One-dimensional strain time-series data only shows the change in amplitude over time, while two-dimensional time-spectrum plots extend this to a three-dimensional space of time-frequency-energy (or amplitude). It clearly demonstrates how the frequency components of the signal evolve over time, which is crucial for analyzing the non-stationary strain signal in this application. Furthermore, two-dimensional time-spectrum plots allow for the direct application of mature and powerful convolutional neural networks to automatically learn local patterns and hierarchical features, opening the door to utilizing modern image recognition technology for efficient and accurate identification of non-non ...

[0105] Step S5: Train the pre-set empty prediction model using the processed binary classification sample set.

[0106] Specifically, in this embodiment, the empty prediction model is built based on a convolutional neural network, coarse and fine-grained convolutional modules, a self-attention mechanism, and a classifier;

[0107] Convolutional neural networks consist of multiple convolutional layers used to extract spatial features from a two-dimensional temporal spectrogram.

[0108] The coarse-grained and fine-grained convolution modules include a coarse-grained convolution module (composed of two layers of 1D convolution and ReLU activation) and a fine-grained convolution module (composed of two layers of 1D convolution and ReLU activation), which are used to extract coarse-grained and fine-grained temporal features from the sequence of spatial features, respectively.

[0109] The self-attention mechanism is used to encode and fuse coarse-grained and fine-grained temporal features with self-attention, capturing long-distance temporal dependencies to generate fused temporal features;

[0110] Spatial features and fused temporal features are globally pooled and concatenated, and then input into a classifier for prediction of whether features are empty or not.

[0111] This embodiment collected 68 sets of non-empty binary classification samples and 53 sets of empty binary classification samples, and randomly divided them into training and validation sets in an 8:2 ratio. The training process of the empty prediction model is as follows: Figure 2 As shown, the training performance is evaluated using both de-emptied and non-de-emptied datasets, with the evaluation metrics as follows: Figure 3 As shown.

[0112] Step S6: Deploy and apply the trained voiding prediction model to achieve intelligent monitoring of voiding in the soil at the bottom of the caisson.

[0113] like Figure 4 The image shown is an application effect diagram of the vacancy prediction model. Figure 4 As can be seen from the data, the vacancy prediction model can accurately predict and identify whether vacancy has occurred or not based on strain time series data.

[0114] In summary, this embodiment provides an intelligent monitoring method for soil voiding at the bottom of a caisson. By deploying angle steel in the bottom structure of the caisson for dense distributed fiber optic strain monitoring, the acquired strain time-series data is incorporated into the analysis. Hidden features are mined through STFT time-frequency conversion, spatial correlation information is extracted by combining CNN, and coarse and fine-grained time features and Transformer features are fused to improve the accuracy and generalization ability of voiding detection, providing technical support for the safety monitoring of caisson construction.

[0115] Example 2

[0116] Based on the intelligent monitoring method for soil voids at the bottom of the caisson proposed in Example 1, in step S4, the time-spectrum diagram can be further converted into graph structure data, specifically including:

[0117] (1) The time spectrum diagram The data is flattened into a one-dimensional vector by frequency and time index, with each pixel corresponding to a graph node. The node feature of the graph node is the spectral value of the corresponding pixel, generating a node feature matrix.

[0118] (2) Create a zero matrix of the target size as the initial adjacency matrix. According to the rule that each graph node only connects to its four neighboring nodes (up, down, left, and right), traverse all graph nodes and update the adjacency matrix.

[0119] (3) Convert the node feature matrix and adjacency matrix into PyTorch tensors and integrate them to form graph structure data that can be directly processed by graph convolutional neural networks.

[0120] At this point, the convolutional neural network in the empty prediction model can be replaced with a graph convolutional neural network to extract high-order features of the graph space from the graph structure data. The subsequent coarse and fine granular convolution modules, self-attention mechanism and classifier are not adjusted.

[0121] In this embodiment, graph convolutional neural networks extract high-order features of the graph space from graph structure data. Compared with extracting spatial features from a two-dimensional time-spectrum graph using a convolutional neural network in Embodiment 1, graph convolutional neural networks extend the concept of "space" from the "regular grid space" of convolutional neural networks to the "relational graph space," thereby providing a more accurate modeling tool for processing complex networked time-series data.

[0122] In summary, the intelligent monitoring method for soil voiding at the bottom of a caisson in this embodiment involves densely distributed fiber optic strain monitoring using angle steel arranged in the bottom structure of the caisson. The acquired strain time-series data is incorporated into the analysis, hidden features are mined through STFT time-frequency conversion, spatial correlation information is extracted by combining GCN, and coarse and fine-grained time features and Transformer features are fused to improve the accuracy and generalization ability of voiding detection, providing technical support for the safety monitoring of caisson construction.

[0123] Example 3

[0124] This invention provides an intelligent monitoring system for soil voids at the bottom of a caisson, comprising:

[0125] The data acquisition module is configured to preset multiple monitoring locations at the bottom of the caisson to acquire strain time-series data at each monitoring location during construction.

[0126] The data processing module is configured to label each strain time series data and standardize its length. The labels include those with and without empty spaces.

[0127] The sample preparation module is configured to combine strain time-series data of uniform length and their labels into binary samples to construct a binary sample set.

[0128] The time-frequency conversion module is configured to perform a short-time Fourier transform on the one-dimensional strain time-series data in the binary classification samples to generate a two-dimensional time-frequency spectrum.

[0129] The model training module is configured to train a pre-set empty prediction model using a processed binary classification sample set.

[0130] The model application module is configured to deploy and apply the trained voiding prediction model to achieve intelligent monitoring of voiding in the soil at the bottom of the caisson.

[0131] Example 4

[0132] Based on the intelligent monitoring method for soil voids at the bottom of the caisson provided in Embodiments 1 and 2, this embodiment of the invention also provides an electronic device, including a processor and a storage medium;

[0133] Storage media are used to store instructions;

[0134] The processor is used to perform operations according to instructions to execute the steps according to the method described above.

[0135] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0137] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0138] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0139] 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 technical principles 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 method for intelligent monitoring of soil voids at the bottom of a caisson, characterized in that, include: Multiple monitoring locations are pre-set at the bottom of the caisson to obtain strain time-series data at each monitoring location during construction. Each strain time series data is labeled and has a uniform length; the labels include those with and without empty spaces. By combining strain time-series data of uniform length and their labels into binary classification samples, a binary classification sample set is constructed. A short-time Fourier transform is performed on the one-dimensional strain time-series data in the binary classification samples to generate a two-dimensional time-spectrum graph. The pre-set empty prediction model is trained using the processed binary sample set. The trained voiding prediction model is deployed and applied to achieve intelligent monitoring of voiding in the soil at the bottom of the caisson; The labeling of each strain time series data includes: Each strain time series data is divided into a front-end data and a back-end data according to the drainage sinking stage and the non-drainage sinking stage; During the drainage and settling stage, if a void is found based on the on-site investigation, the data at the moment the void occurs and the preceding data are extracted and marked as void, and the marking of the current strain time series data is terminated. If the data is not empty, then mark the preceding data as not empty; During the non-drainage settling stage, a sliding window is used to segment the subsequent data, and the strain rate is calculated by taking the first data point of each segment. : ； In the formula, The first The first data in each segment The step size of the sliding window; According to the variable speed Calculate the emptying coefficient : ； In the formula, The first The rate of increase in strain within each segment For the first The emptying coefficient within each segment This represents the total number of units during non-empty periods; Threshold determination steps: like Then cut off the first The data is segmented and marked as empty, and the labeling of the current strain time series data is ended. like and Then the subsequent data will be marked as non-empty; like and ,but Recalculate the devotion coefficient And repeat the threshold determination step; in, The threshold for determining emptiness. This represents the number of segments in the subsequent data.

2. The intelligent monitoring method for soil voids at the bottom of a caisson according to claim 1, characterized in that, The monitoring locations include the footrest surface of the blade, the bottom surface of the partition wall, and the bottom surface of the well wall.

3. The intelligent monitoring method for soil voids at the bottom of a caisson according to claim 1, characterized in that, The acquisition of strain time-series data at each monitoring location during the construction process includes: Deformable mounting plates are pre-embedded at each of the aforementioned monitoring locations; Densely distributed fiber Bragg grating sensors are arranged on each of the deformable mounting plates; The fiber optic sensor, the deformable mounting plate, and the caisson are coupled together by concrete pouring to form a combined structure. During construction, strain timing data is collected and uploaded using the fiber optic grating sensor.

4. The intelligent monitoring method for soil voids at the bottom of a caisson according to claim 3, characterized in that, The deformable mounting plate is made of angle steel.

5. The intelligent monitoring method for soil voids at the bottom of a caisson according to claim 1, characterized in that, The uniform length includes: Set a uniform length benchmark value, and delete strain time series data with lengths smaller than the specified length benchmark value after labeling; The remaining strain time series data after labeling are retained forward from the last digit, taking the length reference value as the starting point.

6. The intelligent monitoring method for soil voids at the bottom of a caisson according to claim 1, characterized in that, The step of generating a two-dimensional time-spectrum image by performing a short-time Fourier transform on the one-dimensional strain time-series data in the binary classification samples includes: Short-time Fourier transform of one-dimensional strain time series data Transformed into a two-dimensional time-frequency matrix And calculate the time-frequency matrix. energy amplitude ; energy amplitude Perform a logarithmic transformation to generate a two-dimensional time-frequency spectrum. .

7. The intelligent monitoring method for soil voids at the bottom of a caisson according to claim 1, characterized in that, The empty prediction model is built based on a convolutional neural network, coarse and fine-grained convolutional modules, a self-attention mechanism, and a classifier; The convolutional neural network includes multiple convolutional layers for extracting spatial features from the two-dimensional time-spectrum map; The coarse-grained and fine-grained convolution modules include a coarse-grained convolution module and a fine-grained convolution module, which are used to extract coarse-grained and fine-grained temporal features from the sequence of spatial features, respectively. The self-attention mechanism is used to perform self-attention encoding and feature fusion on coarse-grained and fine-grained temporal features, capturing long-distance temporal dependencies to generate fused temporal features; The spatial features and the fused temporal features are globally pooled and concatenated, and then input into the classifier for prediction of whether the features are empty or not.

8. An intelligent monitoring system for soil voids at the bottom of a caisson, characterized in that, include: The data acquisition module is configured to preset multiple monitoring locations at the bottom of the caisson and acquire strain time-series data at each monitoring location during construction. The data processing module is configured to label and unify the length of each strain time series data, wherein the labels include those with and without empty spaces; The sample preparation module is configured to combine strain time-series data of uniform length and their labels into binary samples to construct a binary sample set. The time-frequency conversion module is configured to perform a short-time Fourier transform on the one-dimensional strain time-series data in the binary classification samples to generate a two-dimensional time-frequency spectrum. The model training module is configured to train a pre-set empty prediction model using the processed binary classification sample set. The model application module is configured to deploy and apply the trained voiding prediction model to achieve intelligent monitoring of voiding in the soil at the bottom of the caisson. The data processing module performs labeling on each strain time series data point, including: Each strain time series data is divided into a front-end data and a back-end data according to the drainage sinking stage and the non-drainage sinking stage; During the drainage and settling stage, if a void is found based on the on-site investigation, the data at the moment the void occurs and the preceding data are extracted and marked as void, and the marking of the current strain time series data is terminated. If the data is not empty, then mark the preceding data as not empty; During the non-drainage settling stage, a sliding window is used to segment the subsequent data, and the strain rate is calculated by taking the first data point of each segment. : ； In the formula, The first The first data in each segment The step size of the sliding window; According to the variable speed Calculate the emptying coefficient : ； In the formula, The first The rate of increase in strain within each segment For the first The emptying coefficient within each segment This represents the total number of units during non-empty periods; Threshold determination steps: like Then cut off the first The data is segmented and marked as empty, and the labeling of the current strain time series data is ended. like and Then the subsequent data will be marked as non-empty; like and ,but Recalculate the devotion coefficient And repeat the threshold determination step; in, The threshold for determining emptiness. This represents the number of segments in the subsequent data.

9. An electronic device, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-7.

Images