Ground surface settlement control method and system based on multi-source muck properties

Abstract

The invention discloses a ground surface settlement control method based on multi-source muck properties. The method comprises the steps that the strain distribution condition of a ground surface settlement to-be-detected area under an initial grouting material is obtained; based on the strain distribution condition, determining a target connection relation matrix used for describing strain abnormity; if the number of the target connection relation matrixes is 0, ending the step; inputting the target connection relation matrix into a trained graph neural network, and determining a corresponding secondary grouting material; and taking the secondary grouting material as the initial grouting material again. According to the invention, the FBG sensor is used to monitor and accurately describe the stress distribution in the earth surface in real time; a graph neural network is further introduced to carry out optimization selection on the secondary grouting material, and the problems that loss functions are not matched and verification is difficult in the training process are solved.

Description

Technical Field

[0001] The present invention belongs to the field of surface settlement control, and more specifically, relates to a surface settlement control method and system based on multi-source slag properties. Background Art

[0002] Prior art grouting materials often use cement-fly ash-water glass or cement-bentonite systems. These systems offer advantages such as controllable setting speed, high compressive strength, and good fluidity. They can quickly fill shield tail gaps and seal ground cavities during shield tunneling, preventing surface subsidence and soil and water influx. However, traditional grouting materials present challenges such as high cost, high environmental impact, and the tendency to segregate and backflow during construction.

[0003] To improve the resource utilization and environmental friendliness of grouting materials, a new green grouting solution has been proposed in the existing technology: by comprehensively considering the basic working performance and durability of the slurry (such as impermeability, anti-segregation, and water stability in groundwater environments), the natural sand and gravel components in the slurry are partially or completely replaced with slag (derived from solid waste such as pile foundation spoil and shield tunneling slag). More specifically, the researchers introduced appropriate admixtures (such as retarders, dispersants, and thickeners) into the cement-slag system and, through scientific experimental design, constructed an environmentally friendly synchronous grouting material based on the cement-slag-admixture system.

[0004] On this basis, the system reveals the influence relationship between key design parameters such as water-cement ratio, waste sand ratio, fly-lime ratio and slurry workability (such as fluidity, water bleeding rate, setting time) and durability. Combined with the GP model (Goal Programming, multi-objective programming model), the nonlinear programming tool in MATLAB is used to solve the grouting material ratio scheme that takes into account both optimal performance and optimal cost, thus realizing the high-performance and low-cost coordinated optimization of grouting materials.

[0005] However, although the above-mentioned technology has achieved the resource utilization of construction solid waste, the slag as a raw material has problems such as uneven composition, large fluctuations in particle grading, and uncontrollable organic impurity content. This poses great challenges to the grouting stability and filling uniformity during shield excavation, and thus introduces new uncertainty factors for surface settlement control.

[0006] Traditional surface settlement monitoring and prediction techniques typically employ pre-installed fixed monitoring points, comparing changes in their elevation before and after shield tunneling to infer settlement trends across the entire surface. This approach implicitly assumes that the spatial distribution of the ground response is continuous and predictable, ignoring the existence of spatially heterogeneous settlement and, in particular, failing to consider the impact of settlement gradients.

[0007] The settlement gradient, or the rate of change of settlement per unit length, is a direct indicator of surface deformation inconsistency. Its impact on structural safety far exceeds the average settlement value. This is especially true in areas densely populated with buildings, bridges, underground pipeline corridors, shallow foundations, or older structures. Sudden changes in the settlement gradient are often the root cause of engineering problems such as structural cracking, pipeline severance, and foundation slippage.

[0008] However, under the condition of large amount of slag, due to poor uniformity of grouting materials and large fluctuations in filling density, the formation response loses regularity. Traditional monitoring point interpolation or surface fitting methods cannot effectively identify local high gradient areas or abnormal settlement points, the prediction accuracy decreases significantly, and the engineering risk increases significantly. Summary of the Invention

[0009] In order to address the deficiencies in the prior art, the purpose of the present invention is to solve the above-mentioned defects and further propose a surface settlement control method and system based on multi-source slag properties.

[0010] The present invention adopts the following technical solutions.

[0011] The first aspect of the present invention discloses a method for controlling surface settlement based on multi-source slag properties, comprising:

[0012] Obtain the strain distribution of the surface settlement area to be measured under the initial grouting material;

[0013] Based on the strain distribution, determine the target connection relationship matrix used to describe the strain anomaly; if the number of the target connection relationship matrix is 0, end the step;

[0014] The target connection relationship matrix is input into the trained graph neural network to determine the corresponding secondary grouting material; and the secondary grouting material is reused as the initial grouting material.

[0015] Furthermore, a surface settlement control method based on multi-source slag properties is characterized in that the method includes:

[0016] Obtain the strain distribution of the surface settlement area to be measured under the initial grouting material;

[0017] Based on the strain distribution, determine the target connection relationship matrix used to describe the strain anomaly; if the number of the target connection relationship matrix is 0, end the step;

[0018] The target connection relationship matrix is input into the trained graph neural network to determine the corresponding secondary grouting material; and the secondary grouting material is reused as the initial grouting material.

[0019] Furthermore, the strain distribution is the center wavelength vector measured by the FBG sensor.

[0020] Furthermore, the measurement of the central wavelength vector specifically includes:

[0021] Determine multiple target heights in the area to be measured;

[0022] At each target height, a corresponding FBG sensor line is laid out, wherein each FBG sensor line is a plurality of optical fibers with a plurality of pre-written periodic refractive index modulation regions;

[0023] The central wavelength of the reflected light at each target height is obtained in turn and integrated into multiple central wavelength vectors as the strain distribution.

[0024] Furthermore, the target height includes at least a first target height and a second target height, the first target height and the second target height are located at the boundary between the multi-source slag grouting area and the original surface, the first target height is set within the multi-source slag grouting area, and the second target height is set within the original surface.

[0025] Furthermore, the target connection relationship matrix is determined based on the central wavelength vector; the process specifically includes: when the value of an element in the central wavelength vector exceeds a preset threshold range, the adjacent nodes are uniformly classified into the target connection relationship matrix with the strain direction initial vector corresponding to the element as the boundary.

[0026] Furthermore, the target connection relationship matrix is used to determine the injection position of the secondary grouting material.

[0027] Furthermore, the graph neural network takes the strain distribution of this iteration and the strain distribution of the grouting material as input, and the selection probability of the grouting material as output. It determines the loss function through the target control value in the target connection relationship matrix of the next iteration and the probability of whether the secondary grouting material is selected.

[0028] Furthermore, the usage amount of the secondary grouting material is determined based on the probability of selection of the secondary grouting material.

[0029] A second aspect of the present invention discloses a terminal, comprising a processor and a storage medium; the characteristics of the terminal are:

[0030] The storage medium is used to store instructions;

[0031] The processor is configured to operate according to the instructions to execute the steps of the method of the first aspect.

[0032] A third aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, wherein the program implements the steps of the method described in the first aspect when executed by a processor.

[0033] The beneficial effects of the present invention are that, compared with the prior art, the present invention has the following advantages:

[0034] The present invention uses FBG sensors to monitor and accurately describe the internal stress distribution of the surface in real time. Traditional methods often rely on discrete monitoring points or low-resolution equipment, which makes it difficult to capture small changes in local stress concentration. FBG sensors have the advantages of high sensitivity, high stability and resistance to electromagnetic interference. They can be continuously deployed along optical fibers and collect data in real time at multiple key target heights (such as the junction of the multi-source slag grouting area and the original surface). By normalizing the central wavelength vector output by the FBG sensor, the distribution characteristics of the complex stress field inside the surface can be accurately reflected. This improves the credibility of the monitoring data. Based on the accurate description of the surface stress distribution using FBG sensors,

[0035] In the graph neural network, the initial strain direction vector of the grouting material itself is coupled with the initial strain direction vector of the current node, achieving a precise connection between material properties and the real-time strain state in the field. This not only improves the graph neural network's ability to predict the response of various materials in different strata and settlement environments, but also provides a richer basis for material selection at each node, thereby improving the prediction accuracy of the entire system. By incorporating the target control value for the next iteration into the loss function, the graph neural network model implements real-time feedback correction of the output of each iteration. This design enables the loss function to more accurately reflect the difference between the current selection and the actual control effect in the future, alleviating the difficulties of traditional single-variable loss functions in verifying multiple spatial indicators. In addition, each graph neural network focuses on the strain data and settlement characteristics of the corresponding layer (the first target height), making the prediction of the grouting layer distribution more targeted and accurate. Based on the target connection relationship matrix, the customized loss function can optimize the unique characteristics of each layer in terms of stress concentration and settlement anomalies, achieving differentiated configuration of grouting materials and construction parameters for different soil layers, thereby meeting the specific reinforcement and consolidation needs of each layer. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of a surface settlement control method based on multi-source slag properties according to an embodiment of the present invention;

[0037] Figure 2 FIG. 4 is a schematic diagram of the internal structure of the FBG sensor according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The present application will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present application.

[0039] The present invention discloses a surface settlement control method based on multi-source slag properties, such as Figure 1As shown, it includes steps 1 to 3.

[0040] Step 1: Obtain the strain distribution of the surface settlement area to be measured.

[0041] The strain distribution should describe the strain and direction at multiple points in the area to be measured for surface settlement. Strain at each point: Strain describes the degree of deformation per unit length of a material (such as soil) under external load. The strain value can reflect the magnitude of local deformation in the soil. Early detection of abnormal increases in local strain can indicate potential settlement issues or structural weaknesses. The direction of strain can help engineers determine whether the deformation caused by settlement is uniform or has distinct principal stress directions, thereby distinguishing between tensile, compressive, and shear deformation. Understanding the principal strain direction can provide a basis for subsequent support, grouting, or other reinforcement measures. For example, if the strain in one direction is significantly greater than in other directions, more concentrated reinforcement measures may be required in that direction. In some embodiments, the strain distribution is a center wavelength vector measured by an FBG sensor. The center wavelength vector is composed of the center wavelengths of different periodic refractive index modulation zones. Specifically, measuring the center wavelength vector can include steps 1.1 to 1.3.

[0042] Step 1.1, determine multiple target heights in the area to be measured.

[0043] The target height includes at least a first target height and a second target height. The first target height and the second target height are usually located at the boundary between the multi-source slag grouting area and the original ground surface. This is usually an area of stress concentration, which is prone to produce local abnormal settlement and high strain gradient. In some embodiments, the first target height can be set within the multi-source slag grouting area, and the second target height can be set within the original ground surface. At each target height, its height is considered as a key control variable to ensure that the impact of different heights on stress concentration and settlement characteristics can be accurately reflected in subsequent analysis. In addition, setting the second target height below the original ground surface as a natural boundary not only helps to identify the settlement gradient relative to the underground grouting area, but also can capture the most intuitive changes in stress concentration and local abnormal settlement.

[0044] It should be noted that the number of target heights is a fixed value, which can usually be set to 2 to 4; if it is set to 3, the additional third target height should be regarded as an extension of the first target height, that is, the number of first target heights can be one or more, and different first target heights are used to determine different levels of grouting.

[0045] Step 1.2: at each target height, a corresponding FBG sensor line is laid out, wherein each FBG sensor line is a plurality of pre-written optical fibers with a plurality of periodic refractive index modulation regions.

[0046] Considering the problems of uneven composition of multi-source slag, large fluctuations in particle gradation, and uncontrollable organic impurity content, in step 1.2, measurements can be performed based on high-precision fiber Bragg grating (FBG) sensors.

[0047] The working principle of FBG sensor is essentially based on the fiber Bragg grating effect, such as Figure 2 As shown, its core is a periodic refractive index modulation area, which reflects light of a specific wavelength, which is the Bragg wavelength λ B , as shown below:

[0048] λ B =2n c A

[0049] Among them, n c is the effective refractive index in the fiber, and A is the grating period.

[0050] As you can understand, strain causes the fiber to stretch or compress, directly changing the physical period of the grating. That is, in the case of stretching, A increases, while in the case of compression, A decreases. Therefore, the magnitude of the strain can be determined based on the central wavelength of the emitted light.

[0051] exist Figure 2 Figure 1: Fiber cladding: external low-refractive-index material that ensures light transmission in the fiber core; fiber core: internal uniform transmission of broadband optical signals; periodic refractive-index modulation region: periodic refractive-index modulation region within the fiber core, with modulation periods represented by short horizontal lines; incident light: broadband optical signal entering the fiber from the left; reflected light: when a wavelength in the broadband light approaches the Bragg wavelength, constructive interference from multiple modulation periods within the periodic refractive-index modulation region forms a strong reflection, and the reflected light returns to the source; transmitted light: light of non-Bragg wavelengths continues to travel.

[0052] It is understandable that an FBG sensor can essentially only measure the strain change at its location, more precisely, at a single point in the periodic refractive index modulation region.

[0053] It is important to note that both ends of each periodic refractive index modulation region in each optical fiber must be fixed and clipped onto the steel frame, and the distance between the two ends must be consistent. The fixed frame is a rectangular steel structural member, and the cross-section of the fixed frame can be channel steel or I-beam. Each FBG sensor is fixed to the fixed frame with mechanical clamps. Protective rubber pads and weather-resistant protective covers are embedded around the sensor to ensure data stability under the vibration of shield construction and groundwater infiltration. The strain signals collected by the fiber optic sensors are received by a fiber optic data acquisition box integrated into the side of the base. The acquisition box is connected to the central data processing equipment via fiber optic cables for real-time data transmission. The central data processing equipment is usually installed in the shield control center or a nearby monitoring room. The data acquisition box adopts a waterproof and dustproof design and has seismic reinforcement. The fixed frame is fixed to the concrete base with reserved mechanical support bolts. The buried installation method ensures that the sensor network is closely integrated with the stratum in the monitoring area, thereby accurately reflecting the strain amount and direction of the grouting material.

[0054] In step 2, the principle of wavelength division multiplexing (WDM) is essentially implemented. Within the optical fiber, each periodic refractive index modulation region can generate its own central wavelength within the spectral range of a tunable or broadband light source. Crucially, the spacing between adjacent central wavelengths must be large enough to ensure that the maximum spectral shifts of adjacent periodic refractive index modulation regions under the influence of the measured parameter do not overlap, otherwise the measurement results may be ambiguous or erroneous. Based on this, a spectral window can be defined: the span of central wavelengths that an FBG sensor can cover under a specific range of measured variables. Ultimately, the maximum number of periodic refractive index modulation regions in a single optical fiber can be estimated by dividing the overall spectral width or span of the light source by the spectral window width of a single sensor. The number of optical fibers is the quotient obtained by dividing the number of periodic refractive index modulation regions in the FBG sensor circuit by the maximum number of periodic refractive index modulation regions in a single optical fiber.

[0055] In the optical path design, the reflected light is first collected by an optical circulator or optical coupler placed between the light source and the periodic refractive index modulation region, and then transmitted to a photodetector. This allows the peak reflection wavelength generated by each periodic refractive index modulation region to correspond to the peak signal on the detector. Another approach is to use a broadband light source so that the center wavelength of all periodic refractive index modulation regions falls within the spectral range of the light source. The reflected light is then transmitted to a spectrometer, where the spectral data is analyzed to determine the reflection peak, or center wavelength, of each periodic refractive index modulation region.

[0056] It's understandable that a single FBG sensor circuit can determine the spatial distribution of its internal periodic refractive index modulation zones in the horizontal, two-dimensional direction. However, because the actual morphology of the target area to be grouted may be highly uneven, the FBG sensor circuit doesn't necessarily need to remain at the same height. Instead, its vertical position can be adaptively adjusted based on the actual grouting opening to better match on-site construction requirements.

[0057] In step 1.3, the central wavelength of the reflected light at each target height is obtained in turn, and integrated into multiple central wavelength vectors as the strain distribution.

[0058] The output of step 1.3 is multiple center wavelength vectors, including a first center wavelength vector and a second center wavelength vector, corresponding to the first and second FBG sensor lines, respectively. Each element in each center wavelength vector corresponds to a periodic refractive index modulation region. The value of each element is the central wavelength. Because strain can cause the optical fiber to stretch or compress, when preprocessing the center wavelength vector (e.g., in step 2.1 below), the value of the element can be the difference between the central wavelength and the Bragg wavelength of the periodic refractive index modulation region, and then the result is normalized. Normalization here refers to the process of using a proportional factor to correspond unit wavelength change to unit strain change, achieving a one-to-one numerical conversion relationship between the two, thereby ensuring that the wavelength change output by the optical fiber sensor accurately reflects the actual strain change.

[0059] Step 2: Based on the strain distribution, determine the target connection relationship matrix used to describe the strain anomaly; if the number of target connection relationship matrices is 0, end the step.

[0060] In some embodiments, a target connection matrix C can be determined based on the center wavelength vector. Each target connection matrix should have a fixed size, for example, 9*9 or 11*11, but generally should not be smaller than 9*9. Its size depends on the density of periodic refractive index modulation regions within the FBG sensor circuit. It is understood that if the number of target connection matrices is 0, it indicates that the current strain distribution meets the expected requirements and no control is required.

[0061] The target connection matrix is determined based on the center wavelength vector. Specifically, when the value of an element in the center wavelength vector exceeds a preset threshold range (for example, ±0.8nm of the center wavelength), adjacent nodes are uniformly assigned to the target connection matrix, using the strain direction initialization vector corresponding to that element as the boundary. The threshold range can typically be centered around the Bragg wavelength, and the element in the target connection matrix at the center corresponds to the strain direction initialization vector corresponding to that element.

[0062] It is understandable that in one measurement, the number of target connection relationship matrices may be 0 or at least one.

[0063] Step 3: Input the target connection relationship matrix into the trained neural network model to determine the corresponding secondary grouting material; use the secondary grouting material as the initial grouting material again, and then return to step 1.

[0064] The training process of the graph neural network includes steps S21 to S26. The graph neural network is mainly used to analyze the guidance results of the corresponding variables of different secondary grouting materials, especially the strain direction.

[0065] In the graph neural network, an initial strain direction vector represents a node in the graph neural network, denoted as a i , the jth element value of the initial vector of the i-th strain direction is the i-th element value of the j-th central wavelength vector. The value of each element in the target connection relationship matrix is a i [w], the position of the element depends on the initial vector of the strain direction. For the convenience of description, the value of the most central element in the target connection relationship matrix is used as the target control value.

[0066] It is understandable that the strain direction initial vector actually corresponds to a set of central wavelengths vertically distributed along the periodic refractive index modulation area. Therefore, in the above text, adjacent nodes refer to other strain direction initial vectors that are closest in the horizontal two-dimensional direction. It is not difficult to infer that the positional relationship of the elements in the target connection relationship matrix also corresponds to the spatial distribution of the periodic refractive index modulation area in the horizontal two-dimensional direction. Unless otherwise specified, the following a i Or the initial vector of the strain direction, both refer to a defined by the target connection relationship matrix i Or the strain direction initial vector, w represents the number of the element in the strain direction initial vector. In an embodiment of the present invention, the value range of w is limited to the range of the first target height, that is, w∈[0,N-1), where N is the length of the strain direction initial vector.

[0067] The difficulty in training a graph neural network lies in, on the one hand, its goal: to find the most appropriate secondary grouting material; on the other hand, it is difficult to find a suitable loss function to verify the correctness of the graph neural network. Because there is no multi-space, a strict control variable method (the variable is only the secondary grouting material) can be used to verify the performance of each grouting material one by one to verify the output probability of the graph neural network.

[0068] It is understandable that the surface settlement control system includes a database of various slag components, which is used to record the ratios and properties of various available grouting materials. For example, the following different grouting materials can be included: the ratio of the first grouting material includes: cement: 70g; slag: 20g; fly ash: 10g; water: 40g; high-efficiency water reducer (polycarboxylate): 0.2g. The ratio of the second grouting material includes: cement: 60g; slag: 30g; fly ash: 10g; water: 45g; retarder (sodium gluconate): 0.3g. The ratio of the third grouting material includes: cement: 80g; slag: 10g; fly ash: 10g; water: 35g; anti-separation agent (hydroxypropyl methylcellulose, HPMC): 0.15g. The ratio of the fourth grouting material includes: cement: 50g; slag: 40g; fly ash: 10g; water: 50g; waterproofing agent: 0.4g. The proportions of the fifth grouting material include: cement: 65g; slag: 25g; fly ash: 10g; water: 42g; and early strength agent (calcium chloride): 0.7g.

[0069] The surface settlement control system of the present invention actually adjusts the grouting materials (i.e., secondary grouting materials) and corresponding injection positions for the next round according to the target connection relationship matrix based on the real-time strain distribution after the current iteration. It can be understood that the location of the target control value in the target connection relationship matrix of this iteration serves as the injection position (i.e., grouting hole) of the secondary grouting material. In other words, the injection position corresponds one-to-one with the target connection relationship matrix.

[0070] In step S21, based on a preset feature conversion matrix, the initial strain direction vector is mapped to a unified feature space to obtain a space mapping vector, as shown in the following formula:

[0071] s i =V a ·a i

[0072] Among them, V a is the feature transformation matrix, the initial values of its elements are randomly set, s i is the space mapping vector.

[0073] It is understandable that during the training process of the graph algorithm, the initial strain direction vector should be the initial strain direction vector in the historical sample data.

[0074] Step S22: For each non-edge node, calculate the associated embedding representation of its adjacent nodes.

[0075]

[0076] Among them, π iRefers to the set of node numbers of the eight most adjacent nodes of node i. Non-edge nodes refer to the nodes with eight most adjacent nodes in the target connection relationship matrix. C is the target connection relationship matrix, R i is the associative embedding representation, and t and i both represent node numbers. It can be understood that if the target connection relationship matrix is v*v, then the number of non-edge nodes is (v-2)*(v-2).

[0077] In step S23, the maximum value of each non-edge node and its associated embedding representation is calculated as the attention score.

[0078] Attention score g i , which is used to describe the direction of strain, as shown in the following formula:

[0079]

[0080] Where σ(·) represents the activation function, which can be a sigmoid function, || represents the concatenation operation between vectors, and v is the preset attention vector, which is the parameter to be learned and its initial value is randomly set.

[0081] Step S24: Calculate the coupling degree of any node with respect to any grouting material.

[0082] The coupling degree is shown as follows:

[0083] β ik =σ(v·softmax((g i ·s i )||(v a ·A k )))

[0084] Among them, A k is the initial vector of the strain direction of the kth grouting material.

[0085] Understandably, A k In fact, it is obtained in advance based on prior experiments. That is, the kth grouting material is injected into the steel container, and then the same technical means (i.e. the technical means of steps 1.1 to 1.3) are used to obtain the corresponding initial vector of the strain direction. In order to unify the characteristics, usually, A k The vector length is greater than a i Less 1, because a i The last element of (i.e., the element corresponding to the second target height) is embedded in the original surface.

[0086] Step S25: Calculate the coupling characteristic vector according to the coupling degree.

[0087]

[0088] Among them, zk is the coupling eigenvector, and N is the number of elements in the target connection relationship matrix.

[0089] Step S26: Calculate the strain direction vector.

[0090] The strain direction matrix G can be expressed as follows:

[0091] Q k =V Q z k

[0092] Y k =V Y z k

[0093] L k =V L z k

[0094]

[0095] Where d is V Q The dimension, V Q ,V Y ,V L It is the weight matrix in the attention mechanism model, which are all parameters to be learned. Their initial values are randomized and used to calculate the query vector Q k , key vector Y k Sum vector L k , K is the amount of grouting material, T represents the transpose of the matrix, and G is the strain direction vector.

[0096] The loss function Loss is:

[0097] Loss=-{(C m -C′ m )ln(y′ l )+C′ m ln(1-y′ l )}

[0098] y′=[y′1...y′ k ...y′ K ]=softmax(MLP(G))

[0099] Among them, y′ l is an element in y′, l represents the number selected in this iteration, y′ is the prediction result output by the graph neural network, representing the probability of selecting the grouting material; MLP() is a multi-layer perceptron, which is composed of multiple layers of neurons (nodes) and is a feedforward neural network; C m is the target control value of the target connection relationship matrix C of this iteration, C′m is the target control value of the target connection relationship matrix C′, where C′ is the target connection relationship matrix for the next iteration.

[0100] It's important to note that since the number of target connectivity matrices corresponds one-to-one to the number of first target heights, the above approach effectively trains multiple graph neural networks using the same method. The number of graph neural networks should correspond one-to-one to the number of first target heights, differing only in the loss function. Each different graph neural network is used to determine the hierarchical distribution of grouting.

[0101] Taking the first target height as an example, when the number of heights is three, the layers are divided into shallow, medium, and deep layers. The injection depth, hole diameter, grouting pressure, and grouting rate of the grouting holes all vary. Generally, deep layers require higher grouting pressures to overcome the higher compressive strength and pore resistance of the deep soil and ensure that the slurry can penetrate the target area. In addition, the slurry injection rate is usually set faster to quickly fill the soil voids under high pressure, but sufficient time must be ensured for the slurry to spread evenly. Relatively low grouting pressures in shallow layers can meet the requirements, avoiding ground disturbance or local soil damage caused by excessive pressure. A slower grouting rate is set to more accurately control the slurry diffusion range and stabilize the formation, while reducing local disturbances caused by excessive flow rate. It is understandable that during the iterative process, the settings of parameters such as grouting pressure and grouting rate depend on the currently selected secondary grouting material. Their values are determined based on performance indicators such as fluidity, viscosity, and setting time of the secondary grouting material, which can be pre-recorded in the slag database. Since the above details have little relevance to the core invention of this application, they will not be further elaborated here.

[0102] In the specific implementation process, steps 1 to 3 are actually a process of continuous iteration. Therefore, for the loss function, the results of the current round can be evaluated with the help of the results of the next round of iteration. In other words, the graph neural network takes the strain distribution of this iteration and the strain distribution of the grouting material as input, and the selection probability of the grouting material as output, and connects the target control value in the relationship matrix of the next round of iteration, as well as the probability of whether the secondary grouting material is selected (the probability of not selecting the secondary grouting material is 1-y′ l ), determine the loss function.

[0103] It is understandable that during the initial training process, different grouting materials should be selected as dispersedly as possible to enrich the sample data for model training in the slag database. In the later use process, the data generated can also be used as historical data for further model training and continuous iterative optimization.

[0104] In some embodiments, the amount of secondary grouting material used in each iteration may be based on the probability of selection of the secondary grouting material, ie, y′ l , determined by. More specifically, the usage of secondary grouting material can be expressed as follows:

[0105] L′=y′ l ×L(t)

[0106] Where L represents the reference maximum grouting volume, t represents the number of iterations, and the value of L(t) is set based on experience. In principle, it should be as small as possible. When it is set small, the measurement accuracy of the FBG sensor should be taken into consideration. In addition, L(t) should gradually decay with the increase of t, that is, the amount of secondary grouting material used shows an exponential distribution with respect to the number of iterations.

[0107] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation plans of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, and is not a limitation on the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

Claims

1. A surface settlement control method based on multi-source soil properties, characterized in that: The method comprises: Obtain the strain distribution of the surface settlement area to be measured under the initial grouting material; Based on the strain distribution, determine the target connection relationship matrix used to describe the strain anomaly; if the number of the target connection relationship matrix is 0, end the step; The target connection relationship matrix is input into the trained graph neural network to determine the corresponding secondary grouting material; and the secondary grouting material is reused as the initial grouting material.

2. A surface settlement control method based on multi-source slag properties according to claim 1, characterized in that: The strain distribution is the central wavelength vector measured by the FBG sensor.

3. The surface settlement control method based on multi-source soil properties according to claim 2, characterized in that: The measurement of the central wavelength vector specifically includes: Determine multiple target heights in the area to be measured; At each target height, a corresponding FBG sensor line is laid out, wherein each FBG sensor line is a plurality of optical fibers with a plurality of pre-written periodic refractive index modulation regions; The central wavelength of the reflected light at each target height is obtained in turn and integrated into multiple central wavelength vectors as the strain distribution.

4. The surface settlement control method based on multi-source soil properties according to claim 2, characterized in that: The target height includes at least a first target height and a second target height. The first target height and the second target height are located at the boundary between the multi-source slag grouting area and the original surface. The first target height is set within the multi-source slag grouting area, and the second target height is set within the original surface.

5. The surface settlement control method based on multi-source slag properties according to claim 2, characterized in that: The target connection relationship matrix is determined based on the central wavelength vector. The specific process includes: when the value of an element in the central wavelength vector exceeds a preset threshold range, the adjacent nodes are uniformly classified into the target connection relationship matrix using the strain direction initial vector corresponding to the element as the boundary.

6. The surface settlement control method based on multi-source soil properties according to claim 5, characterized in that: The target connection relationship matrix is used to determine the injection location of the secondary grouting material.

7. The surface settlement control method based on multi-source soil properties according to claim 1, characterized in that: The graph neural network takes the strain distribution of this iteration and the strain distribution of the grouting material as input, and the probability of selecting the grouting material as output. It determines the loss function through the target control value in the target connection relationship matrix of the next iteration and the probability of whether the secondary grouting material is selected.

8. The surface settlement control method based on multi-source soil properties according to claim 6, characterized in that: The amount of the secondary grouting material used is determined based on the probability of the secondary grouting material being selected.

9. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images