Multi-layer soft soil foundation settlement prediction method

Abstract

The invention provides a multi-layer soft soil foundation settlement prediction method, and relates to the field of settlement prediction, and the method comprises the steps: obtaining soft soil parameters through arranging a multi-source monitoring device, carrying out inversion to generate multi-dimensional feature vectors containing particle size distribution, void ratio, compression modulus gradient and effective stress, and dividing structure piece elements to construct a feature set; the Euclidean distance between the fragments is calculated and clustered to generate a structural similar region, and a judgment threshold is dynamically adjusted and combined based on the change rate of the structural information entropy to construct a dynamic structural map; newly added or cracked similar areas are identified, a structure sudden change strength index is generated, and when a threshold value is exceeded, a jump prediction model is constructed according to entropy change, pore pressure gradient and deformation rate to output discontinuous settlement increment; and generating an error field by using the difference value of the actually measured settlement and the predicted value, and correcting the structural information entropy and merging the threshold value after weighted averaging according to the similar region, thereby realizing the self-adaptive correction of the structural map and iterative optimization prediction.

Description

Technical Field

[0001] The present invention relates to the field of settlement prediction, in particular to a method for predicting settlement of a multi-layer soft soil foundation. Background Art

[0002] Soft soil foundations are widely used in infrastructure construction, and their settlement characteristics directly affect the stability and safety of structures. With the acceleration of urbanization and the increasing frequency of underground space development, the problem of soft soil foundation settlement prediction has received increasing attention in the fields of foundation treatment and engineering design. There is an urgent need to establish a refined and intelligent settlement evolution analysis model. Traditional methods are unable to fully reflect the nonlinear response characteristics of the complex spatiotemporal deformation process in soft soil multi-layer structures.

[0003] In the existing technology, a method combining field monitoring and numerical simulation is usually used to predict settlement trends. For example, displacement meters and pore pressure gauges are set up to obtain settlement and pore pressure data, and then a numerical model is established in combination with geological parameters for step-by-step calculation. This type of method relies on the accuracy of initial parameters, and the calculation process often adopts a fixed constitutive relationship, which lacks real-time performance and regional structure recognition capabilities. It has a weak response to uneven settlement and microstructural changes in the foundation, making it difficult to accurately capture sudden settlement behavior.

[0004] Although existing technologies can meet the prediction needs in some scenarios, they have obvious limitations when facing complex soil responses under the coupling of structural evolution and disturbances. The main problem is that the cognition of changes in the internal state of the region relies on static assumptions and lacks the ability to abstract dynamic structural evolution. The model is not sensitive and robust enough when dealing with abnormal responses, which can easily lead to prediction lags or misjudgments, making it difficult to meet the actual needs of engineering for high-efficiency and high-precision predictions. Summary of the Invention

[0005] (1) Technical problems solved In response to the shortcomings of the existing technology, the present invention provides a multi-layer soft soil foundation settlement prediction method to solve the obvious limitations raised in the above-mentioned background technology when facing the complex soil response under the coupling of structural evolution and disturbance; the cognition of the internal state changes of the region relies on static assumptions and lacks the ability to abstract dynamic structural evolution. The model is not sensitive and robust enough in dealing with abnormal responses, which can easily lead to prediction lag or misjudgment.

[0006] (2) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for predicting the settlement of a multi-layer soft soil foundation comprises the following steps: S1, by deploying pore water pressure sensors, shear wave velocity monitoring devices and microstructure image acquisition devices in the multi-layer soft soil area, pore water pressure, deformation, shear wave velocity and microstructure image data are obtained; based on the data inversion, a multidimensional element feature vector containing particle size distribution, porosity, compression modulus gradient and effective stress is generated, the soft soil foundation is divided into several structural elements, and a structural element feature set is constructed; S2, based on the structural fragment feature set, calculating the Euclidean distance of the feature vectors between the structural fragments, clustering the structural fragments according to the distance threshold to generate a set of structural similar regions, and calculating the structural information entropy of each structural similar region. Based on the rate of change of the structural information entropy, the similarity merging judgment threshold of the current time step is dynamically adjusted to generate a dynamically updated structural map; S3, based on the structural map in the continuous time step, identifying the newly added or cracked structural similarity area; calculating the rate of change of the structural information entropy of the structural similarity area, and generating a structural mutation strength index in combination with the depth weighting factor; when the structural mutation strength index exceeds the preset mutation judgment threshold, constructing a jump prediction model based on the change of the structural information entropy, the pore pressure gradient and the deformation rate of the structural element in the structural mutation area, and predicting the discontinuous settlement increment value in the current time step; S4, matching the actual monitored settlement value to the corresponding structural fragment according to the spatial position, calculating the difference with the predicted settlement value to generate an error field; and performing weighted averaging on the error field according to the structural similarity area, correcting the structural information entropy and similarity merging judgment threshold of the corresponding similarity area according to the average error, realizing adaptive correction of the structural map, and using the corrected structural map for the jump prediction iteration of the next time step.

[0007] Preferably, the original data required for constructing the structural element feature set is obtained by arranging sensors and acquisition devices on the multi-layer structure of the soft soil foundation. First, pore water pressure sensors are evenly arranged in the target soft soil area. The sensors acquire pore water pressure data at different depths in the soil at an hourly time step. A shear wave velocity monitoring device is arranged in the same spatial area. The device applies high-frequency shear waves and receives the propagation time to invert and obtain the shear wave velocity in each spatial voxel. The shear wave velocity parameters are used to derive the compression modulus of the current voxel. The microstructure image is collected by The device scans the soil sample and uses 3D reconstruction technology to obtain the local microstructure image of each soil body. Then, the image analysis algorithm is used to process the image and extract the information reflecting the soil particle structure characteristics, including indicators such as particle size distribution and porosity. The acquired data is stored in a database and an inversion operation is performed. The shear wave velocity is combined with the pore water pressure data to calculate the compression modulus, and the effective stress state is further derived by combining the soil sample density. The microstructure image analysis outputs the frequency function of the particle size distribution and the porosity as the structural parameter input. After obtaining the particle size distribution, the inversion operation is carried out. After the four key structural features of the structure are determined, such as the distribution, porosity, compression modulus gradient and effective stress, these data are spatially discretized according to three-dimensional cubic voxels. The specific division rules of the spatial regular voxels include but are not limited to: dividing the three-dimensional space into uniform cubic units with a fixed side length of 0.5 meters, and each unit is bound to a unique structural element; adjusting the voxel size according to the stratum characteristics, for example, for shallow soft soil 0-5 meters, a small voxel size of 0.5 meters is used to improve the resolution, and a large voxel size of 2 meters is used for deep layers >10 meters to reduce the amount of data; for undulating areas, such as slopes in reclamation projects, The voxel size is adjusted according to the elevation gradient. The grid width is encrypted to 0.5 meters in the high gradient area and relaxed to 1.5 meters in the low gradient area to ensure sufficient spatial resolution while taking into account data stability. Each voxel is regarded as a structural fragment unit, which is assigned a unique number. At the same time, the four eigenvalues ​​of particle size distribution, porosity, compression modulus gradient and effective stress are combined into a multidimensional fragment feature vector and bound to the number. In this way, a structural fragment feature set containing all structural fragment units and their corresponding eigenvectors is generated in the entire target area, providing a data basis for subsequent structural analysis and settlement prediction.

[0008] Preferably, the Euclidean distance between any two structural fragments is calculated in sequence based on the Euclidean distance between each fragment feature vector through the structural fragment feature set, and a complete distance matrix is ​​formed after the calculation is completed. The distance matrix is ​​used as the input of the similarity clustering operation; then the distance threshold is set to 2.5 in the initial stage, and the fragment feature vectors are clustered according to the threshold. A hierarchical clustering algorithm is used to ensure that the dual constraints of spatial continuity and feature similarity of adjacent feature fragments are established. After the clustering is completed, each group of aggregated structural fragments is identified as a structural similarity area; for each structural similarity area, the feature distribution of all the structural fragments it contains is counted and the information entropy evaluation is performed. The information entropy reflects the degree of discreteness of the structural fragment features within the current similarity area; the information entropy is used as an important factor in the subsequent structural change judgment. According to the information recorded and input into the dynamic merge judgment threshold adjustment step, a set of information entropy change rate sequences at the time step scale is established in this step to judge the dynamics of the structural evolution of adjacent time steps; the specific dynamic threshold update mechanism is as follows: in the current time step, if the structural information entropy change rate increases by 10% compared with the previous time step, the current merge judgment threshold is automatically reduced to enhance the clustering sensitivity, conversely, if the change rate decreases, the threshold is increased to reduce the system noise response interference; the initial merge judgment threshold is set to 2.5, and when the structural information entropy change rate value exceeds 0.1, the threshold is reduced by 0.5; then the structural map of the current time step is constructed, and it is compared with the structural map of the previous time step to update the affiliation of the structural similarity areas, realizing the dynamic evolution process of the structural map over time.

[0009] Preferably, the constituent units of the structural similarity zone of each time step and their corresponding characteristic vectors are extracted based on the structural atlas generated by two consecutive time steps, and the merging and cracking of similarity zones between the current time step and the previous time step are compared by tracking the numbers of the structural similarity zones to identify the newly added or cracked structural similarity zones in the current time step. The newly added similarity zones refer to a set of structural fragments that appear for the first time in the current atlas and have no corresponding numbers in the previous time step. The cracked similarity zones refer to a change state formed by the original structural similarity zones being split into multiple sub-regions. The information entropy of all structural fragments in the newly added or cracked regions is recalculated and the rate of change of the information entropy relative to the previous time step is counted. Different weighting factors are then assigned according to the depth position of the region. The weighting factor is 1 for depths less than 5 meters, 1.5 for depths between 5 and 10 meters, and 2 for regions above 10 meters. The rate of change of the structural information entropy is fused through the weighting mechanism to generate a structural mutation intensity index, which reflects the degree of abnormal change in the mutation region. If the structural mutation intensity index of any mutation region in the current time step exceeds the preset mutation judgment threshold (the default threshold is 1.25), it is considered to be a structural mutation event. On this basis, the jump prediction model construction stage begins. The model input includes three dimensions of data: one is the change in the structural information entropy of the element in the mutation region; the second is the gradient value of the spatial distribution of pore water pressure in the same region. The gradient is obtained by fitting the rate of change of pore pressure along the vertical direction; the third is the deformation rate of the structural element in adjacent time steps. The deformation rate is calculated by inverting the instantaneous settlement increment inferred from the compression modulus and effective stress changes. The three-dimensional data input variables are standardized and used as the input factors of the jump prediction model. The jump prediction response function is constructed to output the discontinuous settlement increment value of the mutation region in the current time step. This value is used to simulate the sudden settlement response caused by structural mutation. The jump prediction model results will participate in the error feedback link in the subsequent structural map update process, completing the logical closed loop from structural mutation to settlement response.

[0010] Preferably, the actual monitored settlement values ​​are compared with the predicted values ​​to form an error field, and the atlas is dynamically corrected through the mapping relationship of the structural similarity zone. First, the measured settlement values ​​obtained from the on-site settlement monitoring device are subjected to position registration processing, and each measured settlement value is accurately matched to a specific structural element number according to the number of the structural element and its three-dimensional coordinate position. The mapping process is realized through the spatial position index; then the predicted settlement increment value of each structural element in the current time step is read, and the difference between the two is obtained to obtain the settlement prediction error value of the element, thereby constructing the error field of the complete three-dimensional space; the error field is aggregated according to the division of the structural similarity zone, that is, the settlement error values ​​of all structural elements in the same similarity zone are weighted averaged, wherein the weighting coefficient is set according to the depth level of each element, the shallow element is weighted as 1, and the deep element is weighted as 0.5, so as to enhance the representation The dominance of layer response on the map evolution; the average settlement prediction error of each structural similarity area is obtained, and the error is marked as the prediction deviation of the area; the deviation is fed back to the structural information entropy calculation module and the similarity merging judgment threshold adjustment module of the corresponding similarity area, the former is used to correct the eigenvector distribution information of the structural fragments to recalculate the information entropy value, and the latter is used to adjust the threshold used for similarity judgment in the next time step. When the average prediction error value is greater than 0.3, the threshold is reduced by 0.5 to enhance the aggregation sensitivity. When the error is less than 0.1, the threshold is increased by 0.5 to improve the clustering stability. The whole process constitutes a structural map adaptive correction mechanism guided by prediction error, realizing the map's self-learning and feedback capabilities for external actual monitoring data, thereby improving the accuracy and stability of the structural map in settlement prediction in subsequent time steps, and completing the closed-loop coupling of monitoring data and map evolution.

[0011] Preferably, the layout method and inversion process of various devices include: arranging pore water pressure sensors, shear wave velocity monitoring devices and microstructure image acquisition devices in the target soft soil foundation area according to three-dimensional spatial rules, and the layout adopts the principle of spatial grid layout, the layout spacing is set to 1 meter, and a group of devices is set every 1 meter in the vertical direction to achieve full coverage monitoring of the entire multi-layer soft soil foundation area in the vertical and horizontal directions; the pore water pressure sensor is fixed at a specified depth by being buried in a pre-drilled hole, and a digital multi-channel sensor is used to collect data once an hour and transmit it to the data center in real time; the shear wave velocity monitoring device adopts a combination of source excitation and multi-point reception, by exciting seismic shear waves and recording the response time difference at receiving points at different depths, and using the ratio of propagation time to distance to derive the shear wave velocity; the shear wave velocity value is then input into the formation inversion module, and the compression modulus is calculated using the foundation physical model in combination with the pore water pressure value, and the compression modulus gradient is then obtained by vertical difference of the modulus; and The microstructure image acquisition device uses a micro-drilling camera probe and a digital image acquisition unit to capture images of the in-situ structure of the soil. After the image data is uploaded, it is processed by an image processing algorithm to perform particle identification, edge extraction and morphological segmentation, and finally obtain the particle size distribution frequency spectrum and void area distribution characteristics. The porosity index is obtained by converting the image pixel distribution probability and morphological size. The four types of structural feature data generated by inversion, namely particle size distribution, porosity, compression modulus gradient and effective stress, are respectively assigned a unified data format and imported into the three-dimensional feature field construction system. The three-dimensional space is divided into voxels, and each voxel is a cubic grid unit with a side length of 1 meter. The various eigenvalues ​​are mapped to the corresponding voxels according to the spatial coordinates, forming a structural fragment unit with a clear position and data binding relationship in space. Each structural fragment is then assigned a unique number and the four types of features are assembled into a set of feature vectors and bound to the number, and finally a structural fragment feature set is generated as the basic data set for subsequent clustering analysis and map construction.

[0012] Preferably, the specific operation process of constructing similar areas and dynamically merging judgment threshold control mechanism in the structural map is described in detail, and its core operation includes the following stages: first, in the initial time step, the initial clustering is performed on the feature set of the structural fragments, and the Euclidean distance is used as the measurement criterion, and similarity clustering is performed under the condition that the distance threshold is 3. The initial clustering method adopts a density-based spatial clustering algorithm to ensure that the result has strong local connectivity and spatial aggregation; after the clustering is completed, each clustering unit is defined as a structural similarity area, and then the entropy value of the feature vectors of all fragments within each structural similarity area is calculated. The information entropy is used to measure the degree of discreteness of the internal structural features of the similarity area. The higher the value, the greater the difference between the fragments; during the evolution of the map, the structural map is updated every time step, and a dynamic adjustment of the merge judgment threshold step is introduced in this update. This step is jointly constructed by the basic rhythm perturbation mechanism and the intermittent perturbation mechanism; the basic rhythm perturbation mechanism takes the baseline threshold as the starting point, and combines the preset periodic perturbation parameters and the information entropy change rate to establish an oscillation response. When the structural information entropy change rate is higher than 0.1, the judgment threshold is automatically lowered. The value is set to 0.5 to improve the sensitivity of the change response. When the change rate is lower than 0.1, the preset disturbance amplitude is increased by 0.3 to improve the analytical ability of the structural stability period. If the structural information entropy remains in a low change state for more than 3 time steps, a random disturbance term is introduced, and its amplitude is controlled between 0.1 and 0.3. The intermittent disturbance mechanism works according to the threshold condition of the information entropy change rate. When the change rate exceeds 0.2, the current baseline threshold is maintained unchanged to ensure the stability of the system judgment in the mutation stage. When it is lower than 0.2, a compensation factor is generated according to the deviation of the prediction result. The compensation factor is determined by the direction and amplitude of the prediction error at the previous moment. Positive deviation corresponds to a lower threshold, while negative deviation corresponds to an increase. The basic rhythmic perturbation mechanism and the intermittent perturbation mechanism work together in the clustering judgment of structural fragments in the current time step. That is, when the Euclidean distance between a structural fragment and any existing similar area is less than the adjusted threshold, it is assigned to the similar area. Otherwise, a new structural similar area is generated separately. Finally, the evolution and update of the structural map in the current time step are completed, and the control of the dynamic merging and cracking mechanism of the structural similar areas and the continuous dynamic construction of the map structure are realized.

[0013] Preferably, in each time step of the structural atlas update, the value of the current structural information entropy change rate is read and compared with the preset judgment threshold of 0.15. When the change rate is greater than the threshold, the current merge judgment threshold is kept unchanged to enhance the response stability to drastic changes; when the change rate is lower than the preset value, the random perturbation mechanism is called to perform parameter correction operation, and the perturbation value is sampled uniformly in the range of 0.1 to 0.3. At the same time, the error deviation direction between the settlement prediction result and the actual monitoring value in the structural mutation area in the previous time step is considered. The deviation value is defined as the offset direction. If the offset is positive, it means that the random perturbation mechanism is over-estimated, and the current merge judgment threshold is reduced by 0.2 to improve the clustering sensitivity; if the offset is negative, it means that the random perturbation mechanism is underestimated, and the 0.2 threshold is increased to enhance the structure retention ability; the offset direction and the perturbation amplitude are combined. The isomorphic offset compensation factor is embedded in the judgment module as part of the dynamic threshold function to achieve linkage control between prediction error and structural clustering feedback. This control method has convergence adjustment capabilities in the stage of frequent changes in prediction deviation and has an adaptability improvement mechanism in the stage of slow changes. In addition, a control cycle evaluation is performed every 5 time steps to analyze the information entropy change trend and the deviation direction fluctuation amplitude in the current time period. When the amplitude standard deviation exceeds 0.4, the threshold nonlinear correction path is automatically enabled. This path gradually adjusts the sensitivity parameters in the evolution of the map to adjust the response threshold of the current map to a maximum of no more than 1.5 times the initial threshold and a minimum of no less than 0.5 times. The introduction of this nonlinear mechanism effectively balances the contradiction between noise suppression and early response to structural mutations, improving the continuity and accuracy of map construction.

[0014] Preferably, the specific operation process of the structural mutation intensity index construction and jump prediction model output mechanism is: first obtain the structural maps constructed in the current time step and the previous time step respectively, and perform number matching and structural fragment attribution tracking on each structural similar area, and analyze whether there is a similar area merger, splitting or new addition phenomenon in the map through the map comparison module. If a similar area is split into two or more similar areas in the current time step, or a new structural fragment clustering unit without a historical number appears, then the area is identified as a mutation candidate area; then enter the mutation evaluation stage, recalculate the structural information entropy of each candidate area and obtain the entropy value difference with the previous time step, and further calculate its change rate, which is obtained by time difference normalization; then assign a weighting factor according to the depth of the regional fragment to distinguish the degree of mutation impact of different layers, specifically setting the weighting factor to 1 within five meters, 1.5 for 5 to 10 meters, and 0.5 for more than 10 meters. The weighting factor is 2. The rate of change of information entropy is multiplied by the depth weighting factor and normalized to form a structural mutation intensity index, which quantifies the severity of structural evolution in the mutation area. If the index value exceeds the preset mutation judgment threshold of 1, the jump prediction model is activated to estimate discontinuous settlement. The model requires three input parameters: the first is the change in structural information entropy; the second is the vertical gradient of pore water pressure in the mutation area, which is obtained by least squares fitting the pore pressure values ​​of consecutive monitoring points; and the third is the difference in deformation rate of the structural element in two consecutive time steps, which is derived from the strain series calculated from the compression modulus and effective stress changes. The three types of data are standardized and input into the jump prediction function for weighted combination operation. The output is the settlement jump increment value of the mutation area in the current time step. This predicted value is recorded as a quantitative expression of the structural sudden response and sent to the settlement prediction output interface, which is used to guide the adjustment of engineering monitoring and control strategies.

[0015] Preferably, the dynamic correction mechanism for the output process of the jump prediction model is as follows: after the jump prediction model completes the output of the settlement increment, the system compares the predicted value with the actual monitored settlement value of the corresponding structural element in the current time step, calculates the deviation value between the two, and forms an error evolution sequence with the deviation value in the order of time steps through the error feedback step; if the sequence shows alternating error signs in multiple consecutive time steps, that is, there is a trend of repeated changes in the positive and negative directions, it is determined that the current prediction has a high-frequency oscillation error, and the jump amplitude coefficient is automatically reduced by 0.5 to suppress unstable fluctuations; conversely, if the error maintains a unidirectional deviation trend in three consecutive time steps, it is considered that the current prediction result has a systematic deviation, and the structural drive weight is increased by 0.3 to enhance the response sensitivity to structural mutations and make the model more sensitive to changes in input parameters; The error feedback parameter is not only used to dynamically adjust the output of the jump prediction model, but also links the feedback weight and merge judgment threshold settings involved in the structural map construction process. When the offset direction is positive and the amplitude is greater than 0.3, the structural information entropy of the corresponding similar area is multiplied by the adjustment factor 0.85 in the next time step, compressing its fragment differences to promote the merging operation; when the offset direction is negative and the amplitude is greater than 0.3, the information entropy is multiplied by 1.15 to improve the ability to identify structural differences; the merge judgment threshold of the corresponding similar area is also adjusted synchronously, decreasing by 0.5 for positive deviations and increasing by 0.5 for negative deviations, forming a feedback linkage control mechanism with prediction error as the core; this mechanism constructs a closed-loop optimization path between the jump prediction model and the structural map construction, while ensuring the convergence of the prediction output, enhancing the system's stable response capability and anti-interference robustness to sudden changes.

[0016] Preferably, after completing the settlement prediction at each time step, the measured settlement values ​​of each monitoring point in the target area are automatically collected, and their positions are spatially grid-aligned with the three-dimensional structural element numbers. The alignment method is based on the principle of minimizing the Euclidean distance between the spatial coordinates of the structural element and the monitoring point position, and each measured point is mapped to the nearest element number; after completing the alignment, the difference between the predicted settlement value and the measured value is calculated to form an element-level error matrix, which is the three-dimensional error field; based on the mapping relationship between the similar area and the structural element in the structural atlas, the error value of each element in the error field is aggregated to the structural similar area to which it belongs, and the average prediction error of the similar area is obtained by weighted averaging the error values ​​of all elements in the similar area; if the average error exceeds the preset threshold of 0.5mm in two consecutive time steps, the structural similar area is automatically marked as a high-risk area; the system It then enters the high-risk response mode, which includes three key mechanisms: first, the structural similarity merging judgment threshold in the next time step of the area is automatically lowered by 0.5 to enhance the splitting sensitivity of the map and improve the early identification ability of potential mutations; second, the weight factor of the change rate of structural information entropy in the jump prediction model is increased to 1.5 times the original value when calculating the mutation intensity index, so that the settlement prediction of the area can more sensitively reflect structural changes; third, the nonlinear adjustment factor in the jump prediction model of the high-risk area is updated, and the correlation coefficient in its adjustment function is adjusted to the numerical configuration in the enhanced mode to enhance the amplitude and recognition ability of the structural mutation response; the numerical configuration in the enhanced mode establishes a dynamic shared channel between the map construction module, mutation recognition module and settlement prediction module to ensure the high-risk response logic that the map evolution and settlement prediction are consistent.

[0017] (3) Beneficial effects The present invention provides a method for predicting the settlement of a multi-layer soft soil foundation. It has the following beneficial effects: 1. The present invention constructs a dynamic structural map and adjusts the adaptive threshold; generates similar areas based on clustering of structural element features; and dynamically adjusts the merging judgment threshold according to the rate of change of structural information entropy; combines basic rhythm and intermittent disturbance mechanism; realizes intelligent adjustment of map sensitivity; enhances random disturbance to improve differential response when information entropy is stable; reduces disturbance amplitude to suppress noise when changes are drastic; enables the system to accurately capture potential structural mutations both in the stable period and the drastic change period of the soil; optimizes the triggering timing of the jump prediction model; and enhances the reliability of discontinuous settlement warning.

[0018] 2. The present invention constructs an error-driven closed-loop collaborative optimization mechanism; generates a spatial error field based on the difference between actual settlement and predicted values; feeds back weighted average of structurally similar areas to corresponding areas; dynamically corrects structural information entropy and merging judgment thresholds; marks regional areas as high-risk when errors continue to exceed the standard; automatically relaxes the merging threshold and improves the sensitivity of the map; synchronously enhances the weight of the structural entropy change rate in the jump model; dynamically adjusts the output amplitude and driving weight of the prediction model through historical error accumulation; forms a "monitoring-prediction-correction" closed loop; and significantly improves the mutation resolution of high-risk areas and the overall adaptability of the system. DETAILED DESCRIPTION

[0019] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0020] Example 1: An embodiment of the present invention provides a method for predicting the settlement of a multi-layer soft soil foundation. The method includes predicting the long-term settlement trend of a multi-layer soft soil foundation formed by land reclamation during the operation period of a foundation reinforcement project in a coastal city port area. First, pore water pressure sensors, shear wave velocity monitoring devices, and microstructure image acquisition devices are arranged in a regular grid manner within an area with a total project area of ​​approximately ten hectares. The sensors and monitoring devices are arranged at intervals of 1 meter in the horizontal and vertical directions, and the sensors are buried at multiple levels ranging from 1 meter to 15 meters based on stratum survey results. The shear wave velocity monitoring device is centered on a source seismic device and is provided with multiple depth receivers for collecting propagation time to obtain shear wave velocity data. The microstructure image acquisition device obtains soil structure images through a borehole camera system and transmits them to a data processing terminal. The terminal automatically extracts particle size distribution information and porosity ratio parameters through an image recognition module. After the acquisition is completed, the pore pressure data and shear wave velocity data are input into the compression modulus inversion model. Combined with the initial effective stress conditions of the formation, the compression modulus and effective stress state of voxels at different depths are calculated. The particle size distribution frequency and porosity ratio obtained from the image analysis results are combined to form a feature vector. The feature vector is bound to each spatial voxel number, which is a cubic unit with a side length of 1 meter, to construct a complete set of structural element features. The system then enters the structural map construction stage, calculates the Euclidean distance for all structural element feature vectors, sets the distance threshold to 3 in the initial time step, performs cluster analysis on the elements and generates several structural similarity areas. At the same time, the information entropy of each structural similarity area is calculated to represent its internal structural differences, and constructs an initial structural map. The system then collects data once every day and updates the structural element features, executing the structural map update process, which introduces a dynamic merging judgment threshold function based on the rate of change of structural information entropy. When the rate of change of information entropy is greater than 0.1, the system generates a clustered data. The threshold is automatically lowered by 0.5 to improve the system's ability to identify structural mutations. At a certain moment, through two consecutive structural map comparisons, a structurally similar area at a depth of about 8 meters was identified to have cracked, with an information entropy change rate of up to 0.25. The system calculated the mutation intensity index value to be 1.7 based on the weighting factor, exceeding the threshold of 1.25, triggering the jump prediction model construction process. After inputting the three variables of information entropy change, pore pressure gradient, and deformation rate, the system outputs a discontinuous settlement increment value of approximately 5mm in the current time step in the area and records it in the database as a predicted value. In the next time step, the measured settlement value collected in the area is 4.4mm, with a calculated error of 0.6mm, exceeding the error field threshold of 0.5mm. The structurally similar area is marked as a high-risk area, and the response mechanism is immediately executed, including lowering the similarity merging threshold of 0.5 for the next time step and increasing the weight of the information entropy change rate in the jump prediction model to 1.5 times. At the same time, the nonlinear parameter coefficient within the model is adjusted to enhance the prediction sensitivity. During subsequent operations, the area continued to be monitored, and its structural map evolution process also responded more frequently to external disturbances. Ultimately, the system achieved high-precision predictions of the area's long-term settlement trends through continuous feedback and correction. The entire implementation process, from the deployment of acquisition equipment to structural map construction, mutation identification, jump prediction, error feedback, and risk response, formed a closed loop, realizing field application and full process implementation in multi-layer soft soil engineering.

[0021] Example 2: The difference between this embodiment and the first embodiment is that this embodiment is applied to a soft foundation site with complex terrain, which has obvious undulations and stratum heterogeneity. During the implementation process, the spatial mapping accuracy and local prediction sensitivity of each structural element feature construction and map evolution operation are enhanced. The specific implementation is as follows: during the target area layout process, a zoning layout strategy is adopted according to the site terrain. The area is divided into three categories: elevation area, depression area and central equilibrium area. Pore water pressure sensors and shear wave velocity monitoring devices are deployed at layout intervals of 0.5 meters, 1 meter and 1.5 meters respectively. Image acquisition devices are also synchronously deployed in representative boreholes. The voxel side lengths of each type of block are set to 0.5 meters, 1 meter and 1.5 meters respectively to meet the spatial resolution requirements of different structural differences. After the data collection is completed, the system establishes a four-dimensional feature vector for each element, and assigns different numbers to them according to regional attributes, thereby realizing multi-scale organization of the structural element set. Then, the map construction stage is entered. Different initial clustering thresholds are set for the elevation area and the depression area. A lower threshold of 2 is used in the elevation area to enhance the identification of local structural changes, while a high threshold of 3.5 is used in the depression area to improve overall stability. After clustering is completed, the structural information entropy of similar areas is calculated and incorporated into the information entropy change monitoring system. During the dynamic evolution of the structural map, in addition to using the standard information entropy change rate to control the merging judgment threshold, this embodiment further introduces a compensation mechanism based on the direction of the prediction deviation of the previous time step. In the elevation area, if the model underestimates the settlement value in two consecutive time steps, the compensation factor is automatically added to the nonlinear adjustment function, making the map more sensitive to structural mutations at the elevation difference position. In a certain prediction, it was found that the structural information entropy change rate increased to 0.3 in the central equilibrium area, triggering the calculation of the mutation intensity index. Because the weighting factor was set to 1.5, the final mutation index reached 1.8. The system then constructed a jump prediction model and output a settlement jump value of 4.8 mm. The monitored value was actually measured to be 5.1 mm, with an error of only 0.3 mm, which did not exceed the threshold, and the map maintained its original stable state. In another area of ​​the elevation zone, the error between the measured and predicted values ​​reached 0.7 mm. The system immediately marked the area as a high-risk area and lowered the structural information entropy to 0.9 of the original value in the positive direction of the error. At the same time, the combined judgment threshold was lowered by 0.5 to improve the sensitivity response of the structural evolution. The information entropy weight in the model was also increased by 1.6 times to strengthen the mutation response. In the subsequent two time steps, the system continued to track its structural mutation and settlement development process. Finally, it was verified in the actual engineering monitoring data that the abnormal settlement trend in the area was accurately identified and predicted in advance by the system. This embodiment effectively improved the prediction accuracy and local mutation identification ability of the system under complex geological conditions by strengthening the spatial resolution control, regional threshold difference configuration and nonlinear compensation path integration on the basis of Example 1, and realized a highly adaptable implementation plan under more complex site conditions, forming a sensitive dynamic settlement prediction process integrating structural mapping, mutation identification and error feedback.

[0022] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting settlement of multi-layer soft soil foundation, characterized in that: The steps include: S1, by deploying pore water pressure sensors, shear wave velocity monitoring devices and microstructure image acquisition devices in the multi-layer soft soil area, pore water pressure, deformation, shear wave velocity and microstructure image data are obtained; based on the data inversion, a multidimensional element feature vector containing particle size distribution, porosity, compression modulus gradient and effective stress is generated, the soft soil foundation is divided into several structural elements, and a structural element feature set is constructed; S2, based on the structural fragment feature set, calculating the Euclidean distance of the feature vectors between the structural fragments, clustering the structural fragments according to the distance threshold to generate a set of structural similar regions, and calculating the structural information entropy of each structural similar region. Based on the rate of change of the structural information entropy, the similarity merging judgment threshold of the current time step is dynamically adjusted to generate a dynamically updated structural map; S3, identifying newly added or cleaved structural similar regions based on the structural profiles in consecutive time steps; The rate of change of the structural information entropy in the structural similarity area is counted, and a structural mutation strength index is generated in combination with a depth weighting factor. When the structural mutation strength index exceeds a preset mutation judgment threshold, a jump prediction model is constructed based on the change of the structural information entropy, the pore pressure gradient and the deformation rate of the structural element in the structural mutation area to predict the discontinuous settlement increment value in the current time step; S4, matching the actual monitored settlement value to the corresponding structural fragment according to the spatial position, calculating the difference with the predicted settlement value to generate an error field; and performing weighted averaging on the error field according to the structural similarity area, correcting the structural information entropy and similarity merging judgment threshold of the corresponding similarity area according to the average error, realizing adaptive correction of the structural map, and using the corrected structural map for the jump prediction iteration of the next time step.

2. A method for predicting settlement of a multi-layer soft soil foundation according to claim 1, characterized in that: The S1 includes: placing the pore water pressure sensor, shear wave velocity monitoring device and microstructure image acquisition device in the target area to obtain pore water pressure, shear wave velocity and microstructure image data, the shear wave velocity is used to invert the compression modulus and determine the effective stress state in combination with the pore water pressure, and the structural image obtains the fragment particle size distribution and porosity ratio parameters through image analysis; after converting the inverted particle size distribution, porosity, compression modulus gradient and effective stress into a three-dimensional feature field, a plurality of structural fragment units are generated by spatial regular voxel division, and a unique number and a corresponding feature vector are bound to each of the structural fragments to construct a structural fragment feature set; the spatial regular voxels refer to voxels arranged and distributed according to certain rules in three-dimensional space, which is a discretized representation of three-dimensional space, specifically, the three-dimensional space is divided into multiple structural fragments according to cubic grid units of fixed size.

3. The method for predicting settlement of a multi-layer soft soil foundation according to claim 2, characterized in that: Said S2 comprises: applying an initial similarity clustering method to said structural fragment feature set to construct a set of structural similar regions, and calculating structural information entropy for each of said similar regions to measure the internal fragment structure difference; During the structural map update process, the dynamic adjustment of the merging judgment threshold includes: constructing an adaptive judgment threshold function jointly generated by a basic rhythmic perturbation mechanism and an intermittent perturbation mechanism; the basic rhythmic perturbation mechanism introduces an oscillatory perturbation based on a preset period parameter and the structural information entropy change rate on the basis of a baseline merging threshold: when the structural information entropy change rate increases, the perturbation amplitude is reduced to reduce sensitivity, and when the change rate decreases, the perturbation amplitude of the preset amplitude is increased; when the structural information entropy is stable, the random perturbation amplitude within a preset variance range is increased to improve the structural difference response sensitivity; the intermittent perturbation mechanism maintains the baseline threshold when the structural information entropy change rate exceeds a preset high threshold, and when it is lower than the preset high threshold, the threshold is nonlinearly adjusted according to the random perturbation amplitude and a compensation factor generated by the prediction deviation at the previous moment. The adaptive judgment threshold function serves as the basis for determining whether the structural fragments are merged or split. Each fragment is judged whether to be included in the original similarity zone or the newly created similarity zone based on the characteristic distance with the current similarity zone and the dynamic threshold, and a new structural similarity zone is generated, thereby completing the update of the similarity zone map at the current time step.

4. A method for predicting settlement of a multi-layer soft soil foundation according to claim 3, characterized in that: The S2 also includes: in the adaptive judgment threshold function, in each time step, when the rate of change of the structural information entropy is higher than a preset threshold, the merge judgment threshold is maintained at the initial baseline value; when the rate of change of the structural information entropy is lower than the preset threshold, the merge judgment threshold amplitude is adjusted according to the random perturbation mechanism; based on the jump prediction result of the previous time step, an offset compensation factor is generated, and when the prediction result is a positive deviation, the threshold perturbation amplitude is reduced, and when it is a negative deviation, the threshold perturbation amplitude is increased; the offset compensation factor is linked with the adaptive judgment threshold function and the random perturbation mechanism to realize nonlinear control of the threshold, and the sensitivity of the graph evolution is gradually adjusted within the control period, so that the system can maintain the ability to respond to potential structural mutations at different stability stages.

5. The method for predicting settlement of a multi-layer soft soil foundation according to claim 4, characterized in that: The S3 includes: obtaining a structural map of two consecutive time steps, calculating and comparing the structural maps of two consecutive time steps in adjacent time steps, identifying newly added or cracked structural similar areas or cracked structural fragments; performing statistical analysis of the change rate of the structural information entropy in the mutation area, and constructing a structural mutation intensity index by assigning weighting factors to different depths or stratum levels; when the structural mutation intensity index exceeds a preset mutation judgment threshold, constructing a jump prediction model based on the change in the structural information entropy, the pore pressure field gradient and the deformation rate of the structural fragment, and the jump prediction model outputs a discontinuous settlement increment value in the current time step to simulate the sudden settlement response of multi-layer soft soil.

6. The method for predicting settlement of a multi-layer soft soil foundation according to claim 5, characterized in that: The S3 also includes: setting an output dynamic correction mechanism for the aforementioned jump prediction model, introducing an error feedback parameter based on the historical error accumulation result in the model output process, reducing the jump amplitude when the historical error oscillates, and enhancing the structural driving weight to improve the prediction sensitivity when the error continuously deviates in one direction. The error feedback parameter is also used to dynamically adjust the feedback weight and the merge judgment threshold benchmark in the atlas construction to achieve closed-loop collaborative optimization of the prediction model and the structural atlas, thereby achieving stability and robustness control of the jump prediction model in the structural mutation area.

7. The method for predicting settlement of a multi-layer soft soil foundation according to claim 5, characterized in that: The S4 includes: aligning the actual monitored settlement value to the structural fragment according to the spatial grid, and on this basis constructing the difference between the actual value and the predicted settlement value as an error field; performing a weighted average operation on the error field in units of similar areas through the fragment space mapping relationship with the similar area to obtain the average prediction error of each similar area, and feeding back the error to the structural information entropy corresponding to the similar area and the aggregation judgment threshold of the current time step of the corresponding similar area, so that the basic data for constructing the map in subsequent time steps can be adaptively updated and corrected according to the prediction accuracy.

8. The method for predicting settlement of a multi-layer soft soil foundation according to claim 7, characterized in that: The S4 also includes: when the average error of any structural similarity area continuously exceeds the preset error threshold, the area is marked as a high-risk area, and the similarity merging judgment threshold of the next time step is relaxed, thereby increasing the weight of the rate of change of structural information entropy in the jump prediction model; when executing the atlas construction in the next time step, the merging judgment threshold is preferentially lowered to enhance the sensitivity of the atlas of the area; at the same time, by modifying the nonlinear adjustment factor of the jump prediction model of the area, the weight of the rate of change of structural information entropy in the prediction model is increased, thereby further improving the prediction resolution of the sedimentation mutation in the high-risk area, and ultimately improving the response accuracy and adaptability of the overall prediction system.