Foundation pit monitoring analysis and geotechnical parameter inversion method and system

Abstract

This invention relates to the field of foundation pit construction safety monitoring technology, and provides a method and system for foundation pit monitoring analysis and geotechnical parameter inversion. The method includes: building a finite element model of the foundation pit to generate a batch of simulation monitoring curve datasets and binding labels; preprocessing the simulation monitoring curve data and field data respectively, and performing logarithmic transformation on the permeability coefficient labels of soft clay and confined aquifers; constructing and training a two-level integrated model composed of a classification neural network and multiple sets of regression subnetworks; inputting field data into the trained model for two-level inference parameter inversion; performing an inverse exponential transformation on the permeability coefficient to restore the true value and performing risk classification and early warning. This invention can replace measured samples with simulation data, improve inversion accuracy through a two-level integrated architecture, and realize intelligent monitoring and parameter inversion of complex strata.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of foundation pit construction safety monitoring, and particularly relates to a foundation pit monitoring analysis and geotechnical parameter inversion method and system. BACKGROUND

[0002] Foundation pit engineering is a key link of underground engineering and construction engineering. In order to ensure construction safety, the current "Technical Standard for Monitoring of Building Foundation Pit Engineering" requires continuous monitoring of indexes such as foundation pit horizontal displacement, ground settlement, supporting structure axial force, underground water level, and soil pressure all day long. At the same time, the design of foundation pit support and the overall stability calculation highly depend on core geotechnical mechanics parameters such as soil cohesion, internal friction angle, elastic modulus, and permeability coefficient. The mainstream technology in the industry at the present stage is divided into three categories: Manual monitoring data interpretation: field sensors collect displacement, stress, water level and other data to generate one-dimensional continuous monitoring curves, which are analyzed, identified and indexed by technical personnel; Traditional geotechnical parameter acquisition method: relying on indoor soil test and in-situ test to obtain geotechnical parameters, or using genetic algorithm, particle swarm algorithm combined with finite element iteration to complete parameter inversion; Conventional machine learning monitoring scheme: using a single neural network to simply identify or predict foundation pit monitoring data to assist in completing monitoring work.

[0003] At present, the above-mentioned traditional technologies have the following specific defects: Manual processing mode is low in efficiency, strong in subjectivity and high in risk of missing detection. The environment of foundation pit construction site is complex, sensors are easily affected by vibration, dust and electromagnetic interference, and the original monitoring data contains a large amount of noise, jump points and outliers. Manual interpretation of a large number of monitoring curves is time-consuming and laborious, and manual judgment has subjective bias. Small deformation hidden dangers are easily missed, and real-time risk warning cannot be realized. Conventional simple filtering algorithms cannot simultaneously consider noise reduction effect and curve original feature preservation.

[0004] Geotechnical parameter acquisition and traditional inversion technology have outstanding shortcomings. Indoor soil test is easy to cause soil disturbance in the process of sampling and sample preparation, and the test parameters deviate greatly from the actual working conditions on site. The number of in-situ test points is limited, and the detection cost is high, which cannot reflect the overall stratum characteristics of the site. The inversion method of traditional intelligent algorithm combined with finite element iteration has a huge amount of calculation, and the time consumption of single inversion is several hours. Moreover, it is easy to fall into local optimal solution, and it cannot fully meet the real-time application requirements of engineering.

[0005] Existing machine learning solutions suffer from poor adaptability and insufficient generalization ability. Current machine learning models in the field of geotechnical engineering generally rely heavily on massive amounts of field-measured samples. However, the collection of effective field-measured data for foundation pits is difficult and samples are scarce. For special strata such as soft clay, silty soil, and confined aquifers, the magnitude of geotechnical parameters varies greatly, and conventional neural networks suffer from prediction distortion due to data scale imbalance. At the same time, existing solutions mostly use a single network structure, which cannot complete the classification of working conditions and parameter regression step by step, resulting in limited accuracy in identification and inversion, and making it difficult to adapt to complex foundation pit working conditions.

[0006] In summary, traditional foundation pit monitoring and parameter inversion technologies suffer from a series of problems, such as a shortage of measured samples, weak anti-interference ability, poor adaptability to special strata, low computational efficiency, and insufficient automation. There is an urgent need for an intelligent integrated analysis solution that can be independent of massive measured samples, adapt to complex construction site environments, and balance computational accuracy and speed. Summary of the Invention

[0007] The purpose of this invention is to solve at least one technical problem in the background art and to provide a method and system for foundation pit monitoring analysis and geotechnical parameter inversion.

[0008] To achieve the above objectives, this invention provides a method for foundation pit monitoring analysis and geotechnical parameter inversion, comprising: S1. Build a finite element model of the foundation pit, generate a batch of simulation monitoring curve datasets, and bind the stratum type label, disease type label and corresponding real soil and rock parameters to each simulation monitoring curve; S2. Preprocess the simulated monitoring curve dataset and the field measured monitoring data respectively; the preprocessing includes: performing a base-10 logarithmic transformation on the permeability coefficient labels of soft clay and confined aquifer in the simulated monitoring curve dataset to unify the data scale; S3. Construct an integrated neural network model consisting of a classification neural network and a regression neural network, wherein the classification neural network is used to identify the stratum type and the working condition of the foundation pit, and the regression neural network includes multiple regression subnetworks, each corresponding to a different stratum type. Based on the identification result of the classification neural network, the corresponding regression subnetwork is called to perform parameter inversion. S4. Train the ensemble neural network model using the preprocessed simulation monitoring curve dataset; S5. Input the pre-processed field measured monitoring data into the trained integrated neural network model. First, the classification neural network identifies the stratum type and disease conditions. Then, according to the identified stratum type, the corresponding regression sub-network is called to invert the geotechnical mechanics parameters. S6. Perform an inverse exponential transformation to the permeability coefficient after the logarithmic transformation to restore the true value of the parameter, and conduct a risk classification and early warning of the foundation pit based on the comparison results of the on-site measured monitoring data and the preset early warning threshold.

[0009] According to one aspect of the present invention, the preprocessing further includes: using the Savitzky-Golay filtering algorithm to filter and reduce noise in the simulated monitoring curve dataset and the field measured monitoring data, using a cubic interpolation algorithm to interpolate and normalize the simulated monitoring curve dataset and the field measured monitoring data to unify the input dimension, and performing outlier truncation.

[0010] According to one aspect of the present invention, the preprocessing further includes: adding Gaussian random noise to the simulation curves in the simulation monitoring curve dataset to achieve data augmentation.

[0011] According to one aspect of the present invention, the classification neural network employs a Softmax activation function and a classification cross-entropy loss function to achieve multi-class classification; the regression sub-network performs MinMax normalization on the output soil and rock parameters during training and saves the normalization scaler for subsequent inverse normalization operations.

[0012] According to one aspect of the present invention, the simulation monitoring curve dataset includes: inclinometer curves, surface subsidence curves, support axial force curves, and groundwater level curves; the number of regression subnetworks is consistent with the number of strata types that the classification neural network can identify.

[0013] According to one aspect of the present invention, in step S4, the preprocessed simulation monitoring curve dataset is divided into a training set and a test set. The integrated neural network model is trained using the training set, and the model accuracy is verified using the test set. When the classification accuracy and parameter inversion error meet the preset convergence conditions, the classification neural network model file, each regression sub-network model file, and the corresponding normalization scaler file are saved.

[0014] According to one aspect of the present invention, step S6 further includes: comparing and displaying the theoretical monitoring curve reconstructed from the geotechnical parameters obtained by the regression subnetwork inversion with the field measured monitoring curve on the same axis, and generating a geotechnical parameter report and a foundation pit monitoring analysis report.

[0015] To achieve the above objectives, the present invention also provides a foundation pit monitoring and analysis and geotechnical parameter inversion system, comprising: The simulation monitoring curve dataset generation module builds a finite element model of the foundation pit, generates a batch of simulation monitoring curve datasets, and binds stratum type labels, disease type labels and corresponding real soil and rock parameters to each simulation monitoring curve. The data preprocessing module preprocesses the simulated monitoring curve dataset and the field measured monitoring data respectively. The preprocessing includes performing a base-10 logarithmic transformation on the permeability coefficient labels of soft clay and confined aquifer in the simulated monitoring curve dataset to unify the data scale. An integrated neural network model construction module is used to construct an integrated neural network model consisting of a classification neural network and a regression neural network. The classification neural network is used to identify the stratum type and foundation pit disease conditions. The regression neural network includes multiple regression subnetworks, each corresponding to a different stratum type. Based on the identification results of the classification neural network, the corresponding regression subnetwork is called to perform parameter inversion. The model training module trains the integrated neural network model using the preprocessed simulation monitoring curve dataset; The analytical inversion module inputs the pre-processed field measured monitoring data into the trained integrated neural network model. First, the classification neural network identifies the stratum type and the fault condition. Then, based on the identified stratum type, the corresponding regression subnetwork is called to invert the geotechnical mechanics parameters. The foundation pit risk classification and early warning module performs an inverse exponential transformation to the permeability coefficient after the logarithmic transformation, restoring the true value of the parameter, and performs foundation pit risk classification and early warning based on the comparison results of on-site measured monitoring data and preset early warning thresholds.

[0016] To achieve the above objectives, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the foundation pit monitoring analysis and geotechnical parameter inversion method as described above.

[0017] To achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the foundation pit monitoring analysis and geotechnical parameter inversion method as described above.

[0018] According to the present invention, the present invention uses finite element software to generate labeled simulation monitoring curve datasets in batches, replacing scarce field measured samples for model training. This effectively solves the problem of insufficient training data for machine learning models in the field of foundation pit engineering, reduces the dependence on massive field monitoring data, and avoids the drawbacks of indoor geotechnical test sampling disturbance and in-situ testing, such as high cost and limited measurement points.

[0019] The two-level integrated architecture constructed in this invention, consisting of a classification neural network and multiple sets of regression sub-networks, realizes a step-by-step operation mode that first identifies the stratum type and disease conditions, and then performs targeted parameter inversion. Compared with a single neural network structure, this architecture reduces the complexity of the nonlinear mapping between soil and rock parameters and monitoring data, and improves the accuracy of identification and inversion under complex stratum conditions.

[0020] This invention addresses the issue of permeability coefficients in special strata such as soft clay and confined aquifers spanning orders of magnitude. It employs a base-10 logarithmic transformation to unify the data scale and performs an inverse exponential transformation in the output stage to restore the true physical values. This avoids prediction distortion caused by data scale imbalance and enables the model to be stably applied to complex engineering scenarios such as urban soft soil foundation pits and water-rich foundation pits.

[0021] The standardized preprocessing workflow of this invention integrates Savitzky-Golay filtering for noise reduction, cubic interpolation normalization, outlier truncation, and Gaussian noise enhancement, enabling the model to adapt to complex noise environments such as construction site vibration, dust, and electromagnetic interference, thereby improving the model's robustness and generalization ability.

[0022] After the model training is completed, the inference of a single monitoring curve can complete parameter inversion and risk assessment within milliseconds, which is far superior to the inversion method of traditional intelligent algorithm combined with finite element iteration. It can meet the timeliness requirements of 24-hour uninterrupted real-time monitoring and dynamic early warning for foundation pit engineering.

[0023] This invention automates the entire process from field data access, curve analysis, parameter calculation to report output, reducing the workload of manually interpreting each monitoring curve, lowering the subjective bias and risk of missed detection in manual judgment, and improving the efficiency and standardization of monitoring work.

[0024] This invention is compatible with various types of foundation pit monitoring data, such as inclination measurement, settlement, axial force, and water level. It can be adapted to different support forms and geological conditions. The technical framework can be further extended to other geotechnical engineering fields such as tunnels, slopes, and roadbeds, and has good engineering application value. Attached Figure Description

[0025] Figure 1 A flowchart illustrating a method for analyzing and inverting geotechnical parameters during foundation pit monitoring, according to an embodiment of the present invention, is shown in the schematic diagram. Figure 2 The diagram illustrates a comparison between the measured values ​​and the reconstructed values ​​of the inclinometer curve of the foundation pit according to Embodiment 1 of the present invention. Detailed Implementation

[0026] The invention will now be discussed with reference to exemplary embodiments. It should be understood that the described embodiments are merely intended to enable those skilled in the art to better understand and thus implement the invention, and are not intended to imply any limitation on the scope of the invention.

[0027] As used herein, the term "comprising" and its variations are to be interpreted as open-ended terms meaning "including but not limited to". The term "based on" is to be interpreted as "at least partially based on". The terms "one embodiment" and "an embodiment" are to be interpreted as "at least one embodiment".

[0028] Figure 1 This diagram illustrates a flowchart of a foundation pit monitoring analysis and geotechnical parameter inversion method according to an embodiment of the present invention. Figure 1 As shown in this embodiment, the method for analyzing and inverting geotechnical parameters during foundation pit monitoring includes: S1. Build a finite element model of the foundation pit, generate a batch of simulation monitoring curve datasets, and bind the stratum type label, disease type label and corresponding real soil and rock parameters to each simulation monitoring curve; S2. Preprocess the simulation monitoring curve dataset and the field measured monitoring data respectively; the preprocessing includes: performing a base-10 logarithmic transformation on the permeability coefficient labels of soft clay and confined aquifer in the simulation monitoring curve dataset to unify the data scale; S3. Construct an integrated neural network model consisting of a classification neural network and a regression neural network. The classification neural network is used to identify the stratum type and the working conditions of the foundation pit. The regression neural network includes multiple regression subnetworks, each corresponding to a different stratum type. The corresponding regression subnetwork is called to perform parameter inversion based on the identification results of the classification neural network. S4. Train the ensemble neural network model using the preprocessed simulation monitoring curve dataset; S5. Input the pre-processed field measured monitoring data into the trained integrated neural network model. First, the classification neural network identifies the stratum type and disease conditions. Then, based on the identified stratum type, the corresponding regression sub-network is called to invert the geotechnical mechanics parameters. S6. Perform an inverse exponential transformation to the base-10 of the permeability coefficient after logarithmic transformation to restore the true value of the parameter, and conduct a risk classification and early warning of the foundation pit based on the comparison results of the on-site measured monitoring data and the preset early warning threshold.

[0029] Furthermore, according to one embodiment of the present invention, the preprocessing further includes: using the Savitzky-Golay filtering algorithm to filter and reduce noise in the simulation monitoring curve dataset and the field measured monitoring data, using a cubic interpolation algorithm to interpolate and normalize the simulation monitoring curve dataset and the field measured monitoring data to unify the input dimension, and performing outlier truncation.

[0030] Furthermore, according to one embodiment of the present invention, the preprocessing further includes: adding Gaussian random noise to the simulation curves in the simulation monitoring curve dataset to achieve data augmentation.

[0031] Furthermore, according to one embodiment of the present invention, the classification neural network adopts the Softmax activation function and the classification cross-entropy loss function to achieve multi-classification; the regression sub-network uses MinMax normalization to process the output soil and rock parameters during training and saves the normalization scaler for subsequent inverse normalization operations.

[0032] Furthermore, according to one embodiment of the present invention, the simulation monitoring curve dataset includes: inclinometer curves, surface subsidence curves, support axial force curves, and groundwater level curves; the number of regression subnetworks is consistent with the number of strata types that the classification neural network can identify.

[0033] Furthermore, according to one embodiment of the present invention, in step S4, the preprocessed simulation monitoring curve dataset is divided into a training set and a test set, the training set is used to integrate the neural network model, and the test set is used to verify the model accuracy. When the classification accuracy and parameter inversion error meet the preset convergence conditions, the classification neural network model file, each regression sub-network model file, and the corresponding normalization scaler file are saved.

[0034] Furthermore, according to one embodiment of the present invention, step S6 further includes: comparing and displaying the theoretical monitoring curve generated by reconstructing the geotechnical parameters obtained by the regression subnetwork inversion with the field measured monitoring curve on the same axis, and generating a geotechnical parameter report and a foundation pit monitoring analysis report.

[0035] According to the above-described scheme of the present invention, the present invention uses finite element software to generate labeled simulation monitoring curve datasets in batches, replacing scarce field measured samples for model training. This effectively solves the problem of insufficient training data for machine learning models in the field of foundation pit engineering, reduces the dependence on massive field monitoring data, and avoids the drawbacks of indoor geotechnical test sampling disturbance and in-situ testing, such as high cost and limited measurement points.

[0036] The two-level integrated architecture constructed in this invention, consisting of a classification neural network and multiple sets of regression sub-networks, realizes a step-by-step operation mode that first identifies the stratum type and disease conditions, and then performs targeted parameter inversion. Compared with a single neural network structure, this architecture reduces the complexity of the nonlinear mapping between soil and rock parameters and monitoring data, and improves the accuracy of identification and inversion under complex stratum conditions.

[0037] This invention addresses the issue of permeability coefficients in special strata such as soft clay and confined aquifers spanning orders of magnitude. It employs a base-10 logarithmic transformation to unify the data scale and performs an inverse exponential transformation in the output stage to restore the true physical values. This avoids prediction distortion caused by data scale imbalance and enables the model to be stably applied to complex engineering scenarios such as urban soft soil foundation pits and water-rich foundation pits.

[0038] The standardized preprocessing workflow of this invention integrates Savitzky-Golay filtering for noise reduction, cubic interpolation normalization, outlier truncation, and Gaussian noise enhancement, enabling the model to adapt to complex noise environments such as construction site vibration, dust, and electromagnetic interference, thereby improving the model's robustness and generalization ability.

[0039] After the model training is completed, the inference of a single monitoring curve can complete parameter inversion and risk assessment within milliseconds, which is far superior to the inversion method of traditional intelligent algorithm combined with finite element iteration. It can meet the timeliness requirements of 24-hour uninterrupted real-time monitoring and dynamic early warning for foundation pit engineering.

[0040] This invention automates the entire process from field data access, curve analysis, parameter calculation to report output, reducing the workload of manually interpreting each monitoring curve, lowering the subjective bias and risk of missed detection in manual judgment, and improving the efficiency and standardization of monitoring work.

[0041] This invention is compatible with various types of foundation pit monitoring data, such as inclination measurement, settlement, axial force, and water level. It can be adapted to different support forms and geological conditions. The technical framework can be further extended to other geotechnical engineering fields such as tunnels, slopes, and roadbeds, and has good engineering application value.

[0042] Furthermore, to achieve the above objectives, the present invention also provides a foundation pit monitoring analysis and geotechnical parameter inversion system, comprising: The simulation monitoring curve dataset generation module builds a finite element model of the foundation pit, generates a batch of simulation monitoring curve datasets, and binds stratum type labels, disease type labels and corresponding real soil and rock parameters to each simulation monitoring curve. The data preprocessing module preprocesses the simulation monitoring curve dataset and the field measured monitoring data respectively. The preprocessing includes performing a base-10 logarithmic transformation on the permeability coefficient labels of soft clay and confined aquifer in the simulation monitoring curve dataset to unify the data scale. An integrated neural network model building module is used to construct an integrated neural network model consisting of a classification neural network and a regression neural network. The classification neural network is used to identify the stratum type and the working conditions of the foundation pit. The regression neural network includes multiple regression subnetworks, each corresponding to a different stratum type. The corresponding regression subnetwork is called to perform parameter inversion based on the identification results of the classification neural network. The model training module uses the preprocessed simulation monitoring curve dataset to train an integrated neural network model. The analytical inversion module inputs the pre-processed field measured monitoring data into the trained integrated neural network model. First, the classification neural network identifies the stratum type and the disease condition. Then, based on the identified stratum type, the corresponding regression sub-network is called to invert the geotechnical mechanics parameters. The foundation pit risk classification and early warning module performs an inverse exponential transformation to the permeability coefficient after logarithmic transformation, restoring the true value of the parameter, and performs foundation pit risk classification and early warning based on the comparison results of on-site measured monitoring data and preset early warning thresholds.

[0043] The foundation pit monitoring analysis and geotechnical parameter inversion system of the present invention can realize the foundation pit monitoring analysis and geotechnical parameter inversion method. The specific process steps are as described above and will not be repeated here.

[0044] Furthermore, to achieve the above objectives, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the foundation pit monitoring analysis and geotechnical parameter inversion method as described above.

[0045] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the foundation pit monitoring analysis and geotechnical parameter inversion method as described above.

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely one preferred embodiment of the invention and are only used to explain the invention. They do not limit the scope of protection of the invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0047] Example 1

[0048] This embodiment uses a deep foundation pit project in soft soil as the application object. The foundation pit has an excavation depth of 8m and adopts a combined support method of pile foundation and steel bracing. The site contains soft clay soil layers and locally confined aquifers. For the above application object, this embodiment provides a method for foundation pit monitoring analysis and geotechnical parameter inversion. The specific implementation steps include: Step S1: Build a finite element model of the foundation pit and generate a batch of simulation monitoring curve datasets.

[0049] First, configure the geometric parameters, stratum parameter ranges, and simulation parameters for the foundation pit project. Specifically, based on the foundation pit design drawings, input geometric parameters such as the excavation depth of 8m, excavation width of 20m, support pile diameter of 0.8m, and horizontal spacing of steel supports of 4m. Set the boundary constraints of the model and the construction sequence of layered excavation and support construction. Simultaneously, referring to the geological survey report, define the value ranges of soil and rock parameters for three types of strata: conventional clay, soft clay, and confined aquifer sand, covering cohesion c, internal friction angle φ, elastic modulus E, and permeability coefficient k. See Table 1 below for specific value ranges.

[0050] Table 1. Comparison of Typical Soil and Rock Parameter Values ​​and Scale Optimization for Foundation Pit

[0051] Subsequently, a two-dimensional numerical model of the foundation pit was built using Midas GTS finite element software. Parallel scheduling of an 8-core CPU was implemented based on the joblib framework. Random sampling was performed within a given range of soil parameters to simulate the excavation, support construction, and seepage conditions of the foundation pit in batches. Each simulation sequentially simulated the complete construction steps, including initial soil stress equilibrium, excavation of the first layer of soil, erection of the first steel support, dewatering to the second excavation face, excavation of the second layer of soil, and erection of the second support. The simulation output depth-horizontal displacement inclinometer curves, time-surface settlement curves, time-support axial force curves, and time-groundwater level curves.

[0052] This embodiment generates 5000 sets of simulation samples, and binds a formation type label (conventional clay, soft clay, confined aquifer), a disease type label (normal working condition, slight deformation, abnormal leakage), and a corresponding real soil and rock parameter label to each simulation monitoring curve. The disease type label is bound by the curve shape obtained by artificially pre-setting abnormal working conditions such as local soil strength reduction, support stiffness loss, and abrupt changes in seepage boundaries in the finite element simulation. The permeability coefficient field of the soft clay and confined aquifer samples is specifically marked to identify large-scale parameter fields.

[0053] Step S2: Preprocess the simulation monitoring curve dataset and the field measured monitoring data respectively.

[0054] First, perform the following standardized preprocessing operations sequentially on all 5000 sets of simulation monitoring curves: (1) Filtering and noise reduction: The Savitzky-Golay filtering algorithm is used to smooth the simulation curves and filter out the high-frequency disturbances introduced by the numerical calculation. The length of the filtering window is adaptively determined according to the number of sampling points of each monitoring curve. (2) Interpolation normalization: The cubic spline interpolation algorithm is used to normalize the discrete monitoring data with different sampling intervals in different simulation schemes into a standard input vector with 256 uniform sampling points, so as to unify the input dimension of the model; (3) Outlier truncation: Set truncation thresholds based on the physical reasonable range of each monitoring indicator to remove obvious outlier data points in the simulation; (4) Data augmentation: Gaussian random noise with a standard deviation of 0.05 is added to all simulation curves to simulate the measured noise introduced by the sensors at the construction site due to complex environmental factors such as vibration, dust and electromagnetic interference, thereby improving the robustness and generalization ability of the model to the harsh environment of the construction site.

[0055] Next, for the soft clay and confined aquifer samples specifically marked in the above simulation monitoring curve dataset, their permeability coefficient label k (in m / d) is transformed logarithmically to base 10 using the following formula: ; Because the permeability coefficient of soft clay is typically on the order of 0.001-0.01 m / d, while that of confined aquifer sand can reach 1.0-10.0 m / d, the difference in magnitude is enormous. Without scale optimization, neural networks will suffer from severe data scale imbalance during regression training, becoming insensitive to small parameters and even failing to converge. By using logarithmic transformation, the permeability coefficients are unified to a similar order of magnitude, fundamentally solving the prediction distortion problem caused by scale imbalance of special geological parameters.

[0056] For the field-measured monitoring data, the original monitoring curves were read from the interfaces of the inclinometers, settlement gauges, water level gauges, and IoT monitoring platform deployed on-site. The same Savitzky-Golay filtering, cubic spline interpolation, and outlier truncation operations were then sequentially performed. The filtering parameters, interpolation dimensions, and truncation thresholds were all kept completely consistent with the preprocessing parameters of the simulation data to ensure that the field-measured data strictly aligned with the simulation data distribution used during model training in terms of input dimensions and data quality. It should be noted that because the field-measured data lacks real soil and rock parameter labels, logarithmic transformation was not performed. Logarithmic transformation was only performed on the labels of the simulation data during the training phase, and the data was restored by inverse transformation after inference.

[0057] Step S3: Construct an integrated neural network model consisting of a classification neural network and a regression neural network.

[0058] In this implementation, an integrated neural network model is built based on the TensorFlow / Keras framework, and the overall architecture adopts a two-level integrated architecture of "classification neural network + multiple sets of regression sub-networks".

[0059] The classification neural network is a four-layer fully connected network with 256, 128, 64, and 32 nodes in each layer, respectively. The hidden layer activation function is ReLU, the output layer activation function is Softmax, and the loss function is classification cross-entropy. This classification neural network is used to simultaneously complete two multi-classification tasks: identifying soil strata types (conventional clay, soft clay, and confined aquifers) and identifying foundation pit defects (normal conditions, slight deformation, and abnormal leakage).

[0060] The regression neural network matches the stratigraphic type output by the classification neural network, setting up three sets of regression subnetworks corresponding to three types of strata: conventional clay, soft clay, and confined aquifers. Each regression subnetwork is a 5-layer fully connected network with 64, 64, 32, 16, and 8 nodes in each layer, respectively. The hidden layer activation function uses ReLU, and the output layer uses a linear activation function. The regression subnetwork receives the preprocessed monitoring curve as input and simultaneously outputs four geotechnical parameters corresponding to this type of stratum: cohesion c, internal friction angle φ, elastic modulus E, and permeability coefficient k (where k for soft clay and confined aquifers is the logarithmically transformed value).

[0061] Before training, the soil and rock parameters output by the regression subnetwork are normalized using two independent MinMaxScaler sets: one set for parameter normalization of conventional clay strata and the other set for parameter normalization of soft clay and confined aquifers. The normalization scalers are saved separately for subsequent inverse normalization operations.

[0062] Step S4: Train an integrated neural network model using the preprocessed simulation monitoring curve dataset.

[0063] The 5000 simulated samples preprocessed in step S2 were divided into a training set (4000 samples) and a test set (1000 samples) at an 8:2 ratio. Training was performed using the Adam optimizer, with the learning rates for both the classification and regression neural networks set to [value missing]. The weight decay coefficient is set to The batch size is set to 32. The classification neural network is trained for 100 epochs, and the regression neural network is trained for 200 epochs.

[0064] During training, the model accuracy is simultaneously verified using a test set. The model training is considered complete when the classification accuracy of the classification neural network reaches the preset convergence condition and the parameter inversion error (measured by root mean square error RMSE) of the regression subnetwork converges to within the acceptable range for engineering applications. After training convergence, the classification neural network model file, the regression subnetwork model files for each group, and the corresponding normalization scaler files are permanently saved for direct loading and use during the inference phase.

[0065] Step S5: Input the preprocessed on-site measured monitoring data into the trained integrated neural network model.

[0066] During the excavation of the foundation pit, the original data of inclination, surface settlement and groundwater level collected by the on-site monitoring instruments are accessed in real time. The standardized preprocessing operation is performed in sequence, which is completely consistent with the simulation data preprocessing in step S2, to obtain the preprocessed on-site measured monitoring curve.

[0067] The preprocessed field-measured curves are input into the trained ensemble neural network model, which performs two-stage inference: First, the classification neural network identifies the input curves, outputting that the current site is a soft clay layer with a local confined aquifer, and classifies the defective working condition as a normal construction condition. Then, based on the classification result, the corresponding regression sub-network for the soft clay layer is automatically invoked to retrieve the cohesion. internal friction angle Elastic modulus And the permeability coefficient value after logarithmic transformation.

[0068] Step S6: Perform an inverse exponential transformation on the permeability coefficient after logarithmic transformation, and conduct a risk classification and early warning for the foundation pit.

[0069] For the logarithmically transformed penetration coefficient of the regressor subnetwork output, perform a base-10 inverse exponential transform using the following formula: ; After inverse transformation, the true permeability coefficient of the soft clay layer is obtained as follows: The deviation from the on-site test results is within the allowable range for engineering.

[0070] Simultaneously, based on the three-level early warning system (yellow, orange, and red) set forth in the "Technical Standard for Monitoring of Building Foundation Pit Engineering" GB 50497, the on-site measured monitoring indicators are compared with this standard. In this implementation, a yellow warning (attention level) occurs when monitoring data reaches 70% of the design control value, or when the rate of change shows an abnormal trend, indicating a need to increase the monitoring frequency and analyze the cause. An orange warning (warning level) occurs when monitoring data reaches 85%-90% of the design control value, or when the rate exceeds the standard for two consecutive days, requiring the initiation of an emergency consultation, limitation of surrounding loads, and increased monitoring frequency to twice or more per day. A red warning (alarm / danger level) occurs when monitoring data reaches 100% of the design control value (i.e., the alarm value), or when the rate increases sharply, requiring immediate work stoppage, evacuation of personnel, and the implementation of emergency measures such as reinforcement. Figure 2 As shown in this embodiment, the measured maximum horizontal displacement was 4.5 mm, which is lower than the yellow warning threshold of 5 mm. Therefore, the foundation pit was determined to be in a safe state and no warning signal was triggered.

[0071] Finally, the theoretical inclination curve reconstructed by the regression subnetwork based on the inverted geotechnical parameters is compared coaxially with the measured inclination curve in the field, providing a visual representation of the curve fitting accuracy. The system automatically generates a comprehensive analysis report on foundation pit monitoring, including geotechnical parameter inversion results, monitoring index statistics, and three-level early warning status indicators, and archives it locally.

[0072] In this embodiment, the model inference time from inputting on-site measured data to parameter inversion and risk warning output is only in milliseconds, far superior to the several hours required by traditional finite element iterative inversion. The curve reconstruction values ​​and measured values ​​have high fitting accuracy, and the deviation between the inverted geotechnical parameters and the on-site in-situ test results is within the allowable range for engineering. The model achieves stable identification and accurate inversion of special strata such as soft clay and confined aquifers, without any parameter prediction distortion issues throughout the process, and the entire process requires no manual intervention, realizing the intelligent and automated analysis of foundation pit monitoring and geotechnical parameter inversion.

[0073] Those skilled in the art will recognize that the modules and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented using electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0074] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and equipment can be referred to the corresponding process in the foregoing method implementation, and will not be repeated here.

[0075] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0076] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the objectives of the embodiments of the present invention, depending on actual needs.

[0077] In addition, the functional modules in the embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0078] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to conventional solutions, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the sending / receiving methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0079] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this application.

[0080] It should be understood that the sequence number of each step in the invention and its embodiments does not absolutely imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

Claims

1. A method for monitoring and analyzing foundation pits and inverting geotechnical parameters, characterized in that, include: S1. Build a finite element model of the foundation pit, generate a batch of simulation monitoring curve datasets, and bind the stratum type label, disease type label and corresponding real soil and rock parameters to each simulation monitoring curve; S2. Preprocess the simulated monitoring curve dataset and the field measured monitoring data respectively; the preprocessing includes: performing a base-10 logarithmic transformation on the permeability coefficient labels of soft clay and confined aquifer in the simulated monitoring curve dataset to unify the data scale; S3. Construct an integrated neural network model consisting of a classification neural network and a regression neural network, wherein the classification neural network is used to identify the stratum type and the working condition of the foundation pit, and the regression neural network includes multiple regression subnetworks, each corresponding to a different stratum type. Based on the identification result of the classification neural network, the corresponding regression subnetwork is called to perform parameter inversion. S4. Train the ensemble neural network model using the preprocessed simulation monitoring curve dataset; S5. Input the pre-processed field measured monitoring data into the trained integrated neural network model. First, the classification neural network identifies the stratum type and disease conditions. Then, according to the identified stratum type, the corresponding regression sub-network is called to invert the geotechnical mechanics parameters. S6. Perform an inverse exponential transformation to the permeability coefficient after the logarithmic transformation to restore the true value of the parameter, and conduct a risk classification and early warning of the foundation pit based on the comparison results of the on-site measured monitoring data and the preset early warning threshold.

2. The method for foundation pit monitoring analysis and geotechnical parameter inversion according to claim 1, characterized in that, The preprocessing also includes: using the Savitzky-Golay filtering algorithm to filter and reduce noise in the simulated monitoring curve dataset and the field measured monitoring data; using a cubic interpolation algorithm to interpolate and normalize the simulated monitoring curve dataset and the field measured monitoring data to unify the input dimension; and performing outlier truncation.

3. The method for analyzing and inverting geotechnical parameters during foundation pit monitoring according to claim 2, characterized in that, The preprocessing also includes adding Gaussian random noise to the simulation curves in the simulation monitoring curve dataset to achieve data augmentation.

4. The method for foundation pit monitoring analysis and geotechnical parameter inversion according to claim 1, characterized in that, The classification neural network uses the Softmax activation function and the classification cross-entropy loss function to achieve multi-class classification; the regression sub-network uses MinMax normalization to process the output soil and rock parameters during training and saves the normalization scaler for subsequent inverse normalization operations.

5. The method for monitoring and analyzing foundation pits and inverting geotechnical parameters according to claim 1, characterized in that, The simulation monitoring curve dataset includes: inclinometer curves, surface subsidence curves, support axial force curves, and groundwater level curves; the number of regression subnetworks is consistent with the number of strata types that the classification neural network can identify.

6. The method for analyzing and inverting geotechnical parameters during foundation pit monitoring according to claim 1, characterized in that, In step S4, the preprocessed simulation monitoring curve dataset is divided into a training set and a test set. The integrated neural network model is trained using the training set, and the model accuracy is verified using the test set. When the classification accuracy and parameter inversion error meet the preset convergence conditions, the classification neural network model file, the regression sub-network model files, and the corresponding normalization scaler files are saved.

7. The method for analyzing and inverting geotechnical parameters during foundation pit monitoring according to claim 1, characterized in that, Step S6 further includes: comparing the theoretical monitoring curve reconstructed from the geotechnical parameters obtained by the regression subnetwork with the field measured monitoring curve on the same axis, and generating a geotechnical parameter report and a foundation pit monitoring analysis report.

8. A foundation pit monitoring and analysis system for geotechnical parameter inversion, characterized in that, include: The simulation monitoring curve dataset generation module builds a finite element model of the foundation pit, generates a batch of simulation monitoring curve datasets, and binds stratum type labels, disease type labels and corresponding real soil and rock parameters to each simulation monitoring curve. The data preprocessing module preprocesses the simulated monitoring curve dataset and the field measured monitoring data respectively; The preprocessing includes performing a base-10 logarithmic transformation on the permeability coefficient labels of soft clay and confined aquifers in the simulation monitoring curve dataset to unify the data scale; An integrated neural network model construction module is used to construct an integrated neural network model consisting of a classification neural network and a regression neural network. The classification neural network is used to identify the stratum type and foundation pit disease conditions. The regression neural network includes multiple regression subnetworks, each corresponding to a different stratum type. Based on the identification results of the classification neural network, the corresponding regression subnetwork is called to perform parameter inversion. The model training module trains the integrated neural network model using the preprocessed simulation monitoring curve dataset; The analytical inversion module inputs the pre-processed field measured monitoring data into the trained integrated neural network model. First, the classification neural network identifies the stratum type and the fault condition. Then, based on the identified stratum type, the corresponding regression subnetwork is called to invert the geotechnical mechanics parameters. The foundation pit risk classification and early warning module performs an inverse exponential transformation to the permeability coefficient after the logarithmic transformation, restoring the true value of the parameter, and performs foundation pit risk classification and early warning based on the comparison results of on-site measured monitoring data and preset early warning thresholds.

9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, it implements the foundation pit monitoring analysis and geotechnical parameter inversion method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the foundation pit monitoring analysis and geotechnical parameter inversion method as described in any one of claims 1-8.

Images