Method for monitoring and analyzing time effect of soil squeezing characteristic of static pressure pile in soft soil area

Abstract

The invention relates to the technical field of soft soil area building engineering construction monitoring, and discloses a soft soil area static pressure pile soil squeezing characteristic time effect monitoring and analysis method, which comprises the following steps: arranging a multi-source sensor array in a construction area, and collecting signals such as soil displacement and pore water pressure; and after noise filtering, time sequence alignment and data normalization preprocessing, a layered soil body finite element model is established. A mathematical model with the soil squeezing effect associated with time is constructed in combination with time sequence analysis, a prediction model based on machine learning is used for predicting soil body response, and model parameters are adjusted through an optimization algorithm. And judging whether convergence occurs or not according to a model prediction error so as to dynamically adjust the sensor array and form a closed-loop iterative analysis process. The method can comprehensively and accurately monitor and analyze the soil squeezing characteristic time effect of the static pressure pile in the soft soil area, provides a scientific basis for construction decision making, and guarantees engineering safety and quality.

Description

Technical Field

[0001] The invention relates to the technical field of construction monitoring of building projects in soft soil areas, in particular to a method for monitoring and analyzing the time effect of soil squeezing characteristics of static pressure piles in soft soil areas. Background Art

[0002] When carrying out construction in soft soil areas, static pressure piles are widely used due to their advantages such as low noise, no vibration, and high construction efficiency. However, during the construction process of static pressure piles, a significant soil squeezing effect will occur, which will have an adverse effect on the stability of the surrounding soil and adjacent buildings, underground pipelines and other facilities. Soft soil has the characteristics of high water content, high compressibility, low strength, and poor permeability, which makes the soil squeezing effect of static pressure piles more complicated in soft soil areas. During construction, the penetration of the pile body will cause displacement and deformation of the surrounding soil, and the pore water pressure will rise sharply and dissipate slowly. This soil displacement may cause the foundation of adjacent buildings to settle, tilt, or even structural damage; changes in pore water pressure may also change the effective stress state of the soil, reduce the shear strength of the soil, and increase construction safety risks.

[0003] At present, there are many deficiencies in the monitoring and analysis methods of static pile soil squeezing characteristics in soft soil areas. Traditional monitoring methods often rely on a single type of sensor, which cannot fully obtain information such as soil displacement, pore water pressure, pile penetration resistance and vibration, and it is difficult to accurately grasp the overall picture of the soil squeezing effect. Moreover, the collected data is easily affected by noise, the time series of data from different sensors are inconsistent, and the data processing methods are simple and crude, resulting in poor accuracy and reliability of monitoring data.

[0004] In terms of analytical methods, most existing models are simplified models that fail to fully consider the complexity and dynamic characteristics of soil layers in soft soil areas. These models cannot accurately simulate the dynamic changes of parameters such as elastic modulus, Poisson's ratio and permeability coefficient of soil layers at different depths, and the description of the relationship between soil squeezing effect and time is not accurate enough, making it difficult to effectively predict the soil response in the subsequent construction stage.

[0005] In addition, during the construction process, due to the lack of effective dynamic monitoring and analysis methods, it is impossible to adjust the construction parameters and monitoring strategies in time according to the actual construction situation. Once a problem occurs, it can only be remedied afterwards, which not only increases the cost of the project, but also may delay the construction period, seriously affecting the quality of the project and construction safety. With the continuous advancement of urban construction in soft soil areas, the scale and complexity of construction projects are increasing, and the need for accurate monitoring and analysis of the soil squeezing characteristics of static pressure piles in soft soil areas is becoming more and more urgent. A more advanced and effective monitoring and analysis method is urgently needed to solve these problems. Summary of the invention

[0006] The purpose of the present invention is to provide a method for monitoring and analyzing the time effect of soil squeezing characteristics of static pressure piles in soft soil areas, so as to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a method for monitoring and analyzing the time effect of soil squeezing characteristics of static piles in soft soil areas, the method comprising: STEP 1: Deploy a multi-source sensor array in the static pile construction area to collect soil displacement, pore water pressure, pile penetration resistance and vibration signals in real time during the construction process; STEP 2: Preprocess the collected raw data, including noise filtering, time series alignment and data normalization, to generate a standardized monitoring data set; STEP3: Based on the standardized monitoring data set, a layered soil finite element model is established, wherein the model includes dynamic parameters of elastic modulus, Poisson's ratio and permeability coefficient of soil layers at different depths in the soft soil area; STEP4: Combine the time series analysis method to extract the time-varying characteristics of soil displacement and pore water pressure, and construct a mathematical model of the correlation between soil squeezing effect and time; STEP5: Use a prediction model based on machine learning, input the time-varying characteristics of STEP4 and the layered soil model parameters of STEP3 to predict the soil response in the subsequent construction stage; STEP 6: Adjust the dynamic parameters of the layered soil model through the optimization algorithm to minimize the error between the prediction results and the actual monitoring data; STEP7: Determine whether the model prediction error after parameter optimization meets the convergence condition; if so, output the optimized soil parameters and the time evolution law of soil squeezing effect; if not, return to STEP6 for re-optimization; STEP8: According to the output results of STEP7, dynamically adjust the sampling frequency and spatial distribution of the sensor array; STEP9: Feed the updated monitoring data back to the prediction model in STEP5 to form a closed-loop iterative analysis process.

[0008] Preferably, in STEP1, the multi-source sensor array includes distributed optical fiber sensors, piezoelectric pore water pressure gauges, strain gauges and acceleration sensors; the sensors are arranged in layers along the axis of the pile body, and the radial arrangement density varies with the gradient of the soil permeability coefficient.

[0009] Preferably, in STEP 2, the noise filtering adopts a wavelet threshold denoising algorithm, the time series alignment adopts a dynamic time warping algorithm, and the data normalization adopts a segmented standardization method based on soil layer depth.

[0010] Preferably, in STEP3, the dynamic parameters of the layered soil finite element model are corrected in real time through a back propagation neural network, the input is the standardized monitoring data of STEP2, and the output is the elastic modulus correction coefficient and permeability attenuation factor of each soil layer.

[0011] Preferably, in STEP 4, the mathematical model is a group of partial differential equations coupling the elastic-plastic deformation of the soil and the seepage field, and its boundary conditions are dynamically updated according to the pile penetration rate and the frequency spectrum characteristics of the vibration signal.

[0012] Preferably, in STEP 5, the prediction model is a long short-term memory neural network, whose input features include historical displacement gradients, pore water pressure change rates and vibration energy distribution, and the output is the soil displacement extremes and pore water pressure peaks in the next three construction stages.

[0013] Preferably, in STEP 6, the optimization algorithm adopts an adaptive particle swarm optimization algorithm, the optimization variables are the cohesion correction factor and the permeability attenuation rate of the layered soil, and the constraint conditions are the engineering experience thresholds of the physical parameters of the soil layer.

[0014] Preferably, in STEP 7, the convergence condition is that the rate of decrease of the root mean square value of the prediction error for three consecutive iterations is less than a set threshold, and the maximum absolute error does not exceed the allowable engineering deviation.

[0015] Preferably, in STEP 8, the sampling frequency adjustment strategy of the sensor array is: dynamically adjusting the sampling interval according to the displacement change rate and the pore water pressure gradient according to an exponential function; and the spatial distribution optimization adopts a monitoring blind spot filling algorithm based on Kriging interpolation.

[0016] Preferably, in STEP 9, in the closed-loop iterative analysis process, the weight parameters of the prediction model are updated through an online learning mechanism, and the online learning mechanism adopts an incremental stochastic gradient descent algorithm, and the learning rate is negatively correlated with the progress of the construction stage.

[0017] Compared with the prior art, the present invention has the following beneficial effects: A multi-source sensor array is used, including distributed fiber optic sensors, piezoelectric pore water pressure gauges, strain gauge displacement gauges and accelerometers, which are arranged in layers along the axis of the pile body and the radial arrangement density varies with the gradient of the soil layer permeability coefficient. This layout can collect soil displacement, pore water pressure, pile penetration resistance and vibration signals in real time from multiple dimensions, and fully reflect the impact of static pressure pile construction on the soil. Compared with traditional single sensor monitoring, the data obtained is richer and more accurate, providing a solid data foundation for subsequent precise analysis.

[0018] In the data preprocessing stage, the wavelet threshold denoising algorithm is used to remove noise, the dynamic time warping algorithm is used to align the time series, and the segmented standardization method based on the soil depth is used to normalize the data. These advanced data processing technologies effectively improve the data quality, eliminate the problems of noise interference and time series inconsistency, make the data of different sensors comparable, and generate standardized monitoring data sets that can more truly reflect the actual state of the soil and improve the reliability of the analysis results.

[0019] The established finite element model of layered soil contains dynamic parameters such as elastic modulus, Poisson's ratio and permeability coefficient of soil layers at different depths in soft soil areas, and is corrected in real time through back propagation neural network. At the same time, a set of partial differential equations for coupling soil elastic-plastic deformation and seepage field is constructed as a mathematical model of soil squeezing effect and time correlation, and its boundary conditions are dynamically updated according to the pile penetration rate and vibration signal spectrum characteristics. This model construction and update method fully considers the complexity of soil layers in soft soil areas and the dynamic changes of the construction process, and can more accurately describe the change law of soil squeezing effect over time, providing a strong model support for accurate analysis and prediction.

[0020] The long short-term memory neural network (LSTM) is used as a prediction model, and the historical displacement gradient, pore water pressure change rate, and vibration energy distribution are input. It can effectively learn the time series characteristics of the soil state and accurately predict the extreme soil displacement and pore water pressure peak values ​​in the next three construction stages. Compared with traditional prediction methods, the LSTM model has a stronger ability to handle complex nonlinear relationships and more reliable prediction results, which helps construction personnel to grasp the soil response trend in advance and take corresponding measures in time to ensure construction safety and project quality.

[0021] The adaptive particle swarm optimization algorithm is used to adjust the dynamic parameters of the layered soil model to minimize the error between the prediction results and the actual monitoring data. The optimization variables are the cohesion correction factor and the permeability attenuation rate of the layered soil, and the constraint conditions are the engineering experience thresholds of the soil layer physical parameters. This optimization algorithm can quickly search for the optimal solution within a reasonable parameter range, making the model continuously close to the actual situation, improving the prediction accuracy and adaptability of the model, and ensuring the accuracy of the analysis results.

[0022] According to the model prediction error and analysis results, the sampling frequency and spatial distribution of the sensor array are dynamically adjusted. By dynamically adjusting the sampling interval according to the displacement change rate and pore water pressure gradient according to the exponential function, and optimizing the spatial distribution by using the monitoring blind area filling algorithm based on Kriging interpolation, the sensor layout can be made more reasonable, and the monitoring efficiency and data representativeness can be improved. The sampling frequency and the number of sensors are increased in areas with drastic soil changes, and reasonably reduced in areas with gentle changes, avoiding waste of resources while ensuring the integrity and accuracy of the monitoring data.

[0023] The updated monitoring data is fed back into the prediction model to form a closed-loop iterative analysis process. The weight parameters of the prediction model are updated through an online learning mechanism (incremental stochastic gradient descent algorithm), and the learning rate is negatively correlated with the progress of the construction phase. As the construction progresses, the model continuously absorbs new monitoring data, continuously optimizes its own performance, and realizes dynamic updates of soil response predictions, so that the monitoring and analysis results are always closely integrated with the actual construction situation, providing timely and accurate basis for construction decisions, and effectively ensuring the smooth progress of the construction process and the overall quality of the project. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a working principle diagram of the method for monitoring and analyzing the time effect of soil squeezing characteristics of static pressure piles in soft soil areas according to the present invention; Figure 2 Diagram of steps for laying out a multi-source sensor array; Figure 3 The diagram is a step-by-step diagram for modifying the dynamic parameters of the finite element model of layered soil; Figure 4 The flowchart of soil response prediction based on LSTM; Figure 5 Dynamic adjustment diagram of the sensor array. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] See also Figure 1-Figure 5 The present invention provides a technical solution: a method for monitoring and analyzing the time effect of soil squeezing characteristics of static piles in soft soil areas, the method comprising: STEP1: Deploy a multi-source sensor array in the static pile construction area to collect soil displacement, pore water pressure, pile penetration resistance and vibration signals in real time during the construction process. By properly arranging sensors, comprehensive and accurate field data can be obtained to provide a basis for subsequent analysis.

[0027] STEP2: Preprocess the collected raw data, including noise filtering, time series alignment and data normalization, to generate a standardized monitoring data set. The preprocessing process removes noise interference in the data, unifies the time series and standardizes the data range, making the data more suitable for subsequent modeling and analysis.

[0028] STEP3: Based on the standardized monitoring data set, a layered soil finite element model is established. The model contains dynamic parameters of elastic modulus, Poisson's ratio and permeability coefficient of soil layers at different depths in soft soil areas. This model can simulate the mechanical response of soil during static pile construction.

[0029] STEP4: Combine the time series analysis method to extract the time-varying characteristics of soil displacement and pore water pressure, and build a mathematical model of the relationship between soil squeezing effect and time. By analyzing the time-varying characteristics, the changing law of soil squeezing effect over time is revealed.

[0030] STEP5: Use a prediction model based on machine learning, input the time-varying characteristics of STEP4 and the layered soil model parameters of STEP3, and predict the soil response in the subsequent construction phase. Use the powerful prediction ability of machine learning to estimate the impact of construction on the soil in advance.

[0031] STEP 6: Adjust the dynamic parameters of the layered soil model through the optimization algorithm to minimize the error between the prediction results and the actual monitoring data. Continuously optimize the model parameters to improve the accuracy of the model prediction.

[0032] STEP7: Determine whether the prediction error of the model after parameter optimization meets the convergence condition; if so, output the optimized soil parameters and the time evolution law of soil squeezing effect; if not, return to STEP6 for re-optimization to ensure that the model achieves satisfactory prediction accuracy.

[0033] STEP8: According to the output results of STEP7, dynamically adjust the sampling frequency and spatial distribution of the sensor array to make the monitoring more targeted and improve the monitoring efficiency and accuracy.

[0034] STEP9: Feed the updated monitoring data back to the prediction model in STEP5 to form a closed-loop iterative analysis process. Continuously update data to continuously improve the performance of the prediction model.

[0035] The present invention will be further described below in conjunction with Examples 1 to 5: Embodiment 1: The multi-source sensor array includes distributed fiber optic sensors, piezoelectric pore water pressure gauges, strain gauges and accelerometers. Distributed fiber optic sensors have the advantages of high precision, long distance and distributed measurement. They can continuously monitor the strain changes of the soil along the axis of the pile body, and then obtain soil displacement information. The piezoelectric pore water pressure gauge uses the piezoelectric effect to quickly and accurately measure the changes in pore water pressure in the soil. It has high measurement accuracy and fast response speed, and can effectively capture the dynamic changes of pore water pressure during construction. The strain gauge converts soil displacement by measuring its own strain. It has the characteristics of simple structure and high measurement accuracy, and can work stably in complex construction environments. The accelerometer is used to monitor the vibration during the pile construction process and provide data support for analyzing the impact of vibration on the soil.

[0036] The sensors are arranged in layers along the axis of the pile. This is because the properties of soil layers at different depths in soft soil areas are different, and the layered arrangement can obtain the response information of soil at different depths. For example, in shallow soft soil, the soil is greatly disturbed by pile construction, and the displacement and pore water pressure change significantly; while in deep soil, although the disturbance is relatively small, there will be a certain response, and the layered arrangement can fully capture these changes. In addition, the radial arrangement density of the sensor changes with the gradient of the soil permeability coefficient. In soil layers with large permeability coefficients, the pore water in the soil flows faster, and the propagation and dissipation of the soil squeezing effect are also relatively fast. Therefore, increasing the radial arrangement density of sensors in these soil layers can more accurately monitor the changes in pore water pressure and soil displacement; on the contrary, in soil layers with small permeability coefficients, the sensor arrangement density is relatively small, which can not only meet the monitoring needs, but also reasonably control costs.

[0037] In actual construction, it is assumed that static pile construction is carried out in a soft soil area. The soil layers of the site are muddy clay, silty clay and sandy silt from top to bottom. According to the permeability coefficient of the soil layer, the radial spacing of the sensors is set to 0.5 meters in muddy clay; in silty clay, the spacing is set to 1 meter; in sandy silt, the spacing is set to 1.5 meters. Along the axis of the pile, a group of sensors is arranged every 1 meter. Each group of sensors includes distributed optical fiber sensors, piezoelectric pore water pressure gauges, strain gauges and accelerometers to ensure that soil information at different depths and radial positions can be fully and accurately collected.

[0038] Embodiment 2: The noise filtering adopts the wavelet threshold denoising algorithm. During the static pile construction process, the data collected by the sensor will inevitably be interfered by various noises, such as environmental noise, noise generated by the vibration of construction machinery, etc. The wavelet threshold denoising algorithm is based on the multi-resolution analysis characteristics of the wavelet transform, and decomposes the signal into different frequency sub-bands. In each sub-band, a threshold is set according to the statistical characteristics of the noise, and the wavelet coefficients less than the threshold are regarded as noise and processed, and the wavelet coefficients greater than the threshold are retained. Then, the signal is reconstructed by inverse wavelet transform, thereby achieving the purpose of denoising. The algorithm can effectively remove noise while retaining the main features of the signal. For example, when processing soil displacement signals, it can not only remove the fluctuations caused by noise, but also accurately retain the trend of displacement changes.

[0039] The dynamic time warping algorithm is used for time series alignment. Since multiple sensors start collecting data at different locations and times, the collected time series may be inconsistent. The dynamic time warping algorithm aligns the asynchronous time series by finding the optimal matching path between the two time series. Specifically, it calculates the cumulative distance of all possible matching paths between the two time series, and selects the path with the smallest cumulative distance as the optimal matching path, thereby achieving the alignment of the time series. Taking soil displacement and pore water pressure data as an example, after being processed by the dynamic time warping algorithm, the data collected by different sensors can be consistent in time, which is convenient for the subsequent simultaneous analysis of the relationship between the two.

[0040] Data normalization uses a segmented standardization method based on soil depth. The physical properties of soil layers at different depths are different, and the range and magnitude of data collected by sensors are also different. The segmented standardization method based on soil depth divides the soil layer into segments according to depth and standardizes the data in each segment. Assuming that the soil layer is divided into three segments: shallow layer (0-5 meters), middle layer (5-10 meters) and deep layer (below 10 meters), for the data of the shallow soil layer, the formula is used:

[0041] in, is the original data, and are the minimum and maximum values ​​of this type of data in the shallow soil layer, This is the normalized data. Through this segmented standardization method, the influence of soil depth differences on the data can be eliminated, making the data of soil layers at different depths comparable, and providing unified standard data for subsequent modeling and analysis.

[0042] In practical applications, a week of data collected from a construction site was processed. After being processed by the wavelet threshold denoising algorithm, the data curve is significantly smoother, and the burr phenomenon caused by noise is greatly reduced; the dynamic time warping algorithm accurately aligns the time series of different sensors, and the data correspondence is accurate when analyzing the relationship between soil displacement and pore water pressure over time; after being processed by the segmented standardization method based on the soil layer depth, the data of soil layers at different depths fluctuate within the range of 0-1, providing a good data basis for the subsequent establishment of a layered soil finite element model.

[0043] Embodiment 3: The dynamic parameters of the finite element model of the layered soil are corrected in real time through a back propagation neural network. The back propagation neural network is a neural network model with a strong learning ability. It adjusts the weight and threshold of the network through the back propagation of errors, so that the output of the network is as close to the true value as possible. In the present invention, the standardized monitoring data of STEP2 is used as the input of the back propagation neural network, and the output is the elastic modulus correction coefficient and the permeability attenuation factor of each soil layer.

[0044] The specific process is to input the soil displacement, pore water pressure and other information in the standardized monitoring data into the back propagation neural network, and the network undergoes multiple forward propagation and back propagation calculations. In the forward propagation process, the data is weighted summed and activated by the neurons in the network to gradually obtain the network output; in the back propagation process, according to the error between the network output and the actual monitoring data, the gradient of the error to the network weight and threshold is calculated, and the weight and threshold are adjusted using the gradient descent method. After multiple iterative training, the elastic modulus correction coefficient and permeability attenuation factor output by the network can more accurately reflect the actual situation of the soil layer.

[0045] For example, in a certain construction stage, the monitoring data showed that there was a deviation between the soil displacement predicted by the model and the actual monitoring value. The standardized monitoring data of this stage was input into the back propagation neural network. After training, the elastic modulus correction factor of a certain soil layer was 1.2, and the permeability attenuation factor was 0.8. This means that the actual elastic modulus of the soil layer is larger than the value set by the original model, and the permeability needs to be appropriately reduced. By real-time correction of the dynamic parameters of the finite element model of the layered soil, the model can be made more in line with the mechanical behavior of the soil during the actual construction process, and the simulation accuracy of the model for the soil squeezing effect can be improved, thereby providing a more reliable basis for subsequent predictions and analysis. In the subsequent construction stage, the prediction was based on the corrected model, and the error between the prediction results and the actual monitoring data was significantly reduced, proving the effectiveness of the dynamic parameter correction method.

[0046] Embodiment 4: In STEP4, the mathematical model constructed is a set of partial differential equations that couple the elastic-plastic deformation of the soil and the seepage field. For the elastic-plastic deformation of the soil, the Mohr-Coulomb yield criterion is used, and the yield function is:

[0047] in, and are the maximum and minimum principal stresses, is the soil cohesion, is the internal friction angle.

[0048] The equilibrium equation of soil is:

[0049] in, is the stress tensor, is the physical force component; in this formula Represents the spatial coordinate variable, used to describe the soil stress tensor In three-dimensional space, j is usually 1, 2, and 3, corresponding to the x, y, and z directions, respectively. about Calculate partial derivatives to reflect the changes in stress in various directions in space, and then reflect the internal stress and body force components of the soil balance relationship.

[0050] For the seepage field, according to Darcy's law, the seepage velocity for:

[0051] in, is the permeability tensor, is the water head, in this formula It is also a spatial coordinate variable used to determine the water head The direction and position of change of j corresponds to different spatial directions. By comparing the water head h with Calculate the partial derivative to obtain the rate of change of the water head in this direction, and then combine it with the permeability tensor To calculate the seepage velocity Component in the i direction.

[0052] The continuity equation is:

[0053] in, is the porosity, is the seepage velocity The value of i determines the spatial direction of the seepage velocity component. is the time variable.

[0054] The elastic-plastic deformation of soil is coupled with the seepage field equation to form a set of partial differential equations describing the soil squeezing effect.

[0055] The boundary conditions are dynamically updated according to the pile penetration rate and the spectral characteristics of the vibration signal. Assuming the pile penetration rate is , at the pile-soil contact boundary, the displacement boundary condition of the soil can be expressed as:

[0056] in, is the soil displacement component, is the boundary normal vector, is the time variable.

[0057] For vibration signals, the main frequency components are obtained through spectrum analysis. , the pore water pressure boundary condition considering the vibration effect on the boundary can be expressed as:

[0058] in, is the pore water pressure, is the initial pore water pressure, is the amplitude, By dynamically updating the boundary conditions, the mathematical model can more accurately reflect the changes of soil squeezing effect over time during static pile construction.

[0059] During the construction of static piles, the soil undergoes elastic-plastic deformation and is accompanied by pore water seepage, and the two affect each other. This set of partial differential equations couples the elastic-plastic mechanical equations of the soil with the seepage equation to fully describe the mechanical response of the soil during the construction process. Its boundary conditions are dynamically updated according to the pile penetration rate and the spectral characteristics of the vibration signal. The pile penetration rate directly affects the deformation rate of the soil and the change of pore water pressure. The spectral characteristics of the vibration signal reflect the distribution of vibration energy during the construction process. These factors will affect the mechanical behavior of the soil. For example, when the pile penetration rate increases, the extrusion effect on the soil increases, and the pore water pressure rises rapidly. At this time, the boundary conditions of the mathematical model will be adjusted accordingly to more accurately simulate the response of the soil.

[0060] In terms of prediction model, long short-term memory neural network (LSTM) is used. LSTM has the ability to process long sequence data and capture long-term data dependencies, which is very suitable for predicting the changes of soil displacement and pore water pressure over time. Its input features include historical displacement gradient, pore water pressure change rate and vibration energy distribution. The historical displacement gradient reflects the trend of soil displacement over time, the pore water pressure change rate reflects the dynamic changes of pore water pressure, and the vibration energy distribution describes the impact of construction vibration on the soil. By inputting these features into LSTM, the model can learn the relationship between them and future soil response, and the output is the extreme soil displacement and pore water pressure peak value in the next three construction stages.

[0061] The LSTM unit contains an input gate , Forget Gate , output gate and memory unit The input features are historical displacement gradient, pore water pressure change rate and vibration energy distribution. These features can fully reflect the change trend of soil state over time and the vibration impact during construction. Assume that the historical displacement gradient is The pore water pressure change rate is , the vibration energy distribution is , the current input vector It consists of three features, namely , the hidden state at the previous moment is .

[0062] Input Gate The calculation formula is:

[0063] in, and The input and hidden state to the weight matrix of the input gate, is the bias of the input gate, is the Sigmoid activation function, which maps the input value to Interval, used to control the extent to which new information enters the memory cell.

[0064] Forget Gate The calculation formula is:

[0065] and is the corresponding weight matrix, is the bias. The forget gate determines which information in the memory unit needs to be retained and which needs to be discarded, and outputs the information in the memory unit through the Sigmoid function. A value close to 1 indicates that the corresponding information is retained, and a value close to 0 indicates that it is discarded.

[0066] Calculating candidate memory cells :

[0067] here, and is the weight matrix, is the bias, The function maps the input to the (-1,1) interval, generating new information that may be updated to the memory cell.

[0068] Update memory unit :

[0069] That is, according to the control of the forget gate and the input gate, the memory unit at the previous moment The information and candidate memory units in Integrate the new information in.

[0070] Output Gate The calculation formula is:

[0071] and is the weight matrix, The output gate determines which information in the memory cell will be used to generate the output at the current moment.

[0072] Finally, calculate the hidden state at the current moment :

[0073] The output of LSTM is the extreme value of soil displacement in the next three construction stages. and peak pore water pressure ,These prediction values ​​provide an important basis for the construction party to understand the soil response in advance.

[0074] In practical applications, a large-scale construction project in a soft soil area is taken as an example. During the construction process, historical displacement gradients, pore water pressure change rates, and vibration energy distribution data are collected in real time and input into the trained LSTM prediction model. The model predicts that the extreme values ​​of soil displacement in the next three construction stages will be 5 cm, 6 cm, and 7 cm, respectively, and the peak values ​​of pore water pressure will be 50 kPa, 55 kPa, and 60 kPa, respectively. Based on these prediction results, construction personnel take corresponding measures in advance, such as adjusting the construction sequence and construction speed of subsequent piles, to avoid the impact of excessive soil displacement and pore water pressure on surrounding buildings and underground pipelines. At the same time, the actual monitoring data is compared with the prediction results, and it is found that the prediction results are close to the actual situation, which verifies the effectiveness and practicality of the mathematical model and prediction model.

[0075] Embodiment 5: In terms of optimization algorithm, an adaptive particle swarm optimization algorithm is adopted. This algorithm simulates the behavior of bird flocks foraging, and finds the optimal solution through continuous search and update of particles in the solution space. In the present invention, the optimization variables are the cohesion correction factor and the permeability coefficient decay rate of the layered soil. During the optimization process, each particle represents a set of values ​​of the cohesion correction factor and the permeability coefficient decay rate. The particle continuously adjusts its position according to its own historical optimal position and the global optimal position of the group to find the parameter value that minimizes the error between the prediction result and the actual monitoring data. The constraint condition is the engineering experience threshold of the physical parameters of the soil layer to ensure that the optimized parameters are within a reasonable engineering range. For example, in a certain optimization iteration, the cohesion correction factor corresponding to a certain particle is calculated to be 1.1, and the permeability coefficient decay rate is 0.05, which meets the engineering experience threshold requirements, and continues to perform subsequent iterations until the optimal solution is found.

[0076] The convergence condition is that the rate of decrease of the root mean square value of the prediction error for three consecutive iterations is less than the set threshold, and the maximum absolute error does not exceed the allowable engineering deviation. The root mean square value of the prediction error can comprehensively reflect the degree of error between the prediction result and the actual data. The rate of decrease is less than the set threshold, indicating that the optimization effect of the model is gradually stabilizing; the maximum absolute error does not exceed the allowable engineering deviation, which ensures the availability of the model prediction results in engineering. Assume that the threshold of the rate of decrease of the root mean square value of the prediction error is set to 0.05, and the allowable engineering deviation is 10%. In the process of model optimization, after multiple iterative calculations, when the rate of decrease of the root mean square value of the prediction error for three consecutive iterations is 0.04, 0.03, and 0.02 respectively, and the maximum absolute error is 8%, the convergence condition is met, the optimization is stopped, and the optimized soil parameters and the time evolution law of the soil squeezing effect are output.

[0077] The sampling frequency adjustment strategy of the sensor array is: according to the displacement change rate and the pore water pressure gradient, the sampling interval is dynamically adjusted according to the exponential function. When the displacement change rate is large or the pore water pressure gradient is large, it means that the soil state changes rapidly, and it is necessary to increase the sampling frequency and shorten the sampling interval; otherwise, the sampling frequency is reduced and the sampling interval is increased. For example, when the displacement change rate reaches 5 mm / hour and the pore water pressure gradient is 10 kPa / m, the sampling interval is adjusted to once every 5 minutes according to the exponential function; when the displacement change rate decreases to 1 mm / hour and the pore water pressure gradient decreases to 5 kPa / m, the sampling interval is adjusted to once every 15 minutes. In this way, the data acquisition volume and processing pressure can be reasonably reduced while ensuring the accuracy of the monitoring data.

[0078] Spatial distribution optimization uses a monitoring blind area filling algorithm based on Kriging interpolation. During the construction process, monitoring blind areas may exist due to the limitations of sensor layout. The Kriging interpolation algorithm uses the data of known monitoring points to estimate the value of the monitoring blind area through the spatial autocorrelation function, and reasonably fills points in the monitoring blind area based on the estimated results. For example, in a certain construction area, the Kriging interpolation algorithm analysis found that a certain area had a monitoring blind area. The soil displacement and pore water pressure in the area were estimated based on the data of the surrounding monitoring points, and sensors were added in the area to improve the comprehensiveness and accuracy of monitoring.

[0079] In the closed-loop iterative analysis process, the weight parameters of the prediction model are updated through an online learning mechanism, which uses an incremental stochastic gradient descent algorithm. As construction progresses, new monitoring data is constantly generated. The incremental stochastic gradient descent algorithm uses these new data and only uses one sample at a time to update the model weights, which can quickly adapt to data changes. The learning rate is negatively correlated with the progress of the construction stage. In the early stages of construction, the learning rate is large, and the model can quickly learn the characteristics of new data; as construction progresses, the learning rate gradually decreases, the model becomes more stable, and overfitting is avoided. By continuously updating the weight parameters of the prediction model, the model can better adapt to changes in the construction process and improve the accuracy of the prediction.

[0080] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0081] Although 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 the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring and analyzing the time effect of soil squeezing characteristics of static piles in soft soil areas, characterized in that: The method comprises: STEP 1: Deploy a multi-source sensor array in the static pile construction area to collect soil displacement, pore water pressure, pile penetration resistance and vibration signals in real time during the construction process; STEP 2: Preprocess the collected raw data, including noise filtering, time series alignment and data normalization, to generate a standardized monitoring data set; STEP3: Based on the standardized monitoring data set, a layered soil finite element model is established, wherein the model includes dynamic parameters of elastic modulus, Poisson's ratio and permeability coefficient of soil layers at different depths in the soft soil area; STEP4: Combine the time series analysis method to extract the time-varying characteristics of soil displacement and pore water pressure, and construct a mathematical model of the correlation between soil squeezing effect and time; STEP5: Use a prediction model based on machine learning, input the time-varying characteristics of STEP4 and the layered soil model parameters of STEP3 to predict the soil response in the subsequent construction stage; STEP 6: Adjust the dynamic parameters of the layered soil model through the optimization algorithm to minimize the error between the prediction results and the actual monitoring data; STEP7: Determine whether the model prediction error after parameter optimization meets the convergence condition; if so, output the optimized soil parameters and the time evolution law of soil squeezing effect; if not, return to STEP6 for re-optimization; STEP8: According to the output results of STEP7, dynamically adjust the sampling frequency and spatial distribution of the sensor array; STEP9: Feed the updated monitoring data back to the prediction model in STEP5 to form a closed-loop iterative analysis process.

2. The method according to claim 1, characterized in that: In STEP1, the multi-source sensor array includes distributed optical fiber sensors, piezoelectric pore water pressure gauges, strain gauges and acceleration sensors; the sensors are arranged in layers along the axis of the pile body, and the radial arrangement density varies with the gradient of the soil permeability coefficient.

3. The method according to claim 1, characterized in that: In the STEP 2, the noise filtering adopts a wavelet threshold denoising algorithm, the time series alignment adopts a dynamic time warping algorithm, and the data normalization adopts a segmented standardization method based on soil layer depth.

4. The method according to claim 2, characterized in that: In the STEP3, the dynamic parameters of the layered soil finite element model are corrected in real time through a back propagation neural network, the input is the standardized monitoring data of STEP2, and the output is the elastic modulus correction coefficient and permeability attenuation factor of each soil layer.

5. The method according to claim 1, characterized in that In the STEP 4, the mathematical model is a set of partial differential equations that couple the elastic-plastic deformation of the soil and the seepage field, and its boundary conditions are dynamically updated according to the pile penetration rate and the frequency spectrum characteristics of the vibration signal.

6. The method according to claim 1, characterized in that In STEP 5, the prediction model is a long short-term memory neural network, whose input features include historical displacement gradients, pore water pressure change rates and vibration energy distribution, and the output is the soil displacement extremes and pore water pressure peaks in the next three construction stages.

7. The method according to claim 1, characterized in that: In the STEP 6, the optimization algorithm adopts an adaptive particle swarm optimization algorithm, the optimization variables are the cohesion correction factor and the permeability attenuation rate of the layered soil, and the constraint conditions are the engineering experience thresholds of the physical parameters of the soil layer.

8. The method according to claim 1, characterized in that In STEP 7, the convergence condition is that the rate of decrease of the root mean square value of the prediction error for three consecutive iterations is less than a set threshold, and the maximum absolute error does not exceed the allowable engineering deviation.

9. The method according to claim 1, characterized in that: In STEP 8, the sampling frequency adjustment strategy of the sensor array is: dynamically adjusting the sampling interval according to the displacement change rate and the pore water pressure gradient according to the exponential function; the spatial distribution optimization adopts the monitoring blind area filling algorithm based on Kriging interpolation.

10. The method according to claim 1, characterized in that In STEP 9, in the closed-loop iterative analysis process, the weight parameters of the prediction model are updated through an online learning mechanism. The online learning mechanism adopts an incremental stochastic gradient descent algorithm, and the learning rate is negatively correlated with the progress of the construction phase.

Images