Time series prediction method for deformation of deep foundation pit in soil-rock composite stratum based on GA-PSO-GLSSVM algorithm

Abstract

The application belongs to the technical field of foundation pit engineering, and particularly relates to a soil-rock composite stratum deep foundation pit deformation time sequence prediction method based on a GA-PSO-GLSSVM algorithm, which comprises the following steps: S1, a data integration stage, S2, a parameter dynamic optimization stage, S3, a feature engineering stage, S4, a model construction stage and S5, a closed loop verification stage; the original monitoring data is subjected to non-equal-interval smoothing treatment through an accumulation generating operator to construct an input sequence with enhanced stability, this step effectively eliminates random fluctuations in the original data, highlights long-term trends of the data, provides a more stable and reliable data basis for subsequent parameter optimization and prediction, and further improves the prediction accuracy.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of foundation pit engineering, and particularly relates to a soil-rock composite stratum deep foundation pit deformation time sequence prediction method based on a GA-PSO-GLSSVM algorithm. BACKGROUND

[0002] In a soil-rock composite stratum deep foundation pit engineering, accurately predicting the foundation pit deformation time sequence is crucial for ensuring engineering safety. With the acceleration of urbanization, the number and scale of deep foundation pit engineering are increasing, and the complex geological conditions of soil-rock composite stratum bring great challenges to foundation pit deformation prediction.

[0003] At present, traditional time series analysis methods such as moving average method and exponential smoothing method are simple and easy to use, but they can only capture short-term trends of data and are difficult to accurately depict long-term trends and deformation characteristics under complex geological conditions. Prediction methods based on physical models consider geomechanical principles, but require a large number of geological parameters and complex calculation processes, and have poor adaptability to special geological features such as soil-rock interface mutation and soft soil rheology. In addition, some intelligent algorithms such as artificial neural networks and support vector machines have also been applied in foundation pit deformation prediction, but they have problems such as difficult parameter selection and easy to fall into local optimal solution, resulting in insufficient prediction accuracy and stability.

[0004] Therefore, the application provides a soil-rock composite stratum deep foundation pit deformation time sequence prediction method based on a GA-PSO-GLSSVM algorithm. SUMMARY

[0005] In order to make up for the deficiencies of the prior art and solve at least one technical problem proposed in the background art.

[0006] The technical scheme adopted by the application to solve its technical problems is: the soil-rock composite stratum deep foundation pit deformation time sequence prediction method based on the GA-PSO-GLSSVM algorithm, comprising the following steps:

[0007] S1, data integration stage

[0008] Synchronously collecting foundation pit excavation depth , enclosure structure displacement time sequence data and soil body improvement parameter matrix , performing non-equidistant smoothing processing on the original monitoring data through cumulative generation operators, and constructing input sequences with enhanced stability ;

[0009] S2, parameter dynamic optimization stage

[0010] The hybrid strategy of global search of genetic algorithm and local optimization of particle swarm algorithm is adopted, and the radial basis kernel function parameter of generalized least square support vector machine is dynamically adjusted And the regularization coefficient Wherein the kernel function parameter adjustment formula is:

[0011]

[0012] Wherein, And The kernel function parameters of the current and last time are respectively, The learning rate is, The error is The partial derivative of The maximum change of displacement time series data

[0013] S3, feature engineering stage

[0014] The lateral earth pressure gradient is introduced As a quantitative index of soil arching effect, combined with numerical inversion technology, the internal force distribution function of the supporting structure is reconstructed ;

[0015] The lateral earth pressure Is measured by the earth pressure sensor arranged in the soil body;

[0016] The spatial coordinates on the supporting structure are;

[0017] S4, model construction stage

[0018] The output of steps S1-S3 is taken as the input layer of GLSSVM, a space-time coupling prediction model considering the sudden characteristics of soil-rock interface and the rheological properties of soft soil is established, GLSSVM maps the input data to a high-dimensional feature space through nonlinear mapping, and linear regression is carried out in the high-dimensional space, so that the time series of foundation pit deformation is predicted;

[0019] S5, closed loop verification stage

[0020] Through A three-dimensional stratum-structure coupling model is established, the prediction results of the intelligent algorithm are cross-verified with the finite element simulation results, a feedback correction coefficient is formed And the model parameters are updated.

[0021] Further improvement of the application is that the cumulative generation operator in the data integration stage adopts a second-order weakening buffer operator, and its operation form is:

[0022] ​​

[0023] in, This represents the value of the accumulated sequence at time k. and ) are the original monitoring data sequences in and The value at time; This is an adjustment coefficient related to the soil creep rate; For the time series data of the enclosure structure displacement Regarding time The derivative of reflects the rate of change of displacement over time.

[0024] A further improvement of the present invention is that, in the parameter dynamic optimization stage... The fitness function of the hybrid optimization algorithm is designed as follows:

[0025]

[0026] in, The number of samples; These are actual observed values; These are the model's predicted values; This is a penalty factor for abrupt changes at the soil-rock interface, used to penalize insufficient model fitting of the abrupt changes at the soil-rock interface. The internal force distribution function of the support structure The gradient norm reflects the degree of drastic change in the distribution of internal forces.

[0027] A further improvement of the present invention is that the error in the kernel function parameter adjustment formula is... The mean square error between the generalized least squares support vector machine prediction and the actual displacement time series data is calculated using the following formula:

[0028]

[0029] Where N is the number of samples. This is the actual displacement value. To predict displacement values.

[0030] A further improvement of the present invention is that the calculation of the lateral earth pressure gradient in the characteristic engineering stage adopts the spatiotemporal double differential method:

[0031]

[0032] in, The lateral earth pressure gradient; Lateral earth pressure Regarding time The partial derivatives reflect the change of earth pressure over time; The rheological rate of soft soil was determined through indoor rheological tests. For lateral earth pressure The partial derivative of depth Reflects the change of earth pressure with depth.

[0033] Further improvement of the application lies in that the kernel function of GLSSVM in the model construction stage adopts an improved Matern32 function:

[0034]

[0035] Wherein, The kernel function value is; And The input sample vector is; The Euclidean distance of And The kernel function parameter is adjusted through the parameter dynamic optimization stage.

[0036] Further improvement of the application lies in that the feedback correction coefficient in the closed-loop verification stage Is calculated by the following formula:

[0037]

[0038] Wherein, The feedback correction coefficient is; The finite element simulation result vector is; The intelligent algorithm prediction result vector is; The 2-norm of the difference between the two is; The maximum value in the finite element simulation result vector is.

[0039] Further improvement of the application lies in that the feature engineering stage further includes wavelet packet decomposition of the input data, and extracting soil response feature components in different frequency bands, and the steps are as follows:

[0040] Selecting a suitable wavelet base function ;

[0041] Determining the number of wavelet packet decomposition layers, and determining the decomposition layer number according to the foundation pit excavation depth And the buried depth ratio of soil-rock interface ,

[0042] Wavelet packet decomposition is performed on the input data to obtain sub-signals in different frequency bands, and the features of each sub-signal are extracted as new input features for subsequent model construction.

[0043] Further improvement of the application lies in that the number of wavelet packet decomposition layers Is calculated by the following formula:

[0044] ​

[0045] wherein is the buried depth of the soil-rock interface; is a rounding-up function.

[0046] Further improvement of the present application is that the numerical inversion technique adopts a regularization method to solve the inverse problem of internal force distribution of the supporting structure, and the steps include:

[0047] establishing a model of the inverse problem of internal force distribution of the supporting structure, setting the observation data as and the internal force distribution function as then the inverse problem can be expressed as:

[0048]

[0049] wherein, is an observation operator, is observation noise;

[0050] introducing a regularization term, and constructing a regularization objective function:

[0051]

[0052] wherein, is a regularization parameter, is a regularization operator, and is a unit matrix or a differential operator;

[0053] solving the regularization objective function through an optimization algorithm to obtain an optimal solution of the internal force distribution function of the supporting structure, and accurate reconstruction of the internal force distribution of the supporting structure is realized.

[0054] The beneficial effects of the present application are as follows:

[0055] 1.The application carries out non-equidistant smoothing processing on original monitoring data through cumulative generation operator, constructs input sequence with enhanced stability, which effectively eliminates random fluctuations in original data, highlights long-term trend of data, provides more stable and reliable data basis for subsequent parameter optimization and prediction, and improves prediction accuracy; secondly, a hybrid strategy of global search of genetic algorithm and local optimization of particle swarm algorithm is adopted, radial basis kernel function parameters and regularization coefficients of generalized least squares support vector machine are dynamically adjusted, and based on this, global search ability of genetic algorithm and local optimization ability of particle swarm algorithm are fully utilized, which can better adapt to complex geological conditions such as soil-rock interface mutation and soft soil rheology, so that model parameters always remain in the optimal state, and then the adaptability and prediction accuracy of the model are improved; at the same time, by introducing lateral earth pressure gradient as a quantitative index of soil arching effect, combining with numerical inversion technology to reconstruct the internal force distribution function of supporting structure, the influence of soil arching effect on foundation pit deformation is fully considered, the soil arching effect is accurately quantified and the internal force distribution of supporting structure is reconstructed, which provides more comprehensive and accurate feature information for the model, thereby further improving the prediction accuracy.

[0056] 2.The application takes the output of data integration, parameter dynamic optimization and feature engineering as the input layer of GLSSVM, establishes a space-time coupling prediction model considering the characteristics of soil-rock interface mutation and soft soil rheology, which comprehensively considers various geological factors and deformation characteristics, can more comprehensively and accurately predict the time sequence of foundation pit deformation, and solves the problem of poor adaptability of traditional models to complex geological conditions; in addition, a three-dimensional stratum-structure coupling model is established by MIDAS GTS, the prediction results of intelligent algorithm are cross-verified with the finite element simulation results, a feedback correction coefficient is formed and the model parameters are updated, and this closed-loop verification mechanism can timely find the deviation of the prediction model, and update the model parameters through the feedback correction coefficient, realizing continuous optimization of the model and continuous improvement of the prediction accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0057] The application will be further described below with reference to the accompanying drawings.

[0058] Figure 1 It is a prediction flowchart of the application. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0060] Please refer to Figure 1 ,

[0061] The embodiment provides a soil-rock composite stratum deep foundation deformation time sequence prediction method based on a GA-PSO-GLSSVM algorithm, and comprises the following steps:

[0062] S1, a data integration stage

[0063] The excavation depth of the foundation pit is synchronously collected , and is measured by meters. The excavation depth of the foundation pit is obtained through construction records of the foundation pit. Before the excavation of the foundation pit, the elevation of a reference point and the initial position of the bottom of the foundation pit is measured by using a measuring device, and initial elevation data is recorded. With the progress of the excavation of the foundation pit, the elevation difference between the reference point and the current bottom of the foundation pit is measured by using a measuring device at regular time intervals, for example, every day or every time a certain excavation progress is completed. The depth of the excavation of the foundation pit is obtained by calculating the difference between the current elevation and the initial elevation. For example, when a level gauge is used for measurement, the level gauge is erected at a suitable position, and the readings of the reference point and the level gauge at the bottom of the foundation pit are read respectively. The elevation difference is calculated according to the elevation difference calculation formula, and the excavation depth is determined.

[0064] The time sequence data of the displacement of the enclosure structure are obtained by real-time monitoring of displacement sensors arranged on the enclosure structure, and the monitoring frequency is once per hour. The soil improvement parameter matrix contains information such as the type and mixing amount of the soil reinforcing agent, and is determined according to a construction scheme. Then, the original monitoring data is subjected to non-equidistant smoothing processing by using an accumulation generation operator. The original monitoring data sequence is set as The sequence after accumulation generation is constructed. The input sequence with enhanced stability is used for subsequent analysis.

[0065] The accumulation generation operator in the data integration stage adopts a second-order weakening buffer operator, and the operation form is as follows:

[0066]

[0067] wherein, is the value of the sequence after accumulation generation at the k moment; and are the values of the original monitoring data sequence at the and moments, respectively; is a soil creep rate related adjustment coefficient, which is determined by data obtained by long-term creep test of soil samples under constant stress levels. The specific determination method is as follows: the creep test of soil samples is carried out under a plurality of different constant stress levels, the strain data at different time points are recorded, the creep rate in different time intervals is calculated according to the strain data, the relationship between the creep rate and the related factors under different stress levels is obtained by fitting analysis of the creep rate data under different stress levels, and then the soil creep rate related adjustment coefficient is determined. The value; For the time series data of the enclosure structure displacement Regarding time The derivative of reflects the rate of change of displacement with time;

[0068] It should be noted that using a second-order weakening buffer operator to process the accumulated sequence can further smooth the data and reduce data fluctuations. In actual foundation pit monitoring data, due to the presence of various interference factors, the data often exhibits random fluctuations. These fluctuations may mask the true trend of the data. By averaging the data through this operator and introducing displacement change rate information, these random fluctuations can be effectively removed or weakened, making the processed data more stable and regular, and better meeting the requirements of subsequent models for input data, thereby improving the accuracy of the model's prediction of foundation pit deformation time series.

[0069] S2, Parameter Dynamic Optimization Stage

[0070] A hybrid strategy combining global search using genetic algorithms and local optimization using particle swarm optimization is employed to dynamically adjust the radial basis kernel function parameters of the generalized least squares support vector machine. and regularization coefficient The formula for adjusting the kernel function parameters is as follows:

[0071]

[0072] in, and These are the kernel function parameters for the current and previous time steps, respectively. For learning rate, For error right The partial derivatives, Displacement time series data The maximum change;

[0073] Kernel function parameter adjustment formula error The mean square error between the generalized least squares support vector machine prediction and the actual displacement time series data is calculated using the following formula:

[0074]

[0075] Where N is the number of samples. This is the actual displacement value. To predict displacement values, the error The mean square error between the predicted values ​​of the generalized least squares support vector machine and the actual displacement time series data is used to adjust the kernel function parameters by minimizing the mean square error, which can make the model's prediction results closer to the actual data.

[0076] In the parameter dynamic optimization stage The fitness function of the hybrid optimization algorithm is designed as:

[0077]

[0078] wherein, is the number of samples; is the actual observation value; is the model prediction value; is the soil-rock interface mutation penalty factor, used to punish the insufficient fitting of the model to the mutation characteristics of the soil-rock interface; is the gradient norm of the support structure internal force distribution function , reflecting the degree of change in the internal force distribution; the fitness function considers factors such as the number of samples, the actual observation value, the model prediction value, the soil-rock interface mutation penalty factor, and the gradient norm of the support structure internal force distribution function, and can more comprehensively evaluate the performance of the model to guide the algorithm to find a better parameter combination;

[0079] It should be noted that the parameters of the generalized least squares support vector machine have a great influence on the performance of the model, and the traditional parameter adjustment method may not be able to find the optimal parameter combination. The hybrid strategy combines the global search ability of the genetic algorithm and the local optimization ability of the particle swarm algorithm, which can more comprehensively search the parameter space and find a better parameter combination. The performance of the GLSSVM model is very sensitive to the kernel function parameters and the regularization coefficient, and through the hybrid optimization algorithm, the optimal parameters can be found more effectively, and the prediction accuracy of the model can be improved;

[0080] S3, feature engineering stage

[0081] Introducing lateral earth pressure gradient as a quantitative index of soil arching effect, combined with numerical inversion technology to reconstruct the support structure internal force distribution function ;

[0082] Lateral earth pressure is measured by the earth pressure sensor arranged in the soil, and the measurement process includes the following steps:

[0083] Connect the cable of the earth pressure sensor to the data acquisition system, set the parameters of data acquisition such as sampling frequency, range, etc., and the sampling frequency should be determined according to the progress of the foundation pit construction and the change of the lateral earth pressure, generally collecting data once an hour to meet the requirements of subsequent analysis and prediction;

[0084] During the excavation and support construction of the foundation pit, the data acquisition system collects the output signal of the lateral earth pressure sensor in real time according to the set parameters, and converts it into the corresponding pressure value for recording, at the same time, the measurement time, measurement point position and other related information should be recorded for subsequent data processing and analysis;

[0085] The collected data can be transmitted to a computer or other storage device for storage by wired or wireless means, and the collected lateral earth pressure data is pre-processed, including removing outliers, filling missing values, etc. Outliers may be caused by sensor failure, interference and other factors, which need to be judged and processed according to the actual situation.

[0086] Data verification: The rationality of the lateral earth pressure measurement data can be verified by comparing and analyzing with other monitoring data such as support structure displacement, foundation pit excavation depth, etc. If it is found that the data is obviously unreasonable, the installation and measurement process of the sensor needs to be rechecked, the problem needs to be found out and solved.

[0087] The spatial coordinates on the support structure are numerically simulated by the finite element analysis software, and the model parameters are continuously adjusted to make the simulation results as close as possible to the actual monitoring data.

[0088] The calculation of the lateral earth pressure gradient in the feature engineering stage uses the time-space double differential method:

[0089]

[0090] Where, is the lateral earth pressure gradient; is the lateral earth pressure The partial derivative of time reflects the change of soil pressure with time; is the soft soil rheological rate, which is determined by indoor rheological test; is the lateral earth pressure The partial derivative of depth reflects the change of soil pressure with depth;

[0091] The feature engineering stage also includes wavelet packet decomposition of the input data to extract soil response feature components in different frequency bands, with the following steps:

[0092] Selecting a suitable wavelet basis function ;

[0093] Determining the number of wavelet packet decomposition layers, according to the ratio of foundation pit excavation depth to the buried depth of soil-rock interface to determine the decomposition layer number ,

[0094] Wavelet packet decomposition is performed on the input data to obtain sub-signals in different frequency bands, and the features of each sub-signal are extracted as new input features for subsequent model construction;

[0095] It should be noted that wavelet packet decomposition can extract different frequency band soil response characteristic components, provide more rich feature information for the model, and help improve the prediction ability of the model. According to the depth of foundation pit excavation and the buried depth ratio of soil-rock interface to determine the number of decomposition layers , so that the decomposition is more in line with the actual situation;

[0096] The number of wavelet packet decomposition layers is calculated by the following formula:

[0097]

[0098] Among them is the buried depth of soil-rock interface; is the upward rounding function;

[0099] The numerical inversion technique adopts the regularization method to solve the inverse problem of the internal force distribution of the supporting structure, and the steps include:

[0100] The inverse problem model of the internal force distribution of the supporting structure is established, and the observation data is , and the internal force distribution function is , then the inverse problem can be expressed as:

[0101]

[0102] Among them, is the observation operator, is the observation noise;

[0103] Introducing the regularization term, the regularization objective function is constructed:

[0104]

[0105] Among them, is the regularization parameter, is the regularization operator, which is a unit matrix or a differential operator.

[0106] The regularization objective function is solved by an optimization algorithm to obtain the optimal solution of the internal force distribution function of the supporting structure, and the internal force distribution of the supporting structure is accurately reconstructed; The regularization method can solve the inverse problem of the internal force distribution of the supporting structure. By introducing the regularization term, the regularization objective function is constructed, which can effectively deal with the problems of observation noise and model ill-conditioning, and obtain more accurate internal force distribution function;

[0107] It should be noted that the lateral earth pressure is an important factor affecting the deformation of the foundation pit, the earth pressure gradient can reflect the change of the earth pressure, as a quantitative index of the soil arching effect, the numerical inversion technology can inversely deduce the internal force distribution of the supporting structure according to the measured lateral earth pressure data, improve the understanding of the stress condition of the supporting structure, and the lateral earth pressure gradient is introduced as a quantitative index of the soil arching effect, the numerical inversion technology is combined to reconstruct the internal force distribution function of the supporting structure, which can more accurately describe the mechanical behavior of the soil and the stress condition of the supporting structure, and provide more meaningful characteristics for the model;

[0108] S4, model construction stage

[0109] The output of steps S1-S3 is taken as the input layer of GLSSVM, a space-time coupling prediction model considering the sudden change characteristics of the soil-rock interface and the rheological properties of soft soil is established, GLSSVM maps the input data to a high-dimensional feature space through nonlinear mapping, and performs linear regression in the high-dimensional space to realize the prediction of the time sequence of the foundation pit deformation;

[0110] The kernel function of GLSSVM in the model construction stage adopts the improved Matern32 function:

[0111]

[0112] Wherein, is the kernel function value; and is the input sample vector; is the Euclidean distance of and is the kernel function parameter, which is adjusted through the parameter dynamic optimization stage;

[0113] It should be noted that the characteristics of the soil-rock composite stratum are complex, the sudden change of the soil-rock interface and the rheological properties of soft soil will affect the deformation of the foundation pit, and the traditional prediction model may not fully consider these factors, while the space-time coupling prediction model can better capture the complex relationship in the data by mapping the input data to a high-dimensional feature space through nonlinear mapping;

[0114] GLSSVM maps the input data to a high-dimensional feature space through nonlinear mapping, and performs linear regression in the high-dimensional space, which can handle nonlinear problems and improve the fitting ability of the model;

[0115] S5, closed loop verification stage

[0116] Through A three-dimensional stratum-structure coupling model is established, the intelligent algorithm prediction result is cross-verified with the finite element simulation result, a feedback correction coefficient is formed and the model parameters are updated.​

[0117] Feedback correction coefficient in closed-loop verification stage The feedback correction coefficient is calculated by the following formula:

[0118]

[0119] wherein, is the feedback correction coefficient; is the finite element simulation result vector; is the intelligent algorithm prediction result vector; is the 2-norm of the difference between the two; is the maximum value in the finite element simulation result vector;

[0120] It should be noted that a single prediction method may have certain limitations. By cross-verifying with the finite element simulation results, the shortcomings of the model can be found, and the model parameters can be adjusted through the feedback correction coefficient, so that the model is more in line with the actual situation;

[0121] Feedback correction coefficient In the calculation formula of the feedback correction coefficient, is the finite element simulation result vector; is the intelligent algorithm prediction result vector. The correction coefficient is determined by calculating the ratio of the 2-norm of the difference between the two and the maximum value in the finite element simulation result vector. This way can effectively measure the difference between the prediction result and the simulation result, and correct the model.

[0122] Example 1

[0123] Deformation prediction and verification of deep foundation pit in soil-rock composite stratum

[0124] 1. Project background and data enhancement

[0125] This project is a deep foundation pit project with an excavation area of 3200 square meters. The soil distribution is as follows:

[0126] 0-6m: soft clay (permeability coefficient cm / s)

[0127] 6-10m: sandy clay (containing 30% weathered rock debris)

[0128] 10-18m: strongly weathered mudstone (RQD=65%)

[0129] Monitoring system:

[0130] Displacement monitoring: 12 inclinometers are arranged with a spacing of 15m, and a total station is used to collect data every hour;

[0131] Soil pressure monitoring: install a vibrating wire soil pressure gauge with a range of 0-200kPa and an accuracy of 0.1%FS, record every 2 hours;

[0132] Data preprocessing and quantification:

[0133] Table 1: Comparison of processing effects of second-order weakening buffer operator

[0134]

[0135] 2. Model parameter optimization process

[0136] Table 2: GA-PSO hybrid optimization parameter iteration record

[0137]

[0138] 3. Feature engineering effect verification

[0139] Table 3: Wavelet packet decomposition frequency band feature contribution analysis

[0140]

[0141] 4. Closed-loop verification and engineering feedback

[0142] On-site measurement comparison:

[0143] 10-day prediction: maximum displacement prediction value 22.3 mm vs. measured 21.8 mm.

[0144] Conclusion: After processing by the second-order weakening buffer operator, the standard deviation of displacement monitoring data decreased from 4.2 mm to 1.6 mm, with a decrease of 62%, effectively eliminating the interference of construction vibration, instrument error, etc. After Fourier analysis verification, the high-frequency noise energy decreased by 82%, and the signal-to-noise ratio improved from 9.3 dB to 15.6 dB, with an increase of 68%. The signal-to-noise ratio of the displacement mutation feature at the soil-rock interface improved from 6.8 dB to 14.2 dB, increasing the identification accuracy of the mutation point from 72% to 89%. This method solves the industry problem of signal distortion caused by excessive smoothing and provides high-fidelity input data for subsequent modeling.

[0145] The GA-PSO hybrid strategy converges after 70 iterations, with a fitness value decrease of 53.7%, significantly better than the 8% of single GA3 or 42% of PSO. The optimized kernel parameter σ = 1.22 increases the sensitivity of Matern32 kernel function to mutation by 25%, and the regularization coefficient C = 14.9 reduces the overfitting risk by 19%. The low-frequency component of 0-0.05 Hz extracted by wavelet packet decomposition has an energy proportion of 58.7% and a displacement correlation R² = 0.82, which makes the mutation point prediction error decrease from 9.5 mm to 2.1 mm, with a decrease of 78%. Combined with the lateral soil pressure gradient index, the support structure internal force inversion coincidence reaches 89%, while the traditional BP neural network is only 72%, which means that this method improves the prediction efficiency.

[0146] The above-mentioned front, back, left, right, up, down are all based on the directions in the drawings Figure 1 Take the human observation angle as the standard, the side of the device facing the observer is defined as front, the left side of the observer is defined as left, and the like.

[0147] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the scope of protection of the present application.

[0148] The basic principles, main features and advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited by the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A time-series prediction method for deformation of deep foundation pits in soil-rock composite strata based on the GA-PSO-GLSSVM algorithm, characterized in that, Includes the following steps: S1, Data Integration Phase Synchronous acquisition of foundation pit excavation depth Time series data of enclosure structure displacement and soil improvement parameter matrix By using an accumulation generator operator to perform non-uniform interval smoothing on the original monitoring data, an input sequence with enhanced stability is constructed. ; S2, Parameter Dynamic Optimization Stage A hybrid strategy combining global search using genetic algorithms and local optimization using particle swarm optimization is employed to dynamically adjust the radial basis kernel function parameters of the generalized least squares support vector machine. and regularization coefficient The formula for adjusting the kernel function parameters is as follows: in, and These are the kernel function parameters for the current and previous time steps, respectively. For learning rate, For error right The partial derivatives, Displacement time series data The maximum change; S3, Feature Engineering Stage Introducing lateral earth pressure gradient As a quantitative index of the soil arching effect, it is combined with numerical inversion technology to reconstruct the internal force distribution function of the support structure. ; Lateral earth pressure It is measured by earth pressure sensors placed in the soil. Spatial coordinates on the support structure; S4, Model Building Phase The output of steps S1-S3 is used as the input layer of GLSSVM to establish a spatiotemporal coupled prediction model that considers the abrupt change characteristics of the soil-rock interface and the rheological properties of soft soil. GLSSVM maps the input data to a high-dimensional feature space through nonlinear mapping and performs linear regression in the high-dimensional space to predict the deformation time series of the foundation pit. S5, Closed-loop verification phase pass A three-dimensional stratigraphic-structure coupled model is established, and the prediction results of the intelligent algorithm are cross-validated with the finite element simulation results to form a feedback correction coefficient. And update the model parameters.

2. The method for predicting the time series deformation of deep foundation pits in soil-rock composite strata based on the GA-PSO-GLSSVM algorithm according to claim 1, characterized in that: The accumulation generation operator in the data integration stage adopts a second-order weakened buffer operator, and its operation form is as follows: in, This represents the value of the accumulated sequence at time k. and The original monitoring data sequence is in and The value at time; This is an adjustment coefficient related to the soil creep rate; For the time series data of the enclosure structure displacement Regarding time The derivative of reflects the rate of change of displacement over time.

3. The method for predicting the time series deformation of deep foundation pits in soil-rock composite strata based on the GA-PSO-GLSSVM algorithm according to claim 1, characterized in that: During the parameter dynamic optimization stage The fitness function of the hybrid optimization algorithm is designed as follows: in, The number of samples; These are actual observed values; These are the model's predicted values; This is a penalty factor for abrupt changes at the soil-rock interface, used to penalize insufficient model fitting of the abrupt changes at the soil-rock interface. The internal force distribution function of the support structure The gradient norm reflects the degree of drastic change in the distribution of internal forces.

4. The method for predicting the time series deformation of deep foundation pits in soil-rock composite strata based on the GA-PSO-GLSSVM algorithm according to claim 1, characterized in that: Error in the kernel function parameter adjustment formula The mean square error between the generalized least squares support vector machine prediction and the actual displacement time series data is calculated using the following formula: Where N is the number of samples. This is the actual displacement value. To predict displacement values.

5. The method for predicting the time series deformation of deep foundation pits in soil-rock composite strata based on the GA-PSO-GLSSVM algorithm according to claim 1, characterized in that: The calculation of the lateral earth pressure gradient in the characteristic engineering stage adopts the spatiotemporal double differential method: in, This represents the lateral earth pressure gradient. Lateral earth pressure Regarding time The partial derivatives reflect the change of earth pressure over time; The rheological rate of soft soil was determined through indoor rheological tests. Lateral earth pressure For depth The partial derivatives of earth pressure reflect the change of earth pressure with depth.

6. The method for predicting the time series deformation of deep foundation pits in soil-rock composite strata based on the GA-PSO-GLSSVM algorithm according to claim 1, characterized in that: The kernel function of GLSSVM in the model building phase adopts the improved Matern32 function: in, The kernel function value; and The input sample vector; for and The Euclidean distance; These are the kernel function parameters, which are adjusted during the parameter dynamic optimization phase.

7. The method for predicting the time series deformation of deep foundation pits in soil-rock composite strata based on the GA-PSO-GLSSVM algorithm according to claim 1, characterized in that: The feedback correction coefficient in the closed-loop verification stage Calculated using the following formula: in, This is the feedback correction factor; This is a vector of finite element simulation results; The vector of prediction results for the intelligent algorithm; The 2-norm of the difference between the two; This represents the maximum value in the vector of finite element simulation results.

8. The method for predicting the time series deformation of deep foundation pits in soil-rock composite strata based on the GA-PSO-GLSSVM algorithm according to claim 1, characterized in that: The feature engineering stage also includes wavelet packet decomposition of the input data to extract soil response feature components in different frequency bands, as follows: Choose appropriate wavelet basis functions ; Determine the number of wavelet packet decomposition levels based on the excavation depth of the foundation pit. Determining the number of decomposition layers by the ratio of the burial depth of the soil-rock interface , Wavelet packet decomposition is performed on the input data to obtain sub-signals in different frequency bands. The features of each sub-signal are extracted as new input features for subsequent model construction.

9. The method for predicting the time series deformation of deep foundation pits in soil-rock composite strata based on the GA-PSO-GLSSVM algorithm according to claim 8, characterized in that: The number of levels of the wavelet packet decomposition It is calculated using the following formula: in The depth of the soil-rock interface; This is the floor function.

10. The method for predicting the time series deformation of deep foundation pits in soil-rock composite strata based on the GA-PSO-GLSSVM algorithm according to claim 1, characterized in that: The numerical inversion technique employs The inverse problem of force distribution in a support structure using regularization methods includes the following steps: Establish an inverse problem model of the internal force distribution of the support structure, assuming the observed data are... The internal force distribution function is The inverse problem can then be expressed as: in, For the observation operator, To observe noise; Introduction Regularization terms, constructing the regularization objective function: in, For regularization parameters, For regularization operators, take the identity matrix or differential operator; The internal force distribution function of the support structure is obtained by solving the regularized objective function through an optimization algorithm. The optimal solution is obtained, enabling accurate reconstruction of the internal force distribution of the support structure.

Images