An artificial intelligence-based numerical simulation software parameter inversion optimization method

Abstract

The present application belongs to the technical field of data processing, and particularly relates to a numerical simulation software parameter inversion optimization method based on artificial intelligence. The steps include: collecting deep foundation pit geological, site, support and deformation monitoring data; preprocessing data; constructing a FLAC3D initial numerical model; building an AI optimization model with the simulation and monitoring deformation error as the objective function, improving the particle swarm global optimization through double closed-loop optimization, dynamically correcting parameters through the inner ring CNN-LSTM, and outputting the optimal parameters; and statistically determining the parameter value range through multiple inversions. The present application improves the parameter optimization accuracy and efficiency, enhances the simulation reliability, and provides technical support for deep foundation pit deformation prediction and safety evaluation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data processing, and particularly relates to a numerical simulation software parameter inversion optimization method based on artificial intelligence. BACKGROUND

[0002] With the increasing development of urban underground space, the scale and complexity of deep foundation pit engineering are continuously improved, and the deformation control during construction directly affects the structural safety of surrounding buildings, underground pipelines and roads. FLAC3D finite element simulation software is currently the mainstream tool for deep foundation pit deformation prediction, but its simulation accuracy completely depends on the accuracy of core parameters. Since these parameters in actual engineering are difficult to obtain through direct measurement on site, they need to be inversely deduced combined with field deformation monitoring data (i.e. parameter inversion), therefore, developing an efficient and accurate parameter inversion optimization method has become a key requirement to improve the accuracy of deep foundation pit deformation prediction and ensure engineering safety. The existing parameter optimization process is prone to local optimal solution, and the iterative convergence speed is slow, making it difficult to efficiently find the global optimal parameters, thereby resulting in insufficient scientificity and reliability of the parameter range, and ultimately leading to the difficulty of meeting the engineering safety requirements for the prediction accuracy of deep foundation pit deformation by numerical simulation software. SUMMARY

[0003] The application proposes a numerical simulation software parameter inversion optimization method based on artificial intelligence to solve the technical problems in the above background technology.

[0004] In order to achieve the above purpose, the technical scheme adopted by the application comprises the following steps:

[0005] S1, collecting geological conditions, site conditions, supporting means and deformation monitoring data of historical deep foundation pit engineering;

[0006] S2, performing data preprocessing to convert the collected unstructured data into a format, and performing outlier processing on the deformation monitoring data;

[0007] S3, constructing an initial numerical model of deep foundation pit engineering based on FLAC3D finite element simulation software combined with the preprocessed data;

[0008] S4, constructing an artificial intelligence optimization model with the error value between the simulation deformation data output by the FLAC3D finite element initial numerical model and the deformation monitoring data as the objective function; the objective function is: wherein M is the total number of detections, is the actual monitoring data value, is the simulation deformation data value output by the finite element model;

[0009] S5, parameter optimization is performed based on the artificial intelligence optimization model using double closed-loop parameter optimization; wherein the outer loop adopts an improved particle swarm algorithm for global parameter optimization, and the inner loop learns the deformation law constraint parameter search method based on a deep learning model to realize dynamic correction of the parameters and output optimal parameters;

[0010] S6, based on the optimal parameters obtained by the artificial intelligence optimization model, the value range of each type of to-be-inverted parameter in the FLAC3D finite element numerical simulation software is determined by combining the results of multiple inversion verifications and statistical analysis.

[0011] As preferred, the geological conditions include soil layer thickness, physical parameters, and soil layer permeability coefficients; the site conditions include site topographic slope and ground elevation; and the supporting means include supporting type and supporting material parameters.

[0012] As preferred, the step S2 of performing format conversion on the collected unstructured data and performing outlier processing on the deformation monitoring data is implemented as follows:

[0013] S21, first, the unstructured data of the supporting type is numbered into structured data for input;

[0014] S22, for the deformation monitoring data, the original deformation monitoring data of each monitoring point in the deep foundation pit project is obtained, wherein the original deformation monitoring data includes displacement values, settlement values, and corresponding sampling time stamps of each monitoring point at different sampling times;

[0015] S23, based on the spatial coordinates of the monitoring points and the sampling time stamps, a deformation monitoring data space-time matrix D is constructed, and the matrix elements represent the deformation monitoring value of the ith monitoring point at the jth sampling time;

[0016] S24, a space-time attention mechanism is introduced to extract features from the space-time matrix D, and spatial attention weights and temporal attention weights are calculated respectively, wherein the spatial attention weights are constructed based on the spatial distance and geological correlation between monitoring points to construct a spatial correlation matrix S, wherein , wherein is the straight-line distance between the ith and kth monitoring points, is the geological correlation coefficient between the ith monitoring point and the kth monitoring point, is a weight coefficient, is an attenuation coefficient, and the spatial correlation matrix S is normalized by row to obtain a spatial attention weight matrix ; the attention weight is constructed based on the sampling time interval to construct a time correlation vector T, wherein , wherein is the time interval between the jth and (j-1)th sampling, is a time interval weight coefficient, For the time attenuation coefficient, the time correlation vector T is normalized to obtain a time attention weight matrix , by obtaining the deformation monitoring matrix after attention weighting , wherein, indicates converting into a diagonal matrix;

[0017] S25, based on the weighted matrix , an LSTM prediction is used to realize preliminary identification of abnormal values, wherein the LSTM prediction model selects the first 70% of the sampling data in the matrix as a training set, and uses the trained LSTM model to predict the remaining 30% of the data to obtain the predicted value of each monitoring point at each sampling time , calculate the prediction residual , and based on the training set residual statistical threshold , wherein is the mean of the training set residual, is the standard deviation of the training set residual; if , the detection value is marked as an abnormal value and removed.

[0018] As a preferred embodiment, the step S3 is based on FLAC3D finite element simulation software, and the implementation manner of constructing the initial numerical model of the deep foundation pit engineering based on the preprocessed data is that: first, a geometric model is constructed according to the actual shape of the object and meshing is performed; the mesh quality evaluation tool of FLAC3D is used to check the unit angle and twist index, and the mesh shape is optimized through a smoothing algorithm or node relocation; different regions are assigned values according to the collected data after preprocessing; the boundary constraints are set to output the simulated deformation data.

[0019] As a preferred embodiment, the error value of the simulated deformation data output by the finite element model and the deformation monitoring data is used as the objective function to construct an artificial intelligence optimization model, and the artificial intelligence optimization model uses double closed-loop parameter optimization for parameter optimization; wherein the specific implementation of the outer loop using the improved particle swarm algorithm for global optimization of parameters is:

[0020] S511, first, population initialization is performed, and ecological niches are divided according to the functional subdomains of the parameters to be optimized: the parameter vector to be inverted , wherein represents the total number of parameters, which is divided into ecological niches according to the function, wherein the parameter subset corresponding to the th ecological niche is , satisfying ;

[0021] S512, the initial particles are generated by improved Logistic chaotic mapping, and for each ecological niche generate For each initial particle, the mapping formula is: ,in, The value of the chaotic sequence at the t-th iteration is 0 to 1. Let be the chaotic sequence value of the (t+1)th iteration. For chaos control parameters, The coefficient is a sinusoidal disturbance.

[0022] S513. Perform fitness variance verification and supplementation, calculate the particle fitness variance within the niche, where fitness is the reciprocal of the objective function, and the fitness variance is calculated as follows: ,like If the value is less than the set threshold, then supplement. A chaotic particle;

[0023] S514, Next, we will perform particle updates, following... ,in For inertial weights, As a learning factor, For the individual's optimal, It is the optimal ecological niche. For global optimality, The values ​​are random numbers, ranging from 0 to 1; calculate the population convergence. ,in For the standard deviation of the parameter, For the first The upper and lower limits of the parameters corresponding to each niche are set. If the convergence C is less than the set threshold, the random particles are perturbed. ,in The range of values ​​for the parameter vector. Let be the parameter vector of the particle after chaotic perturbation. This represents the intensity coefficient of chaotic perturbation; every ten iterations, niches are re-clustered based on particle fitness, with the cluster centers being the top-ranked particles. Preferred particle position;

[0024] S515. Then, adjust the parameters and calculate the error change rate. The error ,in The target error is used as the objective function; the parameters of the improved particle swarm are adjusted using the error and the rate of change of error as the second objective function, and the optimal improved particle swarm parameters are output through a genetic algorithm.

[0025] S516, Finally, perform a loop iteration. Or, when the number of iterations reaches its maximum, output... .

[0026] As preferred, the inner loop is based on the deep learning model to learn the deformation law constraint parameter search method on the basis of the optimization of the outer loop, and the specific implementation of dynamic correction of parameters is:

[0027] S521, extract the trend consistency coefficient of the measured deformation data of the deep foundation pit engineering and the FLAC3D simulation deformation data ; if the trend consistency coefficient is greater than or equal to the set threshold, the inner loop optimization is not performed and is directly retained, otherwise the inner loop optimization is performed;

[0028] S522, the inner loop optimization is to construct a CNN-LSTM deep learning model, input the parameters obtained in the outer loop optimization, and output is a parameter correction coefficient vector , wherein is the correction ratio of the first parameter;

[0029] S523, output the correction coefficient vector according to the characteristic deviation Calculate the corrected parameters , wherein is the parameter obtained by the outer loop optimization, is the corrected parameter, and the physical limit value of the parameter is set to avoid the corrected parameter exceeding the reasonable range.

[0030] As preferred, based on the optimal parameters obtained by the artificial intelligence optimization model, the value range of each type of to-be-inverted parameter in the FLAC3D finite element numerical simulation software is determined by statistical analysis and verification of multiple inversion results, which is to obtain multiple parameters by multiple inversions, calculate the average value and standard deviation of the multiple parameters, and finally use the average value ± 1.96 times the standard deviation to determine the value range of each parameter.

[0031] Compared with the prior art, the advantages and positive effects of the present application are that by introducing the artificial intelligence algorithm to optimize the inversion process of deep foundation pit parameters, the present application can effectively overcome the problem of easily falling into local optimal solution in the traditional optimization method, and improve the precision and efficiency of parameter optimization. Secondly, the double closed loop optimization mechanism is adopted, combined with the global and local optimization strategy, so that the parameter inversion can more accurately approach the true value, avoiding the calculation inefficiency and local optimization problem of the traditional method. Thirdly, through multiple inversion verification and statistical analysis, the reasonable value range of each to-be-inverted parameter is accurately determined, further improving the reliability of numerical simulation. Finally, the adaptability and flexibility of the present application make it suitable for deep foundation pit engineering of different scales and complexities, and it is widely used in practical engineering, providing strong technical support for deep foundation pit deformation prediction and safety evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0033] Figure 1 A flowchart of a parameter inversion optimization method based on artificial intelligence for a numerical simulation software; DETAILED DESCRIPTION

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0035] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application. However, the present application can be practiced in a variety of ways beyond the specific details set forth herein, and the present application is not limited to the specific embodiments described in the following.

[0036] Embodiments, deep foundation pit engineering is widely used in urban underground space development, and with the increase of building scale and complexity, how to accurately predict the deformation during construction, especially how to accurately obtain and inverse the required parameters in numerical simulation software, become the key to ensure the safety of the structure. However, the existing parameter optimization method is easy to fall into local optimal solution, and the iterative convergence speed is slow, which causes the low efficiency of the optimization process, and finally affects the accuracy of numerical simulation and the safety of the project. Therefore, the present application proposes a parameter inversion optimization method based on artificial intelligence for numerical simulation software, and the specific implementation process is as shown in Figure 1

[0037] Firstly, the geological conditions, site conditions, supporting means and deformation monitoring data of the historical deep foundation pit engineering are collected. The geological conditions include soil thickness, physical parameters and soil permeability coefficient; the site conditions include site topographic slope and ground elevation; the supporting means include supporting type and supporting material parameters. Then the data is preprocessed. The data preprocessing steps include converting unstructured data into structured data, and performing outlier processing on the deformation monitoring data to ensure the accuracy and effectiveness of the data. Specifically, firstly, the unstructured data of the supporting type is numbered into structured data for input; for the deformation monitoring data, the original deformation monitoring data of each monitoring point in the deep foundation pit engineering is obtained, which includes the displacement value, settlement value and corresponding sampling timestamp of each monitoring point at different sampling time; based on the spatial coordinates of the monitoring point and the sampling timestamp, a deformation monitoring data space-time matrix D is constructed, and the matrix elements are ​Dij represents the deformation monitoring value of the i th monitoring point at the j th sampling time; a spatiotemporal attention mechanism is introduced to extract features of the spatiotemporal matrix D, and spatial attention weights and temporal attention weights are calculated respectively, the spatial attention weights are constructed based on spatial distances and geological correlations between monitoring points to construct a spatial correlation matrix S, wherein Dij represents the deformation monitoring value of the i th monitoring point at the j th sampling time; a spatiotemporal attention mechanism is introduced to extract features of the spatiotemporal matrix D, and spatial attention weights and temporal attention weights are calculated respectively, the spatial attention weights are constructed based on spatial distances and geological correlations between monitoring points to construct a spatial correlation matrix S, wherein is the straight-line distance between the i th and the k th monitoring points, is the geological correlation coefficient between the i th monitoring point and the k th monitoring point, is a weight coefficient, is a decay coefficient, the spatial correlation matrix S is normalized by row to obtain a spatial attention weight matrix ; the attention weights are constructed based on sampling time intervals to construct a time correlation vector T, wherein Dij represents the deformation monitoring value of the i th monitoring point at the j th sampling time; a spatiotemporal attention mechanism is introduced to extract features of the spatiotemporal matrix D, and spatial attention weights and temporal attention weights are calculated respectively, the spatial attention weights are constructed based on spatial distances and geological correlations between monitoring points to construct a spatial correlation matrix S, wherein is the time interval between the j th and the j-1 th sampling, is a time interval weight coefficient, is a time decay coefficient, the time correlation vector T is normalized to obtain a time attention weight matrix , through a deformation monitoring matrix after attention weighting is obtained, wherein, represents converting into a diagonal matrix; based on the weighted matrix , an LSTM prediction is used to realize preliminary identification of abnormal values, wherein the LSTM prediction model selects the first 70% of the sampling data in the matrix as a training set, and uses the trained LSTM model to predict the remaining 30% of the data to obtain the predicted value of each monitoring point at each sampling time, calculates the prediction residual , and based on the training set residual statistical threshold , wherein is the mean of the training set residual, is the standard deviation of the training set residual; if , the detection value is marked as an abnormal value and is removed. After preprocessing, the collected data is integrated and matched together as the input of the finite element simulation software, ensuring that each input is complete data.

[0038] Secondly, based on the FLAC3D finite element simulation software, an initial numerical model of deep foundation pit engineering is constructed combined with the pre-processed data. First, a geometric model is constructed according to the actual shape of the object and meshing is performed; the mesh quality evaluation tool of FLAC3D is used to check the cell angle and twist index, and the mesh form is optimized through smoothing algorithm or node relocation; different regions are assigned values according to the collected data after preprocessing; the simulated deformation data output by the boundary constraint is set. Specifically, when constructing the initial numerical model of deep foundation pit engineering based on FLAC3D finite element simulation software, first, a geometric model needs to be constructed according to the actual shape of the object. This process involves accurate modeling of the spatial geometry of deep foundation pit engineering, including the shape, size and relative position of the foundation, supporting structure and other related facilities. After the construction of the geometric model, the meshing stage is entered. Meshing is a key step in finite element analysis, which requires reasonable selection of mesh size and shape according to the complexity and precision requirements of the geometric model. FLAC3D provides multiple meshing strategies, which can set different mesh densities according to the needs of different regions to ensure accuracy and computational efficiency. In order to ensure the accuracy and stability of the numerical model, FLAC3D provides a mesh quality evaluation tool to evaluate the quality of the mesh by checking the cell angle, twist index and other indicators. If the mesh quality does not meet the requirements, the mesh form can be optimized through smoothing algorithm or node relocation to eliminate irregular elements or element mismatch, thereby improving the calculation accuracy and convergence. Next, according to the collected data after preprocessing, different regions of the model are assigned values. Specifically, according to the field monitoring data or geological survey data, the model is assigned different regional material properties, such as the mechanical properties of soil and the stiffness of supporting structure. At the same time, reasonable boundary constraints are set to simulate the deformation behavior of deep foundation pit. Finally, the simulated deformation data output by FLAC3D provides the basis data for subsequent parameter inversion and optimization. Through the accurate construction of finite element model combined with actual geological and engineering data, the deformation of deep foundation pit during construction can be effectively predicted, providing reliable basis for engineering design and safety evaluation.

[0039] Then an artificial intelligence optimization model is constructed with the error value of the simulated deformation data output by the FLAC3D finite element initial numerical model and the deformation monitoring data as the objective function; the objective function is: where M is the total number of tests, is the actual monitoring data value, is the simulated deformation data value output by the finite element model. Based on the artificial intelligence optimization model, double closed-loop parameter optimization is used for parameter optimization; where the outer loop uses the improved particle swarm algorithm for global parameter optimization, and the inner loop uses the deep learning model to learn the deformation law to constrain the parameter search method, realize dynamic correction of parameters, and output the optimal parameters.

[0040] Specifically, the outer ring adopts the improved particle swarm algorithm to realize the global optimization of parameters, which includes the following steps: first, population initialization, niche division according to the function sub-domain of the parameters to be optimized: the parameter vector to be inverted is divided into , where represents the total number of parameters, which is divided into niches according to the function, where the parameter subset corresponding to the th niche is , satisfying ; improved Logistic chaotic mapping is used to generate initial particles, and for each niche , generate initial particles, and the mapping formula is: , where is the chaotic sequence value at the tth iteration, with a value range of 0 to 1, is the chaotic sequence value at the t+1th iteration, is the chaotic control parameter, is the sine perturbation coefficient; perform adaptive variance check and supplement, calculate the particle fitness variance in the niche, the fitness is the inverse of the objective function, and the fitness variance calculation method is , if is less than the set threshold, then supplement chaotic particles; next, update the particles, according to , where is the inertia weight, is the learning factor, is the individual optimal, is the niche optimal, is the global optimal, is a random number with a value range of 0 to 1; calculate the population convergence degree , where is the parameter standard deviation, is the upper and lower limits of the parameters corresponding to the th niche, if the convergence degree C is less than the set threshold, then perturb the random particles , where is the value range of the parameter vector, is the parameter vector of the chaotic perturbed particle, is the chaotic perturbation intensity coefficient; re-cluster and divide the niches based on the particle fitness every ten iterations, and the clustering center is the current top optimal particle position; then adjust the parameters, and calculate the error change rate as , where the error , where The target error is taken as the first objective function, and the error and error rate are taken as the second objective function to adjust the parameters of the improved particle swarm, and the optimal improved particle swarm parameters are output by the genetic algorithm . When the number of iterations reaches the maximum, the output is . The outer loop uses the improved particle swarm optimization (PSO) algorithm, which significantly improves the global search ability and convergence speed of the optimization process by introducing chaos mapping, adaptive variance check, particle update mechanism, and random particle disturbance strategies. First, the population is initialized according to the functional sub-domain division niche, and the initial particles are generated by the improved Logistic chaos mapping, which avoids the initialization bias problem in traditional particle swarm optimization and improves the diversity and coverage of the initial solution. The introduction of chaos sequence enhances the exploration ability of the search space, enabling the particle swarm to effectively avoid falling into local optimal solutions. The adaptive variance check mechanism ensures that the population does not lose diversity during the optimization process. If the variance of the fitness of the particles in a niche is less than the threshold, the system automatically supplements chaos particles, further enhancing the exploratory ability of the particle swarm and improving the global optimization ability. During the particle update process, the inertia weight and learning factor are adjusted adaptively to optimize the search efficiency of the particle swarm and ensure rapid approximation of the global optimal solution. The random particle disturbance mechanism and the strategy of re-dividing the niche by clustering effectively avoid premature convergence of the algorithm and improve the flexibility of the search process. In addition, the combination of error rate and genetic algorithm optimization strategies further optimizes the parameter adjustment of the particle swarm, enabling the algorithm to dynamically adapt to the optimization needs at different stages and improving the accuracy and stability of the optimization.

[0041] After the outer loop optimization is completed, it is determined whether to perform the inner loop optimization operation. Based on the deep learning model, the deformation law constraint parameter search method is learned to realize dynamic correction of the parameters. Specifically, the trend consistency coefficient of the measured deformation data and the FLAC3D simulation deformation data of the deep foundation pit engineering is extracted; if the trend consistency coefficient is greater than or equal to the set threshold, the inner loop optimization is not performed and the parameters are directly retained, otherwise the inner loop optimization is performed; the inner loop optimization is to construct a CNN-LSTM deep learning model, input the parameters obtained in the outer loop optimization, and output a parameter correction coefficient vector , where is the correction ratio of the th parameter; the correction coefficient vector is output according to the feature deviation , where is the parameter obtained by the outer loop optimization, The parameters obtained by the outer loop optimization are set as the input of the CNN-LSTM model. The main role of the CNN (Convolutional Neural Network) part in the CNN-LSTM model is to extract local features in the input data. The parameters obtained by the outer loop optimization are subjected to convolution operation to extract local features such as the trend of change of each parameter and the influence on deformation. Through multiple convolution layers and pooling layers, the CNN can extract the spatial features of the deep foundation parameters, helping the LSTM part to better understand the change pattern of the parameters. The main role of the LSTM (Long Short-Term Memory Network) part is to capture the dynamic change pattern of the input parameters over time, especially the time sequence dependence of the foundation deformation. The features extracted by the CNN are input into the LSTM model together with the historical data. The LSTM model outputs a correction coefficient vector representing the correction ratio of each parameter by processing the time sequence dependence in the historical data. Finally, the corrected parameters are calculated using the correction coefficient vector, and the physical limit value of the parameters is set to avoid the corrected parameters exceeding the reasonable range.

[0042] Finally, based on the optimal parameters obtained by the artificial intelligence optimization model, combined with the verification results of multiple inversions, the value range of each type of to-be-inverted parameter in the FLAC3D finite element numerical simulation software is determined through statistical analysis. Specifically, first, multiple parameters are obtained by multiple inversions, and the average value and standard deviation of the multiple parameters are calculated. Finally, the value range of each parameter is determined by using the average value ± 1.96 times the standard deviation. Through statistical analysis of multiple inversion results and confidence interval calculation, this method can effectively determine the reasonable value range of each type of to-be-inverted parameter, reduce uncertainty, improve the accuracy and stability of the numerical simulation results, and provide a scientific parameter basis for engineering practical applications.

[0043] The above is only a preferred embodiment of the present application, and is not intended to limit the present application in other forms. Any person skilled in the art can modify or change the above disclosed technical content to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made on the basis of the technical essence of the present application to the above embodiments still belongs to the protection scope of the present application technical solution.

Claims

1. An artificial intelligence-based numerical simulation software parameter inversion optimization method, characterized in that, The method comprises the following steps: S1, collecting geological conditions, site conditions, supporting means and deformation monitoring data of historical deep foundation pit projects; S2, performing data preprocessing to convert the collected unstructured data into a format and to perform outlier processing on the deformation monitoring data; S3, constructing an initial numerical model of the deep foundation pit project based on the FLAC3D finite element simulation software and in combination with the preprocessed data; S4, the error value of the simulated deformation data output by the FLAC3D finite element initial numerical model and the deformation monitoring data is taken as a target function to construct an artificial intelligence optimization model; the target function is: Wherein M is the total number of detections, is the actual monitoring data value, is the simulated deformation data value output by the finite element model; S5, performing parameter optimization based on an artificial intelligence optimization model using a double closed-loop parameter optimization method; the outer loop uses an improved particle swarm optimization algorithm to perform global optimization of parameters, and the inner loop uses a deep learning model to learn deformation rules to constrain parameter search methods, thereby achieving dynamic correction of parameters and outputting optimal parameters; S6, based on the optimal parameters obtained by the artificial intelligence optimization model, combining multiple inversion verification results, and statistically analyzing and determining the value range of each type of parameter to be inverted in the FLAC3D finite element numerical simulation software; The artificial intelligence optimization model uses a double closed-loop parameter optimization method to perform parameter optimization, and the specific implementation of the outer loop using an improved particle swarm optimization algorithm to perform global optimization of parameters is as follows: S511、First, population initialization is performed, and niches are divided according to functional subdomains of parameters to be optimized: the parameter vector to be inverted is divided into , where represents the total number of parameters, which is divided into niches according to functions, where the first niches correspond to the parameter subset , and satisfy ; S512, generating initial particles by improving Logistic chaotic mapping, for each niche generate initial particles, the mapping formula is: wherein, is the chaotic sequence value at the tth iteration, the value range is 0 to 1, is the chaotic sequence value at the t+1th iteration, is the chaotic control parameter, is the sinusoidal disturbance coefficient; S513、perform adaptive variance check complement, calculate the niche particle fitness variance, the fitness is the inverse of the objective function, the fitness variance calculation method is , if is less than the set threshold, then complement chaotic particles; S514, Next, we will perform particle updates, following... ,in For inertial weights, As a learning factor, For the individual's optimal, It is the optimal ecological niche. For global optimality, The values ​​are random numbers, ranging from 0 to 1; calculate the population convergence. ,in For the standard deviation of the parameter, For the first The upper and lower limits of the parameters corresponding to each niche are set. If the convergence C is less than the set threshold, the random particles are perturbed. ,in The range of values ​​for the parameter vector. Let be the parameter vector of the particle after chaotic perturbation. This represents the intensity coefficient of chaotic perturbation; every ten iterations, niches are re-clustered based on particle fitness, with the cluster centers being the top-ranked particles. Preferred particle position; S515, then the parameter adjustment, the error change rate is Wherein the error Wherein The target error; the error and the error change rate are the second target function for adjusting the parameters of the improved particle swarm, and the optimal improved particle swarm parameters are output through a genetic algorithm; S516, the loop iteration is performed if or the iteration number reaches the maximum, output .

2. The numerical simulation software parameter inversion optimization method based on artificial intelligence according to claim 1, characterized in that, The geological conditions include soil layer thickness, physical parameters, and soil layer permeability coefficients; the site conditions include site terrain slope and ground elevation; and the supporting means include supporting types and supporting material parameters.

3. The numerical simulation software parameter inversion optimization method based on artificial intelligence according to claim 1, characterized in that, In step S2, the unstructured data collected is converted into a format, and the deformation monitoring data is processed for outliers as follows: S21, first, the unstructured data of the supporting type is numbered into structured data for input; S22, for the deformation monitoring data, the original deformation monitoring data of each monitoring point in the deep foundation pit project is obtained, including the displacement value, settlement value, and corresponding sampling timestamp of each monitoring point at different sampling times; S23, based on the monitoring point spatial coordinates and the sampling time stamp, constructing a deformation monitoring data space-time matrix D, the matrix elements representing the deformation monitoring value of the i th monitoring point at the j th sampling time; S24, a spatio-temporal attention mechanism is introduced to extract features from the spatio-temporal matrix D, and spatial attention weights and temporal attention weights are calculated respectively, the spatial attention weights are based on the spatial distance and geological correlation between each monitoring point to construct a spatial correlation matrix S, wherein , wherein is the straight-line distance between the ith and kth monitoring points, is the geological correlation coefficient between the ith monitoring point and the kth monitoring point, is the weight coefficient, is the attenuation coefficient, the spatial correlation matrix S is normalized by row to obtain the spatial attention weight matrix ; the attention weight is based on the sampling time interval to construct a time correlation vector T, wherein , wherein is the time interval between the jth and (j-1)th sampling, is the time interval weight coefficient, is the time attenuation coefficient, the time correlation vector T is normalized to obtain the time attention weight matrix , and the deformation monitoring matrix after attention weighting is obtained by , wherein, indicates that is converted into a diagonal matrix; S25, based on the weighted matrix , the abnormal value preliminary identification is realized by adopting the LSTM prediction, wherein the LSTM prediction model selects the matrix The first 70% of the sampling data are taken as the training set, the remaining 30% of the data are predicted by using the completed LSTM model, and the prediction value of each monitoring point at each sampling time is obtained , the prediction residual is calculated , and the residual threshold of the training set is calculated based on the training set , wherein is the mean of the training set residual, is the standard deviation of the training set residual; if , the detection value is marked as an abnormal value and removed.

4. The numerical simulation software parameter inversion optimization method based on artificial intelligence according to claim 1, characterized in that, In step S3, the FLAC3D finite element simulation software is used to construct an initial numerical model of the deep foundation pit project in combination with the preprocessed data as follows: first, a geometric model is constructed according to the actual shape of the object and meshing is performed; the mesh quality evaluation tool of FLAC3D is used to check the unit angle and twist degree indicators, and the mesh shape is optimized through a smoothing algorithm or node relocation; The preprocessed data is assigned to different regions; the boundary constraints of the output simulated deformation data are set.

5. The numerical simulation software parameter inversion optimization method based on artificial intelligence according to claim 1, characterized in that, The specific implementation of the inner loop based on a deep learning model to learn deformation rules to constrain parameter search methods and achieve dynamic correction of parameters is as follows: S521、extracting a trend consistency coefficient of the measured deformation data of the deep foundation pit engineering and the simulated deformation data of the FLAC3D ; if the trend consistency coefficient is greater than or equal to a set threshold value, no inner loop optimization is performed and it is directly reserved, otherwise the inner loop optimization is performed; S522, the inner loop optimization is to construct a CNN-LSTM deep learning model, input the parameters obtained in the outer loop optimization, and output is a parameter correction coefficient vector wherein is the correction ratio of the first parameter S523、outputting a correction coefficient vector according to the feature bias calculating the corrected parameter wherein is the parameter obtained by outer loop optimization, is the corrected parameter, and a physical limit value of the parameter is set to avoid the corrected parameter from exceeding a reasonable range.

6. The numerical simulation software parameter inversion optimization method based on artificial intelligence according to claim 1, characterized in that, Based on the optimal parameters obtained by the artificial intelligence optimization model, combining multiple inversion verification results, and statistically analyzing and determining the value range of each type of parameter to be inverted in the FLAC3D finite element numerical simulation software, the implementation is as follows: multiple parameters are obtained by performing multiple inversions, the average value and standard deviation of the multiple parameters are calculated, and finally the average value ± 1.96×standard deviation is used to determine the value range of each parameter.

Images