A method for predicting the sinking posture of a caisson

Abstract

The application discloses a kind of sinking well sinking posture prediction method, belong to sinking well sinking prediction technical field, comprising: according to the construction of field measured time series data first-level data matrix;Smooth processing is carried out to first-level data matrix using Savitsky-Golay filter algorithm, and second-level data matrix is obtained;Second-level data matrix and construction of construction excavation information representation matrix are combined to construct third-level data matrix;Correlation analysis is carried out to third-level data matrix using three kinds of analysis method, and correlation degree representation matrix is obtained;Correlation degree representation matrix is used to screen key variables to construct fourth-level data matrix;Dimension reduction processing is carried out to fourth-level data matrix using principal component analysis method, and principal component matrix is obtained;Long short memory network is trained using principal component matrix, and sinking well sinking posture prediction model is obtained to carry out sinking well sinking posture prediction, and sinking well sinking posture prediction result is obtained.The method can improve the accuracy and stability of deep water sinking well sinking posture prediction.

Description

A method for predicting the sinking attitude of a caisson Technical Field

[0001] This invention relates to a method for predicting the sinking attitude of a caisson, belonging to the field of caisson sinking prediction technology. Background Technology

[0002] In recent years, with the development of large-scale projects such as deep-water bridges, port terminals, and underwater tunnels, caissons have been widely used in deep-water foundation engineering due to their advantages such as high load-bearing capacity and wide adaptability. However, the sinking and positioning process of deep-water caissons still faces many complex engineering challenges. Caisson sinking is one of the key points closely monitored during construction, as significant offsets and torsion may pose safety hazards to the superstructure construction.

[0003] During the sinking of caissons in water, the potential influencing factors are complex and diverse, and the interactions between these factors are difficult to clearly define through traditional experience or theoretical analysis. On-site construction typically involves the deployment of diverse and comprehensive monitoring equipment, accumulating a large amount of actual monitoring data, providing a foundation for conducting caisson sinking research based on this data. The monitoring content includes various types of data such as humidity and temperature, flow velocity and direction, caisson sinking depth, water level, water injection volume into compartments, sidewall earth pressure, cutting edge reaction force, and structural internal forces. Grey relational model, maximum mutual information number, and Spearman correlation coefficient are all effective methods for quantifying the correlation between variables and can serve as important steps in feature selection. Long short-term neural networks are a reliable model for handling long-term dependencies in time-series data.

[0004] Currently, research on caisson sinking prediction based on monitoring data is relatively limited, mainly focusing on numerical simulation and soil mechanics theoretical models. Some scholars have conducted research on intelligent caisson models based on monitoring data, but the types of data considered are relatively limited, failing to fully extract information from the monitoring data. Furthermore, the definition of correlated data lacks sufficient theoretical support, and the model data often neglects consideration of construction schemes. Therefore, given sufficient data, combining the characteristics of actual engineering projects and considering as many different types of data as possible is of great significance for further deepening the understanding of the development law of caisson sinking. Summary of the Invention

[0005] The purpose of this invention is to provide a method for predicting the sinking attitude of caissons. This method integrates the spatiotemporal characteristics of construction excavation with multi-source monitoring data, and combines multi-parameter filtering with multi-method correlation analysis. It solves the problems of low prediction accuracy and poor robustness caused by the neglect of the dynamic process of construction and insufficient data utilization in traditional caisson sinking prediction methods, and achieves more accurate and stable prediction of the sinking attitude of deep-water caissons.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for predicting the sinking attitude of a caisson, comprising:

[0008] Acquire on-site measured time-series data during the sinking process of the caisson, and based on the construction sequence, construct a construction excavation information representation matrix reflecting the excavation status of each compartment using the binary representation method;

[0009] A primary data matrix was constructed based on the on-site measured time-series data.

[0010] By using the Savitzky-Gore filtering algorithm and setting several filtering parameters, continuous variables in the first-level data matrix are smoothed to generate a second-level data matrix.

[0011] Based on the secondary data matrix and combined with the construction excavation information representation matrix, a tertiary data matrix is ​​constructed.

[0012] The grey relational analysis method, the maximum mutual information analysis method, and the Spearman correlation coefficient analysis method are used to conduct a correlation analysis on the three-level data matrix to obtain the correlation degree characterization matrix.

[0013] Using the correlation characterization matrix, key variables were selected from the original variables contained in the characterization matrix of the field measured time series data and the construction excavation information, and a four-level data matrix was constructed.

[0014] Principal component analysis was used to reduce the dimensionality of the four-level data matrix to obtain the principal component matrix.

[0015] Using the principal component matrix as training data, the long short-term memory network is trained to obtain a well-trained long short-term memory network, which serves as a prediction model for the sinking attitude of the caisson.

[0016] The sinking attitude prediction model of the caisson was used to predict the sinking attitude of the caisson, and the prediction results of the sinking attitude of the caisson were obtained.

[0017] In conjunction with the first aspect, further, the on-site measured time series data includes:

[0018] Data used to reflect the interaction between the caisson and the surrounding soil include earth pressure on the caisson sidewalls and reaction force at the caisson cutting edge;

[0019] Data used to reflect the internal mechanical state of the caisson structure, including stress at the root of the caisson cutting edge and stress in the concrete of the caisson body;

[0020] Data used to reflect the sinking environment and attitude of the caisson includes the water level in the observation well, the groundwater level, and the sinking depth.

[0021] In conjunction with the first aspect, furthermore, in the construction excavation information representation matrix, 1 indicates that the compartment is in the excavation state, and 0 indicates that the compartment is in the stopped excavation state.

[0022] In conjunction with the first aspect, further, based on the on-site measured time-series data, a primary data matrix is ​​constructed, including:

[0023] The field-measured time-series data are preprocessed, and the preprocessed field-measured time-series data form a first-level data matrix.

[0024] The preprocessing of the on-site measured time series data includes:

[0025] Interpolation was used to fill in missing data in the field measured time series data;

[0026] Calculate the increase data between consecutive data points in the field measured time series data after interpolation and filling, as well as the interquartile range of the increase data;

[0027] Based on the increase data and the interquartile range of the increase data, outliers are selected from the interpolated and filled field measured time series data, and the outliers are replaced with the average value of their adjacent data points.

[0028] The formula for calculating the increase data is:

[0029] ;

[0030] in, This indicates the first time series data obtained from the field measurements after interpolation and filling. Data points With the Data points The growth rate data between;

[0031] The formula for calculating the interquartile range of the increase data is:

[0032] ;

[0033] in, This represents the interquartile range of the growth rate data. , These represent the values ​​at the top 25% and 75% respectively after sorting the growth rate data in ascending order;

[0034] like or If true, then determine This is an outlier.

[0035] Building upon the first aspect, further, by utilizing the Savitzky-Gore filtering algorithm and setting several filtering parameters, the continuous variables in the first-level data matrix are smoothed to generate the second-level data matrix, which includes:

[0036] Select from any column of continuous variables in the first-level data matrix 1 consecutive data point, forming a length of 1 Sliding window;

[0037] use A polynomial of order 1 is used to fit the data points within the sliding window, and the least squares method is used to solve for the fitting coefficients that minimize the residuals, thus obtaining the fitted coefficients. polynomial of order 1;

[0038] The center point of the sliding window is substituted into the fitting coefficients to determine... A polynomial of order 1 is used to calculate the smoothed value of the center point of the sliding window;

[0039] Move the sliding window and traverse all data points in the column of continuous variables. The smoothed value of the center point of the sliding window, calculated in each move, forms a new smoothed sequence of the column of continuous variables.

[0040] Iterate through each column of continuous variables, and construct a second-level data matrix from the new smoothed sequence of each column of continuous variables;

[0041] The order polynomial is:

[0042] ;

[0043] in, Indicates the first [number]th ... Fitted values ​​for each data point This indicates the sequential index of the data point within the sliding window. This represents the order of the polynomial, i.e., the total number of filter parameters. , , … for The fitting coefficients of the order polynomial to be solved;

[0044] The residual is:

[0045] ;

[0046] in, Represents the residual. Indicates the first [number]th ... The original values ​​of each data point This indicates the length of the sliding window.

[0047] Building upon the first aspect, further analysis is conducted using grey relational analysis, maximum mutual information analysis, and Spearman correlation coefficient analysis to perform correlation analysis on the three-level data matrix, resulting in a correlation characterization matrix including:

[0048] Based on a three-level data matrix, the correlation degree matrix factors are calculated using the grey relational model analysis method, the maximum mutual information analysis method, and the Spearman correlation coefficient analysis method, respectively.

[0049] The correlation degree characterization matrix is ​​composed of the correlation degree matrix factors corresponding to the grey relational model analysis method, the maximum mutual information number analysis method, and the Spearman correlation coefficient analysis method.

[0050] The factors used in calculating the correlation matrix using the grey relational analysis model include:

[0051] The monitoring data column representing the sinking attitude of the caisson in the three-level data matrix is ​​used as the reference sequence, and the other monitoring data columns and the construction excavation information data column are used as the comparison sequence.

[0052] Calculate the correlation coefficient between each comparison sequence and the reference sequence across all data points;

[0053] The correlation coefficient between each comparison sequence and the reference sequence is averaged across all data points to obtain the correlation degree between each comparison sequence and the reference sequence.

[0054] The correlation degree between each comparison sequence and the reference sequence is normalized to obtain the correlation degree matrix factor corresponding to the grey relational analysis method.

[0055] The factors for calculating the correlation matrix using the maximum mutual information analysis method include:

[0056] The reference sequence and each comparison sequence are arranged in ascending order. For each sequence pair consisting of the reference sequence and each comparison sequence, the joint probability distribution of each sequence pair is obtained by dividing the grid and calculating the distribution of data points in the grid cells.

[0057] The maximum mutual information of each sequence pair is determined by exhaustively searching different grid partitioning methods;

[0058] The maximum mutual information of each sequence pair is normalized to obtain the correlation matrix factor corresponding to the maximum mutual information analysis method;

[0059] The factors used in calculating the correlation matrix using Spearman correlation coefficient analysis include:

[0060] Calculate the Spearman rank correlation coefficient between each comparison sequence and the reference sequence;

[0061] The degree of association between each sequence pair is determined based on the Spearman rank correlation coefficient between each comparison sequence and the reference sequence.

[0062] Based on the degree of correlation between each sequence pair, the correlation matrix factors corresponding to the Spearman correlation coefficient analysis method are obtained.

[0063] Building upon the first aspect, further, using the correlation characterization matrix, key variables are selected from the original variables contained in the field measured time-series data and construction excavation information characterization matrix, constructing a four-level data matrix including:

[0064] The original variables that are identified as highly correlated in the grey relational analysis method, the maximum mutual information analysis method, and the Spearman correlation coefficient analysis method will be used as key variables;

[0065] And the original variables that are determined to be highly correlated under all filtering parameters in the grey relational model analysis method, the maximum mutual information number analysis method, or the Spearman correlation coefficient analysis method are used as key variables;

[0066] A four-level data matrix is ​​constructed from key variables;

[0067] In the grey relational analysis method, the original variables whose normalized correlation degree is greater than or equal to the correlation degree threshold are judged as original variables with high correlation.

[0068] In the maximum mutual information analysis method, the original variables whose normalized maximum mutual information is greater than or equal to the maximum mutual information threshold are identified as highly correlated original variables.

[0069] In Spearman correlation coefficient analysis, original variables whose absolute value of Spearman rank correlation coefficient is greater than or equal to the Spearman rank correlation coefficient threshold are considered to be highly correlated.

[0070] Building upon the first aspect, further, principal component analysis is used to reduce the dimensionality of the fourth-level data matrix, resulting in the following principal component matrices:

[0071] The fourth-level data matrix is ​​standardized column-wise to eliminate the influence of dimensions, resulting in a standardized fourth-level data matrix.

[0072] Calculate the covariance matrix of the standardized four-level data matrix, and solve for the eigenvalues ​​and eigenvectors of the covariance matrix;

[0073] Based on the eigenvalues ​​in descending order, select several principal components whose cumulative variance contribution rate is greater than or equal to the contribution rate threshold.

[0074] The standardized four-level data matrix is ​​projected onto the eigenspace of several principal components whose cumulative variance contribution rate is greater than or equal to the contribution rate threshold, thus obtaining the principal component matrix.

[0075] Secondly, the present invention provides a caisson sinking attitude prediction system, comprising:

[0076] The data acquisition module is used to acquire on-site measured time-series data during the sinking process of the caisson, and based on the construction procedures, constructs a construction excavation information representation matrix that reflects the excavation status of each compartment through the binary representation method.

[0077] The data analysis module is used to construct a primary data matrix based on field-measured time-series data; to smooth the continuous variables in the primary data matrix using the Savitzky-Gore filtering algorithm with several filtering parameters, generating a secondary data matrix; to construct a tertiary data matrix based on the secondary data matrix and the construction excavation information representation matrix; to perform correlation analysis on the tertiary data matrix using grey relational analysis, maximum mutual information analysis, and Spearman correlation coefficient analysis, obtaining a correlation degree representation matrix; to select key variables from the original variables contained in the field-measured time-series data and the construction excavation information representation matrix using the correlation degree representation matrix, constructing a quaternary data matrix; and to perform dimensionality reduction on the quaternary data matrix using principal component analysis, obtaining the principal component matrix.

[0078] The model training module is used to train the Long Short-Term Memory Network using the principal component matrix as training data, and obtain the trained Long Short-Term Memory Network as a caisson sinking attitude prediction model.

[0079] The attitude prediction module is used to predict the sinking attitude of the caisson using the sinking attitude prediction model, and obtain the sinking attitude prediction results.

[0080] Thirdly, the present invention provides a computer device, comprising:

[0081] Storage medium: used to store computer programs;

[0082] Processor: Used to execute the computer program to implement the caisson sinking attitude prediction method described in the first aspect.

[0083] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the caisson sinking attitude prediction method described in the first aspect.

[0084] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the caisson sinking attitude prediction method described in the first aspect.

[0085] Compared with the prior art, the beneficial effects of the present invention are:

[0086] The caisson sinking attitude prediction method provided by this invention combines on-site measured time-series data and construction excavation information representation matrix during the caisson sinking process. It integrates filtering methods to conduct correlation analysis, and uses grey relational model, maximum mutual information number, and Spearman correlation coefficient as the basic analysis methods to determine the correlation degree representation matrix. It then uses principal component analysis combined with long short-term memory network to construct a caisson sinking attitude prediction model for predicting the caisson sinking attitude. This method can solve the problems of low prediction accuracy and poor robustness caused by traditional caisson sinking prediction methods that ignore the dynamic process of construction and insufficient data utilization, thereby improving the accuracy and stability of deep-water caisson sinking attitude prediction. Attached Figure Description

[0087] Figure 1 is a flowchart of the caisson sinking attitude prediction method provided in an embodiment of the present invention. Detailed Implementation

[0088] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0089] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Unless otherwise specified, embodiments of the present invention and the technical features thereof can be combined with each other.

[0090] This invention provides a method for predicting the sinking attitude of a caisson, comprising:

[0091] Acquire on-site measured time-series data during the sinking process of the caisson, and based on the construction sequence, construct a construction excavation information representation matrix reflecting the excavation status of each compartment using the binary representation method;

[0092] A primary data matrix was constructed based on the on-site measured time-series data.

[0093] By using the Savitzky-Gore filtering algorithm and setting several filtering parameters, continuous variables in the first-level data matrix are smoothed to generate a second-level data matrix.

[0094] Based on the secondary data matrix and combined with the construction excavation information representation matrix, a tertiary data matrix is ​​constructed.

[0095] The grey relational analysis method, the maximum mutual information analysis method, and the Spearman correlation coefficient analysis method are used to conduct a correlation analysis on the three-level data matrix to obtain the correlation degree characterization matrix.

[0096] Using the correlation characterization matrix, key variables were selected from the original variables contained in the characterization matrix of the field measured time series data and the construction excavation information, and a four-level data matrix was constructed.

[0097] Principal component analysis was used to reduce the dimensionality of the four-level data matrix to obtain the principal component matrix.

[0098] Using the principal component matrix as training data, the long short-term memory network is trained to obtain a well-trained long short-term memory network, which serves as a prediction model for the sinking attitude of the caisson.

[0099] The sinking attitude prediction model of the caisson was used to predict the sinking attitude of the caisson, and the prediction results of the sinking attitude of the caisson were obtained.

[0100] The caisson sinking attitude prediction method provided in this invention combines on-site measured time-series data and construction excavation information representation matrix during the caisson sinking process. It integrates filtering methods to conduct correlation analysis, uses grey relational model, maximum mutual information number, and Spearman correlation coefficient as the basic analysis methods to determine the correlation degree representation matrix, and employs principal component analysis combined with long short-term memory network to construct a caisson sinking attitude prediction model for caisson sinking attitude prediction. This method can solve the problems of low prediction accuracy and poor robustness caused by traditional caisson sinking prediction methods that ignore the dynamic process of construction and insufficient data utilization, and improve the accuracy and stability of deep-water caisson sinking attitude prediction.

[0101] Figure 1 is a flowchart of the caisson sinking attitude prediction method provided in this embodiment. This flowchart only shows the logical order of the method in this embodiment. Under the premise that they do not conflict with each other, the steps shown or described can be completed in a different order than that shown in Figure 1.

[0102] The caisson sinking attitude prediction method provided in this embodiment can be applied to a terminal and can be executed by a caisson sinking attitude prediction system. This system can be implemented by software and / or hardware and can be integrated into the terminal, such as any tablet computer or computer device with communication function.

[0103] This invention provides a method for predicting the sinking attitude of a caisson, which specifically includes the following steps:

[0104] Step 1: Obtain on-site measured time-series data during the caisson sinking process, and based on the construction sequence, construct a construction excavation information representation matrix reflecting the excavation status of each compartment using the binary representation method;

[0105] In this embodiment, from the perspectives of sinking control and structural stability, real-time on-site measured data during the sinking process of the caisson are collected. The real-time on-site measured data is time series data with no limit on the types and quantity.

[0106] Specifically, the on-site measured time series data includes:

[0107] Data used to reflect the interaction between the caisson and the surrounding soil include earth pressure on the caisson sidewalls and reaction force at the caisson cutting edge;

[0108] Data used to reflect the internal mechanical state of the caisson structure, including stress at the root of the caisson cutting edge and stress in the concrete of the caisson body;

[0109] Data used to reflect the sinking environment and attitude of the caisson includes the water level in the observation well, the groundwater level, and the sinking depth.

[0110] In this embodiment, the construction excavation information representation matrix includes two aspects: the depth of penetration and the excavation scheme design. The excavation scheme design in the construction excavation information representation matrix is ​​achieved by numbering the caisson compartments and using binary representation to indicate the construction status of each caisson compartment at each time point.

[0111] Specifically, the caisson is a 3×4 specification, and the caisson compartments are numbered from 1 to 12; the construction status of each compartment is represented by binary method. In the construction excavation information representation matrix, 1 indicates that the compartment is in the excavation state, and 0 indicates that the compartment is in the stopped excavation state.

[0112] Step 2: Construct a primary data matrix based on the actual measured time-series data;

[0113] In this embodiment, constructing a primary data matrix based on field-measured time-series data specifically includes: preprocessing the field-measured time-series data, and using the preprocessed field-measured time-series data to form a primary data matrix.

[0114] In this embodiment, the preprocessing of the field measured time series data specifically includes the following steps:

[0115] Step 1: Interpolate and fill in the missing data in the field measured time series data;

[0116] Step 2: Calculate the increase data between consecutive data points in the field measured time series data after interpolation and fill, as well as the interquartile range of the increase data;

[0117] Specifically, the formula for calculating the increase data is as follows:

[0118] ;

[0119] in, This indicates the first time series data obtained from the field measurements after interpolation and filling. Data points With the Data points The growth rate data between them.

[0120] The formula for calculating the interquartile range of the increase data is:

[0121] ;

[0122] in, This represents the interquartile range of the growth rate data. , These represent the values ​​at the top 25% and 75% after the growth rate data is sorted in ascending order.

[0123] Step 3: Based on the increase data and the interquartile range of the increase data, outliers are selected from the interpolated and filled field measured time series data, and the outliers are replaced with the average value of their adjacent data points.

[0124] Specifically, if or If true, then determine This is an outlier.

[0125] After identifying outliers, specific analysis can be conducted in conjunction with the monitored objects. If the outlier is a real extreme case, the data point is retained; if the outlier is a meaningless anomaly, it is replaced with the average value of adjacent data points.

[0126] Step 3: Using the Savitzky-Gore filtering algorithm and setting several filtering parameters, smooth the continuous variables in the first-level data matrix to generate the second-level data matrix;

[0127] In this embodiment, the Savitzky-Gore filtering algorithm is used and several filtering parameters are set to smooth the continuous variables in the first-level data matrix to generate the second-level data matrix. The specific steps include the following:

[0128] Step 1: Select from any column of continuous variables in the first-level data matrix 1 consecutive data point, forming a length of 1 Sliding window;

[0129] Step 2: Utilize A polynomial of order 1 is used to fit the data points within the sliding window, and the least squares method is used to solve for the fitting coefficients that minimize the residuals, thus obtaining the fitted coefficients. polynomial of order 1;

[0130] Specifically, The order polynomial is:

[0131] ;

[0132] in, Indicates the first [number]th ... Fitted values ​​for each data point This indicates the sequential index of the data point within the sliding window. This represents the order of the polynomial, i.e., the total number of filter parameters. , , … for The fitting coefficients of the order polynomial to be solved.

[0133] The residual is:

[0134] ;

[0135] in, Represents the residual. Indicates the first [number]th ... The original values ​​of each data point This indicates the length of the sliding window.

[0136] Step 3: Substitute the center point of the sliding window into the values ​​determined by the fitting coefficients. A polynomial of order 1 is used to calculate the smoothed value of the center point of the sliding window;

[0137] Step 4: Move the sliding window and traverse all data points in the continuous variable column. The smoothed value of the center point of the sliding window calculated in each move constitutes a new smoothed sequence of the continuous variable column.

[0138] Step 5: Traverse each column of continuous variables and construct a second-level data matrix from the smoothed new sequence of each column of continuous variables.

[0139] In this embodiment, the Savitzky-Gore filtering algorithm is used, combined with the actual analysis requirements. Each filtering parameter is used to smooth the continuous variables in the first-level data matrix, and a second-level data matrix is ​​generated for each filtering parameter.

[0140] Specifically, settings The window sizes are 10, 20, and 30 respectively.

[0141] By minimizing the residuals using the least squares method, and setting the derivatives of each fitting coefficient to 0, we have:

[0142] ;

[0143] in, This represents the derivative of the fitting coefficients.

[0144] Once the smoothing coefficient and the size of the sliding window are determined, the data within the window can be imported into the above formula to obtain the fitting coefficient. The expression for the polynomial of order X can then be determined by substituting the coordinates of the point to be found into the expression. The smoothed value can be obtained by using a polynomial of order one, and then filtering calculations can be performed to obtain filtered data with different levels of noise removal. The smoothed continuous variable time series data is represented as follows: , , The continuous variables are recombined to form different analytical matrices.

[0145] Step 4: Based on the secondary data matrix and combined with the construction excavation information representation matrix, construct the tertiary data matrix;

[0146] Step 5: Using the grey relational analysis method, the maximum mutual information analysis method, and the Spearman correlation coefficient analysis method, perform correlation analysis on the three-level data matrix to obtain the correlation degree characterization matrix;

[0147] In this embodiment, the grey relational analysis method, the maximum mutual information analysis method, and the Spearman correlation coefficient analysis method are used to perform correlation analysis on the three-level data matrix to obtain the correlation degree characterization matrix. The specific steps include the following:

[0148] Step 1: Based on the three-level data matrix, calculate the correlation matrix factors using the grey relational analysis method, the maximum mutual information analysis method, and the Spearman correlation coefficient analysis method, respectively;

[0149] In this embodiment, the calculation of the correlation degree matrix factors using the grey relational model analysis method specifically includes the following steps:

[0150] Step ①: Use the monitoring data column representing the sinking attitude of the caisson in the three-level data matrix as the reference sequence, and the other monitoring data columns and the construction excavation information data column as the comparison sequence;

[0151] Step 2: Calculate the correlation coefficient between each comparison sequence and the reference sequence across all data points;

[0152] The formula for calculating the correlation coefficient is:

[0153] ;

[0154] in, Indicates the first The comparison sequence in the th ... The correlation coefficient between each data point and the reference sequence. Indicates the first In the comparison sequence, the th... Data points, Indicates the first in the reference sequence Data points, , These represent the two levels of minimum difference and the two levels of maximum difference, respectively, which are the minimum and maximum differences between all comparison sequences and the reference sequence across all data points. The resolution coefficient is a... The constant within the coefficient, usually taken as 0.5, is used to adjust for the degree of difference between correlation coefficients. The smaller the value, the stronger the amplification effect on differences and the greater the distinguishing ability.

[0155] Step 3: Take the average of the correlation coefficients between each comparison sequence and the reference sequence across all data points to obtain the correlation degree between each comparison sequence and the reference sequence;

[0156] Step 4: Normalize the correlation between each comparison sequence and the reference sequence to obtain the correlation matrix factors corresponding to the grey relational analysis method.

[0157] In this embodiment, the calculation of the correlation matrix factors using the maximum mutual information analysis method specifically includes the following steps:

[0158] Step ①: Sort the reference sequence and each comparison sequence in ascending order. For each sequence pair consisting of the reference sequence and each comparison sequence, obtain the joint probability distribution of each sequence pair by dividing the grid and calculating the distribution of data points in the grid cells.

[0159] Step 2: Determine the maximum mutual information number of each sequence pair by exhaustively searching different grid partitioning methods;

[0160] Step 3: Normalize the maximum mutual information of each sequence pair to obtain the correlation matrix factors corresponding to the maximum mutual information analysis method.

[0161] Specifically, define a The grid, in which one of the partitions divides the data points of the comparison sequence into... Another division divides the data points of the reference sequence into two parts. Find the probability distribution function of all cells in the grid and obtain the maximum mutual information number.

[0162] Since different grids lead to different probability distribution functions, the global optimal grid is locked by exhaustive search of the feature matrix, thereby determining the maximum mutual information number.

[0163] Normalizing the maximum mutual information of each sequence pair yields the correlation matrix factors corresponding to the maximum mutual information analysis method:

[0164] ;

[0165] in, This represents the factors of the correlation matrix corresponding to the maximum mutual information analysis method. Indicates the first The maximum mutual information number of sequence pairs. This represents the total number of sequence pairs.

[0166] In this embodiment, the calculation of the correlation matrix factors using the Spearman correlation coefficient analysis method specifically includes the following steps:

[0167] Step ①: Calculate the Spearman rank correlation coefficient between each comparison sequence and the reference sequence;

[0168] Specifically, the formula for calculating the Spearman rank correlation coefficient between each comparison sequence and the reference sequence is as follows:

[0169] ;

[0170] in, Indicates the first Spearman rank correlation coefficients between the comparison series and the reference series. Indicates the first The rank sequence corresponding to all data points in a comparison sequence after being sorted in ascending order. This represents the rank sequence corresponding to all data points in the reference sequence after being sorted in ascending order. , They represent , The sample standard deviation express and The sample covariance between them.

[0171] Step 2: Determine the degree of association between each sequence pair based on the Spearman rank correlation coefficient between each comparison sequence and the reference sequence;

[0172] Specifically, when When, it indicates the first The comparison sequence is highly correlated with the reference sequence;

[0173] when When, it indicates the first The comparison sequence is strongly correlated with the reference sequence;

[0174] when When, it indicates the first The comparison sequences showed a moderate correlation with the reference sequence;

[0175] when When, it indicates the first The comparison sequence showed a weak correlation with the reference sequence;

[0176] when When, it indicates the first The comparison sequences are either very weakly correlated or uncorrelated with the reference sequence.

[0177] Step 3: Based on the correlation degree of each sequence pair, obtain the correlation matrix factors corresponding to the Spearman correlation coefficient analysis method.

[0178] Step 2: The correlation degree characterization matrix is ​​constructed from the correlation degree matrix factors corresponding to the grey relational model analysis method, the maximum mutual information number analysis method, and the Spearman correlation coefficient analysis method.

[0179] Step 6: Using the correlation characterization matrix, key variables are selected from the original variables contained in the field measured time series data and construction excavation information characterization matrix to construct a four-level data matrix;

[0180] In this embodiment, key variables are selected from the original variables contained in the field measured time series data and construction excavation information representation matrix using the correlation degree representation matrix, and the construction of a four-level data matrix specifically includes the following steps:

[0181] Step 1: Select the original variables that are determined to be highly correlated in the grey relational model analysis, maximum mutual information analysis, and Spearman correlation coefficient analysis as key variables; and select the original variables that are determined to be highly correlated in the grey relational model analysis, maximum mutual information analysis, or Spearman correlation coefficient analysis under all filtering parameters as key variables.

[0182] In this embodiment, in the grey relational model analysis method, the original variables whose normalized correlation degree is greater than or equal to the correlation degree threshold are identified as highly correlated original variables; in the maximum mutual information number analysis method, the original variables whose normalized maximum mutual information number is greater than or equal to the maximum mutual information number threshold are identified as highly correlated original variables; in the Spearman correlation coefficient analysis method, the original variables whose absolute value of the Spearman rank correlation coefficient is greater than or equal to the Spearman rank correlation coefficient threshold are identified as highly correlated original variables.

[0183] Specifically, original variables with a normalized correlation coefficient greater than or equal to 0.7 are considered highly correlated; in the maximum mutual information analysis method, original variables with a normalized maximum mutual information coefficient greater than or equal to 0.7 are considered highly correlated; and in the Spearman correlation coefficient analysis method, original variables with an absolute value of Spearman rank correlation coefficient greater than or equal to 0.7 are considered highly correlated.

[0184] Step 2: Construct a four-level data matrix from key variables.

[0185] Step 7: Use principal component analysis to reduce the dimensionality of the fourth-level data matrix to obtain the principal component matrix;

[0186] In this embodiment, principal component analysis is used to reduce the dimensionality of the fourth-level data matrix to obtain the principal component matrix. The specific steps include the following:

[0187] Step 1: Standardize the fourth-level data matrix column by column to eliminate the influence of dimensions and obtain the standardized fourth-level data matrix;

[0188] Step 2: Calculate the covariance matrix of the standardized four-level data matrix, and solve for the eigenvalues ​​and eigenvectors of the covariance matrix;

[0189] Step 3: Select several principal components whose cumulative variance contribution rate is greater than or equal to the contribution rate threshold, based on the eigenvalues ​​in descending order.

[0190] Specifically, based on the eigenvalues ​​from largest to smallest, select the top eigenvalues ​​with a cumulative variance contribution rate of no less than 80%. Each principal component consists of a cumulative variance contribution rate of at least 80%. The eigenvectors of the principal components constitute the eigenspace.

[0191] Step 4: Project the standardized four-level data matrix onto the eigenspace composed of the eigenvectors of several principal components whose cumulative variance contribution rate is greater than or equal to the contribution rate threshold, to obtain the principal component matrix.

[0192] Step 8: Using the principal component matrix as training data, train the Long Short-Term Memory Network to obtain the trained Long Short-Term Memory Network, which serves as the caisson sinking attitude prediction model.

[0193] Step 9: Use the caisson sinking attitude prediction model to predict the caisson sinking attitude and obtain the caisson sinking attitude prediction results.

[0194] This invention provides a caisson sinking attitude prediction system, comprising:

[0195] The data acquisition module is used to acquire on-site measured time-series data during the sinking process of the caisson, and based on the construction procedures, constructs a construction excavation information representation matrix that reflects the excavation status of each compartment through the binary representation method.

[0196] The data analysis module is used to construct a primary data matrix based on field-measured time-series data; to smooth the continuous variables in the primary data matrix using the Savitzky-Gore filtering algorithm with several filtering parameters, generating a secondary data matrix; to construct a tertiary data matrix based on the secondary data matrix and the construction excavation information representation matrix; to perform correlation analysis on the tertiary data matrix using grey relational analysis, maximum mutual information analysis, and Spearman correlation coefficient analysis, obtaining a correlation degree representation matrix; to select key variables from the original variables contained in the field-measured time-series data and the construction excavation information representation matrix using the correlation degree representation matrix, constructing a quaternary data matrix; and to perform dimensionality reduction on the quaternary data matrix using principal component analysis, obtaining the principal component matrix.

[0197] The model training module is used to train the Long Short-Term Memory Network using the principal component matrix as training data, and obtain the trained Long Short-Term Memory Network as a caisson sinking attitude prediction model.

[0198] The attitude prediction module is used to predict the sinking attitude of the caisson using the sinking attitude prediction model, and obtain the sinking attitude prediction results.

[0199] The caisson sinking attitude prediction system provided in this embodiment of the invention can execute the caisson sinking attitude prediction method provided in this embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0200] This invention provides a computer device, comprising:

[0201] Storage medium: used to store computer programs;

[0202] Processor: Used to execute computer programs to implement the caisson sinking attitude prediction method provided in the embodiments of the present invention.

[0203] This invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the caisson sinking attitude prediction method provided in this invention.

[0204] This invention provides a computer program product, including a computer program that, when executed by a processor, implements the caisson sinking attitude prediction method provided in this invention.

[0205] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0206] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a system for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams.

[0207] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction system that implements the functions specified in one or more flowcharts and / or one or more block diagrams.

[0208] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0209] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting the sinking attitude of a caisson, characterized in that, include: The on-site measured time-series data during the sinking process of the caisson were obtained, and based on the construction sequence, a construction excavation information representation matrix reflecting the excavation status of each compartment was constructed using the binary representation method; a first-level data matrix was constructed based on the on-site measured time-series data; and the continuous variables in the first-level data matrix were smoothed by using the Savitzky-Gore filtering algorithm and setting several filtering parameters to generate a second-level data matrix. Based on the secondary data matrix and combined with the construction excavation information representation matrix, a tertiary data matrix is ​​constructed. The correlation analysis of the third-level data matrix was carried out using the grey relational analysis method, the maximum mutual information number analysis method, and the Spearman correlation coefficient analysis method to obtain the correlation degree characterization matrix. Using the correlation degree characterization matrix, key variables were selected from the original variables contained in the characterization matrix of the field measured time series data and the construction excavation information to construct the fourth-level data matrix. The fourth-level data matrix was then dimensionality-reduced using the principal component analysis method to obtain the principal component matrix. Using the principal component matrix as training data, a long short-term memory network is trained to obtain a trained long short-term memory network, which serves as a caisson sinking attitude prediction model. The caisson sinking attitude prediction model is then used to predict the caisson sinking attitude, and the prediction results are obtained.

2. The method for predicting the sinking attitude of a caisson according to claim 1, characterized in that, The on-site measured time-series data include: data reflecting the interaction between the caisson and the surrounding soil, including earth pressure on the caisson sidewalls and reaction force at the caisson cutting edge; data reflecting the internal mechanical state of the caisson structure, including stress at the root of the caisson cutting edge and stress in the concrete of the caisson body; and data reflecting the sinking environment and attitude of the caisson, including water level in the observation well, groundwater level, and sinking depth.

3. The method for predicting the sinking attitude of a caisson according to claim 1, characterized in that, In the construction excavation information representation matrix, 1 indicates that the compartment is in the excavation state, and 0 indicates that the compartment is in the stopped excavation state.

4. The method for predicting the sinking attitude of a caisson according to claim 1, characterized in that, Based on the field-measured time-series data, the construction of a primary data matrix includes: preprocessing the field-measured time-series data, with the preprocessed data forming the primary data matrix; the preprocessing includes: interpolating and imputing missing data in the field-measured time-series data; calculating the amplification data and interquartile range of consecutive data points in the interpolated and imputed field-measured time-series data; based on the amplification data and the interquartile range, outliers are selected from the interpolated and imputed field-measured time-series data, and these outliers are replaced with the average of their adjacent data points; the formula for calculating the amplification data is: ;in, This indicates the first time series data obtained from the field measurements after interpolation and filling. Data points With the Data points The growth rate data between; the formula for calculating the interquartile range of the growth rate data is: ;in, This represents the interquartile range of the growth rate data. 、 These represent the values ​​at the top 25% and 75% after sorting the growth rate data in ascending order, respectively; if or If true, then determine This is an outlier.

5. The method for predicting the sinking attitude of a caisson according to claim 1, characterized in that, Using the Savitzky-Gore filtering algorithm and setting several filtering parameters, the continuous variables in the first-level data matrix are smoothed to generate the second-level data matrix. This process involves selecting from any column of continuous variables in the first-level data matrix. 1 consecutive data point, forming a length of 1 Sliding window; using A polynomial of order 1 is used to fit the data points within the sliding window, and the least squares method is used to solve for the fitting coefficients that minimize the residuals, thus obtaining the fitted coefficients. The order polynomial; the center point of the sliding window is substituted into the fitting coefficients to determine the order. The first order polynomial is used to calculate the smoothed value of the center point of the sliding window; the sliding window is moved to traverse all data points in the column of continuous variables, and the smoothed value of the center point of the sliding window calculated in each move constitutes a new smoothed sequence of the column of continuous variables; the smoothed sequences of each column of continuous variables are used to construct a second-level data matrix. The order polynomial is: ;in, Indicates the first [number]th ... Fitted values ​​for each data point This indicates the sequential index of the data point within the sliding window. This represents the order of the polynomial, i.e., the total number of filter parameters. 、 、 、…、 for The fitting coefficients of the order polynomial to be solved; the residuals are: ;in, Represents the residual. Indicates the first [number]th ... The original values ​​of each data point This indicates the length of the sliding window.

6. The method for predicting the sinking attitude of a caisson according to claim 1, characterized in that, The correlation analysis of the three-level data matrix was conducted using the grey relational model analysis, maximum mutual information number analysis, and Spearman correlation coefficient analysis to obtain the correlation degree characterization matrix. This included: calculating the correlation degree matrix factors based on the three-level data matrix using the grey relational model analysis, maximum mutual information number analysis, and Spearman correlation coefficient analysis respectively; constructing the correlation degree characterization matrix from the correlation degree matrix factors corresponding to the grey relational model analysis, maximum mutual information number analysis, and Spearman correlation coefficient analysis; calculating the correlation degree matrix factors using the grey relational model analysis included: using the monitoring data column representing the caisson sinking attitude in the three-level data matrix as the reference sequence, and the other monitoring data columns and construction excavation information data columns as comparison sequences; calculating the correlation coefficient between each comparison sequence and the reference sequence at all data points; averaging the correlation coefficients of each comparison sequence and the reference sequence at all data points to obtain the correlation degree between each comparison sequence and the reference sequence; and comparing each comparison sequence with the reference sequence... The correlation degree is normalized to obtain the correlation matrix factor corresponding to the grey relational analysis method. The correlation matrix factor is calculated using the maximum mutual information analysis method, which includes: arranging the reference sequence and each comparison sequence in ascending order; for each sequence pair formed by the reference sequence and each comparison sequence, dividing the data into grids and calculating the distribution of data points in the grid cells to obtain the joint probability distribution of each sequence pair; determining the maximum mutual information number of each sequence pair by exhaustively searching different grid division methods; normalizing the maximum mutual information number of each sequence pair to obtain the correlation matrix factor corresponding to the maximum mutual information analysis method. The correlation matrix factor is calculated using the Spearman correlation coefficient analysis method, which includes: calculating the Spearman rank correlation coefficient between each comparison sequence and the reference sequence; determining the correlation degree of each sequence pair based on the Spearman rank correlation coefficient; and obtaining the correlation matrix factor corresponding to the Spearman correlation coefficient analysis method based on the correlation degree of each sequence pair.

7. The method for predicting the sinking attitude of a caisson according to claim 1, characterized in that, Using a correlation degree representation matrix, key variables are selected from the original variables contained in the field measured time series data and construction excavation information representation matrix, and a four-level data matrix is ​​constructed. This includes: original variables that are determined to have high correlation in all three methods (grey relational model analysis, maximum mutual information number analysis, and Spearman correlation coefficient analysis) are considered key variables; and original variables that are determined to have high correlation in all three methods (grey relational model analysis, maximum mutual information number analysis, or Spearman correlation coefficient analysis) under all filtering parameters are considered key variables. The four-level data matrix is ​​constructed from these key variables. In the grey relational model analysis, original variables with a normalized correlation degree greater than or equal to the correlation degree threshold are considered highly correlated. In the maximum mutual information number analysis, original variables with a normalized maximum mutual information number greater than or equal to the maximum mutual information number threshold are considered highly correlated. In the Spearman correlation coefficient analysis, original variables with an absolute value of the Spearman rank correlation coefficient greater than or equal to the Spearman rank correlation coefficient threshold are considered highly correlated.

8. The method for predicting the sinking attitude of a caisson according to claim 1, characterized in that, Principal component analysis (PCA) is used to reduce the dimensionality of the four-level data matrix to obtain the principal component matrix. This process involves: standardizing the four-level data matrix column-wise to eliminate the influence of dimensions, resulting in a standardized four-level data matrix; calculating the covariance matrix of the standardized four-level data matrix and solving for its eigenvalues ​​and eigenvectors; selecting several principal components whose cumulative variance contribution rate is greater than or equal to a contribution rate threshold based on the eigenvalues ​​in descending order; and projecting the standardized four-level data matrix onto the eigenspace formed by the eigenvectors of these principal components whose cumulative variance contribution rate is greater than or equal to the contribution rate threshold, thus obtaining the principal component matrix.

9. A caisson sinking attitude prediction system, characterized in that, include: The data acquisition module is used to acquire on-site measured time-series data during the caisson sinking process, and based on the construction sequence, constructs a construction excavation information representation matrix reflecting the excavation status of each compartment using a binary representation method; the data analysis module is used to construct a primary data matrix based on the on-site measured time-series data; and uses the Savitzky-Gore filtering algorithm and sets several filtering parameters to smooth the continuous variables in the primary data matrix to generate a secondary data matrix. Based on the secondary data matrix and combined with the construction excavation information representation matrix, a tertiary data matrix is ​​constructed. The correlation analysis of the third-level data matrix was carried out using the grey relational analysis method, the maximum mutual information number analysis method, and the Spearman correlation coefficient analysis method to obtain the correlation degree characterization matrix. Using the correlation degree characterization matrix, key variables were selected from the original variables contained in the characterization matrix of the field measured time series data and the construction excavation information to construct the fourth-level data matrix. The fourth-level data matrix was then dimensionality-reduced using the principal component analysis method to obtain the principal component matrix. The model training module is used to train the Long Short-Term Memory Network using the principal component matrix as training data, and obtain the trained Long Short-Term Memory Network as a caisson sinking attitude prediction model. The attitude prediction module is used to predict the sinking attitude of the caisson using the sinking attitude prediction model, and obtain the sinking attitude prediction results.

10. A computer device, characterized in that, Includes: Storage media: used to store computer programs; Processor: Used to execute the computer program to implement the caisson sinking attitude prediction method according to any one of claims 1 to 8.

Images