A geotechnical engineering investigation data analysis method and system based on edge computing

Abstract

The application discloses a kind of geotechnical engineering investigation data analysis method and system based on edge computing, comprising the following steps: collecting geotechnical engineering investigation data and preprocessing;Sparse representation base is constructed;Adaptive observation matrix is generated;Compressed sampling processing is carried out to real-time collected geotechnical investigation data, and compressed observation data are obtained;Sparse reconstruction processing is carried out to compressed observation data using CoSaMP algorithm, and reconstruction geotechnical investigation data sequence is obtained;Stratigraphic structure evolution characteristics are extracted, and stratigraphic structure evolution characteristic vector is constructed;Stratigraphic state discrimination processing is carried out, and stratigraphic state discrimination result is obtained;Sparse representation base and adaptive observation matrix are updated, and compressed observation data, reconstruction geotechnical investigation data sequence and stratigraphic state discrimination result are uploaded to cloud server for collaborative processing, efficient, accurate real-time data analysis and processing are realized, and the accuracy and real-time performance of investigation data analysis are improved.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering investigation data analysis, and in particular to a geotechnical engineering investigation data analysis method and system based on edge computing. Background Technology

[0002] With the continuous development of infrastructure construction in my country, geotechnical engineering investigation and analysis has become a crucial link in ensuring project quality and safety. Geotechnical engineering investigation mainly includes the measurement and analysis of various aspects such as soil mechanical properties, groundwater level, and stratigraphic structure. These data provide important basis for engineering design and construction. However, existing geotechnical engineering investigation technologies still have certain shortcomings in data acquisition, processing, and analysis, affecting the accuracy and real-time nature of the investigation results. Traditional geotechnical engineering investigation data processing methods usually rely on ground monitoring instruments and manual analysis, resulting in slow data transmission speeds, low analysis efficiency, and susceptibility to external environmental influences, leading to significant errors in the investigation data.

[0003] In the field of geotechnical engineering investigation, many technical challenges remain in data acquisition and processing, particularly in areas such as data compression, sparse reconstruction, and analysis. Currently, most geotechnical investigation data processing relies on centralized computing platforms, where the real-time performance and accuracy are often limited by computing power. Traditional methods often overlook how to improve the real-time performance and accuracy of data processing through edge computing. Therefore, how to efficiently process large amounts of geotechnical engineering investigation data in real time and adjust data analysis strategies according to actual conditions has become a core issue that urgently needs to be addressed. Summary of the Invention

[0004] One objective of this invention is to propose a method and system for analyzing geotechnical engineering investigation data based on edge computing. This invention fully utilizes edge computing, sparse representation bases, and the CoSaMP algorithm, detailing the process of real-time acquisition, processing, and analysis of geotechnical engineering investigation data through edge computing nodes. It also employs a sparse reconstruction method for efficient processing of compressed observation data, possessing advantages such as high real-time data processing performance, high computational efficiency, high accuracy, and strong adaptability.

[0005] A geotechnical engineering investigation data analysis method based on edge computing according to an embodiment of the present invention includes the following steps: Collect and preprocess geotechnical engineering investigation data, and then transmit the preprocessed geotechnical engineering investigation data to the edge computing node; Based on the preprocessed geotechnical engineering investigation data, stratigraphic structure description parameters are extracted, and a sparse representation basis is constructed in the edge computing nodes; An adaptive observation matrix is ​​generated in the edge computing nodes based on sparse representation base and stratigraphic structure description parameters; Compressed observation data is obtained by performing compressed sampling processing on the real-time acquired geotechnical investigation data using an adaptive observation matrix. Based on a sparse representation basis, the CoSaMP algorithm is used to perform sparse reconstruction processing on compressed observation data in edge computing nodes to obtain a reconstructed geotechnical investigation data sequence. Based on the reconstructed geotechnical exploration data sequence, stratigraphic structure evolution features are extracted, and stratigraphic structure evolution feature vectors are constructed. Based on the stratigraphic structure evolution feature vector, stratigraphic state discrimination processing is performed in the edge computing nodes to obtain the stratigraphic state discrimination result; The sparse representation base and adaptive observation matrix are updated based on the stratigraphic state discrimination results, and the compressed observation data, reconstructed geotechnical exploration data sequence and stratigraphic state discrimination results are uploaded to the cloud server for collaborative processing.

[0006] Optionally, at the geotechnical engineering investigation site, borehole parameters, pore water pressure, soil stress, and displacement data at different investigation locations and times are collected using borehole detection devices, pore water pressure sensors, soil pressure sensors, displacement monitoring sensors, and connected data acquisition terminals to form geotechnical engineering investigation data. The geotechnical engineering investigation data is then preprocessed, including data denoising, outlier removal, missing value completion, time synchronization, and data standardization, to obtain preprocessed geotechnical engineering investigation data. The preprocessed geotechnical engineering investigation data is then transmitted to the edge computing node via a wireless communication module.

[0007] Optionally, the construction of the sparse representation basis specifically includes: The preprocessed geotechnical engineering investigation data are arranged sequentially according to the sampling time sequence to construct a data matrix composed of column vectors of geotechnical engineering investigation data corresponding to the sampling time. The data matrix is ​​divided into sliding segments according to a preset window length to obtain a local data window; Extracting stratigraphic descriptive parameters based on local data windows; Build a basic atomic library; Calculate the atomic weight coefficients corresponding to each basic atom based on the stratigraphic structure description parameters; A weight matrix is ​​constructed based on the atomic weight coefficients corresponding to each basic atom, and a corresponding sparse representation basis is generated based on the basic atom library and the weight matrix.

[0008] Optionally, the generation of the adaptive observation matrix specifically includes: The compressed sampling dimension is determined by setting a preset compressed sampling ratio, and an initial observation matrix is ​​constructed. Calculate the structural modulation coefficients corresponding to each stratigraphic structural description parameter based on the stratigraphic structural description parameters; Construct the modulation matrix based on the structural modulation coefficients; The initial observation matrix is ​​weighted and modulated using the modulation matrix to obtain the modulated observation matrix; An adaptive observation matrix is ​​obtained by matching and adjusting the modulation observation matrix using a sparse representation basis.

[0009] Optionally, obtaining the compressed observation data specifically includes: The edge computing node receives real-time collected geotechnical engineering investigation data and constructs the geotechnical engineering investigation data corresponding to the current sampling time into a real-time sampling data vector. The adaptive observation matrix is ​​used to perform a linear projection operation on the real-time sampled data vector to obtain the compressed observation vector corresponding to the current sampling time; Compressed observation data is constructed by arranging the compressed observation vectors corresponding to consecutive sampling times according to the sampling time sequence.

[0010] Optionally, obtaining the reconstructed geotechnical investigation data sequence specifically includes: In the edge computing node, the compressed observation data is processed by the CoSaMP algorithm to perform sparse reconstruction. The compressed observation data is used as the current residual data. The initial state of the support index set, the initial value of the iteration count and the target sparsity are set. The sparse reconstruction iterative solution environment of the CoSaMP algorithm is established. The CoSaMP algorithm includes a candidate support identification unit, a constrained least squares estimation unit and a sparse pruning update unit. In the candidate support identification unit, the correlation between each column of the adaptive observation matrix and the current residual data is calculated, and a preset number of candidate indices with the highest correlation are selected. The candidate indices are then merged with the current support index set to obtain the candidate support index set. In the constrained least squares estimation unit, corresponding columns are extracted from the adaptive observation matrix based on the candidate support index set to construct a candidate observation submatrix. Under the constraint of the candidate observation submatrix, Tikhonov regularized least squares estimation is performed on the compressed observation data to obtain candidate sparse coefficients. In the sparse pruning update unit, the candidate sparse coefficients are sorted by absolute value, the coefficient component corresponding to the target sparsity with the largest absolute value is retained, the other coefficient components are removed, and the updated support index set is obtained according to the index corresponding to the retained coefficient components. Based on the updated support index set, the corresponding columns in the adaptive observation matrix are re-extracted, an updated observation sub-matrix is ​​constructed, and Tikhonov regularized least squares estimation is re-executed based on the updated observation sub-matrix and compressed observation data to obtain the reconstruction coefficient results corresponding to the current iteration. The difference between the current reconstruction result and the compressed observation data is calculated based on the reconstruction coefficient result corresponding to the current iteration, and the updated residual data is obtained. When the iteration count of the updated residual data reaches the preset upper limit, the final sparse coefficient result is output, and the final sparse coefficient result is mapped and reconstructed using the sparse representation basis to obtain the reconstructed geotechnical exploration data sequence.

[0011] Optionally, the construction of the stratigraphic structure evolution feature vector specifically includes: The reconstructed geotechnical investigation data sequence is arranged sequentially according to the sampling time order to construct a data matrix; Based on the preset time window length, the data matrix is ​​divided into sliding segments to obtain a local time window; For each local time window, extract the gradient change characteristics of the stratigraphic structure evolution; For each local time window, extract the data dispersion characteristics; For each local time window, calculate the corresponding correlation characteristics; For each local time window, extract the distribution entropy feature; The gradient change features, data dispersion features, correlation features, and distribution entropy features extracted from each local time window are concatenated in a unified order to form a stratigraphic evolution feature vector for the local time window.

[0012] Optionally, obtaining the formation state determination result specifically includes: In the edge computing nodes, the stratigraphic state deviation is calculated based on the difference between the stratigraphic structure evolution feature vector at the current sampling time and the reference stratigraphic structure evolution feature vector. Define a first state threshold and a second state threshold in the edge computing node; In the edge computing node, the formation state deviation is compared with the first state threshold and the second state threshold. The formation state at the current sampling time is determined based on the magnitude of the formation state deviation, and a formation state discrimination result is formed.

[0013] Optionally, the updating of the sparse representation basis and the adaptive observation matrix specifically includes: Based on the stable state, transitional state and anomalous state in the stratigraphic state discrimination results, the corresponding sparse representation base is updated; The adaptive observation matrix is ​​updated based on the stratigraphic state discrimination results; The updated sparse representation base and the updated adaptive observation matrix are uploaded to the cloud server for collaborative processing.

[0014] According to an embodiment of the present invention, a geotechnical engineering investigation data analysis system based on edge computing includes: The data acquisition module is used to collect geotechnical engineering investigation data and perform preprocessing to generate preprocessed geotechnical engineering investigation data. The sparse representation basis construction module is used to extract stratigraphic structure description parameters through edge computing nodes and construct sparse representation basis. The adaptive observation matrix module is used to generate an adaptive observation matrix based on a sparse representation base and stratigraphic structure description parameters. The compressed sampling module is used to perform compressed sampling processing on the real-time acquired geotechnical investigation data using an adaptive observation matrix to obtain compressed observation data; The reconstruction module is used to perform sparse reconstruction processing on compressed observation data based on sparse representation basis and CoSaMP algorithm to obtain reconstructed geotechnical exploration data sequence. The feature extraction module is used to extract stratigraphic evolution features based on the reconstructed geotechnical exploration data sequence and construct stratigraphic evolution feature vectors. The state discrimination module is used to perform stratigraphic state discrimination processing based on the stratigraphic structure evolution feature vector to obtain the stratigraphic state discrimination result; The update module is used to update the sparse representation base and adaptive observation matrix based on the stratigraphic state discrimination results, and upload the compressed observation data, reconstructed geotechnical exploration data sequence and stratigraphic state discrimination results to the cloud server for collaborative processing.

[0015] The beneficial effects of this invention are: This invention, by introducing edge computing technology, shifts data processing from traditional centralized computing platforms to edge computing nodes, effectively solving the problems of low computational efficiency, large data transmission latency, and poor real-time performance in existing geotechnical engineering investigation data processing methods. Edge computing nodes can perform real-time processing near the data acquisition point, reducing time consumption during data transmission and improving the real-time performance and accuracy of data analysis. This innovative design enables more flexible and timely response to changes in the field when processing and analyzing geotechnical engineering investigation data, providing more reliable real-time data support for engineering decision-making.

[0016] This invention employs sparse representation bases and the CoSaMP algorithm for compressed sampling and sparse reconstruction of geotechnical engineering investigation data, significantly reducing the computational load and improving processing efficiency. Traditional methods suffer from massive computational demands and long processing times when handling large-scale data. However, through sparse reconstruction technology, this invention can reduce unnecessary computational work and quickly obtain the desired results while maintaining data reconstruction accuracy. This not only improves computational efficiency but also reduces the system load and avoids resource waste.

[0017] By dynamically updating the formation state discrimination results, the system of this invention can adjust the sparse representation basis and adaptive observation matrix according to the actual situation, thereby adapting to data changes under different formation states. This dynamic adjustment mechanism enables the system to be flexibly optimized according to different geotechnical engineering investigation conditions, further improving the accuracy and reliability of data analysis.

[0018] This invention significantly improves the real-time performance, accuracy, and efficiency of geotechnical engineering investigation data analysis through edge computing technology and advanced data reconstruction methods, providing more intelligent and efficient technical support for the safety and stability of geotechnical engineering. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a geotechnical engineering investigation data analysis method and system based on edge computing proposed in this invention; Figure 2 This is a schematic diagram illustrating the construction of a sparse representation basis for a geotechnical engineering investigation data analysis method and system based on edge computing proposed in this invention. Figure 3 This is a schematic diagram illustrating the construction of a reconstructed geotechnical investigation data sequence based on an edge computing-based geotechnical engineering investigation data analysis method and system proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figures 1-3 A method for analyzing geotechnical engineering investigation data based on edge computing includes the following steps: Collect and preprocess geotechnical engineering investigation data, and then transmit the preprocessed geotechnical engineering investigation data to the edge computing node; Based on the preprocessed geotechnical engineering investigation data, stratigraphic structure description parameters are extracted, and a sparse representation basis is constructed in the edge computing nodes; An adaptive observation matrix is ​​generated in the edge computing nodes based on sparse representation base and stratigraphic structure description parameters; Compressed observation data is obtained by performing compressed sampling processing on the real-time acquired geotechnical investigation data using an adaptive observation matrix. Based on a sparse representation basis, the CoSaMP algorithm is used to perform sparse reconstruction processing on compressed observation data in edge computing nodes to obtain a reconstructed geotechnical investigation data sequence. Based on the reconstructed geotechnical exploration data sequence, stratigraphic structure evolution features are extracted, and stratigraphic structure evolution feature vectors are constructed. Based on the stratigraphic structure evolution feature vector, stratigraphic state discrimination processing is performed in the edge computing nodes to obtain the stratigraphic state discrimination result; The sparse representation base and adaptive observation matrix are updated based on the stratigraphic state discrimination results, and the compressed observation data, reconstructed geotechnical exploration data sequence and stratigraphic state discrimination results are uploaded to the cloud server for collaborative processing.

[0022] In this embodiment, at the geotechnical engineering investigation site, borehole detection devices, pore water pressure sensors, soil pressure sensors, displacement monitoring sensors, and connected data acquisition terminals are used to collect borehole parameters, pore water pressure, soil stress, and displacement data at different investigation locations and times, forming geotechnical engineering investigation data. Preprocessing is performed on the geotechnical engineering investigation data, including data denoising, outlier removal, missing value completion, time synchronization, and data standardization, to obtain preprocessed geotechnical engineering investigation data. The preprocessed geotechnical engineering investigation data is then transmitted to the edge computing node via a wireless communication module.

[0023] In this embodiment, the construction of the sparse representation base specifically includes: The preprocessed geotechnical engineering investigation data are arranged sequentially according to the sampling time sequence, and a data matrix is ​​constructed consisting of column vectors of geotechnical engineering investigation data corresponding to the sampling time. The number of rows in the data matrix represents the feature dimension of a single sampling, and the number of columns represents the sampling time sequence length. The data matrix is ​​divided into sliding segments according to a preset window length to obtain a local data window; Stratigraphic structure description parameters are extracted based on local data windows. These parameters include average gradient change, data dispersion, average correlation, and distribution entropy. The average gradient change is obtained by accumulating the absolute values ​​of the differences between adjacent sampling times for each feature within the local data window and normalizing them according to the feature dimension and window length. The data dispersion is obtained by calculating the cumulative squared deviations of all data within the local data window relative to the mean of the local data window and normalizing them according to the feature dimension and window length. The average correlation is obtained by calculating the correlation coefficient between any two different features within the local data window and averaging the correlation coefficients of all features. The distribution entropy is obtained by dividing the data within the local data window into intervals, calculating the data distribution probability of each interval, and then calculating the entropy value of the data distribution probability of each interval. Build a basic atomic library; The process of constructing the basic atom library is as follows: local data windows are clustered and divided according to the similarity of stratigraphic structure description parameters to obtain multiple data categories corresponding to different stratigraphic structure patterns; local data windows that characterize the typical variation features of the data category are selected from each data category as the corresponding basic atoms, and the basic atoms corresponding to each data category are summarized to construct the basic atom library. The atomic weight coefficients corresponding to each basic atom are calculated based on the stratigraphic structure description parameters. The specific calculation process of the atomic weight coefficients is as follows: calculate the distance between the current stratigraphic structure description parameter and the center of the structural parameter corresponding to each basic atom, generate the unnormalized weight of each basic atom based on the distance, accumulate the unnormalized weights of all basic atoms, and divide the unnormalized weight of each basic atom by the accumulated result to obtain the atomic weight coefficients corresponding to each basic atom. A weight matrix is ​​constructed based on the atomic weight coefficients corresponding to each basic atom. The weight matrix is ​​a diagonal matrix formed with the atomic weight coefficients corresponding to each basic atom as diagonal elements. A sparse representation basis is generated based on the basic atom library and the weight matrix. The specific process of generating the sparse representation basis is as follows: a weight matrix is ​​constructed based on the atomic weight coefficients corresponding to each basic atom; the weight matrix is ​​used to weight each basic atom in the basic atom library to obtain a weighted set of basic atoms that matches the current stratigraphic structure description parameters; the weighted basic atoms are combined according to the arrangement order in the basic atom library to generate the corresponding sparse representation basis.

[0024] In this embodiment, the generation of the adaptive observation matrix specifically includes: The compressed sampling dimension is determined by setting a preset compressed sampling ratio, and an initial observation matrix is ​​constructed. The process of constructing the initial observation matrix is ​​as follows: the matrix size of the initial observation matrix is ​​determined based on the compressed sampling dimension and the characteristic dimension of the geotechnical engineering investigation data. The number of rows of the initial observation matrix is ​​set to the compressed sampling dimension, and the number of columns is set to the characteristic dimension of the geotechnical engineering investigation data. Matrix elements are generated according to the determined matrix size, and each matrix element is normalized. The normalized matrix is ​​then used as the initial observation matrix. Calculate the structural modulation coefficients corresponding to each stratigraphic structural description parameter based on the stratigraphic structural description parameters; The calculation process of the structure modulation coefficient is as follows: read the average gradient change, data dispersion, average correlation and distribution entropy in the stratigraphic structure description parameters; perform dimensionless normalization on the average gradient change, data dispersion, average correlation and distribution entropy respectively to obtain the corresponding normalized parameter values; divide each normalized parameter value by the sum of all normalized parameter values ​​to obtain the structure modulation coefficient corresponding to each stratigraphic structure description parameter. A modulation matrix is ​​constructed based on the structural modulation coefficients. The modulation matrix is ​​a diagonal matrix formed with the structural modulation coefficients corresponding to the structural description parameters of each stratum as diagonal elements. The initial observation matrix is ​​weighted and modulated using the modulation matrix to obtain the modulated observation matrix; The process of obtaining the modulation observation matrix is ​​as follows: read the initial observation matrix and the modulation matrix, and establish a one-to-one correspondence between each diagonal element in the modulation matrix and each row of the initial observation matrix; use each diagonal element in the modulation matrix to perform weighted adjustment on the matrix elements of each corresponding row of the initial observation matrix, and rearrange the matrix elements of each row after weighted adjustment according to the original row and column order in the initial observation matrix to obtain the modulation observation matrix. An adaptive observation matrix is ​​obtained by matching and adjusting the modulation observation matrix using a sparse representation basis; The matching adjustment process is as follows: determine the arrangement order and corresponding weight distribution of each basic atom in the sparse representation basis; based on the distribution characteristics of each basic atom in the sparse representation basis, adjust the sampling direction and sampling intensity of each row in the modulation observation matrix row by row, enhance the response of high-weight basic atoms in the sparse representation basis, and weaken the response of low-weight basic atoms; recombine the adjusted rows according to the original matrix structure to obtain an adaptive observation matrix that matches the sparse representation basis.

[0025] In this embodiment, obtaining compressed observation data specifically includes: The edge computing node receives real-time collected geotechnical engineering investigation data and constructs the geotechnical engineering investigation data corresponding to the current sampling time into a real-time sampling data vector. The adaptive observation matrix is ​​used to perform a linear projection operation on the real-time sampled data vector to obtain the compressed observation vector corresponding to the current sampling time. The compressed observation vector is obtained by multiplying the adaptive observation matrix with the real-time sampled data vector. Compressed observation data is constructed by arranging the compressed observation vectors corresponding to consecutive sampling times according to the sampling time sequence. The specific process of constructing compressed observation data is as follows: read the compressed observation vectors corresponding to each sampling time in sequence according to the sampling time sequence, and arrange the compressed observation vectors corresponding to each sampling time as column vectors in the compressed observation data matrix in sequence. The number of rows in the compressed observation data matrix corresponds to the compressed sampling dimension, and the number of columns corresponds to the sampling time sequence length. After arranging the compressed observation vectors corresponding to all sampling times within the preset time sequence length, the compressed observation data is obtained.

[0026] In this embodiment, the specific steps for reconstructing the geotechnical investigation data sequence include: In edge computing nodes, the CoSaMP algorithm is used to perform sparse reconstruction processing on compressed observation data. The compressed observation data is used as the current residual data. The initial state of the support index set, the initial value of the iteration count and the target sparsity are set. The sparse reconstruction iterative solution environment of the CoSaMP algorithm is established. The CoSaMP algorithm includes a candidate support identification unit, a constrained least squares estimation unit and a sparse pruning update unit. The process of setting the initial state of the support index set, the initial value of the iteration count, and the target sparsity is as follows: Read the compressed sampling dimension, the number of columns in the adaptive observation matrix, and the total number of basic atoms in the sparse representation basis corresponding to the compressed observation data in the edge computing node; initialize the support index set to an empty set, indicating that no index corresponding to any basic atom is selected at the start of the current iteration; set the initial value of the iteration count to zero, as the initial iteration round for the CoSaMP algorithm's sparse reconstruction processing; determine the target sparsity based on the total number of basic atoms in the sparse representation basis, the compressed sampling dimension, and the preset compressed sampling ratio, and set the target sparsity as the upper limit of the number of coefficient components allowed to be retained in each iteration. In the candidate support identification unit, the correlation between each column of the adaptive observation matrix and the current residual data is calculated, and a preset number of candidate indices with the highest correlation are selected. The candidate indices are then merged with the current support index set to obtain the candidate support index set. The process of obtaining the candidate support index set is as follows: Read all column vectors in the adaptive observation matrix and calculate the correlation between each column vector and the current residual data to obtain the correlation value for each column vector; sort the column vectors according to their absolute values ​​from largest to smallest, and select the indices corresponding to the first preset number of column vectors to form candidate indices; read the current support index set and merge the candidate indices with the indices in the current support index set; perform duplicate index removal and sorting on all merged indices to obtain the candidate support index set. In the constrained least squares estimation unit, corresponding columns are extracted from the adaptive observation matrix based on the candidate support index set to construct a candidate observation submatrix. Under the constraint of the candidate observation submatrix, Tikhonov regularized least squares estimation is performed on the compressed observation data to obtain candidate sparse coefficients. The process of obtaining candidate sparse coefficients is as follows: Based on the candidate support index set, determine the column vectors in the adaptive observation matrix that correspond one-to-one with each candidate support index, and extract the corresponding column vectors according to the order of the candidate support indices in the candidate support index set, combining them to form a candidate observation submatrix; use the compressed observation data as the observation target vector, and use the coefficients corresponding to each column in the candidate observation submatrix as the parameters to be solved; establish a Tikhonov normal coefficient model that minimizes the sum of squared deviations between the compressed observation data and the product of the candidate observation submatrix and the parameters to be solved, while ensuring that the sum of squared deviations of the parameters to be solved is subject to regularization constraints. The equation is solved using least squares. A first intermediate matrix is ​​obtained by multiplying the transpose of the candidate observation submatrix with the candidate observation submatrix. The first intermediate matrix is ​​then weighted and summed with the identity matrix according to preset regularization parameters to obtain a regularization coefficient matrix. The regularization coefficient matrix is ​​inverted, and the inverted result is then multiplied sequentially with the transpose of the candidate observation submatrix and the compressed observation data to obtain the coefficient values ​​corresponding to each candidate support index. The coefficient values ​​corresponding to each candidate support index are arranged in order of their index positions in the candidate support index set to form candidate sparse coefficients. In the sparse pruning update unit, the candidate sparse coefficients are sorted by absolute value, the coefficient component corresponding to the target sparsity with the largest absolute value is retained, the other coefficient components are removed, and the updated support index set is obtained according to the index corresponding to the retained coefficient components. Based on the updated support index set, the corresponding columns in the adaptive observation matrix are extracted again, an updated observation submatrix is ​​constructed, and Tikhonov regularized least squares estimation is re-executed based on the updated observation submatrix and compressed observation data to obtain the reconstruction coefficient results corresponding to the current iteration. The difference between the current reconstruction result and the compressed observation data is calculated based on the reconstruction coefficient result corresponding to the current iteration, and the updated residual data is obtained. The process of obtaining the updated residual data is as follows: Read the reconstruction coefficient results corresponding to the current iteration, and determine the one-to-one correspondence between each coefficient component in the reconstruction coefficient results and the corresponding column vector of the adaptive observation matrix based on the updated support index set; use each coefficient component in the reconstruction coefficient results as a weighting coefficient for the corresponding column vector, and perform a weighted summation on each column vector in the adaptive observation matrix corresponding to the updated support index set to obtain the current reconstruction result; subtract the current reconstruction result from the compressed observation data element by element to obtain the residual components corresponding to each compressed sampling dimension; combine all residual components according to the original arrangement order in the compressed observation data to obtain the updated residual data. When the iteration count of the updated residual data reaches the preset upper limit, the final sparse coefficient result is output, and the sparse representation basis is used to map and reconstruct the final sparse coefficient result to obtain the reconstructed geotechnical investigation data sequence. The mapping and reconstruction process is as follows: Each basic atom in the sparse representation basis is read, and a one-to-one correspondence is established between each basic atom and each coefficient component in the final sparse coefficient result according to the original arrangement order in the sparse representation basis; each coefficient component in the final sparse coefficient result is used as a scaling factor for the corresponding basic atom, and each basic atom is weighted to obtain a weighted result; the weighted results of all basic atoms are accumulated according to the original arrangement order in the sparse representation basis to generate a reconstruction result vector corresponding to the compressed observation data; the reconstruction result vectors corresponding to each sampling time are arranged sequentially according to the sampling time sequence to obtain the reconstructed geotechnical exploration data sequence.

[0027] In this embodiment, the construction of the stratigraphic structure evolution feature vector specifically includes: The reconstructed geotechnical investigation data sequence is arranged sequentially according to the sampling time order to construct a data matrix; The data matrix construction process is as follows: The reconstructed geotechnical investigation data sequence is arranged according to the sampling time sequence to obtain a sequence composed of reconstructed data vectors corresponding to each sampling time; the reconstructed data vectors corresponding to each sampling time are placed as columns in the data matrix according to the sampling time sequence, with each column representing the reconstructed data at one sampling time, and the number of columns being the length of the sampling time sequence; each eigenvalue in the reconstructed data vector corresponding to each sampling time is arranged sequentially in the corresponding row of the data matrix to obtain the data matrix. Based on the preset time window length, the data matrix is ​​divided into sliding segments to obtain a local time window; The sliding segmentation process is as follows: the number of sampling moments contained in each local time window is determined according to the preset time window length; starting from the first column of the data matrix, each time window is slid sequentially with the time window length as the step size to obtain the reconstructed data vector corresponding to the continuous sampling moments in each local time window; during each slide, the starting position of the window will move backward by one sampling moment until the entire data matrix is ​​divided into multiple local time windows. For each local time window, the gradient change features of the stratigraphic structure evolution are extracted. The gradient change features are obtained by accumulating the absolute values ​​of the differences between each feature at adjacent sampling times within the local time window and normalizing them according to the feature dimension and window length. For each local time window, the data dispersion feature is extracted. The data dispersion feature is obtained by calculating the sum of squared deviations of all data within the local time window relative to the mean of the data within the window, and normalizing the value according to the feature dimension and the window length. For each local time window, the corresponding correlation feature is calculated. The correlation is obtained by calculating the correlation coefficient between any two different features within the local time window and averaging the correlation coefficients. For each local time window, the distribution entropy feature is extracted. The distribution entropy feature is obtained by dividing the local data into intervals, calculating the distribution probability of the data in each interval, and then calculating the entropy value of the data distribution probability in each interval. The gradient change features, data dispersion features, correlation features, and distribution entropy features extracted from each local time window are concatenated in a unified order to form a stratigraphic evolution feature vector for the local time window.

[0028] In this embodiment, obtaining the formation state determination result specifically includes: In the edge computing node, based on the difference between the stratigraphic structure evolution feature vector at the current sampling time and the reference stratigraphic structure evolution feature vector, the stratigraphic state deviation is calculated. The stratigraphic deviation is the weighted sum of each feature difference and the corresponding weight coefficient, which represents the degree of deviation between the current sampling time and the reference stratigraphic state. Define a first state threshold and a second state threshold in the edge computing node; The definition process of the first and second state thresholds is as follows: In the edge computing node, based on historical stratigraphic data, the reasonable range of the two thresholds is determined by analyzing the deviation range of stratigraphic structure evolution characteristics under different stratigraphic states; representative data samples of stable, transitional, and anomalous states in historical stratigraphic states are selected, and the distribution range of the deviation values ​​corresponding to the samples is calculated; according to the deviation distribution of the samples, the first state threshold is set as the lower limit of the deviation value range, and the first state threshold corresponds to the boundary condition for classifying the stratigraphic state as a stable state; the second state threshold is set as the upper limit of the deviation value range, and the second state threshold corresponds to the boundary condition for classifying the stratigraphic state as an anomalous state; the interval between the first and second state thresholds corresponds to the transition region of the stratigraphic state. In the edge computing node, the formation state deviation is compared with the first state threshold and the second state threshold. The formation state at the current sampling time is determined based on the magnitude of the formation state deviation, forming a formation state discrimination result. The specific discrimination process is as follows: when the deviation is less than or equal to the first state threshold, it is judged as a stable state; when the deviation is greater than the first state threshold and less than or equal to the second state threshold, it is judged as a transitional state; when the deviation is greater than the second state threshold, it is judged as an abnormal state.

[0029] In this embodiment, the updating of the sparse representation basis and the adaptive observation matrix specifically includes: Based on the stable state, transitional state, and anomalous state in the formation state discrimination results, the corresponding sparse representation base is updated. The update process includes: when the state is determined to be stable, the current sparse representation base remains unchanged; when the state is determined to be transitional, fine-tuning is performed based on the current sparse representation base, adjusting the weight coefficients in the sparse representation base; when the state is determined to be anomalous, the sparse representation base is reconstructed, new basic atoms are selected for combination, and the sparse representation base is retrained based on the new formation state data. The adaptive observation matrix is ​​updated based on the formation state discrimination results. The update process includes: weighting and adjusting the observation matrix elements corresponding to the features of stable, transitional, and anomalous states in the formation state discrimination results; maintaining the original structure of the adaptive observation matrix in the stable state; adjusting the weights of some elements in the adaptive observation matrix according to the changes in formation structure in the transitional state, with the adjustment magnitude based on the degree of formation change, using a linear change method to increase or decrease the weight coefficient of each matrix element by a certain proportion; and adjusting the weights of each element in the adaptive observation matrix by introducing the weights of new observation features and using a nonlinear adjustment method to change the weights of the matrix elements in the anomalous state. The updated sparse representation base and the updated adaptive observation matrix are uploaded to the cloud server for collaborative processing.

[0030] A geotechnical engineering investigation data analysis system based on edge computing includes: The data acquisition module is used to collect geotechnical engineering investigation data and perform preprocessing to generate preprocessed geotechnical engineering investigation data. The sparse representation basis construction module is used to extract stratigraphic structure description parameters through edge computing nodes and construct sparse representation basis. The adaptive observation matrix module is used to generate an adaptive observation matrix based on a sparse representation base and stratigraphic structure description parameters. The compressed sampling module is used to perform compressed sampling processing on the real-time acquired geotechnical investigation data using an adaptive observation matrix to obtain compressed observation data; The reconstruction module is used to perform sparse reconstruction processing on compressed observation data based on sparse representation basis and CoSaMP algorithm to obtain reconstructed geotechnical exploration data sequence. The feature extraction module is used to extract stratigraphic evolution features based on the reconstructed geotechnical exploration data sequence and construct stratigraphic evolution feature vectors. The state discrimination module is used to perform stratigraphic state discrimination processing based on the stratigraphic structure evolution feature vector to obtain the stratigraphic state discrimination result; The update module is used to update the sparse representation base and adaptive observation matrix based on the stratigraphic state discrimination results, and upload the compressed observation data, reconstructed geotechnical exploration data sequence and stratigraphic state discrimination results to the cloud server for collaborative processing.

[0031] Example 1: Taking the geotechnical engineering investigation of a large-scale infrastructure project as an example, this invention applies the edge computing-based geotechnical engineering investigation data analysis method and system to solve the problems of poor real-time data processing, low computational efficiency, and insufficient analysis accuracy in traditional geotechnical investigation. The project requires geotechnical investigation of a large tunnel located underground in the city, necessitating high-precision real-time monitoring of the strata along the tunnel route, especially in terms of groundwater level, soil stress, and stratum changes, to ensure construction safety and stability.

[0032] In this scenario, geotechnical investigation data is collected using devices such as borehole probes, pore water pressure sensors, earth pressure sensors, and displacement monitoring sensors. All collected data is preprocessed and then transmitted to edge computing nodes via wireless communication modules. During data transmission, the sensors record data every hour and transmit it to the edge computing nodes in real time via wireless communication. Traditional geotechnical engineering investigation methods rely on centralized computing platforms for data analysis, which suffers from problems such as transmission delays and insufficient processing capacity. Especially when dealing with massive amounts of data, the computing platform is overloaded, making real-time analysis and rapid feedback impossible. This invention, by moving data processing to edge computing nodes, processes the collected data in real time, enabling rapid responses, reducing latency, and improving system responsiveness.

[0033] In edge computing nodes, processed data is categorized into different stratigraphic states, and the compressed sampled data is sparsely reconstructed using the CoSaMP algorithm to reconstruct the geotechnical exploration data sequence and further extract stratigraphic structure evolution characteristics. These characteristics include gradient variation, data dispersion, correlation, and distribution entropy, all of which play a crucial role in determining the stratigraphic state. By comparing the determination results of different stratigraphic states, the system dynamically updates the sparse representation basis and adaptive observation matrix according to the preset target sparsity and stratigraphic state to adapt to different data acquired in real time.

[0034] Specific data processing and analysis include discrimination processing under different stratigraphic states. For stable stratigraphic states, the system maintains the current sparse representation basis unchanged and performs routine data processing based on the characteristics of the stratigraphy. For transitional states, the system fine-tunes the observation matrix, adjusting some of its weights to adapt to new stratigraphic changes. When an abnormal state is identified, the system reconstructs the sparse representation basis and retrains based on the new stratigraphic state data.

[0035] Table 1 Comparison of the data processing effects before and after the present invention method

[0036] As shown in Table 1, the application of this invention significantly improves the efficiency and accuracy of geotechnical engineering investigation data processing. Regarding data processing latency, traditional methods require 3 hours to complete, while the method of this invention can complete processing within 15 minutes, greatly improving the real-time performance of data analysis. Especially in complex geological environments, timely data processing is crucial for engineering safety. Real-time analysis through edge computing nodes enables immediate feedback of investigation results, providing reliable data support for subsequent engineering design and construction decisions.

[0037] In terms of accuracy, the sparse reconstruction accuracy and formation state discrimination accuracy of the method of this invention are significantly higher than those of traditional methods, reaching 98% and 95% respectively, while the sparse reconstruction accuracy and formation state discrimination accuracy of traditional methods are only 92% and 89%. This difference indicates that the present invention can more accurately reflect the true situation of the formation, especially in complex environments, it can better identify the changing trends of the formation, thereby improving the accuracy and reliability of exploration.

[0038] This invention, by introducing edge computing technology and sparse reconstruction algorithms, successfully solves the problems of prolonged data processing time, low computational efficiency, and poor accuracy in existing technologies, providing a real-time, accurate, and efficient data analysis solution for geotechnical engineering investigation. Comparative studies show that data processing after implementing this invention achieves significant improvements in time, accuracy, and efficiency, greatly optimizing the entire process of data analysis in geotechnical engineering investigation.

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for analyzing geotechnical engineering investigation data based on edge computing, characterized in that, Includes the following steps: Collect and preprocess geotechnical engineering investigation data, and then transmit the preprocessed geotechnical engineering investigation data to the edge computing node; Based on the preprocessed geotechnical engineering investigation data, stratigraphic structure description parameters are extracted, and a sparse representation basis is constructed in the edge computing nodes; An adaptive observation matrix is ​​generated in the edge computing nodes based on sparse representation base and stratigraphic structure description parameters; Compressed observation data is obtained by performing compressed sampling processing on the real-time acquired geotechnical investigation data using an adaptive observation matrix. Based on a sparse representation basis, the CoSaMP algorithm is used to perform sparse reconstruction processing on compressed observation data in edge computing nodes to obtain a reconstructed geotechnical investigation data sequence. Based on the reconstructed geotechnical exploration data sequence, stratigraphic structure evolution features are extracted, and stratigraphic structure evolution feature vectors are constructed. Based on the stratigraphic structure evolution feature vector, stratigraphic state discrimination processing is performed in the edge computing nodes to obtain the stratigraphic state discrimination result; The sparse representation base and adaptive observation matrix are updated based on the stratigraphic state discrimination results, and the compressed observation data, reconstructed geotechnical exploration data sequence and stratigraphic state discrimination results are uploaded to the cloud server for collaborative processing. The construction of the sparse representation basis specifically includes: The preprocessed geotechnical engineering investigation data are arranged sequentially according to the sampling time sequence to construct a data matrix composed of column vectors of geotechnical engineering investigation data corresponding to the sampling time. The data matrix is ​​divided into sliding segments according to a preset window length to obtain a local data window; Extracting stratigraphic descriptive parameters based on local data windows; Based on the similarity of stratigraphic structure description parameters, local data windows are clustered to obtain multiple data categories corresponding to different stratigraphic structure patterns. Local data windows that represent typical variation characteristics of each data category are selected as the corresponding basic atoms. The basic atoms corresponding to each data category are summarized to construct a basic atom library. Calculate the distance between the current stratigraphic structure description parameter and the center of the corresponding structural parameter of each basic atom. Generate the unnormalized weight of each basic atom based on the distance. Accumulate the unnormalized weights of all basic atoms and divide the unnormalized weight of each basic atom by the accumulated result to obtain the atomic weight coefficient corresponding to each basic atom. A weight matrix is ​​constructed based on the atomic weight coefficients corresponding to each basic atom. The weight matrix is ​​a diagonal matrix formed with the atomic weight coefficients corresponding to each basic atom as diagonal elements. A sparse representation basis is generated based on the basic atom library and the weight matrix. The specific process of generating the sparse representation basis is as follows: a weight matrix is ​​constructed based on the atomic weight coefficients corresponding to each basic atom; the weight matrix is ​​used to weight each basic atom in the basic atom library to obtain a weighted set of basic atoms that matches the current stratigraphic structure description parameters; the weighted basic atoms are combined according to the arrangement order in the basic atom library to generate the corresponding sparse representation basis. The generation of the adaptive observation matrix specifically includes: The compressed sampling dimension is determined by setting a preset compressed sampling ratio, and an initial observation matrix is ​​constructed. Read the average gradient change, data dispersion, average correlation, and distribution entropy from the stratigraphic structure description parameters; perform dimensionless normalization on the average gradient change, data dispersion, average correlation, and distribution entropy respectively to obtain the corresponding normalized parameter values; divide each normalized parameter value by the sum of all normalized parameter values ​​to obtain the structural modulation coefficients corresponding to each stratigraphic structure description parameter. A modulation matrix is ​​constructed based on the structural modulation coefficients. The modulation matrix is ​​a diagonal matrix formed with the structural modulation coefficients corresponding to the structural description parameters of each stratum as diagonal elements. The initial observation matrix is ​​weighted and modulated using the modulation matrix to obtain the modulated observation matrix; An adaptive observation matrix is ​​obtained by matching and adjusting the modulation observation matrix using a sparse representation basis; The updating of the sparse representation basis and the adaptive observation matrix specifically includes: Based on the stable state, transitional state and anomalous state in the stratigraphic state discrimination results, the corresponding sparse representation base is updated; The adaptive observation matrix is ​​updated based on the stratigraphic state discrimination results; The updated sparse representation base and the updated adaptive observation matrix are uploaded to the cloud server for collaborative processing.

2. The geotechnical engineering investigation data analysis method based on edge computing according to claim 1, characterized in that, At the geotechnical engineering investigation site, borehole detection devices, pore water pressure sensors, soil pressure sensors, displacement monitoring sensors, and connected data acquisition terminals are used to collect borehole parameters, pore water pressure, soil stress, and displacement data at different investigation locations and times, forming geotechnical engineering investigation data. The geotechnical engineering investigation data is then preprocessed, including data denoising, outlier removal, missing value completion, time synchronization, and data standardization, to obtain preprocessed geotechnical engineering investigation data. The preprocessed geotechnical engineering investigation data is then transmitted to the edge computing node via a wireless communication module.

3. The method for analyzing geotechnical engineering investigation data based on edge computing according to claim 1, characterized in that, The acquisition of the compressed observation data specifically includes: The edge computing node receives real-time collected geotechnical engineering investigation data and constructs the geotechnical engineering investigation data corresponding to the current sampling time into a real-time sampling data vector. The adaptive observation matrix is ​​used to perform a linear projection operation on the real-time sampled data vector to obtain the compressed observation vector corresponding to the current sampling time; Compressed observation data is constructed by arranging the compressed observation vectors corresponding to consecutive sampling times according to the sampling time sequence.

4. The geotechnical engineering investigation data analysis method based on edge computing according to claim 1, characterized in that, The specific steps involved in obtaining the reconstructed geotechnical investigation data sequence are as follows: In the edge computing node, the compressed observation data is processed by the CoSaMP algorithm to perform sparse reconstruction. The compressed observation data is used as the current residual data. The initial state of the support index set, the initial value of the iteration count and the target sparsity are set. The sparse reconstruction iterative solution environment of the CoSaMP algorithm is established. The CoSaMP algorithm includes a candidate support identification unit, a constrained least squares estimation unit and a sparse pruning update unit. In the candidate support identification unit, the correlation between each column of the adaptive observation matrix and the current residual data is calculated, and a preset number of candidate indices with the highest correlation are selected. The candidate indices are then merged with the current support index set to obtain the candidate support index set. In the constrained least squares estimation unit, corresponding columns are extracted from the adaptive observation matrix based on the candidate support index set to construct a candidate observation submatrix. Under the constraint of the candidate observation submatrix, Tikhonov regularized least squares estimation is performed on the compressed observation data to obtain candidate sparse coefficients. In the sparse pruning update unit, the candidate sparse coefficients are sorted by absolute value, the coefficient component corresponding to the target sparsity with the largest absolute value is retained, the other coefficient components are removed, and the updated support index set is obtained according to the index corresponding to the retained coefficient components. Based on the updated support index set, the corresponding columns in the adaptive observation matrix are re-extracted, an updated observation sub-matrix is ​​constructed, and Tikhonov regularized least squares estimation is re-executed based on the updated observation sub-matrix and compressed observation data to obtain the reconstruction coefficient results corresponding to the current iteration. The difference between the current reconstruction result and the compressed observation data is calculated based on the reconstruction coefficient result corresponding to the current iteration, and the updated residual data is obtained. When the iteration count of the updated residual data reaches the preset upper limit, the final sparse coefficient result is output, and the final sparse coefficient result is mapped and reconstructed using the sparse representation basis to obtain the reconstructed geotechnical exploration data sequence.

5. The method for analyzing geotechnical engineering investigation data based on edge computing according to claim 1, characterized in that, The construction of the stratigraphic structure evolution feature vector specifically includes: The reconstructed geotechnical investigation data sequence is arranged sequentially according to the sampling time order to construct a data matrix; Based on the preset time window length, the data matrix is ​​divided into sliding segments to obtain a local time window; For each local time window, extract the gradient change characteristics of the stratigraphic structure evolution; For each local time window, extract the data dispersion characteristics; For each local time window, calculate the corresponding correlation characteristics; For each local time window, extract the distribution entropy feature; The gradient change features, data dispersion features, correlation features, and distribution entropy features extracted from each local time window are concatenated in a unified order to form a stratigraphic evolution feature vector for the local time window.

6. The method for analyzing geotechnical engineering investigation data based on edge computing according to claim 1, characterized in that, The specific steps involved in obtaining the formation state determination result are as follows: In the edge computing nodes, the stratigraphic state deviation is calculated based on the difference between the stratigraphic structure evolution feature vector at the current sampling time and the reference stratigraphic structure evolution feature vector. Define a first state threshold and a second state threshold in the edge computing node; In the edge computing node, the formation state deviation is compared with the first state threshold and the second state threshold. The formation state at the current sampling time is determined based on the magnitude of the formation state deviation, and a formation state discrimination result is formed.

7. A geotechnical engineering investigation data analysis system based on edge computing, executing the geotechnical engineering investigation data analysis method based on edge computing as described in any one of claims 1 to 6, characterized in that, include: The data acquisition module is used to collect geotechnical engineering investigation data and perform preprocessing to generate preprocessed geotechnical engineering investigation data. The sparse representation basis construction module is used to extract stratigraphic structure description parameters through edge computing nodes and construct sparse representation basis. The adaptive observation matrix module is used to generate an adaptive observation matrix based on a sparse representation base and stratigraphic structure description parameters. The compressed sampling module is used to perform compressed sampling processing on the real-time acquired geotechnical investigation data using an adaptive observation matrix to obtain compressed observation data; The reconstruction module is used to perform sparse reconstruction processing on compressed observation data based on sparse representation basis and CoSaMP algorithm to obtain reconstructed geotechnical exploration data sequence. The feature extraction module is used to extract stratigraphic evolution features based on the reconstructed geotechnical exploration data sequence and construct stratigraphic evolution feature vectors. The state discrimination module is used to perform stratigraphic state discrimination processing based on the stratigraphic structure evolution feature vector to obtain the stratigraphic state discrimination result; The update module is used to update the sparse representation base and adaptive observation matrix based on the stratigraphic state discrimination results, and upload the compressed observation data, reconstructed geotechnical exploration data sequence and stratigraphic state discrimination results to the cloud server for collaborative processing.

Images