Geological settlement monitoring data feature extraction method, system, equipment and medium

Abstract

The invention discloses a geological settlement monitoring data feature extraction method, system and device, and a medium. The method comprises the following steps: obtaining an instantaneous monitoring sequence in a monitoring time window; decomposing the instantaneous monitoring sequence into a plurality of intrinsic mode functions by adopting a first-level decomposition algorithm; classifying the plurality of decomposed intrinsic mode functions, and identifying a plurality of mode feature types; calculating the plurality of modal feature types to obtain corresponding preset type features; and inputting the obtained features of the preset type into a downstream processing model as independent features. According to the method, a technical route of decomposition first and calculation second is introduced, a first-level decomposition algorithm is adopted to deconstruct a mixed original instantaneous monitoring sequence into a plurality of independent intrinsic mode functions representing long-term trend, periodic fluctuation and sudden disturbance respectively, and then characteristics are calculated on pure modes which are clear in physical significance and independent from one another respectively.

Description

Technical Field

[0001] This invention relates to the field of geological subsidence observation technology, and in particular to a method, system, equipment and medium for extracting features from geological subsidence monitoring data. Background Technology

[0002] In fields such as structural health monitoring and geological disaster early warning, it is often necessary to analyze monitoring data over long periods of time to extract key features that reflect changes in the system's state. These features are typically used as input for subsequent machine learning or deep learning models to perform state assessments or risk predictions.

[0003] Using linear fitting of data within a sliding time window to extract and analyze features such as "settlement trend" and its corresponding "abnormal fluctuations" can reflect the long-term development trend and short-term drastic fluctuations of the data. The "settlement trend" can be represented by the "slope of the linear fitting of data within the window," while "abnormal fluctuations" can be represented by the "variance or standard deviation of data within the window." However, this method of directly calculating features on the original mixed signal has inherent technical contradictions. For example, some schemes use "abnormal fluctuations" to calculate a confidence weight and use this weight to adjust the "settlement trend feature value." The logic is that when fluctuations are large, the confidence of the trend feature should decrease.

[0004] This logic fails in certain scenarios. For example, a real and severe geological subsidence event may simultaneously exhibit a significant trend and strong fluctuations in monitoring data. In this case, an increased "abnormal fluctuation characteristic value" might incorrectly lower the confidence weight, weakening the identification of this real subsidence event and potentially leading to missed reports, i.e., insufficient robustness. On the other hand, for a very weak and gradually changing early subsidence, its "abnormal fluctuation characteristic value" is small, and its confidence weight is high, but its "subsidence trend characteristic value" is also very small. Even after weighting, it may not be enough to trigger an early warning, reflecting a lack of sensitivity to details. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides a method, system, equipment, and medium for extracting features from geological subsidence monitoring data to address the problems of existing detection and monitoring methods being limited by timing, requiring advance cooling preparation of the unit, resulting in high time and material costs; and the detection results being static data, which cannot predict wall thickness changes and lifespan in real time, nor provide corresponding visual early warnings.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for feature extraction from geological subsidence monitoring data, comprising the following steps: acquiring an instantaneous monitoring sequence within a monitoring time window; decomposing the instantaneous monitoring sequence into multiple intrinsic mode functions using a first-order decomposition algorithm; classifying the multiple decomposed intrinsic mode functions to identify multiple modal feature types; obtaining corresponding preset type features by calculating the multiple modal feature types; and inputting the obtained preset type features as independent features into a downstream processing model.

[0008] As a preferred embodiment of the geological subsidence monitoring data feature extraction method of the present invention, the step of decomposing the instantaneous monitoring sequence into multiple intrinsic mode functions includes: decomposing the instantaneous monitoring sequence into a preset number of intrinsic mode functions by solving a constrained variational problem through a first-order decomposition algorithm, until the sum of the bandwidths of each intrinsic mode function is minimized and the sum of each intrinsic mode function can reconstruct the instantaneous monitoring sequence.

[0009] The beneficial effects of this preferred technical solution are as follows: by solving the constrained variational problem, the sum of the bandwidths of each intrinsic mode function can be minimized, so that each mode component after decomposition has a clear frequency band boundary and avoids mode aliasing.

[0010] As a preferred embodiment of the geological subsidence monitoring data feature extraction method of the present invention, the method further includes, before using the first-level decomposition algorithm, using a genetic algorithm to optimize the modal decomposition number and the secondary penalty factor of the first-level decomposition algorithm; wherein, the fitness function of the genetic algorithm is to minimize the sum of the envelope spectrum entropy of all intrinsic modal functions after decomposition.

[0011] As a preferred embodiment of the geological subsidence monitoring data feature extraction method of the present invention, the step of optimizing the modal decomposition number and the secondary penalty factor of the first-level decomposition algorithm includes: encoding the modal decomposition number and the secondary penalty factor as chromosomes of a genetic algorithm and initializing the population; evaluating each chromosome in the population according to the fitness function; performing selection, crossover, and mutation operations on the population to generate a new generation population; determining whether the termination condition is met; if not, returning to the step of evaluating according to the fitness function; if met, outputting the current optimal modal decomposition number and secondary penalty factor.

[0012] The beneficial effects of this preferred technical solution are as follows: by optimizing the modal decomposition number and the secondary penalty factor through a genetic algorithm, and using the sum of the envelope spectrum entropy as the fitness function, the optimal combination of decomposition parameters can be found for different instantaneous monitoring sequences, so that each intrinsic mode function after decomposition has the strongest sparsity and the highest purity.

[0013] As a preferred embodiment of the geological subsidence monitoring data feature extraction method of the present invention, the step of classifying the decomposed intrinsic mode functions includes: calculating the center frequency of each intrinsic mode function and calculating the Spearman rank correlation coefficient between each intrinsic mode function and the time vector; identifying the intrinsic mode function with the lowest center frequency and the absolute value of the Spearman rank correlation coefficient greater than a preset trend threshold as a trend mode; calculating the normalized Fourier spectrum of the remaining intrinsic mode functions and calculating the spectral kurtosis, identifying the intrinsic mode functions with the spectral kurtosis greater than a preset periodicity threshold as periodic modes; and identifying the remaining unclassified intrinsic mode functions as anomalous modes.

[0014] The beneficial effects of this preferred technical solution are as follows: trend modes are identified by center frequency and Spearman-level correlation coefficient, periodic modes are identified by normalized Fourier spectrum and spectral kurtosis, and the remaining unclassified intrinsic mode functions are identified as anomalous modes. This classification method is based on clear physical quantification indicators and can accurately separate signal components from different physical sources.

[0015] As a preferred embodiment of the geological subsidence monitoring data feature extraction method of the present invention, the step of calculating multiple modal feature types includes: calculating a first preset type feature through the trend mode, wherein the first preset type feature is used to characterize the long-term change trend of the instantaneous monitoring sequence; and calculating a second preset type feature through the abnormal mode, wherein the second preset type feature is used to characterize the sudden disturbance of the instantaneous monitoring sequence.

[0016] The beneficial effects of this preferred technical solution are as follows: by calculating the corresponding preset type features of the classified trend mode and abnormal mode respectively, the long-term trend features are extracted only from the trend mode without being affected by sudden disturbances, and the sudden disturbance features are extracted only from the abnormal mode, so that the disturbance intensity can be accurately quantified.

[0017] As a preferred embodiment of the geological subsidence monitoring data feature extraction method of the present invention, the modal feature types include trend modes, periodic modes, and abnormal modes.

[0018] In a second aspect, the present invention provides a geological subsidence monitoring data feature extraction system, comprising: a sequence acquisition module, used to acquire instantaneous monitoring sequences within a monitoring time window; The signal decomposition module is used to decompose the instantaneous monitoring sequence into multiple intrinsic mode functions using a first-level decomposition algorithm; The modality classification module is used to classify the multiple intrinsic mode functions decomposed and identify multiple modality feature types; The feature calculation module is used to obtain the corresponding preset type features by calculating multiple modal feature types; The feature output module is used to input the obtained preset type features as independent features into the downstream processing model.

[0019] Thirdly, the present invention provides an electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the geological subsidence monitoring data feature extraction method.

[0020] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the geological subsidence monitoring data feature extraction method.

[0021] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention introduces a decomposition-then-calculation approach, employing a first-level decomposition algorithm to deconstruct the mixed original instantaneous monitoring sequence into multiple independent intrinsic mode functions (EMFs) representing long-term trends, periodic fluctuations, and sudden disturbances. Then, features are calculated separately on these physically distinct and independent pure modes. This approach fundamentally solves the problem of feature interference caused by the aliasing of different components in the original signal. Trend features are extracted only from trend modes and are no longer affected by sudden disturbances, while abnormal features are extracted only from abnormal modes, enabling more accurate quantification of disturbance intensity.

[0022] This invention achieves both high sensitivity to weak signal variations and strong robustness to drastically disturbed environments, overcoming the contradiction between robustness and detail sensitivity in feature extraction caused by signal aliasing in existing technologies. By optimizing decomposition parameters using a genetic algorithm and classifying intrinsic mode functions based on physical quantification indices, the accuracy, reliability, and physical interpretability of features are improved, thereby enhancing the predictive performance of downstream processing models. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the overall process of a geological subsidence monitoring data feature extraction method according to an embodiment of the present invention. Detailed Implementation

[0025] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0026] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for feature extraction from geological subsidence monitoring data is provided, comprising the following steps: S100: Obtain the instantaneous monitoring sequence within the monitoring time window.

[0027] S200. The instantaneous monitoring sequence is decomposed into multiple intrinsic mode functions using a first-level decomposition algorithm.

[0028] S300. Classify the multiple intrinsic mode functions decomposed and identify multiple modal feature types.

[0029] S400. By calculating multiple modal feature types, the corresponding preset type features are obtained.

[0030] S500: The obtained preset type features are input as independent features into the downstream processing model.

[0031] It should be noted that in fields such as structural health monitoring and geological disaster early warning, it is necessary to analyze long-term monitoring data to extract key features that reflect changes in the system's state. These features are typically used as input for subsequent machine learning or deep learning models for state assessment or risk prediction. However, existing feature extraction methods usually perform calculations directly on the raw mixed signals, which presents inherent technical contradictions. For example, when calculating settlement trend feature values ​​and anomalous fluctuation feature values ​​within a sliding time window, when a real and severe geological settlement event occurs, it will simultaneously exhibit a huge trend and strong fluctuations in the monitoring data. In this case, the increased anomalous fluctuation feature value may incorrectly reduce the reliability, weakening the identification of this real settlement event and potentially leading to missed reports. For a very weak and gradually changing early settlement, its anomalous fluctuation feature value is small, but its settlement trend feature value itself is also small, which may not be sufficient to trigger an early warning, reflecting a lack of sensitivity to details. Therefore, existing technologies suffer from an inherent contradiction between the robustness of feature extraction and the sensitivity to details due to signal aliasing.

[0032] Therefore, to address the aforementioned issues of signal aliasing and feature interference, the steps S100-S500 are implemented as follows: First, the instantaneous monitoring sequence within the monitoring time window is acquired. Then, a first-level decomposition algorithm is used to deconstruct the mixed original monitoring sequence into multiple independent intrinsic mode functions (IMFs). These IMFs are then classified to identify different modal feature types, such as trend modes, periodic modes, and anomalous modes. Subsequently, corresponding preset type features are calculated for these physically distinct and independent pure modes. Finally, the calculated features are input as independent features to the downstream processing model. This decomposition-then-calculation approach ensures that trend features are extracted only from trend modes, unaffected by sudden disturbances, and anomalous features are extracted only from anomalous modes, allowing for more accurate quantification of disturbance intensity. This fundamentally solves the problem of feature interference caused by aliasing of different components in the original signal.

[0033] Example 2, refer to Figure 1 As an embodiment of the present invention, a method for extracting features from geological subsidence monitoring data is provided based on the above embodiment.

[0034] In this embodiment of the application, S100 acquires the instantaneous monitoring sequence within the monitoring time window.

[0035] Specifically, in a continuous monitoring data stream, a segment of data is extracted as an instantaneous monitoring sequence through a fixed sliding window. The length of the monitoring time window is determined based on actual monitoring needs and data sampling frequency to ensure that the captured instantaneous monitoring sequence can fully reflect the changes in geological subsidence during that period.

[0036] In this embodiment of the application, S200 uses a first-level decomposition algorithm to decompose the instantaneous monitoring sequence into multiple intrinsic mode functions.

[0037] The specific steps include A1~A2: A1. A genetic algorithm is used to optimize the number of mode decompositions and the second-order penalty factor of the first-order decomposition algorithm, wherein the fitness function of the genetic algorithm is to minimize the sum of the envelope spectrum entropy of all intrinsic mode functions after decomposition.

[0038] In this embodiment, the first-level decomposition algorithm employs variational mode decomposition (VMD). The performance of VMD is highly dependent on the selection of the number of mode decompositions and the second-order penalty factor; fixed parameters cannot adapt to varying monitoring signals. Therefore, a genetic algorithm is used to optimize these two parameters, using the sum of envelope spectrum entropies as the fitness function. An excellent decomposition should result in the purest possible intrinsic mode functions, with the strongest sparsity of the envelope spectrum and the lowest entropy. Through iterative optimization using the genetic algorithm, the most suitable combination of decomposition parameters can be found for any input instantaneous monitoring sequence, thereby ensuring the accuracy of mode decomposition.

[0039] The steps for optimizing the mode decomposition number and the quadratic penalty factor in the first-level decomposition algorithm include A1.1 to A1.4: A1.1 Encode the mode decomposition number and the quadratic penalty factor as chromosomes for the genetic algorithm, and initialize the population.

[0040] The modality decomposition number and the quadratic penalty factor are encoded as gene positions on chromosomes in the genetic algorithm, and the parameter value range is represented using binary encoding or real number encoding. Based on a preset population size, an initial population is randomly generated, with each individual representing a set of parameter combinations for the modality decomposition number and the quadratic penalty factor.

[0041] A1.2. Evaluate each chromosome in the population according to the fitness function.

[0042] For each individual in the population, the instantaneous monitoring sequence is decomposed using its corresponding mode decomposition number and a quadratic penalty factor. The sum of the envelope spectral entropies of all intrinsic mode functions obtained from the decomposition is calculated, and this entropy value is used as the fitness value of that individual. The smaller the fitness value, the better the decomposition effect and the higher the purity of each intrinsic mode function.

[0043] A1.3 Perform selection, crossover, and mutation operations on the population to generate a new generation of population.

[0044] Specifically, based on the fitness values ​​of each individual, selection operators such as roulette wheel selection, tournament selection, or elite retention are used to select superior individuals from the current population to enter the mating pool. A crossover operation is performed on the selected individuals, exchanging some gene loci to generate new individuals. A mutation operation is then performed on the new individuals, randomly changing the values ​​of gene loci with a certain probability to increase population diversity. Through these genetic operations, a new generation of the population is generated.

[0045] A1.4 Determine whether the termination condition is met. If not, return to the step of evaluating according to the fitness function. If the condition is met, output the current optimal modality decomposition number and the quadratic penalty factor.

[0046] Specifically, the termination conditions include reaching a preset maximum number of iterations or the fitness function value converging to a preset threshold. If the termination condition is not met, return to step A1.2 to continue iterative optimization; if the termination condition is met, output the modality decomposition number and quadratic penalty factor corresponding to the individual with the best fitness value in the current population, as input parameters for subsequent decomposition algorithms.

[0047] A2. By solving a constrained variational problem through a first-order decomposition algorithm, the instantaneous monitoring sequence is decomposed into a preset number of intrinsic mode functions until the sum of the bandwidths of all intrinsic mode functions is minimized and the sum of all intrinsic mode functions can reconstruct the instantaneous monitoring sequence.

[0048] Specifically, the optimal mode decomposition number and quadratic penalty factor output in step A1 are used to perform a first-level decomposition algorithm on the instantaneous monitoring sequence. The essence of this first-level decomposition algorithm is to solve a constrained variational problem, the goal of which is to find a predetermined number of eigenmode functions (EMFs) such that the sum of the bandwidths of each EMF is minimized, while the sum of all EMFs can accurately reconstruct the original instantaneous monitoring sequence. This constrained variational problem is solved iteratively using the alternating direction multiplier method, ultimately yielding a set of mutually independent EMFs. Each EMF has a clear frequency band boundary, avoiding mode aliasing.

[0049] In this embodiment, variational mode decomposition (VMD) algorithm is preferably used. The essence of the VMD algorithm is to solve a constrained variational problem, and its goal is to find... Modal components This minimizes the sum of the bandwidths of each mode, while ensuring that their sum can accurately reconstruct the original signal. This constrained variational problem can be expressed as: Its constraints are: .in, It is the preset mode decomposition number; yes One modal component; These are the center frequencies corresponding to each mode; It is the Dirac function; It is the imaginary unit; This represents the convolution operation. This problem is typically solved iteratively using the Alternating Direction Multiplier Method (ADMM), ultimately yielding a set of eigenmode functions. .

[0050] In one optional implementation, the first-level decomposition algorithm in step S200 adopts the variational mode decomposition algorithm. The variational mode decomposition algorithm transforms the signal decomposition problem into a variational optimization problem by constructing a variational model. It can effectively avoid the endpoint effect and mode mixing problem existing in traditional empirical mode decomposition, and has strong noise resistance and mode separation accuracy.

[0051] In an optional implementation, the first-level decomposition algorithm in step S200 employs a synchronous compression transform algorithm, which includes the following steps: performing continuous wavelet transform or short-time Fourier transform on the instantaneous monitoring sequence to obtain an initial time-frequency representation; calculating the instantaneous frequency of each time-frequency point in the initial time-frequency representation; compressing the energy of the initial time-frequency representation in the frequency domain based on the instantaneous frequency to obtain a sharpened time-frequency representation; extracting mode trajectories from the sharpened time-frequency representation, and reconstructing multiple intrinsic mode functions based on the mode trajectories. This synchronous compression transform algorithm does not require a preset number of modes, and has advantages in processing non-stationary signals and when the number of modes is unknown.

[0052] In the embodiments of this application, S300 classifies the multiple intrinsic mode functions decomposed and identifies multiple modal feature types.

[0053] The steps for classifying the multiple intrinsic mode functions of the decomposition include B1 to B4: B1. Calculate the center frequency of each intrinsic mode function and the Spearman rank correlation coefficient between each intrinsic mode function and the time vector.

[0054] A Fourier transform is performed on each intrinsic mode function (IMF) to obtain its spectral distribution. The center frequency of the IMF is calculated based on the spectral energy distribution. The center frequency reflects the main frequency components of the IMF; a lower center frequency indicates a slower mode change. Simultaneously, a time vector of the same length as the instantaneous monitoring sequence is constructed, and the Spearman rank correlation coefficient between each IMF and the time vector is calculated. The Spearman rank correlation coefficient measures the degree of monotonic correlation between the IMF and time; a larger absolute value indicates a more pronounced monotonic trend over time.

[0055] B2. Identify the intrinsic mode function with the lowest center frequency and an absolute value of the Spearman rank correlation coefficient greater than a preset trend threshold as the trend mode.

[0056] Among all intrinsic mode functions (IMFs), the IMF with the lowest center frequency is first selected. Then, it is determined whether the absolute value of the Spearman rank correlation coefficient between this IMF and the time vector is greater than a preset trend threshold. The preset trend threshold is set according to the actual monitoring scenario, for example, 0.9. When both conditions are met—lowest center frequency and absolute value of correlation coefficient greater than the preset trend threshold—the IMF is identified as a trend mode. The trend mode represents the long-term trend component in the instantaneous monitoring sequence, which changes slowly and has obvious temporal monotonicity.

[0057] B3. Calculate the normalized Fourier spectrum for the remaining intrinsic mode functions and calculate the spectral kurtosis. Identify the intrinsic mode functions with spectral kurtosis greater than the preset periodic threshold as periodic modes.

[0058] For the remaining intrinsic mode functions (IMFs) not identified as trend modes in step B2, Fourier transforms are performed to obtain their spectra, and these spectra are then normalized to obtain normalized Fourier spectra. Based on the normalized Fourier spectra, the spectral kurtosis of each IMF is calculated. The spectral kurtosis reflects the concentration of spectral energy. Higher spectral kurtosis indicates that the energy of the IMF is concentrated in a few narrowband frequencies, a typical characteristic of periodic signals. IMFs with spectral kurtosis greater than a preset periodicity threshold are identified as periodic modes. This preset periodicity threshold is set according to the actual monitoring scenario, for example, 4. The periodic modes represent periodic fluctuation components in the instantaneous monitoring sequence, such as regular fluctuations caused by daily temperature variations, tidal influences, or seasonal factors.

[0059] B4. Identify the remaining unclassified intrinsic mode functions as anomalous modes.

[0060] Specifically, after the classification in steps B2 and B3, all remaining unclassified intrinsic mode functions are uniformly identified as anomalous modes. These anomalous modes typically manifest as high-frequency, non-periodic, or impulsive signal components, representing sudden disturbances in the instantaneous monitoring sequence, such as abnormal fluctuations caused by construction vibrations, seismic activity, or equipment failures. By separating anomalous modes from trend modes and periodic modes, interference from sudden disturbances in the extraction of long-term trend features can be avoided, while also enabling more accurate quantification of disturbance intensity.

[0061] In an optional implementation, when classifying the decomposed multiple intrinsic mode functions (IMFs) in step S300, a machine learning-based classification method can also be used. Specifically, for each IMF, a meta-feature vector containing time-domain features, frequency-domain features, and trend features is extracted. The time-domain features include sample entropy and permutation entropy; the frequency-domain features include spectral kurtosis, spectral entropy, and center frequency; and the trend features include the Spearman correlation coefficient. The meta-feature vector is input into a pre-trained machine learning classifier, which outputs a class label for the IMF. The class label includes trend, periodicity, or anomaly. The machine learning classifier can employ models such as support vector machines, gradient boosting decision trees, or multilayer perceptrons, and is pre-trained using a large sample library of IMFs containing labeled categories. This method transforms the classification problem into a pattern recognition problem, learns classification rules through a data-driven approach, overcomes the problem of poor robustness with fixed thresholds, and has stronger accuracy.

[0062] In this embodiment of the application, S400 obtains the corresponding preset type feature by calculating multiple modal feature types.

[0063] The steps involved in calculating multiple modal feature types include C1~C2: C1. Calculate a first preset type feature through the trend mode, wherein the first preset type feature is used to characterize the long-term change trend of the instantaneous monitoring sequence.

[0064] Specifically, a first preset type feature is calculated based on the trend mode identified in step S300. This first preset type feature can be the linear fitting slope of the trend mode or the difference between the first and last points of the trend mode within the monitoring time window. When using the linear fitting slope, the trend mode is subjected to least-squares linear fitting to obtain the slope value of the fitted line. This slope value reflects the rate of change in geological subsidence; a positive value indicates an upward trend, a negative value indicates a downward trend, and a larger absolute value indicates a more drastic change. When using the difference between the first and last points, the numerical difference between the first and last sampling points of the trend mode within the monitoring time window is calculated. This difference directly reflects the cumulative change during the monitoring period. Since the trend mode has already removed high-frequency fluctuations and periodic components, the first preset type feature can purely reflect the true long-term deformation trend, unaffected by sudden disturbances.

[0065] C2. Calculate a second preset type feature through the abnormal mode, the second preset type feature being used to characterize the sudden disturbance of the instantaneous monitoring sequence.

[0066] Specifically, a second preset type feature is calculated based on the anomalous mode identified in step S300. This second preset type feature can be either the energy or the kurtosis of the anomalous mode. When energy is used, the sum of squares of the anomalous mode signal is calculated; this energy value reflects the overall intensity of the sudden disturbance, with a larger energy value indicating a more severe disturbance. When kurtosis is used, the ratio of the fourth central moment to the square of the second central moment of the anomalous mode is calculated. The kurtosis is used to quantify non-Gaussian impulsive disturbances in the anomalous mode; a larger kurtosis value indicates a more pronounced spike or impulsive component in the signal. Since the anomalous mode has been separated from the trend and periodic components, the second preset type feature can accurately quantify the disturbance intensity and will not cause misjudgment due to the existence of long-term trends.

[0067] In an optional implementation, step S400 may further calculate a third preset type feature based on the periodic mode. Specifically, a Fourier transform is performed on the periodic mode to extract its dominant frequency and average amplitude as the third preset type feature. The dominant frequency reflects the main periodic characteristics of the periodic fluctuation, and the average amplitude reflects the intensity of the periodic fluctuation. The third preset type feature can be used as an additional feature to assist in judging the stability of the monitoring environment or identifying abnormal changes in periodic patterns.

[0068] In this embodiment of the application, S500 inputs the obtained preset type features as independent features to the downstream processing model.

[0069] Specifically, the first preset type feature, the second preset type feature, and other additional features calculated in step S400 are combined to form a feature vector, which is then input into the downstream processing model for state prediction or anomaly detection. The downstream processing model can employ neural networks, support vector machines, random forests, or other machine learning models.

[0070] In summary, this invention introduces a decomposition-then-calculation approach. A first-level decomposition algorithm deconstructs the mixed original instantaneous monitoring sequence into multiple independent intrinsic mode functions (EMFs) representing long-term trends, periodic fluctuations, and sudden disturbances. Then, features are calculated separately for these physically distinct and independent pure modes. This approach fundamentally solves the problem of feature interference caused by the aliasing of different components in the original signal. Trend features are extracted only from trend modes and are no longer affected by sudden disturbances, while anomaly features are extracted only from anomaly modes, enabling more accurate quantification of disturbance intensity.

[0071] This invention achieves both high sensitivity to weak signal variations and strong robustness to drastically disturbed environments, overcoming the contradiction between robustness and detail sensitivity in feature extraction caused by signal aliasing in existing technologies. By optimizing decomposition parameters using a genetic algorithm and classifying intrinsic mode functions based on physical quantification indices, the accuracy, reliability, and physical interpretability of features are improved, thereby enhancing the predictive performance of downstream processing models.

[0072] Example 3 illustrates a schematic scheme for a method of extracting features from geological subsidence monitoring data. It should be noted that the technical solution of this geological subsidence monitoring data feature extraction system belongs to the same concept as the technical solution of the geological subsidence monitoring data feature extraction method described above. Details not described in detail in this embodiment can be found in the description of the technical solution of the geological subsidence monitoring data feature extraction method described above.

[0073] This embodiment also provides a geological subsidence monitoring data feature extraction system, including: The sequence acquisition module is used to acquire instantaneous monitoring sequences within the monitoring time window; The signal decomposition module is used to decompose the instantaneous monitoring sequence into multiple intrinsic mode functions using a first-level decomposition algorithm; The modality classification module is used to classify the multiple intrinsic mode functions decomposed and identify multiple modality feature types; The feature calculation module is used to obtain the corresponding preset type features by calculating multiple modal feature types; The feature output module is used to input the obtained preset type features as independent features into the downstream processing model.

[0074] This embodiment also provides an electronic device suitable for feature extraction of geological subsidence monitoring data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the feature extraction method for geological subsidence monitoring data proposed in the above embodiment.

[0075] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the method for extracting features from geological subsidence monitoring data as proposed in the above embodiments.

[0076] The storage medium proposed in this embodiment and the method for extracting features from geological subsidence monitoring data proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0077] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0078] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for feature extraction from geological subsidence monitoring data, characterized in that, Includes the following steps: Obtain the instantaneous monitoring sequence within the monitoring time window; The instantaneous monitoring sequence is decomposed into multiple intrinsic mode functions using a first-order decomposition algorithm; The multiple intrinsic mode functions of the decomposed data are classified to identify multiple modal feature types; By calculating multiple modal feature types, the corresponding preset type features are obtained; The obtained preset type features are input as independent features into the downstream processing model.

2. The method for extracting features from geological subsidence monitoring data as described in claim 1, characterized in that, The step of decomposing the instantaneous monitoring sequence into multiple intrinsic mode functions includes: The instantaneous monitoring sequence is decomposed into a predetermined number of intrinsic mode functions by solving a constrained variational problem using a first-level decomposition algorithm, until the sum of the bandwidths of all intrinsic mode functions is minimized and the sum of all intrinsic mode functions can reconstruct the instantaneous monitoring sequence.

3. The method for extracting features from geological subsidence monitoring data as described in claim 2, characterized in that, Before employing the first-level decomposition algorithm, the following steps are also included: A genetic algorithm is used to optimize the modal decomposition number and the second-order penalty factor of the first-order decomposition algorithm. The fitness function of the genetic algorithm is to minimize the sum of the envelope spectrum entropy of all intrinsic mode functions after decomposition.

4. The method for extracting features from geological subsidence monitoring data as described in claim 3, characterized in that, The steps for optimizing the mode decomposition number and the quadratic penalty factor of the first-order decomposition algorithm include: The modality decomposition number and the quadratic penalty factor are encoded as chromosomes in the genetic algorithm, and the population is initialized. The chromosomes in the population are evaluated according to the fitness function. The population is subjected to selection, crossover, and mutation operations to generate a new generation of population; Determine whether the termination condition is met. If not, return to the step of evaluating according to the fitness function. If the condition is met, output the current optimal modality decomposition number and the quadratic penalty factor.

5. The method for extracting features from geological subsidence monitoring data as described in claim 4, characterized in that, The steps for classifying the multiple intrinsic mode functions from the decomposition include: Calculate the center frequency of each intrinsic mode function and the Spearman rank correlation coefficient between each intrinsic mode function and the time vector; The intrinsic mode function with the lowest center frequency and an absolute value of the Spearman rank correlation coefficient greater than a preset trend threshold is identified as the trend mode. Calculate the normalized Fourier spectrum for the remaining intrinsic mode functions and calculate the spectral kurtosis. Identify the intrinsic mode functions with spectral kurtosis greater than the preset periodic threshold as periodic modes. The remaining unclassified intrinsic mode functions are identified as anomalous modes.

6. The method for extracting features from geological subsidence monitoring data as described in claim 5, characterized in that, The steps involved in calculating multiple modal feature types include: The first preset type feature is calculated using the trend mode, and the first preset type feature is used to characterize the long-term change trend of the instantaneous monitoring sequence; The second preset type feature is calculated using the abnormal mode, and the second preset type feature is used to characterize the sudden disturbance of the instantaneous monitoring sequence.

7. The method for extracting features from geological subsidence monitoring data as described in claim 6, characterized in that, The modal feature types include trend modes, periodic modes, and anomalous modes.

8. A geological subsidence monitoring data feature extraction system, using the method described in any one of claims 1-7, characterized in that, include: The sequence acquisition module is used to acquire instantaneous monitoring sequences within the monitoring time window; The signal decomposition module is used to decompose the instantaneous monitoring sequence into multiple intrinsic mode functions using a first-level decomposition algorithm; The modality classification module is used to classify the multiple intrinsic mode functions decomposed and identify multiple modality feature types; The feature calculation module is used to obtain the corresponding preset type features by calculating multiple modal feature types; The feature output module is used to input the obtained preset type features as independent features into the downstream processing model.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the geological subsidence monitoring data feature extraction method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the geological subsidence monitoring data feature extraction method according to any one of claims 1 to 7.

Images