A geological disaster early warning method and system based on dynamic data monitoring

Abstract

The present application relates to the technical field of geological disaster early warning, in particular to a geological disaster early warning method and system based on dynamic data monitoring, comprising the following steps: obtaining geological displacement monitoring data, setting a time window range, calculating local extreme points and local zero points of the monitoring data in the time window respectively to obtain multi-scale reference values; based on the multi-scale reference values, calculating a comprehensive statistic, generating data quality evaluation parameters. The present application dynamically calculates local extreme points and zero points by setting a time window range, constructs multi-scale reference values for comprehensive statistic generation and data evaluation, performs collaborative discrimination and threshold comparison of segmented linear regression to form critical indicators, establishes an early warning feature sequence through characteristic quantity recursive residual spectrum analysis and matrix decomposition operation based on power spectrum density distribution and spectral density matrix, and realizes multi-level early warning indication output based on hierarchical rules.

Description

Technical Field

[0001] This invention relates to the field of geological disaster early warning technology, and in particular to a geological disaster early warning method and system based on dynamic data monitoring. Background Technology

[0002] The field of geological disaster early warning technology is a comprehensive research area, primarily focusing on the monitoring, identification, assessment, and early warning of geological disasters. This field involves multiple disciplines such as geology, hydrology, remote sensing, sensor technology, and data analysis. By establishing a geological disaster monitoring network, key parameters such as geological body displacement, groundwater level, rainfall, and crack opening are collected in real time to ultimately achieve early identification and warning of geological disasters such as landslides, collapses, and debris flows. However, current technologies use simple trend-judgment methods for deformation analysis and lack a complete feature extraction process, resulting in the omission of crucial early warning information. Furthermore, the early warning indicator system is too simplistic and lacks a tiered early warning mechanism, making it difficult to implement differentiated management for different risk levels. Therefore, improvements are needed. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a geological disaster early warning method and system based on dynamic data monitoring.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a geological disaster early warning method based on dynamic data monitoring, comprising the following steps:

[0005] Geological displacement monitoring data is acquired, a time window range is set, and local extreme points and local zero points of the monitoring data within the time window are calculated to obtain multi-scale benchmark values. Based on the multi-scale benchmark values, a comprehensive statistic is calculated to generate data quality assessment parameters.

[0006] Based on the data quality assessment parameters, the objective function value under each set of iterative parameters is calculated to obtain the window optimization parameters; the window optimization parameters are substituted into piecewise linear regression, and the regression results are subjected to collaborative discrimination and threshold comparison to generate deformation critical index.

[0007] Based on the aforementioned deformation critical index, periodic analysis is performed on the monitoring data sequence at each time scale, the power spectral density distribution and spectral density matrix in the frequency domain are calculated, frequency domain characteristic quantities are obtained, recursive residual spectrum analysis and matrix decomposition operations are performed on the frequency domain characteristic quantities, and the residual spectrum and matrix eigenvalues ​​are accumulated to establish an early warning characteristic sequence.

[0008] Based on the warning feature sequence, the feature sequence at each time scale is compared with the corresponding threshold, and the time points exceeding the threshold are marked to obtain the scale warning identifier. The distribution of the scale warning identifier at each time scale is statistically analyzed, and the distribution is matched with the preset classification rules to output the multi-scale warning indication.

[0009] Preferably, the steps for obtaining the multi-scale reference values ​​are as follows:

[0010] Acquire geological displacement monitoring data, set time window range parameters and time window movement step parameters, analyze the variation characteristics of monitoring data at different scales within the time window, and generate multi-scale decomposition results;

[0011] Based on the multi-scale decomposition results, local extreme points and local zero points in the monitoring data are analyzed to obtain local feature points;

[0012] Based on the local feature points, the relative relationships and trends of change between the feature points are calculated to form multi-scale benchmark values.

[0013] Preferably, the steps for obtaining the data quality assessment parameters are as follows:

[0014] Based on the multi-scale benchmark value, a difference calculation is performed to obtain the difference results between local extreme points and local zero points, forming local difference values;

[0015] Based on the local difference values, the comprehensive statistic is calculated using the following formula:

[0016]

[0017] Where, Δx i The i-th term in the local extremum difference result, where k is the number of local extrema, Δy j is the j-th term in the local zero-point difference result, m is the number of local zeros, and S is the comprehensive statistic;

[0018] Based on the comprehensive statistics and their characteristic distribution, data quality assessment parameters are obtained.

[0019] Preferably, the steps for obtaining the window optimization parameters are as follows:

[0020] Based on the data quality assessment parameters, initialize the time window range parameter and the time window moving step size parameter, construct a recursive iterative process, and gradually adjust the value range of the time window range and the time window moving step size parameter to form an iterative parameter set;

[0021] Based on the set of iteration parameters, the objective function value for each set of iteration parameters is calculated using the following formula:

[0022]

[0023] Among them, T i R represents the squared value of the time window range parameter in the i-th iteration. i The reference value represents the time window range, n represents the total number of calculations for the time window range parameter, and S represents the reference value. j M represents the squared value of the time window step size parameter in the j-th iteration. j The target reference value for the time window step size is represented by xm, the total number of calculations for the time window step size parameter is represented by F, and the objective function value is represented by the current set of iteration parameters.

[0024] Based on the objective function value, the time window range parameters and time window step size parameters after recursive filtering are analyzed, and all filtered parameters are optimized and combined to generate window optimization parameters.

[0025] Preferably, the step of obtaining the critical deformation index is as follows:

[0026] Based on the window optimization parameters, the regression residuals and fitting coefficients within each segment are calculated to generate piecewise regression analysis results.

[0027] Based on the results of the piecewise regression analysis, the co-discriminant value is calculated using the following formula:

[0028]

[0029] Among them, R i Let Q be the fitting residual for the i-th segment. i For the corresponding window optimization parameters, p is the total number of segments, and T is the total number of segments. j Let q be the test value of the j-th order distribution in the rank-sum test, q be the total number of test distributions, and I be the cooperative discriminant value.

[0030] Based on the collaborative discrimination value, the deformation distribution is identified by comparing it with a set threshold, and a deformation critical index is generated.

[0031] Preferably, the steps for obtaining the frequency domain features are as follows:

[0032] Based on the aforementioned deformation critical index, the monitoring data sequence at each time scale is extracted, and Fourier transform is performed to convert it into a frequency domain signal, thereby generating a frequency domain signal distribution.

[0033] Based on the frequency domain signal distribution, the power spectral density distribution value is calculated using the following formula:

[0034]

[0035] Where P is the power spectral density distribution value, S(f) is the frequency response function of the frequency domain signal, f1 and f2 are the upper and lower limits of the frequency range, and XM k Let N be the amplitude of the k-th component in the frequency domain signal. k tm represents the total number of frequency points of this component, and tm represents the total number of components.

[0036] Based on the power spectral density distribution value, a spectral density matrix is ​​established. Through correlation analysis between matrix elements, characteristic frequencies are extracted and frequency domain characteristic quantities are generated.

[0037] Preferably, the step of obtaining the warning feature sequence is as follows:

[0038] Based on the frequency domain feature quantity, the residual value of each frequency domain data segment is calculated in segments, and the residual values ​​are accumulated sequentially to form a recursive residual spectrum and generate a recursive residual spectrum.

[0039] Based on the recursive residual spectrum, the residual spectrum matrix is ​​decomposed, and the eigenvalues ​​of the matrix are extracted one by one. All eigenvalues ​​are arranged in an ordered manner and merged into a set to obtain the matrix eigenvalue set.

[0040] Based on the set of matrix eigenvalues, all residual values ​​in the recursive residual spectrum are accumulated with the extracted matrix eigenvalues. The accumulated results are then rearranged according to the time series to establish an early warning feature sequence.

[0041] Preferably, the steps for obtaining the scale warning identifier are as follows:

[0042] Based on the aforementioned warning feature sequence, the feature sequence at each time scale is segmented into data segments, arranged according to the time sequence, and the feature values ​​of each segment are compared with the corresponding thresholds. All feature points exceeding the thresholds are extracted to generate over-threshold feature records.

[0043] Based on the above-threshold feature records, the time points corresponding to the feature points are marked, and the record data within all time scales are called up. All time points exceeding the threshold are marked in segments according to the order of the time points to obtain the time point identifiers.

[0044] Based on the above-threshold time point identifiers, the distribution of time point identifiers across all time scales is analyzed hierarchically. All identifiers are merged according to the established identifier classification rules, and the merged results are integrated according to time scales to generate scale warning identifiers.

[0045] Preferably, the steps for obtaining the multi-scale early warning indication are as follows:

[0046] Based on the aforementioned scale-based early warning markers, the marker distribution is extracted from the data at each time scale. The quantity and distribution characteristics of each marker are counted one by one according to the time scale order. The time scale marker distribution statistics are generated by combining the time span and distribution characteristics of the markers.

[0047] Based on the statistical results of the time scale identifier distribution, the preset hierarchical rules are called one by one to perform matching operations on the identifier distribution of each time scale. Each group of identifier distributions is compared with the hierarchical rules item by item, and hierarchical matching results are generated according to the hierarchical standards.

[0048] Based on the hierarchical matching results, the matching categories at all time scales are integrated, and the hierarchical matching categories are analyzed and merged in chronological order. The merged results are then output as a unified multi-scale early warning indication.

[0049] This invention provides a geological disaster early warning system, comprising:

[0050] The data acquisition and preliminary processing module collects geological displacement monitoring data, sets the time window range, calculates the local extreme points and local zero points of the monitoring data within the time window, and generates multi-scale benchmark values.

[0051] The data quality assessment module calculates comprehensive statistics based on multi-scale benchmark values ​​to obtain data quality assessment parameters.

[0052] The parameter optimization and regression analysis module uses data quality assessment parameters to iteratively calculate the objective function value, obtain window optimization parameters, substitute the window optimization parameters into the piecewise linear regression model, perform collaborative discrimination and threshold comparison, and generate deformation critical indicators.

[0053] The frequency domain feature analysis module performs periodic analysis on the monitoring data sequence based on the deformation critical index, including the calculation of power spectral density distribution and spectral density matrix, recursive residual spectrum analysis and matrix decomposition, accumulation of residual spectrum and matrix eigenvalues, and establishment of early warning feature sequence;

[0054] The warning signal generation and output module compares the feature sequences at each time scale with the set thresholds based on the warning feature sequences, marks the time points that exceed the thresholds, generates scale warning labels, statistically analyzes the distribution of scale warning labels at each time scale, performs matching calculations with preset grading rules, and outputs multi-scale warning indications.

[0055] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0056] This invention dynamically calculates local extreme points and zero points within a set time window range, constructs multi-scale benchmark values ​​for comprehensive statistical generation and data evaluation, performs piecewise linear regression for collaborative discrimination and threshold comparison to form critical indicators, and establishes early warning feature sequences by combining power spectral density distribution and spectral density matrix for recursive residual spectral analysis of feature quantities and matrix decomposition operations. Multi-level early warning indication output is achieved based on hierarchical rules. Multi-scale time window adaptive adjustment eliminates external interference factors, piecewise regression and collaborative discrimination mechanisms accurately identify geological deformation trend changes, and spectral analysis and matrix operations deeply mine hidden data features to achieve early prediction of risk factors. Frequency domain feature quantity recursive residual analysis evaluates the periodicity of monitoring sequences and accurately identifies geological deformation evolution characteristics. Rule matching operation mechanism ensures the rationality of early warning output, and the multi-level early warning indication system effectively improves the effectiveness of early warnings, laying a data foundation for disaster prevention and control decision-making. The early warning sequence constructed based on the accumulation of matrix feature values ​​can locate key time nodes, enabling location and rapid response to disaster precursor signals. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0059] Please see Figure 1 This invention provides a technical solution: a geological disaster early warning method based on dynamic data monitoring, comprising the following steps:

[0060] Geological displacement monitoring data is acquired, a time window range is set, and local extreme points and local zero points of the monitoring data within the time window are calculated to obtain multi-scale benchmark values. Based on the multi-scale benchmark values, a comprehensive statistic is calculated to generate data quality assessment parameters.

[0061] Based on the data quality assessment parameters, the objective function value under each set of iterative parameters is calculated to obtain the window optimization parameters; the window optimization parameters are substituted into piecewise linear regression, and the regression results are jointly discriminated and compared with the threshold to generate the deformation critical index.

[0062] Based on the deformation critical index, periodic analysis is performed on the monitoring data sequence at each time scale, the power spectral density distribution and spectral density matrix in the frequency domain are calculated, frequency domain characteristic quantities are obtained, recursive residual spectrum analysis and matrix decomposition operations are performed on the frequency domain characteristic quantities, and the residual spectrum and matrix eigenvalues ​​are accumulated to establish an early warning characteristic sequence.

[0063] Based on the early warning feature sequence, the feature sequence at each time scale is compared with the corresponding threshold, and the time points that exceed the threshold are marked to obtain the scale early warning identifier. The distribution of the scale early warning identifier at each time scale is statistically analyzed, and the distribution is matched with the preset classification rules to output the multi-scale early warning indication.

[0064] The steps for obtaining multi-scale benchmark values ​​are as follows:

[0065] Acquire geological displacement monitoring data, set time window range parameters and time window movement step parameters, analyze the variation characteristics of monitoring data at different scales within the time window, and generate multi-scale decomposition results;

[0066] Based on the multi-scale decomposition results, local extreme points and local zero points in the monitoring data are analyzed to obtain local feature points;

[0067] Based on local feature points, the relative relationships and trends of change between feature points are calculated to form multi-scale benchmark values.

[0068] Specifically, several displacement sensors are first set up in the target area to continuously collect geological displacement monitoring data. A sample is acquired every 10 seconds and used as the basis for subsequent analysis. Then, the time window range parameter is set by combining the observation results of surface displacement changes in historical data. For example, a 24-hour window can be set in areas with relatively frequent surface activity, and a 48-hour window can be set in areas with relatively stable activity. Then, the time window is rotated and compared every 6 hours. To determine whether the geological displacement amplitude in each time period is in an abnormal state, the monitored values ​​can be compared with those obtained in advance through long-term observation and statistics of different stratigraphic types. The threshold values ​​are compared. For example, for loose geological areas, displacement values ​​recorded between 0 cm and 5 cm are considered the normal range, 5 cm to 10 cm are considered the higher range of variation, and values ​​exceeding 10 cm are considered the area of ​​significant fluctuation. If the recorded values ​​exceed the corresponding threshold, they are marked as the target period for subsequent in-depth analysis. After all periods are recorded, the monitoring values ​​in each period are split into corresponding windows for multi-scale difference and trend analysis. By comparing the merging results of high-frequency fluctuation areas and low-frequency fluctuation areas, significant change patterns are identified, and the fluctuation characteristics at these different scales are summarized together. The summarized results are regarded as the multi-scale decomposition results.

[0069] According to the time periods distinguished in the multi-scale decomposition results, monitoring data are extracted separately. To further locate local extreme points, several adjacent sampled values ​​within the same time period are selected and compared sequentially. When a sampled value is greater than its two adjacent sampled values ​​and the difference exceeds the reference difference (e.g., 0.5 cm) calculated based on the natural variation range in past observations, it is marked as a local maximum point. When a sampled value is less than its two adjacent sampled values ​​and is lower than the difference established by the same rule, it is marked as a local minimum point. The determination of local zero points requires recording the monitoring value as it changes from positive to negative or from negative to positive over time. The transition points are identified, and the values ​​before and after the transition are required to change continuously within a range of 0.1 cm. These reference differences and ranges are derived from the average distribution and maximum and minimum interval assessment of multi-year surface deformation data (for example, setting the transition interval from 0.3 cm to 1 cm based on different strata characteristics). After completing the search for high and low values ​​and zero-crossing points in each time period, all identified extreme points and zero points are listed in the same sequence and stored in the corresponding records. By arranging these two types of points in the time domain, a more intuitive trajectory of fluctuation can be obtained. Finally, this sequence is called the local feature points.

[0070] Using local feature points as the main reference, the displacement difference between adjacent extreme points is calculated one by one and the specific values ​​are recorded. Then, the trend of change is analyzed by combining the order of the extreme points and the zero point on the time axis. For displacement increments greater than 1 cm that span at least 3 sampling periods, they can be regarded as an accelerating upward trend, while displacement decreases less than -1 cm that span the same sampling period can be regarded as a significant downward trend. The specific values ​​and the spanning periods mentioned above are all derived from the statistical results of historical data over many years and the field experience. Then, the positive and negative intervals where the zero point is located are distinguished and processed. By summarizing the distribution of these displacement differences and positive and negative intervals in each time period and combining them with the relative positions of the extreme points and the zero point, a set of multi-scale displacement increase and decrease patterns can be obtained in the time domain. The key parameters of these increase and decrease patterns are listed and stored in a unified record. All these summarized parameter sets are regarded as the final multi-scale benchmark values.

[0071] The steps for obtaining data quality assessment parameters are as follows:

[0072] Based on the multi-scale benchmark value, differential calculation is performed to obtain the difference results between local extreme points and local zero points, forming local difference values;

[0073] Based on the local differences, the comprehensive statistic is calculated using the following formula:

[0074]

[0075] Where, Δx i The i-th term in the local extremum difference result, where k is the number of local extrema, Δy jis the j-th term in the local zero-point difference result, m is the number of local zeros, and S is the comprehensive statistic;

[0076] Based on the comprehensive statistics and their characteristic distribution, data quality assessment parameters are obtained.

[0077] Specifically, the multi-scale benchmark value has been obtained in the previous steps. Based on this benchmark value, adjacent sampling points in any monitoring time series are compared and subtracted to generate differential results between each sampling pair. To accurately record minute changes in geological displacement in real-world scenarios, a preliminary range applicable to the current strata characteristics is first statistically determined using pre-accumulated historical geological sampling data (e.g., a single displacement change of 0 cm to 1 cm is considered normal, exceeding 1 cm to 2 cm is considered moderate, and exceeding 2 cm is considered high). These ranges are then divided into multiple intervals to establish a one-to-one correspondence between the differential results and the monitoring time. The process involves reading previously obtained multi-scale benchmark values ​​to determine the interval in which the difference exists. If the difference result exceeds the established upper limit of the medium amplitude, it is directly marked as a potential anomaly. In subsequent statistics, the differences within the same time period are accumulated and summed. If the difference values ​​of three consecutive sampling pairs all exceed the upper limit of the medium amplitude, they are upgraded to high change records. At the same time, it is determined whether there is a trend of maximum or minimum change at the sampling point corresponding to each difference. By continuously tracking these markers, the deviation of local extreme points and zero points is located, thereby obtaining the set of difference values ​​between each local extreme point and local zero point and integrating them to form the final local difference value.

[0078] The advantage of the formula is that by jointly calculating the sum of squares of the differences at local extreme points and the average of the differences at local zero points, it can reflect the comprehensive state of large-scale changes and zero-crossing distributions in the same index, thereby summarizing the overall characteristics of different types of differences in a single value.

[0079] Δx i The acquisition steps are as follows: During 72 hours of continuous displacement observation, the geological displacement data recorded by the sensor is segmented, with each hour selected as a sampling interval. The locations of possible local extreme points within each interval are counted, and the difference between adjacent extreme points is calculated. Each difference is denoted as Δx. i For example, monitoring the same stratum records displacements of 2.2 cm and 3.0 cm at the 5th and 6th hours respectively, with a difference of 0.8 cm. This difference is recorded as Δx1 and stored in the sequence. If the same measuring point records a displacement of 2.7 cm again at the 7th hour, the difference between the 6th and 7th hours is approximately -0.3 cm, recorded as Δx2. The same method is repeated to process the differences of all adjacent extreme points within 72 hours and form a sequence {Δx1, Δx2, ..., Δx...}. k In this case, k is determined by the number of extreme points identified during monitoring.

[0080] The steps to obtain k are as follows: During the above 72-hour observation period, based on the displacement sampling value monitored every hour, all time series data are traversed to identify maximum or minimum values, and the total number of such extreme points is counted to form the corresponding number k. If 12 extreme points are confirmed in the entire monitoring process, then k equals 12. This number is taken from the complete geological displacement sequence uploaded by the sensor, and combined with the interactive inspection results of technicians at the time of each maximum or minimum peak, the final set of extreme point values ​​is formed, thereby obtaining the actual value of k.

[0081] Δy j The steps for obtaining the zero point are as follows: In the same 72-hour observation period, to identify the location where the zero point appears, the point where the adjacent sample value changes from positive to negative or from negative to positive is recorded as the zero point, and the difference between the samples before and after the zero point is called Δy. j The specific method involves comparing the signs of adjacent sampled values ​​and recording their magnitudes. If the sampled value in the previous time period is 0.6 cm and the sampled value in the next time period is -0.1 cm, a zero point is identified between these two samplings, and the absolute value of the difference before and after this zero point is recorded as Δy. j After collecting all such zero-point differences, a {Δy1,Δy2,...,Δy} is formed. m In this process, m corresponds to the number of zero points discovered.

[0082] The steps for obtaining m are as follows: In the aforementioned process, whenever a sample value change from positive to negative or from negative to positive is detected, the position is recorded as a zero point. After a full detection of the entire 72-hour data, all zero point positions are marked one by one, and then the total number of zero points is obtained. This total number is used to represent m. For example, if 9 zero points are finally calibrated, then m is 9. The final confirmation of this value requires comparison with the actual situation of the sensor's timestamp and displacement direction change. Each zero point is matched with the sampling record at a specific time to ensure the accuracy of the statistics.

[0083] The steps to obtain S are as follows: S represents the final comprehensive statistic after combining the sum of squares of the differences of local extreme points and the average of the differences of local zero points, and performing a logarithmic operation. For ease of comparison in the output, its numerical range is set within a logarithmic mapping space, using the previously obtained Δx. i Δy j By performing calculations with k and m, S can become an indicator for measuring the difference distribution between each extreme value and the zero point. When combining this with subsequent data quality analysis, S needs to be written into the same record along with other statistical parameters.

[0084] Calculation process:

[0085] Step 1, Calculation For example, when k = 12, if the difference sequence {Δx1, Δx2, ... Δx} 12Given the lengths of the centimeters {0.8, -0.3, 0.5, 1.2, -0.9, 1.1, 0.6, -0.4, 0.2, 1.4, -0.8, 0.7} cm, their square values ​​are {0.64, 0.09, 0.25, 1.44, 0.81, 1.21, 0.36, 0.16, 0.04, 1.96, 0.64, 0.49}. Summing these values ​​yields:

[0086]

[0087] The second step, using k=12 as the denominator, yields:

[0088]

[0089] The third step is to calculate. Divide by m. If m = 9, and the zero-point difference sequence {Δy1, Δy2, ... Δy9} is {0.5, 0.2, -0.3, 0.6, -0.7, 1.0, 0.9, -0.4, 0.3} cm respectively, then its absolute values ​​are {0.5, 0.2, 0.3, 0.6, 0.7, 1.0, 0.9, 0.4, 0.3}. Summing these values:

[0090]

[0091] Dividing by m=9 gives:

[0092]

[0093] Fourth step, take ln of the second term in the product of the previous two terms, then:

[0094] S=(0.674)·ln(0.544)≈(0.674)·(-0.608)

[0095] S≈-0.410

[0096] The results indicate that when S is negative, it means that the squared value of the local extreme point difference accounts for a relatively high proportion and the average value of the zero point difference is relatively small. If more extreme point differences appear with large amplitude values ​​in the sampling, while the overall zero point difference is not significant, the value of S may decrease further. If S is greater than a certain defined limit (e.g., 0.5), it means that both the extreme points and the zero point differences in this region show greater fluctuations, and the dispersion of the monitoring data can be further confirmed in the future.

[0097] The comprehensive statistics have been calculated above. Combining their distribution in different time periods, the overall fluctuation of the monitoring data can be analyzed. To further develop data quality assessment parameters, it is necessary to find whether there are multiple concentrated distribution intervals in the characteristic distribution. The specific approach is to first establish a set of distinction criteria based on the size of the S value. For example, an S value between -1.0 and 0.3 is considered a normal interval, 0.3 to 0.7 is a critical interval, and greater than 0.7 is an abnormal interval. The above interval range is determined by technicians by summarizing the minimum and maximum S values ​​obtained from previous monitoring and after multiple verifications. By traversing the S values ​​corresponding to all monitoring times, the interval in which the value belongs is determined and the distribution status of each time point is recorded. At the same time, the frequency of the S value in each interval is recorded. If the cumulative period in the critical interval exceeds 5 hours and the S value is greater than 0.3 three times in a row, it is marked as a key concern. For cases where the S value is greater than 0.7, it is marked as abnormal. After the judgment is completed, all the markings are correlated with the original time series and geological displacement, and the marking results are compiled into data quality assessment parameters.

[0098] The steps to obtain window optimization parameters are as follows:

[0099] Based on data quality assessment parameters, initialize time window range parameters and time window step size parameters, construct a recursive iterative process, and gradually adjust the value range of time window range and time window step size parameters to form an iterative parameter set;

[0100] Based on the set of iteration parameters, calculate the objective function value for each set of iteration parameters using the following formula:

[0101]

[0102] Among them, T i R represents the squared value of the time window range parameter in the i-th iteration. i The reference value represents the time window range, n represents the total number of calculations for the time window range parameter, and S represents the reference value. j M represents the squared value of the time window step size parameter in the j-th iteration. j The target reference value for the time window step size is represented by xm, the total number of calculations for the time window step size parameter is represented by F, and the objective function value is represented by the current set of iteration parameters.

[0103] Based on the objective function value, the time window range parameters and time window step size parameters after recursive filtering are analyzed, and all filtered parameters are optimized and combined to generate window optimization parameters.

[0104] Specifically, data quality assessment parameters have been obtained in advance. When setting the time window range and time window movement step based on these parameters, a preliminary summary is made by referring to the periods in previous monitoring records where geological displacement changes were relatively concentrated. These monitoring records are then compared one by one with the pre-established effective range. For example, the geological displacement rate is compared to the range of 0 cm / h to 2 cm / h, and the cumulative displacement is compared to the range of 0 cm to 20 cm. For periods with more frequent geological activity, the time window range parameter can be shortened to 6 hours and the time window movement step can be appropriately increased to 3 hours. For relatively stable periods, the time window range parameter can be extended to 24 hours and the time window movement step can be set to 6 hours. By dividing the data into different time periods... Batch processing of all recorded data forms multiple candidate lists of iterative parameters. For each candidate list, recursive iteration is performed under the current environment to test its adaptability to the displacement distribution of each time period. If the record corresponding to the time window range parameter exceeds 2 cm / hour in a certain segment, it is marked as a high-activity period, and the time window range parameter in this candidate list is retained for subsequent statistics. If the segmentation method defined by the time window movement step parameter does not show significant fluctuations in displacement amplitude in multiple segments, it is determined that the parameter combination is applicable to a relatively stable area, and it continues to be compared with other candidate combinations in the next recursive step. After multiple rounds of statistical comparison of all candidate lists, a complete set of iterative parameters covering the geological activity characteristics of each time period is formed.

[0105] The advantage of the formula is that it measures the deviation of the time window range parameter by comparing the squared difference of the first part with the reference value, and at the same time, it combines the product of the squared value of the step size parameter and the target reference value in the second part to perform an average calculation, so that both the time window range and the time window moving step size parameters can be comprehensively evaluated in the same index.

[0106] T i The steps to obtain it are as follows:

[0107] T i This represents the squared value of the time window range parameter in the i-th iteration. The time window range parameter is derived from historical data records of surface displacement monitoring. Technicians divide the monitoring period over a duration of 360 hours (approximately 15 days) using multiple different time slice methods and mark the daily geological activity periods to form a series of candidate periods as preliminary time windows. These candidate periods are quantified in hours to form the original window sequence. Then, based on the distribution characteristics of peak and low geological activity periods, effective windows are selected and numbered. For example, different ranges such as 6 hours, 12 hours, 18 hours, and 24 hours are selected as core windows, denoted as T1, T2, T3, T4, etc. During the iteration process, these window ranges are squared to obtain T. iThe value is T1 = 36 if a 6-hour time window is used in the first iteration, and T2 = 144 if a 12-hour time window is used in the second iteration, and so on for T3, T4, etc.

[0108] R i The steps to obtain it are as follows:

[0109] R i Reference values ​​representing the time window range are formed based on comparative analysis of stable and active zones in the actual monitoring environment. Technicians break down the 360-hour monitoring record into whole days and extract segments with high displacement rates (e.g., exceeding 4 cm / h). The start and end times of these segments are then used to measure the ideal time window size, thus calculating the average time span for each segment. For example, if multiple high displacement rate segments are found to last approximately 12 hours, then 12 is recorded as one of the values, R. i After combining more monitoring records and classifying and summarizing them, several reference value sequences are formed, such as R1=12, R2=18, R3=24 hours, which are then converted into T... i Store the same order of magnitude (i.e., number of hours) required for comparison, to correspond to T at each iteration. i The difference of squares operation.

[0110] The steps to obtain n are as follows:

[0111] n represents the total number of calculations for the time window range parameter. Within a 360-hour monitoring period, technicians divide the time window into spans of 6 hours, 12 hours, 18 hours, 24 hours, etc., and conduct multiple rounds of recursive attempts. Each round generates a specific candidate window range, squares it, and compares it with the corresponding reference value. If four possible time window ranges are allocated in one iteration, then n = 4. In subsequent iterations, more subdivisions can be introduced, such as splitting 12 hours into 9 hours and 15 hours. If there are a total of 8 feasible window ranges in this iteration, then n = 8. This method generates the final value of n by comparing the window range parameters in the iteration sequence and comparing it with the aforementioned T. i R i This will help complete the subsequent accumulation and average calculations.

[0112] S j The steps to obtain it are as follows:

[0113] S jThis represents the squared value of the time window step size parameter in the j-th iteration. Obtaining the time window step size parameter also relies on observing the characteristics of geological activity. Technicians will examine whether there are continuous large displacement differences between geologically active and less active sections within a certain time interval, thereby determining the appropriate step size interval. For example, the step size for areas with more frequent geological activity is set to 2 hours, while the step size for relatively stable areas is set to 4 hours or 6 hours, etc. Several candidate step size sequences are obtained and squared successively to form S. j If the first step size is 2 hours, then S1 = 4; if the second step size is 4 hours, then S2 = 16. Different step sizes are used to construct a set of candidate parameters for iteration.

[0114] M j The steps to obtain it are as follows:

[0115] M j The target reference value representing the time window step size is determined by comprehensively considering the dynamic changes in geological displacement monitoring and the key time periods confirmed manually. The average step size of these key sections is statistically analyzed. If it is found that the interval of most displacement abrupt changes is about 3 hours, then the value 3 is listed in the candidate reference value list as M. j Meanwhile, technicians may also list hour values ​​of 4 or 6 based on activity records of different regions over a long period. If three target reference values ​​of 3, 4, and 6 are finally selected, j can be accumulated from 1 to 3 in the iteration. Each time, the extracted specific step size is matched with the target reference value to quantify their degree of deviation.

[0116] The steps to obtain xm are as follows:

[0117] xm represents the total number of calculations for the time window step size parameter. Similar to the concept of n, it focuses on the step size as an independent dimension. Each iteration selects a step size from multiple options such as 2 hours, 4 hours, and 6 hours and performs a square operation. If three feasible step sizes are confirmed, then xm = 3. When engineers deem a finer step size granularity necessary, intermediate values ​​such as 3 hours or 5 hours can be inserted into the aforementioned step sizes, increasing xm to 5 or more in the same iteration to better integrate with S. j With M j Perform product operations one by one and finally achieve a comprehensive evaluation of all step size candidates.

[0118] The steps to obtain F are as follows:

[0119] F represents the objective function value under the current set of iteration parameters. The formula consists of two parts: the first part measures the difference between the squared value of the time window range parameter and the reference value; the second part is the product evaluation of the squared value of the step size parameter and the target reference value. F combines these two results to express the suitability of the time window range and step size in the same quantitative index. To obtain the specific value of F, the previously described T is needed. i R i S j M j Both n and xm are defined first.

[0120] Calculation process:

[0121] Step 1, Calculation

[0122] In one iteration, the time window ranges are taken as 6, 12, 18, and 24 hours, then squared to obtain 36, 144, 324, and 576, respectively. The corresponding reference values ​​are denoted as 6, 12, 24, and 24. These values ​​(6, 12, 24, and 24) are extracted from actual geological activity records using the method described earlier. At this point, n = 4. Calculating the absolute value of each difference yields {|36-6|}. , |144-12| , |324-24| , |576-24|}={30, 132, 300, 552}, summing them as 1014, and dividing by n=4, we get:

[0123]

[0124] Taking the square root of this value and rounding it to the nearest decimal, we get approximately 15.93, which forms the result of the first part.

[0125] The second step involves setting the time window step size to three steps: 2, 4, and 6, and then squaring them to obtain 4, 16, and 36, which correspond to the target reference values ​​of 3, 4, and 6. The three sets of multiplications are (4+3), (16+4), and (36+6) = {7, 20, 42}, and their product is 7 × 20 × 42 = 5880. After dividing by xm = 3, we get approximately 1960, which serves as an intermediate result for the second part.

[0126] The third step is to add the result of the first part (approximately 15.93) to the result of the second part (1960) to obtain F≈1975.93. This is the objective function value when the window range of 6, 12, 18, and 24 hours and the step size of 2, 4, and 6 hours are selected in this iteration.

[0127] The results indicate that a smaller value of 1975.93 suggests a better match between the time window range and step size for the monitoring data. A value greater than 2000 indicates significant parameter discrepancies in the current combination, potentially requiring further screening or the introduction of a more suitable combination of time window range and step size. A value between 1500 and 2000 is generally considered a moderate deviation, allowing technicians to assess whether further refinement or local adjustments are necessary. A value below 1500 indicates a good match between the parameter combination and the coverage and step size segmentation of the monitoring period.

[0128] The current objective function value has been calculated. A series of time window range parameters and time window step size parameters, after recursive filtering, need to be compared. The F-values ​​of each group of parameters need to be sorted by size and recorded in the same sequence. Then, check if there are any values ​​higher than 2000 in the sorted list. If so, this group of parameters will be marked as an exclusion type in subsequent filtering. Next, select parameter combinations with F-values ​​between 1500 and 2000 and compare them with existing combinations below 1500. Record the specific distribution of the two types of combinations in each monitoring period. Through this inductive filtering, obtain a set of parameters considered to best meet monitoring needs and with good fit in coverage and step size. For combinations with duplicate or similar F-values, decide whether to retain or discard them based on the comparison of geological activity distribution. Finally, after comprehensive analysis of all parameter combinations that meet the screening criteria, generate window optimization parameters.

[0129] The steps for obtaining the critical deformation index are as follows:

[0130] Based on the window optimization parameters, the regression residuals and fitting coefficients within each segment are calculated to generate piecewise regression analysis results.

[0131] Based on the piecewise regression analysis results, the co-discriminant value is calculated using the following formula:

[0132]

[0133] Among them, R i Let Q be the fitting residual for the i-th segment. i For the corresponding window optimization parameters, p is the total number of segments, and T is the total number of segments. j Let q be the test value of the j-th order distribution in the rank-sum test, q be the total number of test distributions, and I be the cooperative discriminant value.

[0134] Based on the collaborative discriminant value, the deformation distribution is identified by comparing it with a set threshold, and a critical deformation index is generated.

[0135] Specifically, after obtaining the window optimization parameters, it is necessary to extract observation data in different time periods. First, determine the start and end times of each segment and compare them with the previously recorded geological displacement values. Arrange the sampled values ​​within each segment in order and use time and displacement as the independent and dependent variables for regression. Then, perform a univariate linear fitting operation on each segment, extracting the corresponding slope and intercept as fitting coefficients. The slope can be obtained by dividing the displacement difference between two selected monitoring times within the segment by the monitoring time difference. The intercept is calculated using least squares operation from the minimum deviation between all sampled values ​​and the fitted line. After completing the linear fitting, the difference between the predicted and measured values ​​is recorded to obtain the regression residuals. If the residual value at any time is greater than 2 cm or exceeds 1 cm three times consecutively, the segment is marked as an abnormal segment. These thresholds are determined by the long-term statistical results of historical observation data and corrected in combination with the actual field measurement. The regression residuals and fitting coefficients of all segments are recorded and organized in sequence after traversing all monitoring periods. In this process, the residual distribution of each segment can be further identified, and segments with consecutive large errors are marked as high-priority areas to be checked. After all records are summarized, the segmented regression analysis results are formed.

[0136] The advantage of the formula is that by multiplying and summing the squared residuals in each segment with the window optimization parameters and taking the average, and then calculating the difference with the square root of the sum of squares of the rank-sum test distribution, the squared magnitude of the regression residuals and the overall deviation reflected by the test distribution can be comprehensively measured in the same index. This is beneficial for achieving the correlation assessment of multi-dimensional information with fewer calculation steps.

[0137] R i The steps to obtain it are as follows:

[0138] R i Let Ri represent the fitting residual for the i-th segment, referring to the difference between the regression prediction and the measured value within the segment. When performing linear regression on each segment, technicians compare the numerical difference between the measured data and the fitted line point by point. Then, they average or weight all these differences within the same segment to obtain the average residual or weighted residual for that segment. To maintain consistency in subsequent calculations, this residual is often recorded in centimeters and stored in the sequence {R1, R2, ..., Ri}. p In}, p is the total number of segments, and each R iAll of these originate from the regression process of corresponding segments. If a 40-hour geological displacement monitoring sequence is divided into four segments of 10 hours each, then p = 4 can be obtained. The regression residuals of the corresponding segments can be recorded as R1, R2, R3, and R4, respectively. In actual implementation, if a segment contains 60 sampled data points, each of the 60 sampled values ​​is compared with the regression line of that segment, and the difference is calculated. Then, the differences are summarized according to a predetermined weighting method to finally form R. i .

[0139] Q i The steps to obtain it are as follows:

[0140] Q i To optimize the corresponding window parameters, in the previous process of screening and combining time window range and step size parameters, each regression segment corresponds to a specific final confirmed time window size or window movement step size. The technicians then quantify this time window or step size data to form a Q value corresponding to the segment number. i For example, in one calculation, the time window is optimized to 8 hours and the step size is optimized to 4 hours. These combined values, obtained after multiple rounds of recursive iteration and filtering, can be labeled on each segmented regression task, forming {Q1, Q2, ..., Q...} p In this sequence, each Q i All data is recorded in hours, and in some scenarios, window parameters with longer time spans are converted to correspond to the residual units, so that multiplication operations can be performed in the same unit system. If the window optimization parameter corresponding to the first segment is 8, then Q1 = 8; if the window optimization parameter corresponding to the second segment is 6, then Q2 = 6, and so on.

[0141] The steps to obtain p are as follows:

[0142] p represents the total number of segments. When the data monitoring period is divided into several time blocks by technicians, a regression analysis is performed on each time block, resulting in an R-value. i With the corresponding Q i When the entire monitoring period is divided into p segments, there will be p residual values ​​paired with p window optimization parameters. The value of p is related to the monitoring duration and the segmentation strategy. For example, if the observation period is 40 hours and each segment is 10 hours, then p = 4. If the segment is divided into 5 hours in a more detailed way, then p = 8. Technicians will determine the final number of segments based on the range of window optimization parameters obtained above and the actual characteristics of the monitored object. Once p is determined, it can be used for iterative calculations and subsequent statistical analysis of discrimination indicators.

[0143] T j The steps to obtain it are as follows:

[0144] T jThis is the test value for the j-th order distribution in the rank-sum test. This test value comes from the process of comparing multiple distributions of the observed data. Technicians will sort the residual data in each segment and compare them with the rank information of the known distributions, and assign test statistics one by one. In this process, they will also refer to historical geological displacement records or the results of multiple repeated samplings in the same area to obtain a set of T1, T2, ..., T q These test values ​​are usually presented as dimensionless numbers. To facilitate subsequent calculations, they can be extended into a multinomial sequence that is not directly related to the number of segments but can reflect the degree of dispersion of the error distribution. If q = 3 different levels of distribution are adopted in a rank-sum test, T1, T2, and T3 can be obtained according to the statistical ranking. The specific values ​​are determined according to the ranking of each segment residual in all observed samples.

[0145] The steps to obtain q are as follows:

[0146] q represents the total number of the test distribution. In a complete rank-sum test procedure, technicians may perform multiple distribution comparisons on the monitored samples. Each comparison yields a T value associated with a specific rank. j When using k contrastive distributions for testing, we can obtain q = k T values. j If the geological activity in a certain area is complex, three to five dispersion levels may be set, and q=3 or q=5 may be obtained in the test. The specific value of q is determined by the scale and accuracy requirements of the field observation, and is related to the selected statistical strategy and the target geological range. All T j During the testing process, the sequence is stored in the same sequence and its corresponding order j is labeled.

[0147] Calculation process:

[0148] Step 1, Calculation For example, if p = 4, and the piecewise regression residual sequence {R1, R2, R3, R4} is set to {1.2, 0.8, 1.0, 1.5} cm, and the window optimization parameters {Q1, Q2, Q3, Q4} are set to {8, 6, 8, 4} hours, then we get... Summing the above values ​​gives 32.36.

[0149] The second step is to divide the above results by p=4 to obtain the average value.

[0150] The third step is to calculate the sum of squares for the test value of the j-th order distribution in the rank-sum test. Let q = 3, {T1, T2, T3} = {2.0, 3.0, 1.5}.

[0151] The fourth step is to take the square root of the sum of squares, which is 15.25.

[0152] The fifth step is to calculate the difference between the two results and take the absolute value, i.e., |8.09-3.905|=4.185, thus obtaining I=4.185.

[0153] The results show that the difference between the current piecewise regression residual squared value and the mean and rank-sum test distribution values ​​reflected by the window optimization parameters is 4.185. When I exceeds 5, it indicates that there is a greater distribution deviation between the regression residual level and the rank-sum test value. When I is between 2 and 5, it indicates that the degree of deviation is in the moderate range. When it is less than 2, it means that the distribution is relatively close. Technicians can use this to judge the degree of agreement between the piecewise regression and the rank-sum test ranking and combine it with subsequent steps to further identify the deformed distribution.

[0154] The collaborative discrimination value has been obtained. To identify the deformation distribution, the I value at each moment in the monitoring sequence needs to be read sequentially. The I values ​​at all moments are compared with the thresholds formed by statistical analysis of similar geological environments in previous historical monitoring. Technicians will first distinguish the segments with values ​​below 2.0 and mark them as low deviation intervals, the segments between 2.0 and 5.0 as medium deviation intervals, and the segments with values ​​above 5.0 as high deviation intervals. These thresholds are derived from the distribution characteristic range obtained by summarizing the surface monitoring results from multiple geological surveys. When I is greater than 5.0 for three consecutive moments, the segment is considered a potentially significant deformation segment. In subsequent operations, the displacement curves corresponding to these moments need to be compared with the historical case database, and it is necessary to check whether the ground-deployed monitoring instruments have malfunctioned or exceeded the prescribed safety redundancy value. When the cumulative period of medium deviation intervals exceeds 6 hours, they will also be additionally marked and included in the priority investigation scope. Through the above comparison records, it is possible to quickly identify which time points or segments have deformation amplitudes that deserve special attention. Finally, all deviation intervals are integrated with timestamps to generate deformation critical indicators.

[0155] The steps for obtaining frequency domain features are as follows:

[0156] Based on the critical deformation index, the monitoring data sequence at each time scale is extracted, and Fourier transform is performed to convert it into a frequency domain signal, generating a frequency domain signal distribution.

[0157] Based on the frequency domain signal distribution, the power spectral density distribution value is calculated using the following formula:

[0158]

[0159] Where P is the power spectral density distribution value, S(f) is the frequency response function of the frequency domain signal, f1 and f2 are the upper and lower limits of the frequency range, and XM k Let N be the amplitude of the k-th component in the frequency domain signal. ktm represents the total number of frequency points of this component, and tm represents the total number of components.

[0160] Based on the power spectral density distribution value, a spectral density matrix is ​​established. Through correlation analysis between matrix elements, characteristic frequencies are extracted and frequency domain characteristic quantities are generated.

[0161] Specifically, the deformation critical index has been obtained through the previous steps. It is necessary to extract monitoring data sequences for different time scales and discretize these sequences. First, the number of sampling points and the sampling time interval are counted within each time scale, and the time corresponding to the deformation critical index is marked. By summarizing the displacement records of ground monitoring instruments at multiple measuring points, the possible ranges of different frequency components are distinguished. Then, the segmented time series are read, and a Fourier transform is performed on each sequence to extract the amplitude and phase information of the monitoring values ​​in the frequency domain. Since the monitoring frequencies at different locations and under different geological conditions may be distributed in the range of 0.001Hz to 10Hz or higher, it is necessary to first divide several characteristic frequency bands based on the field observation history and established geological activity classification standards. For example, 0.001Hz to 1Hz is considered the low-frequency band, 1Hz to 5Hz is considered the mid-frequency band, and 5Hz to 10Hz is considered the high-frequency band. During the analysis, the monitoring sequences are projected one by one into the corresponding frequency intervals to record the correspondence between amplitude information and frequency. After the Fourier transform is completed, the frequency values ​​need to be sorted and the amplitude of the main peak of each frequency component needs to be listed. This is used to identify whether characteristic peaks appear. When the amplitude exceeds the reference threshold determined in advance based on years of monitoring patterns (for example, the conversion value corresponding to an amplitude higher than 5 cm is considered a significant fluctuation), it is marked in the frequency domain result. At the same time, the amplitude intervals close to the threshold are also recorded to finely distinguish the distribution of frequency components at the critical boundary. For multiple sampling sequences at the same time scale, several frequency domain curves can be obtained through multiple Fourier transforms. The amplitude differences of each curve at the same frequency are compared. If the amplitude distribution of a curve is continuously greater than the reference threshold in the high-frequency range, it is judged as a suspicious fluctuation segment. If multiple curves maintain similar values ​​in the low-frequency range without large abrupt changes, the low-frequency range is marked as a relatively stable region. All classification, comparison and labeling information in the processing process is recorded on a frequency distribution table. Finally, the frequency domain signal distribution at different time scales can be obtained by comprehensively using this distribution table.

[0162] The advantage of the formula is that by integrating the energy of the frequency response function obtained by Fourier transform and combining the amplitude of each frequency component with the number of frequency points into a square root operation of a sum of squares, it can simultaneously reflect the power accumulation of the frequency domain signal in the main frequency band and multiple sub-components.

[0163] The steps to obtain P are as follows:

[0164] P represents the power spectral density distribution value, which consists of two parts: one part comes from the distribution of |S(f)|. 2 The integral value over the frequency range f1 to f2 is derived from the summation and square root of the product terms consisting of the amplitudes of all components and the total number of their corresponding frequency points. Technicians obtain S(f) by performing a Fourier transform on the monitoring sequence and divide the frequency range into at least 10 sub-intervals, such as 0.001Hz to 1Hz, 1Hz to 2Hz, etc. Each sub-interval has corresponding f1 and f2 values. Subsequently, the frequency domain energy of these sub-intervals is calculated during integration. For the amplitude peak portion, the energy is calculated in {XM}. k The sequence additionally records the amplitude values ​​and the total number of corresponding frequency points {N}. k}, tm represents the total number of components, and each component consists of amplitude and frequency points.

[0165] The steps to obtain S(f) are as follows:

[0166] S(f) is the frequency response function of the frequency domain signal. Technicians first acquire a multi-channel displacement data sequence and sample the data once per second for 48 hours, obtaining 172,800 raw data points. These data are then windowed in the time domain. The data within each window undergoes a Fast Fourier Transform (FFT) to generate complex frequency domain coefficients. The real and imaginary parts of these coefficients are calculated to form the amplitude and phase, thus obtaining the corresponding values ​​of S(f) at each frequency point. If, in a certain transformation, the real part is identified as 30 and the imaginary part as 40 at f = 2Hz, then its amplitude is... This magnitude can then be directly used for |S(f)| 2 The subsequent integral calculations combine the S(f) values ​​obtained at all frequency points into a complete frequency response curve.

[0167] The steps to obtain f1 and f2 are as follows:

[0168] f1 and f2 are the upper and lower limits of the frequency range. They need to be defined in the actual monitoring environment based on the periodicity of geological activity and the sampling frequency. After multiple field samplings, technicians will accumulate enough data to estimate the minimum and maximum frequencies that the monitoring system can cover. For example, in the low-frequency geological activity range, it can start from 0.001Hz up to 10Hz. Then, it will be adjusted when coordinating with the range of the field equipment. 0.001Hz is recorded as f1 and 10Hz is recorded as f2. During integration, the range will be further subdivided into several segments, and each segment can perform the corresponding integration operation. If higher frequency components appear in the monitoring, f2 needs to be re-statistically calculated and adjusted. In some earthquake monitoring, it can reach tens of Hz or even hundreds of Hz or more.

[0169] XM k The steps to obtain it are as follows:

[0170] XM k The amplitude of the k-th component in the frequency domain signal is represented by a local peak or significant energy point extracted from the Fast Fourier Transform (FFT) result. Technicians iterate through the values ​​of S(f) at each frequency point, identifying values ​​whose amplitude exceeds a certain reference threshold (e.g., a 3 cm conversion increment) or is located in a local peak region. These values, along with their corresponding frequencies, are recorded to form {XM1, XM2, ..., XM...} tm In this type of sequence, tm represents the total number of components. If five peaks are identified in a geological monitoring sequence, tm = 5. Based on the peak amplitude, the high-amplitude components and medium-amplitude components are further distinguished.

[0171] N k The steps to obtain it are as follows:

[0172] N k This represents the total number of frequency points corresponding to the k-th component. To more accurately measure the actual energy around a certain frequency peak, technicians will group several sampling points whose amplitudes are close to that peak into the same component, thus obtaining the number of samples N for that component. k For example, if there are 5 adjacent sampling points around 2Hz that all show a relatively high value between 40 and 50, then these 5 points are considered as the same component and denoted as N. k =5, each component corresponds to a peak center and several adjacent points. Technicians can determine whether adjacent points belong to the same component based on the smoothing coefficient. Finally, these components are labeled with serial numbers k = 1, 2, ..., tm and their N is recorded. k .

[0173] The steps to obtain tm are as follows:

[0174] tm represents the total number of components, referring to the number of peaks or significant energy segments confirmed in a single frequency domain analysis. Technicians browse the Fourier transform results using both threshold detection and amplitude continuity detection methods, treating each peak or continuous high-amplitude segment as a single component. If the entire S(f) curve has 8 obvious peak segments, then tm = 8, and subsequently, {XM} will be formed. k} and {N k The associated sequence, tm, varies depending on the characteristics of the monitoring data. It may be large when the data sampling is dense enough and the geological activity is rich, or it may be small due to low amplitude changes in stable areas.

[0175] Calculation process:

[0176] Step 1, Calculation

[0177] Assuming the frequency range is from 0.001Hz to 10Hz, technicians divide it into 20 segments, with each segment uniformly approximated as |S(f)|. 2 Multiply the average value by the bandwidth. If the bandwidth of the first segment is 0.5Hz and the average value is |S(f)| 2 If the integral is 1600, then the integral of the first segment is 0.5 × 1600 = 800. This process is repeated for all segments, and the sum of all segments gives the total integral value T. sum .

[0178] The second step is processing. Assume tm = 3, XM1 = 40, XM2 = 30, XM3 = 50, corresponding to N1 = 5, N2 = 3, N3 = 4, then:

[0179]

[0180] The third step is to take the square root of 20700.

[0181] Fourth step, T sum Adding it to 143.88 forms P, if T sum =2000, then:

[0182] P = 2000 + 143.88 = 2143.88

[0183] The results indicate that 2143.88 can be considered as the sum of the frequency domain energy within the integration range of 0.001Hz to 10Hz, plus the contribution from multiple significant peaks. When the P value is greater than 3000, it often means that there are more high-energy components and the peak amplitude is significant, while when the P value is around 1500, it indicates that the overall energy is at a medium level. Technicians will archive different P values ​​and combine them with geological monitoring records to track the differences in energy levels during certain specific periods.

[0184] After obtaining the power spectral density distribution value, the energy values ​​within the corresponding frequency range need to be arranged in matrix form. First, the monitoring data is divided into several time periods. The frequency distribution and amplitude peak information of each time period are calculated in the same way. Then, for each time period, the corresponding energy level is marked at multiple frequency points, and the frequency-energy correspondence of different time periods is concentrated to form a spectral density matrix. Then, the arrangement value of each frequency point in the matrix in different time periods is read. When the amplitude continuously exceeds the threshold determined in advance based on historical records (e.g., the conversion amount corresponding to the amplitude higher than 6 cm), it is recorded as an object of extra attention. The distribution of these matrix elements in rows and columns and whether they show repeated peaks are counted. By comparing the correlation between matrix elements, it is determined whether the characteristic frequencies in several time periods are consistent or whether new frequency components have appeared. If it is found that some characteristic frequencies have a relatively obvious amplitude distribution in multiple time periods, the frequency value is extracted and compared with the monitoring information of different measuring points. All extracted characteristic frequencies are recorded as key frequency domain information. Finally, after comparing with existing geological movement models, the extraction results are recorded as frequency domain feature quantities.

[0185] The steps for obtaining the early warning feature sequence are as follows:

[0186] Based on frequency domain features, the residual value of each frequency domain data segment is calculated in segments, and the residual values ​​are accumulated sequentially to form a recursive residual spectrum, thus generating a recursive residual spectrum.

[0187] Based on the recursive residual spectrum, the residual spectrum matrix is ​​decomposed, and the eigenvalues ​​of the matrix are extracted one by one. All eigenvalues ​​are arranged in an ordered manner and merged into sets to obtain the set of matrix eigenvalues.

[0188] Based on the matrix eigenvalue set, all residual values ​​in the recursive residual spectrum are accumulated with the extracted matrix eigenvalues. The accumulated results are then rearranged according to the time series to establish an early warning feature sequence.

[0189] Specifically, based on frequency domain characteristics, the frequency domain data segments within each time period are divided according to fixed time intervals and frequency ranges. Then, the residual values ​​of each data segment are statistically recorded one by one. This identifies the differences between each data segment and a predetermined monitoring reference value. The residual sequence within that data segment is then obtained through point-by-point comparison. To correlate the residual sequences between different data segments on the same coordinate system, the start and end times of the data segments are first compared with the applicable range of the reference value. For example, the time period from 0 to 24 hours is divided into multiple intervals, and several sampling points are selected for each interval. These sampling points are then compared one by one with a pre-defined reasonable displacement variation range. If a single record exceeds the empirical range... The threshold is marked as a high discrete value in the residual sequence. Then, these residual values ​​are accumulated sequentially, and the time point information of the occurrence of large discrete values ​​is retained. If three consecutive records exceed the empirical threshold (for example, the reference value for a certain stratigraphic activity is that the displacement exceeds 2 cm and is considered large discrete), a segment mark is added in the subsequent processing stage. This threshold is generally obtained by combining the mean and variance of displacement changes obtained from multiple monitoring in the same area to form an empirical range that conforms to geological characteristics. When the residual values ​​of all segments have been recorded, these differences are arranged in the same sequence according to time, and accumulation is performed within the same segment. The accumulation curve is mapped to the corresponding time series to construct a complete recursive residual spectrum.

[0190] Based on the generated recursive residual spectrum, a two-dimensional residual spectrum matrix is ​​formed in its matrix representation, using the time axis as the row index and the identifiers of different frequency domain data segments as the column index. The residual value corresponding to each matrix unit is read, and then the residual spectrum matrix is ​​decomposed using a numerical decomposition method. Technicians will check each row and column of the matrix individually, and aggregate the highly similar parts of the row or column vectors into submatrices for subsequent extraction of their respective feature values. To determine the decomposition dimension, a minimum submatrix size (e.g., containing at least 1) is set based on past monitoring experience. (0 time segments or 10 frequency domain segments) After statistically organizing the values ​​of each sub-matrix, the values ​​of key positions are obtained by linear decomposition or matrix partitioning. Then, the corresponding eigenvalues ​​are calculated and all eigenvalues ​​are arranged in ascending or descending order. The distribution of these values ​​in each sub-matrix is ​​then statistically analyzed. Similar eigenvalues ​​are merged into one category and their row and column ranges are recorded. If an eigenvalue appears consecutively in multiple sub-matrices, it is included in the merge set. These merged results correspond to the corresponding time period or frequency segment, thus forming the final matrix eigenvalue set.

[0191] Based on the obtained set of matrix eigenvalues, residual points with the same range or correlation with these eigenvalues ​​are found from the recursive residual spectrum. The two are numerically added to obtain new sequence elements. To avoid the addition of interference terms, it is necessary to check the label information of each residual value on the time series and reconfirm the timestamp of its segment or sampling point. For example, the start time order of different segments is checked to ensure that the final accumulation result can be accurately connected. In this process, the previously obtained eigenvalue interval division will be used. For example, if a certain eigenvalue region is mainly concentrated in the 2nd to 3rd hour and the amplitude is distributed in a high range, it is necessary to focus on weighting or directly accumulating all residual values ​​in that range. If the residual value is lower than the preset threshold in some time periods, the accumulation operation is still completed but it is marked as a low amplitude segment. After all the accumulation processing is completed, the new accumulation sequence is rearranged according to the time sequence, and the times of the high amplitude or high eigenvalue superposition part that has been marked are listed in advance to form an index. Finally, the warning feature sequence is sorted out and the accumulation results of each time point are collected as the basic data for subsequent interpretation.

[0192] The steps to obtain the scale warning indicator are as follows:

[0193] Based on the early warning feature sequence, the feature sequence at each time scale is segmented one by one, arranged according to the time series, and the feature values ​​of each segment are compared with the corresponding thresholds. All feature points exceeding the threshold are extracted to generate over-threshold feature records.

[0194] Based on the record of the over-threshold feature, the time point corresponding to the feature point is marked. Record data within all time scales is called up, and all over-threshold time points are marked in segments according to the order of the time points to obtain the over-threshold time point identifier.

[0195] Based on the time point markers exceeding the threshold, the distribution of time point markers across all time scales is analyzed hierarchically. All markers are merged according to the established marker classification rules, and the merged results are integrated according to the time scale to generate scale warning markers.

[0196] Specifically, based on the early warning feature sequence, the feature sequence at each time scale is broken down into several segments. First, the segment boundaries are determined by combining the previously acquired monitoring duration and sampling frequency. Then, feature values ​​are retrieved one by one within each segment. These feature values ​​are compared item by item with the thresholds obtained by referring to geological activity records over many years. The thresholds may be set as specific values ​​such as surface displacement exceeding 2 cm or displacement rate exceeding 1 cm / hour. These values ​​are obtained by technicians through continuous observation in the same area. When any feature value in a segment exceeds the threshold, it is marked as an anomaly point, and the corresponding time index and feature value are recorded. If multiple adjacent sampling points in a segment exceed the threshold, these points are aggregated and a high-priority mark is added. To avoid missing instantaneous peaks or small-scale abrupt changes, the connection interval between two segments is also checked after each segment is processed to confirm whether there are cross-segment anomalies exceeding the threshold. Finally, all these anomaly feature value information are merged and summarized to form a list of exceeding feature points, called the over-threshold feature record.

[0197] Based on the threshold-exceeding feature records, the exact time corresponding to each feature point at different time scales is located one by one. For example, when mapping minute-level records to hourly or daily levels, it is necessary to first read the previously generated multi-scale time series data and match the occurrence time of the feature points according to the interval of each time scale. Then, these marked times are found sequentially in the record data across all time scales. To maintain consistency in time order, the markers can be inserted in segments according to the absolute time sequence of the feature points. When segmenting, a minimum interval can be set with reference to the sampling frequency and the activity rate of the geological region (e.g., every 2 hours is a segment). If multiple feature points from different time scales are present in the same time slice, they are listed in the marker list of that time slice. When the feature value corresponding to a feature point exceeds the key setting (e.g., a displacement increment greater than 2 cm) and appears consecutively, a high-frequency warning label can be added. After all segmentation and marking operations are completed, the time slices and feature point information of these markers are integrated sequentially to finally obtain the threshold-exceeding time point identifiers for subsequent analysis.

[0198] Based on the acquired threshold-exceeding time point markers, the distribution of each time slice across all time scales is categorized and summarized. Different thresholds for different categories are set to determine the classification. For example, in statistical historical activities, if the displacement exceeds 5 cm, it is considered extremely high level; the range of 2 cm to 5 cm is considered medium level. Therefore, during merging, situations where multiple time scales show records exceeding 2 cm and are continuously distributed for more than 4 hours can be marked as medium-level warnings, while continuous displacement exceeding 5 cm can be marked as extremely high-level warnings. These values ​​are also determined by years of observation data and actual on-site assessments. Then, the cumulative number and distribution range of markers within each time scale are checked. Markers with high time overlap are judged to appear in clusters and written into high-priority categories. At the same time, markers distributed on different time scales but pointing to the same time point are merged. If three consecutive scales show threshold-exceeding situations in the same time window, they are merged. Finally, all merged markers are arranged in order of time scale to obtain the final scale warning markers.

[0199] The steps for obtaining multi-scale early warning indications are as follows:

[0200] Based on scale-based early warning labels, the distribution of labels is extracted from data at various time scales. The quantity and distribution characteristics of each label are counted one by one according to the time scale order. The time scale label distribution statistics are generated by combining the time span and distribution characteristics of the labels.

[0201] Based on the statistical results of the time scale identifier distribution, the preset classification rules are called one by one to perform matching calculations on the identifier distribution of each time scale. Each group of identifier distributions is compared with the classification rules item by item, and classification matching results are generated according to the classification standards.

[0202] Based on the hierarchical matching results, the matching categories at all time scales are integrated, and the hierarchical matching categories are analyzed and merged in chronological order. The merged results are then output as a unified multi-scale early warning indicator.

[0203] Specifically, based on scale-based early warning markers, the distribution of all markers is extracted from data at various time scales. By reading the previously obtained time period divisions and marker information, the occurrence time of the markers is matched with the corresponding time scale. Technicians first select record entries from different time periods, such as hourly or daily periods, and arrange them at the same intervals. Then, the number of markers appearing within each interval is counted. If a marker appears more than three times consecutively within a certain time period with an interval of less than two hours, it is considered a high-density marker distribution and is marked separately in subsequent steps. These settings are all based on the analysis results of previous monitoring experience and field observation data in the same area. For example, if in In the past two years, when the hourly marker density in this area exceeded 3, obvious signs of geological activity were often observed. This was set as a reference threshold. Based on this, the distribution characteristics were summarized into a time period and marker statistics table. The number of markers, intervals, and overlapping markers between adjacent time periods were recorded for each time period. If it was found that the markers at a single time scale were concentrated in a specific time period, that time period was marked as a potential high-distribution area in the statistics table. If the number of markers in some time periods at the same time scale was less than 1, it was marked as a low-distribution area. After completing the statistics one by one, the distribution of the number of markers at all time scales was summarized and the statistical results of the marker distribution at all time scales were output.

[0204] Based on the statistical results of the time-scale identifier distribution, the identifier distribution obtained at each time scale is matched with the existing classification rules. The classification rules are usually set with reference to multiple past monitoring cases and based on displacement amplitude or frequency thresholds. For example, it is stipulated that an hourly distribution with more than 3 identifiers per hour is considered high-level, a medium distribution is 1 to 3 identifiers per hour, and a low distribution is no more than 1 identifier per hour. In the analysis, the number of identifiers in the statistical table is compared with the above classification thresholds item by item. When the number of distributions falls into the interval boundary, it is necessary to supplement the judgment of the identifier trend of the time period before and after the record. For example, if the number of hourly identifiers reached 3 in the previous time period and reached 3 in the current time period, it is directly regarded as high-level and the high-level time period is recorded continuously. Based on this, the distribution records are classified and matched. If the records of a certain time scale are all high-level in 3 adjacent time periods, the cumulative high-level identifiers are displayed in the list. After the classification is completed, the corresponding classification matching results are generated for each time scale.

[0205] Based on the hierarchical matching results, the matching categories at all time scales are combined in chronological order. First, the hourly and daily level identifiers are arranged on the same time axis and compared with the time periods in which their high or medium levels appear. If there are consecutive high-level records at the hourly level and corresponding high-distribution records at the daily level, the two are merged into a high-level category and their respective time periods are uniformly labeled. If the daily level identifier shows a high distribution in multiple consecutive day periods while the hourly level only reaches a high level in a few time periods, then during the merging process, only those time points that simultaneously meet the high level are recorded as the highest category identifier. For the remaining time periods that do not overlap or have lower levels, they are saved as lower category identifiers according to the results of their respective scales. Finally, this multi-level merging result is output as a multi-scale early warning indication and presented in a chronological list.

[0206] This invention provides a geological disaster early warning system, comprising:

[0207] The data acquisition and preliminary processing module collects geological displacement monitoring data, sets the time window range, calculates the local extreme points and local zero points of the monitoring data within the time window, and generates multi-scale benchmark values.

[0208] The data quality assessment module calculates comprehensive statistics based on multi-scale benchmark values ​​to obtain data quality assessment parameters.

[0209] The parameter optimization and regression analysis module uses data quality assessment parameters to iteratively calculate the objective function value, obtain window optimization parameters, substitute the window optimization parameters into the piecewise linear regression model, perform collaborative discrimination and threshold comparison, and generate deformation critical indicators.

[0210] The frequency domain feature analysis module performs periodic analysis on the monitoring data sequence based on the deformation critical index, including the calculation of power spectral density distribution and spectral density matrix, recursive residual spectrum analysis and matrix decomposition, accumulation of residual spectrum and matrix eigenvalues, and establishment of early warning feature sequence;

[0211] The warning signal generation and output module compares the feature sequences at each time scale with the set thresholds based on the warning feature sequences, marks the time points that exceed the thresholds, generates scale warning labels, statistically analyzes the distribution of scale warning labels at each time scale, performs matching calculations with preset grading rules, and outputs multi-scale warning indications.

[0212] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A geological disaster early warning method based on dynamic data monitoring, characterized in that, Includes the following steps: Geological displacement monitoring data is acquired, a time window range is set, and local extreme points and local zero points of the monitoring data within the time window are calculated to obtain multi-scale benchmark values. Based on the multi-scale benchmark values, a comprehensive statistic is calculated to generate data quality assessment parameters. Based on the data quality assessment parameters, the objective function value under each set of iterative parameters is calculated to obtain the window optimization parameters; Substitute the window optimization parameters into piecewise linear regression, perform collaborative discrimination and threshold comparison on the regression results, and generate the critical deformation index. Based on the aforementioned deformation critical index, periodic analysis is performed on the monitoring data sequence at each time scale, the power spectral density distribution and spectral density matrix in the frequency domain are calculated, frequency domain characteristic quantities are obtained, recursive residual spectrum analysis and matrix decomposition operations are performed on the frequency domain characteristic quantities, and the residual spectrum and matrix eigenvalues ​​are accumulated to establish an early warning characteristic sequence. Based on the warning feature sequence, the feature sequence at each time scale is compared with the corresponding threshold, and the time points that exceed the threshold are marked to obtain the scale warning identifier. The distribution of the scale warning identifier at each time scale is statistically analyzed, and the distribution is matched with the preset classification rules to output the multi-scale warning indication. The steps for obtaining the multi-scale reference values ​​are as follows: Acquire geological displacement monitoring data, set time window range parameters and time window movement step parameters, analyze the variation characteristics of monitoring data at different scales within the time window, and generate multi-scale decomposition results; Based on the multi-scale decomposition results, local extreme points and local zero points in the monitoring data are analyzed to obtain local feature points; Based on the local feature points, the relative relationships and trends of change between the feature points are calculated to form multi-scale benchmark values.

2. The geological disaster early warning method based on dynamic data monitoring according to claim 1, characterized in that, The steps for obtaining the data quality assessment parameters are as follows: Based on the multi-scale benchmark value, a difference calculation is performed to obtain the difference results between local extreme points and local zero points, forming local difference values; Based on the local difference values, the comprehensive statistic is calculated using the following formula: in, The first local extremum point difference result item, This represents the number of local extrema. The first in the local zero-point difference result item, S represents the number of local zero points, and S represents the comprehensive statistic. Based on the comprehensive statistics and their characteristic distribution, data quality assessment parameters are obtained.

3. The geological disaster early warning method based on dynamic data monitoring according to claim 1, characterized in that, The steps for obtaining the window optimization parameters are as follows: Based on the data quality assessment parameters, initialize the time window range parameter and the time window moving step size parameter, construct a recursive iterative process, and gradually adjust the value range of the time window range and the time window moving step size parameter to form an iterative parameter set; Based on the set of iteration parameters, the objective function value for each set of iteration parameters is calculated using the following formula: in, The time window range parameter indicates the number of... The squared value in the next iteration Reference values ​​indicating the range of a time window. This indicates the total number of calculations for the time window range parameter. The time window step size parameter indicates the time window step size in the first... The squared value in the next iteration The target reference value represents the time window step size. This indicates the total number of calculations for the time window step size parameter. This represents the objective function value given the current set of iteration parameters; Based on the objective function value, the time window range parameters and time window step size parameters after recursive filtering are analyzed, and all filtered parameters are optimized and combined to generate window optimization parameters.

4. The geological disaster early warning method based on dynamic data monitoring according to claim 1, characterized in that, The steps for obtaining the critical deformation index are as follows: Based on the window optimization parameters, the regression residuals and fitting coefficients within each segment are calculated to generate piecewise regression analysis results. Based on the results of the piecewise regression analysis, the co-discriminant value is calculated using the following formula: in, For the first Piecewise fitting residuals Optimize parameters for the corresponding window. The total number of segments, For the rank-sum test, the first The test value of the order distribution, To test the total number of the distribution, For collaborative discriminant values; Based on the collaborative discrimination value, the deformation distribution is identified by comparing it with a set threshold, and a deformation critical index is generated.

5. The geological disaster early warning method based on dynamic data monitoring according to claim 1, characterized in that, The steps for obtaining the frequency domain feature quantity are as follows: Based on the aforementioned deformation critical index, the monitoring data sequence at each time scale is extracted, and Fourier transform is performed to convert it into a frequency domain signal, thereby generating a frequency domain signal distribution. Based on the frequency domain signal distribution, the power spectral density distribution value is calculated using the following formula: in, This represents the power spectral density distribution value. Let be the frequency response function of the frequency domain signal. and These are the upper and lower limits of the frequency range. For the frequency domain signal, the first The amplitude of each component, This represents the total number of frequency points for this component. The total number of components; Based on the power spectral density distribution value, a spectral density matrix is ​​established. Through correlation analysis between matrix elements, characteristic frequencies are extracted and frequency domain characteristic quantities are generated.

6. The geological disaster early warning method based on dynamic data monitoring according to claim 1, characterized in that, The steps for obtaining the early warning feature sequence are as follows: Based on the frequency domain feature quantity, the residual value of each frequency domain data segment is calculated in segments, and the residual values ​​are accumulated sequentially to form a recursive residual spectrum and generate a recursive residual spectrum. Based on the recursive residual spectrum, the residual spectrum matrix is ​​decomposed, and the eigenvalues ​​of the matrix are extracted one by one. All eigenvalues ​​are arranged in an ordered manner and merged into a set to obtain the matrix eigenvalue set. Based on the set of matrix eigenvalues, all residual values ​​in the recursive residual spectrum are accumulated with the extracted matrix eigenvalues. The accumulated results are then rearranged according to the time series to establish an early warning feature sequence.

7. The geological disaster early warning method based on dynamic data monitoring according to claim 1, characterized in that, The steps for obtaining the scale warning identifier are as follows: Based on the aforementioned warning feature sequence, the feature sequence at each time scale is segmented into data segments, arranged according to the time sequence, and the feature values ​​of each segment are compared with the corresponding thresholds. All feature points exceeding the thresholds are extracted to generate over-threshold feature records. Based on the above-threshold feature records, the time points corresponding to the feature points are marked, and the record data within all time scales are called up. All time points exceeding the threshold are marked in segments according to the order of the time points to obtain the time point identifiers. Based on the above-threshold time point identifiers, the distribution of time point identifiers across all time scales is analyzed hierarchically. All identifiers are merged according to the established identifier classification rules, and the merged results are integrated according to time scales to generate scale warning identifiers.

8. The geological disaster early warning method based on dynamic data monitoring according to claim 1, characterized in that, The steps for obtaining the multi-scale early warning indication are as follows: Based on the aforementioned scale-based early warning markers, the marker distribution is extracted from the data at each time scale. The quantity and distribution characteristics of each marker are counted one by one according to the time scale order. The time scale marker distribution statistics are generated by combining the time span and distribution characteristics of the markers. Based on the statistical results of the time scale identifier distribution, the preset hierarchical rules are called one by one to perform matching operations on the identifier distribution of each time scale. Each group of identifier distributions is compared with the hierarchical rules item by item, and hierarchical matching results are generated according to the hierarchical standards. Based on the hierarchical matching results, the matching categories at all time scales are integrated, and the hierarchical matching categories are analyzed and merged in chronological order. The merged results are then output as a unified multi-scale early warning indication.

9. A geological disaster early warning system based on a dynamic data monitoring method for geological disaster early warning according to any one of claims 1-8, characterized in that, include: The data acquisition and preliminary processing module collects geological displacement monitoring data, sets the time window range, calculates the local extreme points and local zero points of the monitoring data within the time window, and generates multi-scale benchmark values. The data quality assessment module calculates comprehensive statistics based on multi-scale benchmark values ​​to obtain data quality assessment parameters. The parameter optimization and regression analysis module uses data quality assessment parameters to iteratively calculate the objective function value, obtain window optimization parameters, substitute the window optimization parameters into the piecewise linear regression model, perform collaborative discrimination and threshold comparison, and generate deformation critical indicators. The frequency domain feature analysis module performs periodic analysis on the monitoring data sequence based on the deformation critical index, including the calculation of power spectral density distribution and spectral density matrix, recursive residual spectrum analysis and matrix decomposition, accumulation of residual spectrum and matrix eigenvalues, and establishment of early warning feature sequence; The warning signal generation and output module compares the feature sequences at each time scale with the set thresholds based on the warning feature sequences, marks the time points that exceed the thresholds, generates scale warning labels, statistically analyzes the distribution of scale warning labels at each time scale, performs matching calculations with preset grading rules, and outputs multi-scale warning indications.

Images