Highway slope geological disaster real-time early warning method based on multi-source sensor fusion

Abstract

The application discloses a multi-source sensing fusion expressway slope geological disaster real-time early warning method and concretely relates to the technical field of safety early warning monitoring, and is used for solving the problem of false consistency of multi-source data leading to missed reports in the early stage of deep concealed disasters of the existing fusion system; abnormal correlation marks are generated by analyzing energy conversion efficiency differential entropy mutation through collecting time series data of slope multi-source geological parameters; a cooperative evolution matrix is constructed under the mark state; the space-time overlapping area of the zero point of the second derivative of the main deformation parameter and the extreme value point of the frequency domain of the auxiliary parameter in the matrix is searched, and the phase change critical state is compared by comparing the geological history envelope line mark; the hidden variable decoupling is carried out on the critical state matrix, and the concealed disaster risk index is generated based on the spatial similarity of the main feature vector and the disaster mode vector; when the risk index continuously exceeds the dynamic critical value and the frequency domain characteristics show non-steady-state evolution, an early warning signal is generated, the deep disaster critical state is effectively identified, and the missed report caused by the safety illusion is avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of safety early warning monitoring, more specifically, the present application relates to a real-time early warning method for highway slope geological disasters based on multi-source sensor fusion. BACKGROUND

[0002] In the field of highway slope geological disaster monitoring, multi-source sensor fusion technology has become the mainstream real-time early warning method. The existing technology generally deploys multiple sensors (such as displacement meters, rain gauges, and groundwater level meters) on the slope, collects geological environmental parameters, and transmits them to the central processing system. The data fusion algorithm is used to comprehensively judge the disaster risk. Such systems rely on independent threshold alarms of various sensors or fusion mechanisms based on weighted decision-making, aiming to improve the reliability of early warning through cross-verification of multi-source data. The related methods have been widely applied in the safety monitoring of slopes in linear engineering such as highways.

[0003] However, when a deep and concealed disaster occurs in the slope (such as the early stage of a deep landslide), a single sensor can only capture partial information due to physical limitations, and multi-source data will form a "false safety" when they do not reach their own independent alarm thresholds. The false consistency of multi-source data caused by the combined effects of geological body disaster mechanism and sensor characteristics makes it impossible for existing fusion systems to identify critical disaster states, resulting in sudden false negatives. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the present application provides a real-time early warning method for highway slope geological disasters based on multi-source sensor fusion to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0006] The real-time early warning method for highway slope geological disasters based on multi-source sensor fusion comprises:

[0007] S1, collecting multi-source geological parameter monitoring data of the slope monitoring point to form a time series data set;

[0008] S2, analyzing the energy conversion efficiency between two types of parameters with physical coupling relationship in the time series data set, and generating an abnormal association mark when the differential entropy mutation amplitude of the energy conversion efficiency exceeds the amplitude threshold;

[0009] S3, under the abnormal association mark state, constructing a co-evolution matrix based on the frequency domain distribution characteristics of the auxiliary monitoring parameters and the acceleration change direction of the main deformation parameters;

[0010] S4, searching for the spatiotemporal overlap region of the second derivative zero point of the main deformation parameters and the frequency domain extreme point of the auxiliary monitoring parameters in the co-evolution matrix, and marking the phase change critical state when the spatiotemporal overlap region exceeds the geological history envelope.

[0011] S5, decoupling the hidden variable characteristics of the collaborative evolution matrix marking the phase transition critical state region, generating a hidden disaster risk index based on the spatial similarity between the principal eigenvectors after decoupling and the preset disaster mode vector;

[0012] S6, when the hidden disaster risk index continuously exceeds the dynamic threshold value and the frequency domain distribution characteristics of the auxiliary monitoring parameters present non-steady state evolution, a geological disaster warning signal is generated.

[0013] Further, the multi-source geological parameter monitoring data of the slope monitoring point is collected to form a time series data set, including:

[0014] Collecting displacement monitoring data, groundwater level monitoring data and rainfall monitoring data of the slope monitoring point;

[0015] Timestamp alignment of displacement monitoring data, groundwater level monitoring data and rainfall monitoring data according to a unified time reference;

[0016] The missing monitoring values in the timestamp aligned displacement monitoring data, groundwater level monitoring data and rainfall monitoring data are completed by using linear interpolation method;

[0017] The completed displacement monitoring data, groundwater level monitoring data and rainfall monitoring data are integrated into a time series data set according to the collection time sequence.

[0018] Further, the energy conversion efficiency between the two types of parameters with physical coupling relationship in the time series data set is analyzed, and an abnormal correlation marker is generated when the differential entropy mutation amplitude of the energy conversion efficiency exceeds the amplitude threshold, including:

[0019] Extracting displacement monitoring data and groundwater level monitoring data with physical coupling relationship from the time series data set as analysis objects;

[0020] Based on the change rate of displacement monitoring data and the change amount of groundwater level monitoring data, an energy conversion efficiency sequence is calculated;

[0021] The sliding window method is used to calculate the Shannon entropy value of the probability distribution in the window to form a differential entropy change sequence;

[0022] Detecting the event that the entropy value change rate of the three consecutive sampling points in the differential entropy change sequence exceeds 3 times the standard deviation of the historical average change rate as the differential entropy mutation;

[0023] When the differential entropy mutation amplitude exceeds the amplitude threshold calibrated based on historical disaster data, an abnormal correlation marker is generated.

[0024] Further, in the abnormal correlation mark state, a cooperative evolution matrix is constructed based on the frequency domain distribution characteristics of the auxiliary monitoring parameters and the acceleration change direction of the main deformation parameters, including:

[0025] The displacement monitoring data is extracted as the main deformation parameters within the time interval corresponding to the abnormal correlation mark, and the groundwater level monitoring data and rainfall monitoring data are extracted as auxiliary monitoring parameters;

[0026] The groundwater level monitoring data and rainfall monitoring data are respectively subjected to windowed Fourier transform, and the amplitude spectrum is extracted as the frequency domain distribution characteristics;

[0027] The second derivative of the displacement monitoring data is calculated, and the time period is marked as a positive acceleration change direction or a negative acceleration change direction according to the sign of the second derivative;

[0028] The frequency domain distribution characteristic vector of the auxiliary monitoring parameters in each time window is combined with the acceleration change direction projection value of the main deformation parameters to form a row vector;

[0029] The row vectors of all time windows are summarized to form a cooperative evolution matrix.

[0030] Further, the time period with a positive second derivative sign is marked as a positive acceleration change direction, and the time period with a negative second derivative sign is marked as a negative acceleration change direction.

[0031] Further, in the cooperative evolution matrix, the spatiotemporal coincidence region of the zero point of the second derivative of the main deformation parameters and the extreme value point of the frequency domain of the auxiliary monitoring parameters is searched, and when the spatiotemporal coincidence region exceeds the geological history envelope, the phase transition critical state is marked, including:

[0032] The time point where the second derivative of the displacement monitoring data changes from negative to positive or from positive to negative is identified as the zero point of the second derivative in the row vector of the cooperative evolution matrix;

[0033] The time point corresponding to the maximum value of the amplitude spectrum of the groundwater level monitoring data and the time point corresponding to the maximum value of the amplitude spectrum of the rainfall monitoring data are located as the extreme value point of the frequency domain in the row vector of the cooperative evolution matrix;

[0034] The time point with a time difference between the zero point of the second derivative and the extreme value point of the frequency domain less than the sampling interval is marked as a spatiotemporal coincidence time point;

[0035] The spatial coordinates of the spatiotemporal coincidence time point are compared with the boundary coordinates of the geological history envelope generated based on the historical data statistics of the monitoring points;

[0036] When the spatial coordinates exceed the boundary coordinates of the geological history envelope, the region corresponding to the spatiotemporal coincidence time point in the cooperative evolution matrix is marked as a phase transition critical state.

[0037] Further, the cooperative evolution matrix marked by the phase transition critical state region is decoupled from the hidden variable characteristics, and a hidden disaster risk index is generated based on the spatial similarity between the decoupled principal eigenvector and the preset disaster mode vector, including:

[0038] A row vector subset corresponding to the phase transition critical state marker is extracted from the cooperative evolution matrix to form a critical state matrix;

[0039] The critical state matrix is singular value decomposed, and the left singular vector corresponding to the maximum singular value is taken as the principal eigenvector;

[0040] The cosine similarity between the principal eigenvector and the preset disaster mode vector generated based on the historical disaster period data is calculated;

[0041] The cosine similarity value is mapped to the interval of 0 to 1 and taken as the reciprocal to generate the hidden disaster risk index.

[0042] Further, when the hidden disaster risk index continuously exceeds the dynamic threshold value and the frequency domain distribution characteristics of the auxiliary monitoring parameter present non-steady state evolution, a geological disaster warning signal is generated, including:

[0043] Obtain the hidden disaster risk index sequence and the frequency domain distribution characteristic sequence of the auxiliary monitoring parameter;

[0044] Detect whether the hidden disaster risk index is continuously higher than the dynamic threshold value in the continuous sampling period;

[0045] Analyze whether the variance of the frequency domain distribution characteristics of the auxiliary monitoring parameter is monotonically increasing in the continuous sampling period;

[0046] When the hidden disaster risk index is continuously higher than the dynamic threshold value in the continuous sampling period and the variance of the frequency domain distribution characteristics is monotonically increasing in the continuous sampling period, a geological disaster warning signal is generated.

[0047] Compared with the prior art, the present application has the following beneficial effects:

[0048] 1. By accurately capturing the energy conversion efficiency differential entropy mutation, the abnormal linkage characteristics between displacement, water level and rainfall parameters are identified at the physical mechanism level. This mutation detection mechanism based on the energy transfer efficiency between parameters can penetrate the appearance of a single parameter not reaching the threshold, revealing early associated signals of deep geological body instability. Further, through the construction of a cooperative evolution matrix, the time domain acceleration direction and the frequency domain distribution characteristics are projected and associated in a unified matrix, forming a dynamic coupling expression framework of multiple dimensions, which lays a structured analysis foundation for subsequent critical state identification.

[0049] 2. Based on the spatiotemporal overlap area comparison of the geological history envelope and the real-time evolution track, the dual verification of the geological experience model and the real-time data driving is realized, and the normal fluctuation and the disaster precursor are effectively distinguished; the linear constraint of the traditional weighted fusion is broken through by the hidden variable decoupling, the spatial similarity quantification of the principal feature vector and the disaster mode vector, and the complex system evolution is converted into a calculable hidden risk index; finally, through the double condition judgment mechanism of the dynamic critical value and the non-steady-state frequency domain feature, the monotonicity verification of the frequency domain evolution is increased on the basis of the continuous risk index exceeding the standard, an anti-interference early warning decision chain is formed, a closed-loop logic from abnormal association detection, critical state identification to risk quantization decision is formed, while the advantages of multi-source data fusion are maintained, the false consistency caused by the false alarm risk is avoided. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 A flowchart of the real-time early warning method for the highway slope geological disaster of the multi-source sensing fusion of the present application;

[0051] Figure 2 A flowchart of the phase transition critical state marking. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0053] Embodiment: Figure 1 The real-time early warning method for the highway slope geological disaster of the multi-source sensing fusion of the present application is given, which comprises:

[0054] S1, collecting multi-source geological parameter monitoring data of the slope monitoring point to form a time series data set;

[0055] S2, analyzing the energy conversion efficiency between two types of parameters with physical coupling relationship in the time series data set, and generating an abnormal association mark when the differential entropy mutation amplitude of the energy conversion efficiency exceeds an amplitude threshold;

[0056] S3, under the abnormal association mark state, constructing a cooperative evolution matrix based on the frequency domain distribution characteristics of the auxiliary monitoring parameters and the acceleration change direction of the main deformation parameters;

[0057] S4, searching for the spatiotemporal overlap area of the zero point of the second derivative of the main deformation parameters and the frequency domain extreme point of the auxiliary monitoring parameters in the cooperative evolution matrix, and marking the phase transition critical state when the spatiotemporal overlap area exceeds the geological history envelope;

[0058] S5, decoupling the hidden variable characteristics of the cooperative evolution matrix of the marked phase transition critical state region, generating a hidden disaster risk index based on the spatial similarity between the principal eigenvectors after decoupling and the preset disaster mode vectors;

[0059] S6, when the hidden disaster risk index continuously exceeds the dynamic threshold value and the frequency domain distribution characteristics of the auxiliary monitoring parameters present non-steady state evolution, a geological disaster warning signal is generated.

[0060] When collecting multi-source geological parameter monitoring data of the slope monitoring point, displacement monitoring sensors, underground water level monitoring sensors and rainfall monitoring sensors are deployed at typical monitoring positions of the highway slope. The displacement monitoring sensors use high-precision GNSS receivers or crack meters, which are installed at key points in the easy-sliding area of the slope, continuously collect the displacement change of the slope surface or deep part at the set sampling frequency, and record them as displacement monitoring data. The underground water level monitoring sensors use drop-in pressure water level meters, which are buried in the internal boreholes of the slope, collect pore water pressure data at fixed periods and convert them into water level elevation values, and record them as underground water level monitoring data. The rainfall monitoring sensors use tipping-bucket rain gauges, which are installed in the open areas around the slope, and record the rainfall intensity per unit time in real time, and record them as rainfall monitoring data. All sensors are configured with a unique device identification code, and the original monitoring data is transmitted to the data processing center in real time through the Internet of Things communication module.

[0061] After the data processing center receives the original monitoring data, it first performs timestamp alignment. Based on international coordinated time as a unified time reference, the collection timestamps of the displacement monitoring data, the underground water level monitoring data and the rainfall monitoring data are standardized. Specifically, it includes identifying the time marker field of each data stream, converting non-standard time format to unified time encoding in the format of year-month-day hour: minute: second; for devices with different sampling frequencies, the sampling interval of the highest frequency device is used as the reference, and the low-frequency data is interpolated to the unified time sequence by linear resampling method. For example, when the displacement monitoring data sampling frequency is once per minute and the underground water level monitoring data is once every five minutes, the underground water level monitoring data is interpolated and aligned at one-minute intervals.

[0062] After completing the timestamp alignment, missing values in the three types of monitoring data are detected. Missing cases include continuous data loss due to communication interruption and invalid data points output by abnormal sensors. Linear interpolation method is used to complete the missing values: when the adjacent data points before and after a single missing value are valid, the missing position is completed by the arithmetic mean of the two valid data points; when the continuous missing values do not exceed three sampling periods, a linear function is established based on the valid data at the beginning and end of the missing section for interpolation completion. The completion operation needs to meet the physical characteristic constraint of continuous and slow change of slope geological parameters, and if the missing duration exceeds three sampling periods, it is marked as an invalid section and not completed.

[0063] The completed displacement monitoring data, groundwater level monitoring data and rainfall monitoring data are integrated in chronological order. Each sampling time point corresponds to a data record, and each record contains a timestamp field, a displacement monitoring data value field, a groundwater level monitoring data value field and a rainfall monitoring data value field. All records are sorted in order of collection time from early to late to form a structured time series data set. The time series data set is stored in a two-dimensional table form, with the table row index being a strictly monotonically increasing timestamp sequence, and the table columns corresponding to the displacement monitoring data sequence, the groundwater level monitoring data sequence and the rainfall monitoring data sequence respectively. The final output time series data set meets the data integrity requirement, and there is no missing value for the three types of monitoring data at any sampling time point.

[0064] The displacement monitoring data is unified in dimension as millimeter, reflecting the absolute displacement of the slope surface or deep rock-soil mass; the groundwater level monitoring data is unified in dimension as meter, reflecting the water level elevation converted from pore water pressure; and the rainfall monitoring data is unified in dimension as millimeter per hour, reflecting the rainfall intensity per unit time. All monitoring data retains the original measurement accuracy, with the displacement monitoring data accurate to one decimal place, the groundwater level monitoring data accurate to two decimal places, and the rainfall monitoring data accurate to the integer place. When storing the time series data set, a lossless compression format is used, and a data integrity check code is generated at the same time for subsequent process verification.

[0065] The linear resampling method used in the timestamp alignment process dynamically determines the interpolation weight according to the time distance of adjacent data points. Let the time point to be interpolated be t, the time of the previous valid data point be t1, and the value be v1, and the time of the next valid data point be t2, and the value be v2, then the interpolation formula is: v t =v1+(v2-v1)×(t-t1) / (t2-t1), v t represents the interpolation result of the target time point t. The calculation process is automatically executed in the data processing center, ensuring the numerical accuracy of different frequency data streams under the unified time reference.

[0066] The linear interpolation method for missing value completion strictly follows the physical law of continuous change of rock-soil mass parameters. Before completion, the validity of the data points before and after the missing position needs to be verified: when the data fluctuation range exceeds 20% of the historical maximum amplitude, it is determined as an abnormal fluctuation point and is not used. The completed data needs to pass the slope continuity test, i.e. the change rate of adjacent data points does not exceed twice the historical maximum change rate of the sensor, otherwise the manual review process is started. The final time series data set is verified for spatio-temporal continuity, and the displacement monitoring data, groundwater level monitoring data and rainfall monitoring data have physical meaning correlation at the same timestamp.

[0067] When extracting displacement monitoring data and groundwater level monitoring data with physical coupling relationship from time series dataset, based on the coupling mechanism of seepage action and deformation of rock-soil mass in geotechnical mechanics, displacement monitoring data is selected to represent the deformation response of slope, and groundwater level monitoring data is selected to represent the action of pore water pressure. The extraction process is as follows: in the two-dimensional table structure of the time series dataset, the values of the displacement monitoring data sequence and the values of the groundwater level monitoring data sequence are read in time stamp index order respectively, forming two time series data vectors with equal length. The elements of each vector are strictly aligned by time stamp, ensuring that the displacement monitoring data value and the groundwater level monitoring data value at the same time index position have physical correlation.

[0068] When calculating the energy conversion efficiency sequence, first process the rate of change of displacement monitoring data: take the difference between the values of displacement monitoring data at two adjacent sampling time points, and divide by the time interval between the two sampling points to obtain the displacement change rate sequence, which has the dimension of millimeters per hour. At the same time, process the change amount of groundwater level monitoring data: take the absolute difference between the values of groundwater level monitoring data at adjacent sampling time points to obtain the water level change amount sequence, which has the dimension of meters. The specific calculation of energy conversion efficiency is defined as the ratio of the absolute value of displacement change rate to the absolute value of water level change amount, i.e. energy conversion efficiency equals the absolute value of displacement change rate divided by the absolute value of water level change amount. This ratio eliminates the dimensional influence and forms a dimensionless sequence, with each time point corresponding to an energy conversion efficiency value, forming a complete energy conversion efficiency sequence.

[0069] The sliding window method is used to calculate the differential entropy change sequence from the energy conversion efficiency sequence. A fixed length sliding window is set, and the window length is set to a time span of 24 hours, for example, according to the deformation response characteristics of the slope. The window slides along the time axis with a single sampling interval as the step size. At each window position, the following operations are performed: divide the energy conversion efficiency values covered by the window into multiple intervals according to the value range, and set the number of intervals to 10 equal-width intervals, for example, according to the data distribution characteristics; count the number of data points in each interval; calculate the probability value of each interval, which is equal to the number of data points in the interval divided by the total number of data points in the window; calculate the entropy value of the current window according to the Shannon entropy formula, and the entropy value calculation formula is the negative sum of the product of each interval probability value and the logarithm of the corresponding probability value, where only the intervals with probability values greater than zero are involved in the calculation. After each window slide, the above process is repeated to output the entropy value corresponding to the center time point of the window, forming a differential entropy change sequence.

[0070] When detecting the mutation event in the differential entropy change sequence, the following steps are sequentially performed: calculating the change rate of adjacent entropy values in the differential entropy change sequence, the change rate being equal to the difference between adjacent entropy values divided by the sampling time interval; calculating the average value and the standard deviation of the change rate under the historical normal working condition, the historical data being taken as the reference, for example, the differential entropy change sequence in the last 30 days; when the absolute values of the change rates of three consecutive sampling points all exceed the average value plus three times the standard deviation, it is determined that a differential entropy mutation event occurs; and calculating the amplitude value of the mutation event, the amplitude value being defined as the maximum change amount of the entropy value in the duration of the mutation event, that is, the maximum absolute value of the difference between the entropy values at the start and end time points of the event.

[0071] The amplitude threshold is obtained by calibrating the historical disaster data: collecting the differential entropy mutation event samples in a specific period before the occurrence of the historical slope instability event, the length of the specific period being set to, for example, 72 hours; extracting the amplitude values of the mutation events to form a sample data set; taking a specific percentile value of the sample data set as the amplitude threshold, the specific percentile being set to, for example, the 85th percentile. When it is detected that the amplitude value of the current differential entropy mutation event exceeds the amplitude threshold, an abnormal correlation marker is generated. The marker contains three parts of information: the start time stamp of the mutation event, the time stamp format being consistent with the original data; the mutation duration, expressed in the number of sampling points or time units; and the mutation amplitude specific value. The marker is stored in a structured record form, while the corresponding displacement monitoring point position number and the underground water level monitoring point position number are associated.

[0072] Data preprocessing is required in the energy conversion efficiency calculation process: the displacement monitoring data sequence is smoothed by moving average filtering, and the filtering window length is set to, for example, 5 sampling points; when calculating the underground water level change, if the absolute value of the water level change is less than the sensor measurement accuracy value, it is considered as no change, and the sensor measurement accuracy value is set to, for example, 0.01 meters. In the entropy value calculation of the sliding window, when the standard deviation of the data in the window is less than a specific threshold, it is considered as uniform distribution, and the theoretical maximum entropy value is directly returned, and the specific threshold is set to, for example, 0.05.

[0073] A dynamic updating mechanism is established in the historical disaster data calibration process: the historical disaster event samples are periodically recalculated, and the update period is set to, for example, once every month; when the number of newly added disaster events exceeds a set number, the percentile value is recalculated, the set number being, for example, 3; the normal working condition data in the non-disaster period is periodically updated to calculate the change rate, and the update period is set to, for example, every 24 hours; the average value and the standard deviation are updated by using the exponential weighted moving average method, and the smoothing coefficients are set to, for example, 0.2 and 0.15 according to the data stability requirement.

[0074] Abnormal association flag generation and post-processing data verification: plot the original displacement monitoring data and groundwater level monitoring data time history curve corresponding to the flag period; check whether there are device abnormalities or environmental interference factors by manual review; verify the passed flag into the subsequent analysis process; the unpassed flag is stored in the false alarm database for threshold parameter optimization. All intermediate data including energy conversion efficiency sequence and differential entropy change sequence in the whole processing flow are stored with time stamp, and the correlation with the original monitoring data is established through time index.

[0075] Processing of time interval in displacement change rate calculation: when the sampling frequency is not fixed, the actual time difference is used to calculate the change rate; when the sampling frequency is fixed, the time interval is a fixed value. Groundwater level change calculation considers sensor drift compensation: record the baseline value at zero o'clock in the daily rainfall-free period, and deduct the baseline drift when calculating the daily water level change. Sliding window boundary processing: when the window exceeds the start or end position of the data sequence, mirror padding method is used to extend the data. The base of logarithm in entropy value calculation is uniformly 2, and when the probability value is zero, zero multiplied by logarithm is defined as zero.

[0076] Historical disaster data sample collection standard: only collect the slope instability events confirmed by geological survey; the event occurrence time is accurate to the minute level; exclude events induced by external factors such as earthquakes and blasting. The percentile calculation uses linear interpolation method. Amplitude threshold update trigger condition: update immediately when the newly added disaster event sample causes the 85 percentile value to change by more than 5%.

[0077] Manual review operation specification: the reviewer should have professional qualifications in geological monitoring; focus on checking the sensor power supply state and data transmission log; environmental interference factors include heavy rain erosion, mechanical vibration, etc.; the review conclusion is divided into three categories: confirmed valid, device failure, and environmental interference. False alarm database record content: false alarm flag original data, detection intermediate data, review conclusion, and processing personnel information. Threshold optimization method: analyze the false alarm database records every quarter, adjust the amplitude threshold percentile value or standard deviation multiple coefficient.

[0078] Data storage index structure: time stamp as the primary key; displacement monitoring data value, groundwater level monitoring data value, energy conversion efficiency value, and differential entropy value as data fields; abnormal association flag as a status flag bit. All data storage is attached with data check code, and the check algorithm uses cyclic redundancy check method to ensure data transmission integrity.

[0079] Dimensionality control: displacement monitoring data unit remains millimeter; groundwater level monitoring data unit remains meter; time unit is unified as second; energy conversion efficiency is dimensionless; entropy value unit is bit. Numerical precision retention rule: original data maintains the original precision of the sensor; intermediate calculation result retains four decimal places; final flag data retains two decimal places.

[0080] After detecting the abnormal correlation marker, the time interval to be analyzed is determined according to the start time stamp and the duration recorded by the abnormal correlation marker. The displacement monitoring data sequence in the time interval is extracted from the time series data set as the main deformation parameter, and the groundwater level monitoring data sequence and the rainfall monitoring data sequence in the same time interval are extracted as auxiliary monitoring parameters. The extraction operation is specifically: taking the start time point of the abnormal correlation marker as the left boundary and the end time point of the marker as the right boundary, all displacement monitoring data values, groundwater level monitoring data values and rainfall monitoring data values in the time period are intercepted from the two-dimensional table of the time series data set to form three equal-length data sequence vectors, and the number of elements of each vector is equal to the total number of sampling points in the time interval.

[0081] When performing frequency domain analysis on the auxiliary monitoring parameters, the groundwater level monitoring data sequence and the rainfall monitoring data sequence are processed by windowed Fourier transform. The specific execution process includes the following steps: setting the time window length and the window function type, the time window length is set to a fixed time length of, for example, 2 hours according to the hydrological response characteristics of the slope, and the Hanning window function is selected; sliding the window along the time axis with a fixed step length, for example, 1 minute; at each window position, performing discrete Fourier transform on the groundwater level monitoring data sequence in the window; calculating the amplitude spectrum of the transform result, taking the first N main frequency components in the amplitude spectrum, and setting N to, for example, 10; repeating the same operation on the rainfall monitoring data sequence; each window outputs two feature vectors: the groundwater level amplitude spectrum vector and the rainfall amplitude spectrum vector; finally forming a frequency domain distribution feature set corresponding to the time window sequence, and each time window center point is associated with a group of frequency domain features.

[0082] When processing the main deformation parameter, the second derivative of the displacement monitoring data sequence is calculated: the central difference method is used to calculate the first derivative sequence, taking the values of the displacement monitoring data at the adjacent three sampling points, and the derivative value at the middle point is equal to the difference between the value at the rear point and the value at the front point divided by twice the sampling interval; the central difference method is repeated to calculate the second derivative sequence of the first derivative sequence; setting a zero value threshold, when the absolute value of the second derivative is less than the threshold, it is considered as zero, and the threshold is set to, for example, 0.01 mm per hour square; marking the sign of the second derivative, the time period with a positive derivative value is marked as a positive acceleration change direction, and the time period with a negative derivative value is marked as a negative acceleration change direction; quantizing the direction state into a projection value, marking the positive direction as a value of 1 and the negative direction as a value of negative 1; forming an acceleration change direction projection value sequence corresponding to the time points of the displacement monitoring data sequence.

[0083] When constructing the collaborative evolution matrix row vector, the data is combined according to the following rules: taking the time window of the frequency domain distribution characteristics as the basis, the center time point of the current window is determined; the acceleration change direction projection value corresponding to the center time point of the displacement monitoring data is searched; if the center time point has no projection value, the arithmetic mean of the nearest projection values before and after is taken; the groundwater level amplitude spectrum vector, the rainfall amplitude spectrum vector and the acceleration change direction projection value are spliced into a composite vector; the dimension of the composite vector is the number of groundwater level amplitude spectrum components plus the number of rainfall amplitude spectrum components plus 1, for example, when N is equal to 10, the vector dimension is 10+10+1=21; each time window generates a row vector, and the arrangement order of the vector elements is fixed as the first 10 elements are the groundwater level amplitude spectrum components, the middle 10 elements are the rainfall amplitude spectrum components, and the last 1 element is the acceleration change direction projection value.

[0084] The row vectors of all time windows are summarized to form a collaborative evolution matrix: the row vectors are arranged in order of the center point time stamp of the time window; the number of rows of the matrix is equal to the number of time windows, and the number of columns is equal to the dimension of the composite vector; missing data processing, when the data missing rate in the window is more than 20%, the row vector is filled with zero value; the matrix storage format is a two-dimensional floating point array, the row index corresponds to the time sequence, and the column index corresponds to the feature type; the finally generated collaborative evolution matrix contains complete spatio-temporal evolution information, and records the matrix construction parameters including time window length, window sliding step, amplitude spectrum component number and zero value threshold.

[0085] The preprocessing operation of windowed Fourier transform includes: performing detrending processing on the data in the window, subtracting the linear fitting baseline; amplitude spectrum normalization processing, making the sum of the amplitudes of each component equal to 1. The boundary processing of the second derivative calculation includes: using forward difference method or backward difference method at the beginning and end of the sequence. The special case processing of the projection value quantization includes: when the second derivative of the displacement monitoring data is zero, the previous non-zero value direction state is inherited. The time window length setting principle includes: adjusting according to the characteristic frequency range of the auxiliary monitoring parameter, ensuring that the window contains at least two complete periods. The amplitude spectrum component selection standard includes: arranging in descending order of amplitude value, selecting the first N components whose cumulative energy proportion is more than 80%.

[0086] The metadata record of the co-evolution matrix includes: storing the starting time stamp, center time stamp and ending time stamp of each row vector corresponding to the time window; the dimension unified processing includes: the displacement monitoring data second derivative unit conversion is converted into millimeters per hour square; the amplitude spectrum component is dimensionless; the projection value is a pure numerical value; the data precision control includes: the original displacement data is retained to one decimal place; the second derivative calculation result is retained to three decimal places; the amplitude spectrum component is retained to four decimal places; the abnormal situation processing mechanism includes: when the window data variance is less than the noise threshold, it is determined as an invalid window; when the acceleration change direction continuously jumps more than the frequency threshold, the data review process is started; after the matrix is generated, the singular value detection is performed, and the all-zero row vector is removed.

[0087] The frequency domain feature vector splicing rule includes: the groundwater level amplitude spectrum component is arranged in ascending order of frequency, and the rainfall amplitude spectrum component is arranged in ascending order of frequency, so as to ensure that the feature position is fixed. The projection value interpolation method includes: when there is no projection value at the center time point, the time distance weighted average is used, and the weight is inversely proportional to the time difference. The matrix storage optimization includes: the sparse matrix format is used for storage, and when the similarity of the continuous row vectors is more than the threshold, the compressed storage is started. The verification co-evolution matrix quality includes: the condition number of the matrix is calculated, when the condition number is more than the threshold, the feature scaling processing is triggered; the time continuity is checked, the row vector time interval should be equal to the window sliding step; the feature value range is verified, the amplitude spectrum component value domain should be between 0 and 1. The final output matrix labels the abnormal association mark number of the data source, and establishes the reverse tracing path.

[0088] All intermediate data are stored with time stamps: the amplitude spectrum vector output by the windowed Fourier transform is stored as a time series array; the displacement second derivative sequence is stored as a time-stamped floating point array; the acceleration change direction projection value sequence is stored as an integer array. The data association mechanism includes: the frequency domain features, the projection value and the original monitoring data are indexed by a unified time reference. The parameter dynamic adjustment includes: the feature scaling coefficient is automatically adjusted according to the matrix condition number detection result; the zero value threshold is dynamically updated according to the data quality report. The final output co-evolution matrix file header includes: the number of rows and columns of the matrix, the time interval range, the data source identifier and the check code.

[0089] Figure 2 A flowchart for marking the phase change critical state of the application is given, and the specific facts of marking the phase change critical state are:

[0090] When identifying the second derivative zero points of displacement monitoring data in the co-evolution matrix, the acceleration change direction projection value field of the matrix row vector sequence is traversed. The projection value sign change event is detected: when the projection value of the adjacent two time points changes from -1 to +1, it is recorded as a negative-to-positive zero point event; when the projection value changes from +1 to -1, it is recorded as a positive-to-negative zero point event. A change tolerance threshold is set during the identification process, which is set according to the calculation error range of the second derivative of displacement monitoring data, specifically the maximum value of the fluctuation range near the historical data second derivative zero point, for example, take 0.1. When the absolute value of the projection value is less than the threshold, it is considered as a zero point that does not trigger event recording. Each zero point event record contains an accurate timestamp and a change type identifier, forming a complete second derivative zero point sequence.

[0091] When positioning the auxiliary monitoring parameter frequency domain extreme point, the groundwater level amplitude spectrum vector field and the rainfall amplitude spectrum vector field in the co-evolution matrix row vector are processed respectively. For each time window corresponding row vector, the operation is performed: scan the groundwater level amplitude spectrum component field, find out the frequency component index corresponding to the maximum amplitude value, record the frequency value and amplitude value of the component as the groundwater level frequency domain extreme point; the same operation is performed by synchronously scanning the rainfall amplitude spectrum component field. When there are multiple same maximum amplitude values, the component with the lowest frequency is selected as the representative extreme point. Each extreme point record contains a timestamp field, a frequency value field and an amplitude value field, forming two independent frequency domain extreme point sequences.

[0092] When marking the spatiotemporal coincidence time point, a three-way time matching mechanism is established: traverse each zero point timestamp in the second derivative zero point sequence; find the extreme point record with the smallest time difference in the groundwater level frequency domain extreme point sequence; find the extreme point record with the smallest time difference in the rainfall frequency domain extreme point sequence; when both time differences are less than the data sampling interval, mark the zero point timestamp as a spatiotemporal coincidence time point. The sampling interval value is taken from the time series data set metadata, for example, when the original data sampling frequency is 1 per minute, the sampling interval is 60 seconds. Each coincidence point record contains four elements: a zero point timestamp field, a groundwater level extreme point timestamp field, a rainfall extreme point timestamp field, and a spatial coordinate value field.

[0093] The geological history envelope generation process is: collecting the time and space coincident point records in the historical database of the monitoring point in the past three years under normal working conditions; extracting the spatial coordinate values of all records and projecting them into a three-dimensional coordinate system, the coordinate system definition rules are: the origin is set at the position of the geological reference pile of the monitoring point, the X axis points to the main sliding direction of the slope, the Y axis is vertically upward, and the Z axis represents the relative time (with the starting time of the analysis period as zero); calculate the statistical boundary for each coordinate dimension respectively: take the mean value ± 3 times the standard deviation in the X axis direction to form the boundary values Xmin and Xmax, take the historical minimum value Ymin and the maximum value Ymax in the Y axis direction, and take the 5th percentile value Zmin and the 95th percentile value Zmax after converting the time value into the number of seconds from the reference point in the Z axis direction; finally form a three-dimensional envelope surrounded by six planes. The envelope boundary coordinates are updated once every quarter, and the new data uses a rolling update mechanism, retaining the last 36 months of data.

[0094] The spatial coordinate comparison operation is specifically: extracting three-dimensional spatial coordinate values (X, Y, Z) from the time and space coincident time point records; substituting the coordinate values into the boundary condition group of the geological history envelope; checking whether the coordinate values satisfy the following conditions at the same time: the X coordinate is between Xmin and Xmax, the Y coordinate is between Ymin and Ymax, and the Z coordinate is between Zmin and Zmax; when any coordinate value exceeds the corresponding boundary range, it is determined to be outside the envelope. A tolerance coefficient is set for the boundary condition judgment, which is set according to the measurement error analysis, for example, allowing a measurement error tolerance of 0.5%.

[0095] When marking the phase transition critical state, locate the row vector index corresponding to the time and space coincident time point in the co-evolution matrix; write a specific code in the state marker field of the row vector, for example, set the code value to 2; simultaneously generate a phase transition critical state record containing five elements: the coincident timestamp field, the coordinate dimension identifier of the boundary exceeded, the corresponding boundary threshold value, the actual coordinate value, and the deviation percentage. Mark the information in the matrix metadata area, which is associated with the original monitoring data through the timestamp index.

[0096] The tolerance processing method for time and space matching is: when there is a deviation between the frequency domain extreme point timestamp and the zero point timestamp, linear interpolation is used to compensate for the time difference. The spatial coordinate mapping rule is: the X coordinate value is taken from the real-time monitoring value of the horizontal displacement of the displacement monitoring point, the Y coordinate value is taken from the elevation monitoring value of the underground water level monitoring point, and the Z coordinate value is calculated as the number of seconds of the current timestamp minus the starting timestamp of the analysis period. The envelope boundary calculation uses a robust statistical method: first calculate the interquartile range IQR of the coordinate values, and then recalculate the statistics after removing the abnormal values less than (Q1-1.5×IQR) or greater than (Q3+1.5×IQR).

[0097] The invalid data processing mechanism includes: when the amplitude spectrum maximum value of the underground water level is less than the noise threshold, the invalid extreme point is considered, the noise threshold is calculated according to the signal-to-noise ratio of the sensor, and specifically, the average value of the amplitude spectrum maximum value in the period without rainfall is taken, for example, 0.05; when the rainfall amplitude spectrum maximum value is less than the noise threshold, it is also considered invalid; when the acceleration direction stability duration before and after the zero point event is less than the set threshold, the invalid zero point is considered, and the stability threshold is set according to the response time characteristics of the slope deformation, specifically, the minimum stable time length of the historical effective zero point event is taken, for example, 3 sampling periods; all invalid data points do not participate in the space-time matching calculation. The boundary condition updating trigger rule is: when the newly added historical data causes the standard deviation of any coordinate dimension to change by more than a set percentage, the envelope line update is immediately started, and the percentage is set to 5% according to the stability requirement of the monitoring system.

[0098] The operation of verifying the marking result includes: visualizing the coordinate points exceeding the envelope line in the three-dimensional scatter plot; superimposedly displaying the boundary of the historical envelope; confirming the marking validity by geological experts; and establishing an error marking database to collect error marking cases. The phase change critical state record storage adopts a structured format, including a timestamp field, a three-dimensional coordinate field, a boundary threshold field, and a state marking field. All intermediate results are persistently stored, including complete second derivative zero point sequence, frequency domain extreme point sequence, space-time coincidence point list, and envelope line parameter table.

[0099] The statistical optimization method in the envelope body generation is: using kernel density estimation method to smooth the boundary, and the bandwidth parameter is automatically adjusted according to the data distribution density, and the bandwidth calculation formula is 1.06 x σ x n -0.2 , wherein σ is the standard deviation, n is the data amount, and the determination of 1.06 is based on the optimal bandwidth theory of normal distribution, which can reduce the false negative rate through slope engineering data verification. The marking data storage optimization adopts the run-length encoding technology: the marking points of the same state for more than 10 consecutive times are merged into a period record. The final output phase change critical state marking set includes three parts: a detailed list of space-time coincidence points exceeding the envelope line, a corresponding cooperative evolution matrix row index mapping table, and a coordinate dimension deviation amount statistical table.

[0100] The dimension control rule is: the space coordinates X / Y are unified to meters; the time dimension Z is unified to seconds; and the amplitude value maintains the dimensionless characteristic. The numerical precision rule is: the coordinate value is kept to three decimal places; and the boundary threshold is kept to two decimal places. The data verification mechanism includes: randomly extracting 5% of the marking points for reverse verification to review whether the coordinate calculation process is compliant; and generating a marking quality report every month, including effective detection rate, error marking rate, and boundary coverage rate indicators.

[0101] The marked visualization of the co-evolution matrix is implemented by highlighting the marked rows in a specific color in the matrix browsing interface; the marked records can be searched by time range, with a time range accuracy of seconds; a marked data export interface is provided, with export formats including CSV and JSON. The historical envelope data is stored in a separate database, and the version management uses a timestamp marking mechanism. The marked results are pushed to the geological disaster monitoring platform in real time, triggering the yellow warning review process. All operation logs are recorded completely, including input data version number, parameter configuration table, processing time, and abnormal event code.

[0102] When extracting the row vector subset corresponding to the phase transition critical state marker from the co-evolution matrix, all rows with a specific value (e.g., 2) in the state marker field are selected according to the matrix state marker field. The extraction operation is as follows: traverse each row of the co-evolution matrix, check the value of the state marker field; when the marker value meets the critical state condition, copy all feature field values of the row to a new matrix; the row index of the new matrix remains the same as the original matrix, and the column index is completely consistent with the original matrix; finally, a critical state matrix is formed, with the number of rows equal to the number of phase transition critical state markers and the number of columns equal to the number of columns of the co-evolution matrix minus 1 (excluding the state marker column). The critical state matrix is accompanied by metadata, including the original timestamp sequence, spatial coordinate information, and abnormal association marker number of the marker source. The number of rows of the critical state matrix is required to be no less than 10, and when it is insufficient, the subsequent processing is suspended and a data deficiency warning is generated.

[0103] When performing singular value decomposition on the critical state matrix, the following core steps are executed: represent the critical state matrix as a real matrix A with m rows and n columns; calculate the product of matrix A and its transpose to obtain the transpose of square matrix AA; solve the eigenvalues and eigenvectors of the transpose of AA; take the unit eigenvector corresponding to the maximum eigenvalue as the left singular vector U; simultaneously calculate the singular value matrix Σ, whose diagonal elements are the square roots of the eigenvalues and are arranged in descending order; the final decomposition result is A equal to U multiplied by Σ multiplied by the transpose of V. The main eigenvector selection rule is: take the first column vector of the left singular vector matrix U, which corresponds to the maximum singular value and represents the main variation mode of the critical state matrix. Data standardization is performed before singular value decomposition: subtract the mean value of each column of the matrix from the eigenvalue of the column and divide by the standard deviation to ensure that the dimensions of each feature are unified.

[0104] The preset disaster mode vector generation method is: collecting critical state matrix samples of a specific period (for example, 72 hours) before a historical disaster event occurs; performing the same singular value decomposition operation on each sample matrix to extract the respective principal eigenvector; performing weighted average processing on all principal eigenvectors, and the weight is set according to the disaster event level. The disaster level division standard is: according to the landslide volume, less than 100,000 cubic meters is small, the weight is 1, 100,000 to 1,000,000 cubic meters is medium, the weight is 1.5, and more than 1,000,000 cubic meters is large, the weight is 2; the weighted average result is unitized (each component is divided by the vector length) to form the preset disaster mode vector. The vector is updated every half year, and the new disaster event data is included in the training set after being confirmed by a geology expert, and the update trigger condition is that more than 3 new disaster events or 6 months of time.

[0105] When calculating the cosine similarity, the vector space model operation is performed: the principal eigenvector of the current critical state matrix is denoted as u, and the preset disaster mode vector is denoted as v; the dot product of u and v is calculated, that is, the sum of the products of the corresponding components of the two vectors is calculated, and the component index range is from 1 to n (n is the vector dimension); the length of u (the square root of the sum of the squares of the components) and the length of v are calculated respectively; the similarity value calculation formula is the result of u dot v divided by the product of the length of u and the length of v. The similarity value range is -1 to 1, and a positive value indicates that the directions are similar. When the principal eigenvector is all zero, the similarity is 0, and when the mode vector is not initialized, the default value is 0.5.

[0106] When generating the hidden disaster risk index, value domain conversion and inverse mapping are performed: the cosine similarity value is linearly mapped from the interval of -1 to 1 to the interval of 0 to 1, and the conversion formula is: the mapped value is equal to the original similarity value plus 1 and then divided by 2; the inverse of the mapped value is taken as the risk index: the risk index is equal to 1 divided by the mapped value. When the mapped value is less than 0.01, the upper limit truncation processing is adopted, and the maximum risk index value is set to 100. The final risk index value is kept to two decimal places, and a confidence evaluation is attached: the confidence is set according to the number of rows of the critical state matrix, and the number of rows is greater than 30, the confidence is high, 10 to 30 is medium, and less than 10 is low. The confidence evaluation is based on statistical significance test, and the calculation of p value evaluates the reliability: p value less than 0.05 is high confidence, 0.05 to 0.1 is medium confidence, and greater than 0.1 is low confidence.

[0107] Data preprocessing for matrix decomposition: missing value processing of the critical state matrix uses column mean filling, and before filling, the missing rate is detected, and when the column missing rate exceeds 20%, the column is discarded. Boundary processing for similarity calculation: when the vector dimension is detected to be mismatched, it is automatically truncated to the minimum common dimension. Dynamic adjustment of risk index output: adjust the index threshold according to the recent warning accuracy, establish a threshold optimization mechanism: statistics the warning accuracy every month, when the accuracy is less than 85% for three times in a row, start the threshold review process, and recalibrate the risk index classification standard.

[0108] Historical catastrophe event data screening criteria: Only use geological exploration confirmed catastrophe events; The event occurrence time is accurate to the minute level; Exclude cases caused by external factors such as earthquakes; Require data completeness of 72 hours before the catastrophe to be more than 90%. Weight distribution coefficient verification: Determine the optimal weight ratio through cross-validation, take 10 groups of historical data, compare the prediction accuracy under different weights, and select the weight scheme with the highest accuracy. Pattern vector storage format: Store using floating point array, retain 8 decimal precision, with version identification and timestamp.

[0109] Abnormal processing of value range mapping: When the similarity exceeds the range of -1 to 1, it is forced to be truncated to the boundary value (less than -1 takes -1, greater than 1 takes 1). Numerical stability guarantee of inverse mapping: Set the minimum mapping value threshold to 0.01 to prevent division by zero errors. Risk index classification standard: Less than 2 is low risk, 2 to 5 is medium risk, and greater than 5 is high risk. The classification threshold is calibrated once every quarter, and the calibration basis is the risk index distribution percentile value 72 hours before the historical catastrophe event occurs.

[0110] Result visualization output: Risk index is displayed in the form of a dashboard, including current value, historical trend curve, and comparison data under similar geological conditions. Storage optimization: Store feature vectors in binary floating point format to compress storage space. Update and review of catastrophe pattern vector training set: Newly added event data is verified by three or more senior geology engineers independently, and consistent confirmation is required before it can be stored.

[0111] Calculation process verification mechanism: Randomly select 10% of the critical state matrix samples for manual review of feature vector calculation correctness; Generate a quality report every month, including decomposition residual analysis, similarity distribution statistics, and early warning accuracy curve. Performance optimization measures: When the number of matrix rows exceeds 1000, use block singular value decomposition algorithm; Set the processing timeout interrupt mechanism, the longest processing time does not exceed 5 minutes; When memory usage exceeds 80% of the system, start disk caching.

[0112] Dimension consistency guarantee: All vectors are dimensionless, standardized in the preprocessing stage; Risk index is a dimensionless ratio. Numerical processing specification: Floating point calculations follow the IEEE 754 standard; Similarity calculation uses double-precision floating point; Set the maximum number of iterations to 100. Safety fault tolerance mechanism: Automatically enable disk caching when memory is insufficient; Save intermediate results to a recovery file when the calculation is interrupted; Abnormal input detection records detailed error logs.

[0113] Risk index association with raw data: By time stamp chain tracing to displacement monitoring data and groundwater level monitoring data original collection value, the association path is: risk index → critical state matrix → co-evolution matrix → anomaly association mark → original monitoring data. Output document structure: the header contains the calculation timestamp, input matrix row and column number summary; the main body is the risk index value and confidence level; the tail part is attached with the main similar dimension analysis report and characteristic vector visualization chart.

[0114] History version management: a unique hash value is generated each time the calculation is performed, which is stored together with the parameter configuration in the version database. Real-time monitoring board: dynamically display the current risk index and 72-hour trend, support drilling analysis of each dimension contribution. Index application rules: high risk index triggers red warning, automatically pushes to the emergency command system and starts on-site verification; medium risk triggers orange warning and starts data review process. The final output includes: risk index value, confidence level, main abnormal dimension identifier, historical comparison data.

[0115] When obtaining the hidden catastrophe risk index sequence and the frequency domain distribution characteristic sequence of auxiliary monitoring parameters, the complete data record in the recent set period is extracted from the data storage system. The hidden catastrophe risk index sequence acquisition method is: query the risk index database, arrange in ascending order of time stamp, extract the time stamp field and index value field to form a sequence, and the time interval is consistent with the original data sampling interval. The frequency domain distribution characteristic sequence acquisition method of auxiliary monitoring parameters is: extract the groundwater level amplitude spectrum vector sequence and rainfall amplitude spectrum vector sequence corresponding to each time window in the same period from the co-evolution matrix storage library. The sequence time alignment mechanism is: taking the risk index time stamp as the reference, find the nearest time window center time stamp corresponding to the frequency domain distribution characteristic vector. Data integrity check rule: when the sequence missing rate exceeds 5%, start the data repair process, and use time series interpolation method to complete the missing points.

[0116] When detecting whether the hidden catastrophe risk index is continuously higher than the dynamic threshold value in the continuous sampling period, a multi-stage judgment process is performed. The dynamic threshold value calculation method is: taking the moving average value of the risk index sequence in the recent set period plus K times the standard deviation, the moving window length is set to be 24 hours for example, and the K value is set to be 2 to 3 according to the warning level requirement. The continuous higher than the determination condition is: in the continuous M sampling periods, the risk index value of each sampling point is greater than the dynamic threshold value at the corresponding time, and the M value is set to be 6 periods according to the geological disaster response characteristics. The dynamic threshold value updating mechanism is: recompute every N sampling periods, and the N value is set to be 12. Boundary processing: when the risk index value is equal to the threshold value, it is considered not to meet the higher condition.

[0117] When analyzing whether the variance of the frequency domain distribution characteristics of the auxiliary monitoring parameters is monotonically increasing within the continuous sampling period, the groundwater level amplitude spectrum vector sequence and the rainfall amplitude spectrum vector sequence are processed respectively. The variance calculation method is: for the amplitude spectrum vector of each time window, the variance value of all components is calculated, and the formula is: the square sum of each component value minus the mean divided by the number of components. The monotonically increasing determination method is: within the continuous P sampling periods, the variance value of each subsequent period is greater than that of the previous period, and the allowable error tolerance is set to, for example, 0.5%. The P value setting principle is: not less than the risk index continuous period number M, usually set to be equal. When at least one of the two auxiliary parameters meets the monotonically increasing condition, it is determined that the variance of the frequency domain distribution characteristics presents non-steady state evolution.

[0118] When generating a geological disaster warning signal, a double condition joint judgment is performed: when the concealed disaster risk index is continuously higher than the dynamic threshold value in the continuous M sampling periods and the variance of the frequency domain distribution characteristics of the auxiliary monitoring parameters is monotonically increasing in the continuous P sampling periods, a warning signal is generated. The warning signal contains three types of information: the basic information part contains the warning timestamp, the duration, and the current risk index value; the condition information part contains the amplitude of the risk index exceeding the threshold value and the variance growth rate; the location information part is associated with the original monitoring point spatial coordinates. After the signal is generated, a review verification is immediately performed: automatically check the recent warning records at the same location, and if there is an existing warning of the same level within 72 hours, do not generate a repeated warning.

[0119] The parameter setting basis in dynamic threshold value calculation: K value is determined through historical disaster event back analysis, taking the median of the standard deviation multiple distribution of the risk index fluctuation characteristics of 10 historical events. The setting basis of the moving window length of 24 hours is the typical duration of the slope deformation acceleration stage. The determination method of the continuous period number M is: analyze the historical disaster precursor data, and take the sampling period number corresponding to the minimum time interval from the first anomaly to the disaster occurrence.

[0120] Verification mechanism of variance monotonically increasing: linear trend test is used to calculate the slope of the continuous P variance values; when the slope is greater than zero and the determination coefficient exceeds 0.8, the monotonicity is confirmed. Tolerance processing: when the variance value fluctuates within the tolerance range, the monotonically increasing determination is not interrupted. Auxiliary determination of non-steady state evolution: simultaneously monitor the amplitude spectrum main frequency migration rate, and when the main frequency migration speed exceeds the threshold value, the warning level is strengthened.

[0121] Warning signal grading rules: according to the risk index exceeding amplitude and the variance growth rate, three levels of warning are divided. First level warning (yellow) condition: risk index exceeds within 5% and variance growth rate is less than 10%; second level warning (orange) condition: exceeds 5%~10% or growth rate 10%~20%; third level warning (red) condition: exceeds more than 10% or growth rate more than 20%. Each level of warning corresponds to different emergency response processes.

[0122] Signal generation time precision control: Warning timestamps are accurate to the second, and duration calculations are accurate to the sampling period. Location association method: Spatial coordinate fields in the co-evolution matrix metadata are associated with monitoring point numbers in the geographic information system. Signal storage format: A structured message body is used, including header information, conditional data, and verification signature.

[0123] Early warning verification mechanism: Level 1 warnings are generated automatically; Level 2 warnings require confirmation by the on-duty engineer; Level 3 warnings require double confirmation by two senior engineers. False alarm suppression measures: When a monitoring point is under construction interference, the warning level is automatically lowered; the trigger threshold is raised within 72 hours after rain. Signal transmission protocol: Encrypted data packets are used and pushed to the emergency command platform via a dedicated network.

[0124] Adaptive adjustment of dynamic threshold: A feedback learning mechanism is established to analyze the accuracy of early warnings monthly. When the false alarm rate exceeds 15%, the K value is automatically increased; when the false alarm rate exceeds 10%, the K value is automatically decreased. The adjustment increment is set to 0.1 steps, with upper and lower limits controlled between 1.5 and 3.5.

[0125] Special handling for auxiliary parameter variance calculation: When the amplitude spectrum vector contains zero values, this component is not included in the variance calculation; when the effective components are less than 50%, it is considered an invalid window. Lifecycle management of warning signals: The warning signal is valid for 24 hours after generation and will automatically expire; its status is updated every 2 hours during this period.

[0126] Visualization Output: Warning signals are displayed as flashing icons on the geological monitoring platform, with colors corresponding to the warning level; a pop-up details window displays a risk index trend chart, variance change curve, and spatial location map. Historical Warning Repository: All warning records are stored by time index, supporting multi-dimensional retrieval and analysis.

[0127] Boundary security mechanism: Set a daily warning limit, with a maximum of 3 warnings generated from the same monitoring point within 24 hours; the minimum interval between consecutive warnings is no less than 4 hours. Hardware-level protection: When the system load exceeds 80%, suspend non-critical computing to ensure that warning tasks are executed with priority.

[0128] The final warning signal includes the following metadata: generation timestamp, sequence number, data source hash value, and issuer identifier (if required). The signal verification code is calculated based on the MD5 digest of all conditional data to prevent tampering. The push confirmation process involves the recipient returning a digitally signed receipt to ensure information delivery.

[0129] Dimension consistency guarantee: risk index is a dimensionless quantity; variance is a dimensionless quantity of amplitude square; time unit is second. Numerical accuracy: risk index is retained to 2 decimal places; variance is retained to 4 decimal places; coordinate is retained to 3 decimal places. Audit tracking: complete record of intermediate calculation results and decision logic path of signal generation process.

[0130] Emergency linkage trigger: three-level early warning automatically activates on-site monitoring equipment to the highest inspection frequency; starts unmanned aerial vehicle patrol plan; notifies rescue team to enter standby state. System self-check: automatically runs diagnostic program after each early warning to verify data pipeline integrity.

[0131] The embodiment scheme realizes the cooperative evolution analysis of the geological disaster early warning mechanism through multi-step coupling. The abnormal association mark established in step S1 provides a time and space benchmark for subsequent analysis, and by fusing the dynamic coupling relationship of displacement, water level and rainfall parameters, a physically meaningful abnormal event representation is formed; the cooperative evolution matrix constructed in steps S2 to S3 breaks through the limitation of traditional single parameter analysis, projects and correlates the time domain acceleration characteristics and the frequency domain distribution characteristics in a unified matrix, realizes the structural coupling of multi-source heterogeneous data; the phase transition critical state recognition of step S4 is based on the deviation detection of the geological history envelope and the real-time evolution trajectory, and reflects the critical state migration characteristics of the rock and soil mass instability process; step S5 extracts the main feature vector and the spatial similarity of the disaster mode through the hidden variable decoupling technology, and converts the complex system evolution into a quantifiable risk index; the double condition early warning mechanism of step S6 comprehensively considers the risk index dynamic threshold and the frequency domain non-steady state evolution characteristics, and excludes incidental interference through the double verification of persistence and monotonicity.

[0132] The cooperativity is reflected in three aspects: 1) the time and space projection rule of the displacement acceleration direction and the hydrological frequency domain characteristics establishes a physical correlation basis; 2) the comparison mechanism of the historical envelope and the real-time evolution matrix realizes the fusion of geological experience and data-driven; 3) the joint determination of the dynamic threshold of the risk index and the non-steady state frequency domain characteristics forms an anti-interference verification chain, and the overall scheme solves the problem of lagging behind in identifying hidden disasters in traditional methods through structured coupling and progressive analysis of multi-dimensional parameters.

[0133] The calculations involved in the embodiment are all dimensionless numerical calculations, and the preset parameters and threshold values in the calculations are set by a person skilled in the art according to the actual situation.

[0134] It should be noted that the application can be deployed in the device itself to realize embedded application, or run on PC terminal or other terminal with user interface, so as to meet various hardware environments and use requirements.

[0135] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center through wireless or wired transmission. The wired transmission includes optical fiber, twisted pair, coaxial cable, etc. The wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0136] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described system, device and module can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0137] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0138] The modules described as separate components can or can not be physically separated, and the components displayed as modules can or can not be physical modules, which can be located in one place or distributed on a plurality of network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments.

[0139] In addition, each functional module in the various embodiments of the present application can be integrated in one processing module, or each module can exist physically independently, or two or more modules can be integrated in one module.

[0140] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to make a computer device (which can be a personal computer, a server or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0141] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0142] Finally: the above is only the preferred embodiments of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A real-time warning method for highway slope geological disasters based on multi-source sensor fusion, characterized in that, The method comprises the following steps: S1, collecting multi-source geological parameter monitoring data of the slope monitoring point to form a time series data set; S2, analyzing the energy conversion efficiency between two types of parameters with physical coupling relationship in the time series data set, and generating an abnormal correlation mark when the differential entropy mutation amplitude of the energy conversion efficiency exceeds the amplitude threshold, including: extracting displacement monitoring data and groundwater level monitoring data with physical coupling relationship from the time series data set as analysis objects; calculating the energy conversion efficiency sequence based on the change rate of the displacement monitoring data and the change amount of the groundwater level monitoring data; calculating the Shannon entropy value of the probability distribution in the window by using the sliding window method on the energy conversion efficiency sequence to form a differential entropy change sequence; detecting an event that the entropy value change rate of three consecutive sampling points in the differential entropy change sequence exceeds 3 times the standard deviation of the historical average change rate as a differential entropy mutation; generating an abnormal correlation mark when the differential entropy mutation amplitude exceeds the amplitude threshold calibrated based on historical disaster data; S3, under the state of the abnormal correlation mark, constructing a co-evolution matrix based on the frequency domain distribution characteristics of the auxiliary monitoring parameters and the acceleration change direction of the main deformation parameters, including: extracting displacement monitoring data as the main deformation parameter and extracting groundwater level monitoring data and rainfall monitoring data as auxiliary monitoring parameters within the time interval corresponding to the abnormal correlation mark; performing windowed Fourier transform on the groundwater level monitoring data and the rainfall monitoring data respectively, and extracting the amplitude spectrum as the frequency domain distribution characteristics; calculating the second derivative of the displacement monitoring data, and marking the time period as a positive acceleration change direction or a negative acceleration change direction according to the sign of the second derivative; combining the frequency domain distribution characteristic vector of the auxiliary monitoring parameters and the acceleration change direction projection value of the main deformation parameters in each time window into a row vector; summarizing the row vectors of all time windows to form a co-evolution matrix; S4, searching for the spatiotemporal overlap region of the second derivative zero point of the main deformation parameter and the frequency domain extreme point of the auxiliary monitoring parameter in the co-evolution matrix, and marking the phase transition critical state when the spatiotemporal overlap region exceeds the geological history envelope; S5, decoupling the hidden variable characteristics of the co-evolution matrix in the phase transition critical state region, and generating a hidden disaster risk index based on the spatial similarity between the decoupled main feature vector and the preset disaster mode vector; S6, generating a geological disaster warning signal when the hidden disaster risk index continuously exceeds the dynamic threshold value and the frequency domain distribution characteristics of the auxiliary monitoring parameters present non-steady state evolution.

2. The multi-source sensing fusion real-time early warning method for highway slope geological disasters according to claim 1, characterized in that, Collecting multi-source geological parameter monitoring data of the slope monitoring point to form a time series data set, including: collecting displacement monitoring data, groundwater level monitoring data and rainfall monitoring data of the slope monitoring point; aligning the time stamps of the displacement monitoring data, the groundwater level monitoring data and the rainfall monitoring data according to a unified time reference; completing the missing monitoring values in the time-stamped displacement monitoring data, groundwater level monitoring data and rainfall monitoring data by using a linear interpolation method; integrating the completed displacement monitoring data, groundwater level monitoring data and rainfall monitoring data into a time series data set in the order of collection time.

3. The multi-source sensing fusion real-time early warning method for highway slope geological disasters according to claim 1, characterized in that, The time period with the second derivative sign positive is marked as a positive acceleration change direction, and the time period with the second derivative sign negative is marked as a negative acceleration change direction.

4. The multi-source sensing fusion real-time early warning method for highway slope geological disasters according to claim 2, characterized in that, Search for the spatiotemporal coincidence area of the second derivative zero point and the frequency domain extreme point of the auxiliary monitoring parameter in the co-evolution matrix, and mark the phase transition critical state when the spatiotemporal coincidence area exceeds the geological history envelope, including: Identify the time point when the second derivative of displacement monitoring data changes from negative to positive or from positive to negative in the row vector of the co-evolution matrix as the second derivative zero point; Locate the time point corresponding to the maximum value of the amplitude spectrum of the groundwater level monitoring data and the time point corresponding to the maximum value of the amplitude spectrum of the rainfall monitoring data in the row vector of the co-evolution matrix as the frequency domain extreme point; Mark the time point with a time difference between the second derivative zero point and the frequency domain extreme point less than the sampling interval as the spatiotemporal coincidence time point; Compare the spatial coordinates of the spatiotemporal coincidence time point with the boundary coordinates of the geological history envelope generated based on historical monitoring data statistics; When the spatial coordinates exceed the boundary coordinates of the geological history envelope, mark the area corresponding to the spatiotemporal coincidence time point in the co-evolution matrix as the phase transition critical state.

5. The multi-source sensing fusion real-time early warning method for highway slope geological disasters according to claim 4, characterized in that, Decouple the hidden variable characteristics of the co-evolution matrix marked with the phase transition critical state region, and generate a hidden disaster risk index based on the spatial similarity between the principal eigenvector after decoupling and the preset disaster mode vector, including: Extract the row vector subset corresponding to the phase transition critical state marker from the co-evolution matrix to form a critical state matrix; Perform singular value decomposition on the critical state matrix, and take the left singular vector corresponding to the maximum singular value as the principal eigenvector; Calculate the cosine similarity between the principal eigenvector and the preset disaster mode vector generated based on historical disaster period data training; Map the cosine similarity value to the interval [0, 1] and take the reciprocal to generate the hidden disaster risk index.

6. The multi-source sensing fusion expressway slope geological disaster real-time early warning method according to claim 5, characterized in that, When the hidden disaster risk index continuously exceeds the dynamic threshold and the frequency domain distribution characteristics of the auxiliary monitoring parameter present non-stationary evolution, generate a geological disaster warning signal, including: Obtain the sequence of hidden disaster risk indexes and the sequence of frequency domain distribution characteristics of auxiliary monitoring parameters; Detect whether the hidden disaster risk index is continuously higher than the dynamic threshold in the continuous sampling period; Analyze whether the variance of the frequency domain distribution characteristics of the auxiliary monitoring parameter is monotonically increasing in the continuous sampling period; When the hidden disaster risk index is continuously higher than the dynamic threshold in the continuous sampling period and the variance of the frequency domain distribution characteristics is monotonically increasing in the continuous sampling period, generate a geological disaster warning signal.

Images