A loosening type deformation identification and early warning method for reservoir bank landslide, a storage medium and equipment

Abstract

This invention discloses a method, storage medium, and device for identifying and warning of loosening deformation in reservoir bank landslides, relating to the field of landslide early warning. The method includes: collecting multi-source data on surface displacement, deep displacement, reservoir water level, rainfall, and acoustic emission; extracting candidate anomalous jump points from the surface displacement data using a sliding window local anomaly factor detection method, and simultaneously extracting event rate and energy release rate surge points from the acoustic emission data; calculating the spatiotemporal matching degree between displacement jump points and acoustic emission surge points to determine preliminary events exhibiting brittle loosening characteristics; constructing a deformation spatial gradient field, calculating a local deformation mutation index, and combining it with time-domain anomaly markers to obtain a spatiotemporal joint anomaly index to confirm the final loosening event; calculating the trigger response ratio considering reservoir water level, rainfall, and hysteresis effects; and adaptively correcting a multi-level early warning threshold matrix based on the recovery damping coefficient to output the early warning level. This invention achieves a mechanical distinction between brittle loosening and creep deformation, with high early warning accuracy and low false alarm rate.

Description

Technical Field

[0001] This invention relates to the field of landslide early warning, and in particular to a method, storage medium and device for identifying and warning of loose deformation in reservoir bank landslides. Background Technology

[0002] Reservoir bank landslides refer to the geological hazard caused by the overall sliding of soil and rock masses along weak structural surfaces on the bank slope after the reservoir has been impounded and is in operation, due to factors such as the periodic rise and fall of the reservoir water level and rainfall infiltration. Monitoring and early warning of landslide disasters is one of the core tasks of geological disaster prevention and mitigation. Currently, landslide monitoring mainly relies on GNSS monitoring points deployed on the slope surface to obtain surface displacement data, supplemented by equipment such as deep displacement gauges, inclinometers, rain gauges, and water level gauges. By analyzing the deformation characteristics of the displacement-time curve, the deformation stage of the landslide (initial deformation, isotropic deformation, accelerated deformation) is determined, and corresponding early warning levels are issued. Typical early warning criteria include displacement rate thresholds, tangent angles, and acceleration indicators. However, the deformation of reservoir bank landslides often exhibits a "step-like" characteristic, that is, a sudden jump in the displacement curve followed by a period of stabilization. This deformation pattern is closely related to the brittle fracture, loosening, and sliding of local blocks within the landslide body, and is usually referred to as "loosening deformation."

[0003] Existing landslide deformation identification and early warning methods have the following main shortcomings: First, traditional methods typically rely solely on displacement data, using threshold or tangent angle methods to determine the deformation stage, making it difficult to distinguish between deformations with different mechanical mechanisms such as "brittle loosening" and "overall creep." For example, when a landslide exhibits uniform creep, the displacement rate may briefly exceed the threshold, leading to false alarms; conversely, rapid loosening of localized blocks may be missed if masked by overall smoothing.

[0004] Second, existing technologies do not adequately integrate and utilize multi-source data. Although some studies have incorporated inducing factors such as rainfall and reservoir water levels, few have combined acoustic emission / microseismic signals with displacement data for spatiotemporal analysis. Acoustic emission technology can capture high-frequency elastic waves generated by micro-fractures within rock and soil in real time, making it a direct means of monitoring brittle fracture. However, it is currently mainly used in rock mechanics laboratory tests or mine rockburst early warning systems, and has not yet been systematically integrated into the on-site monitoring and early warning of reservoir bank landslides.

[0005] Third, existing warning thresholds are mostly static empirical values, which cannot reflect the "deterioration" or "fatigue" effects of landslides after multiple deformation events. For example, after multiple loosening events, the structural surfaces of a landslide gradually become damaged, and its residual strength decreases. It is more likely to deform under the same external factors in the future, but the static threshold cannot be adjusted adaptively, resulting in a decrease in warning sensitivity or an increase in false alarm rate.

[0006] Fourth, most existing methods ignore the spatial differences in landslide deformation and rely solely on single-point data for global judgment, making it difficult to identify the location and depth of loosening of local blocks. Summary of the Invention

[0007] The technical problems to be solved by this invention include: how to distinguish between brittle loosening deformation and overall creep deformation from a mechanical perspective based on the spatiotemporal matching of displacement and acoustic emission, thereby reducing the false alarm rate; how to construct the deformation spatial gradient field of the monitoring section, and accurately locate the specific monitoring point where loosening deformation occurs by calculating the local deformation mutation index and fusing time-domain anomaly markers; how to establish a trigger response ratio model that considers reservoir water level, rainfall and historical lag effects, and quantitatively evaluate the contribution of inducing factors to loosening events; and how to adaptively and dynamically correct the multi-level early warning threshold matrix based on the recovery damping coefficient and early warning accuracy after each loosening event of the landslide body, so as to realize the closed-loop self-optimization of the early warning system.

[0008] Therefore, this invention proposes a method for identifying and warning of loosening deformation in reservoir bank landslides, comprising the following steps: S1. Collect multi-source monitoring data on the monitoring section of the reservoir bank landslide. The multi-source monitoring data includes surface displacement data, deep displacement data, reservoir water level data, rainfall data, and acoustic emission signal data inside the landslide body. S2. The sliding window local anomaly factor detection algorithm is used to extract candidate abnormal jump points in the time domain from the surface displacement data. At the same time, the sudden increase points of event rate and energy release rate are extracted from the acoustic emission signal data to obtain the candidate abnormal jump point set and the acoustic emission sudden increase point set. S3. Calculate the event matching degree between the candidate abnormal jump point set and the acoustic emission surge point set within the same spatiotemporal window. When the matching degree exceeds the first threshold, the corresponding spatiotemporal position is determined as a loosening deformation event with brittle loosening mechanical characteristics, and a loosening deformation preliminary judgment feature set is obtained. S4. Construct the deformation spatial gradient field of adjacent monitoring points on the same monitoring section, calculate the local deformation mutation index of each monitoring point, and determine the time domain anomaly identifier of each monitoring point based on the candidate anomaly jump point set. Combine the local deformation mutation index to calculate the spatiotemporal joint anomaly index. When the index exceeds the second threshold, the monitoring point is confirmed as the final loosening deformation event, and a loosening deformation confirmation feature set is obtained. S5. Perform correlation analysis between the loosening deformation confirmation feature set and the synchronized reservoir water level data and rainfall data, and calculate the trigger response ratio; S6. Based on the trigger response ratio and the adaptive multi-level early warning threshold matrix, a dynamic comparison is made to output the loosening landslide early warning level, wherein the adaptive multi-level early warning threshold matrix is ​​iteratively corrected according to the recovery damping coefficient determined by the ratio of residual displacement rate to background displacement rate after historical loosening events.

[0009] Preferably, the sliding window local anomaly detection algorithm in step S2 specifically includes: S21. Set an analysis window with a length of L and a sliding step size of ΔL, slide it on the time series of the surface displacement data, and calculate the local outlier factor of the data points in each window. S22. Mark the points where the local outlier exceeds the mean plus k times the standard deviation within the window as candidate outlier jump points, where k is a preset sensitivity coefficient. S23. Perform a time sequence continuity check on the candidate abnormal jump points, remove isolated noise points, and retain the cluster of continuously occurring abnormal jump points to form the candidate abnormal jump point set.

[0010] Preferably, the extraction process of sudden increases in the event rate and energy release rate of the acoustic emission signal data in step S2 specifically includes: S24. Perform short-time Fourier transform or wavelet packet transform on the acoustic emission signal to decompose it into at least 4 frequency bands; S25. Select a preset frequency band as the characteristic frequency band of brittle fracture, and calculate the acoustic emission event rate and energy release rate in the frequency band; S26. Define the event rate spike point as: the event rate at the current moment exceeds N times the standard deviation of the moving average event rate of the previous T hours, and at the same time the energy release rate exceeds A times the background value. S27. Define the point of sudden increase in energy release rate as: the energy release rate at the current moment exceeds B times the standard deviation of the moving average energy release rate of the previous T hours, where T, N, A, and B are all preset values.

[0011] Preferably, the event matching degree in step S3 is calculated according to the following formula:

[0012] in, The set of candidate abnormal jump points, Centered on the candidate abnormal jump point time, the time window The set of acoustic emission abrupt increase points within, The time decay coefficient, Let be the absolute value of the time difference between the displacement jump point and the matched acoustic emission surge point; when The event was determined to be a loosening deformation event. The threshold is the first threshold.

[0013] Preferably, the construction of the deformation space gradient field and the calculation of the spatiotemporal joint anomaly index in step S4 specifically include: S41. Extract the surface displacement rate values ​​of all monitoring points on the same monitoring section at the same time, and use inverse distance weighted interpolation to generate a continuous spatial distribution map of displacement rate. S42. For each monitoring point i Take all neighboring monitoring points within a radius R centered on it, and calculate the neighborhood displacement rate. mean and standard deviation ; S43. Calculate the local deformation abruptness index. and will Normalize to the [0,1] interval; S44. Define the time-domain anomaly identifier. If monitoring point i If the candidate abnormal jump point set is included, the value is 1; otherwise, it is 0. S45. Calculate the spatiotemporal joint anomaly index using the following formula:

[0014] in, This is the spatial mutation threshold. For indicator functions, and For the weighting coefficients, satisfying And all are greater than 0; when This was confirmed as the final loosening deformation event. This is the second threshold.

[0015] Preferably, the trigger response ratio in step S5 is calculated according to the following formula:

[0016] in, The jump amplitude of the deformation jump event in the feature set of the loosened deformation confirmation; The effective range of reservoir water level within N days prior to the event, calculated as follows: calculate, For the first Daily changes in reservoir water level The attenuation coefficient is the effect of water level. The cumulative rainfall within the same time period, calculated as follows: calculate, For the first Daily rainfall, This is the rainfall attenuation coefficient; These are the weighting coefficients for reservoir water level and rainfall, respectively. It is the sum of the lag times of all historical events of detachment. For the total observation time, This represents the lag penalty coefficient.

[0017] Preferably, the process of constructing and dynamically correcting the adaptive multi-level early warning threshold matrix in step S6 specifically includes: S61. Based on the inversion analysis of historical landslide events, construct an initial multi-level early warning threshold matrix. The rows correspond to different reservoir water level operating conditions, and the columns correspond to different loosening and deformation confirmation index threshold levels. S62. Record the residual displacement rate after each loosening deformation event ends. Compared with the background displacement rate before the event occurred Calculate the recovery damping coefficient And store the coefficient in the historical damping sequence; S63. Statistically calculate the average damping coefficient of the most recent M loosening events. And calculate the early warning accuracy of the current early warning system. ,in TP To ensure the correct number of warnings, FP This represents the number of false alarms. S64. Correct the current warning threshold using the following formula:

[0018] in, The initial warning threshold, As a damping factor, The number of loosening events that have occurred. For accuracy feedback coefficient, For target accuracy; S65. When the mean of the historical damping sequence exceeds the preset upper limit. If the residual displacement rate is still greater than the background displacement rate after three consecutive loosening events, the subsequent warning level will be forcibly increased by one level.

[0019] Preferably, the acoustic emission signal data is collected by broadband accelerometers buried at different depths in the landslide body. The depths include a shallow layer of 0-5m, a middle layer of 5-15m, and a deep layer of 15-30m. The signals of each channel are sampled synchronously, and the sampling frequency is not lower than a preset value. In step S3, the matching degree at each depth is calculated. When the difference between the matching degree of the deep layer and the matching degree of the shallow layer is greater than the preset value, it is determined that the deep slip zone is loose, and the output warning level is accompanied by a depth indicator.

[0020] A second aspect of the present invention provides a storage medium that stores instructions and data for implementing the aforementioned method for identifying and warning of loosening deformation in reservoir bank landslides.

[0021] A third aspect of the present invention provides a loosening deformation identification and early warning device for reservoir bank landslides, comprising: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium for the aforementioned loosening deformation identification and early warning method for reservoir bank landslides.

[0022] The present invention has the following beneficial effects: 1. By introducing acoustic emission signals and performing spatiotemporal matching with displacement anomaly jump points, the mechanical mechanism identification of brittle loosening deformation events was achieved for the first time, effectively eliminating false anomalies caused by overall creep or instrument noise, and significantly reducing the false alarm rate.

[0023] 2. By constructing a spatial gradient field and a local deformation mutation index, and combining it with time-domain anomaly markers to form a spatiotemporal joint anomaly index, the monitoring point location of loosening deformation can be accurately located. Based on the difference in matching degree of acoustic emission sensors at different depths, the depth of loosening (shallow weathered layer or deep slip zone) can be determined, providing clear guidance information for on-site inspection and engineering reinforcement.

[0024] 3. The designed trigger response ratio model comprehensively considers reservoir water level, rainfall and historical lag effects, which is more physically reasonable than the traditional single threshold criterion.

[0025] 4. The proposed adaptive threshold correction method based on the recovery damping coefficient and early warning accuracy enables the early warning system to dynamically adjust its sensitivity according to the degree of "fatigue deterioration" of the landslide and the real-time feedback performance, thus avoiding the limitations of static thresholds.

[0026] 5. This invention fully utilizes existing monitoring equipment, such as GNSS, inclinometers, rain gauges, and water level gauges, and is compatible with low-cost acoustic emission sensors. Hardware investment is controllable, making it suitable for universal monitoring and early warning of landslides on reservoir banks. Experiments show that this method accurately predicted deep slip zone loosening events approximately one day in advance in a landslide application on a reservoir bank, verifying its reliability and practicality. Attached Figure Description

[0027] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0028] Figure 1 This is a simplified flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the hardware device of the present invention. Detailed Implementation

[0029] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0030] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0031] Before formally describing the present invention, a general description of the solution of the present invention will be given first to facilitate understanding.

[0032] Example 1: Please refer to Figure 1 The present invention provides a method for identifying and warning of loosening deformation in reservoir bank landslides, comprising the following steps: S1. Collect multi-source monitoring data on the monitoring section of the reservoir bank landslide. The multi-source monitoring data includes surface displacement data, deep displacement data, reservoir water level data, rainfall data, and acoustic emission signal data inside the landslide body. It should be noted that the acoustic emission signal data is collected by broadband accelerometers buried at different depths in the landslide body. The depths include shallow (0-5m), middle (5-15m), and deep (15-30m). The signals of each channel are sampled synchronously, and the sampling frequency is not lower than a preset value. In step S3, the matching degree at each depth is calculated. When the difference between the matching degree of the deep layer and the matching degree of the shallow layer is greater than the preset value, it is determined that the deep slip zone is loose, and the output warning level is accompanied by a depth indicator.

[0033] As one embodiment, step S1 specifically includes: Multiple sensors were deployed on typical monitoring sections of the reservoir bank landslide. Surface displacement data was acquired through Global Navigation Satellite System (GNSS) monitoring stations, with monitoring points arranged along the longitudinal and transverse directions of the landslide, with adjacent points spaced 20–50 meters apart. Data was sampled at 1-hour intervals and transmitted to the data center in real time.

[0034] Deep displacement data are acquired using fixed inclinometers or array-type displacement gauges. The borehole depth must penetrate the potential slip zone by 3–5 meters, with a measurement accuracy of no less than ±0.1 mm / 0.5 m. Reservoir water level data are obtained from the reservoir management system or self-built water level gauges, with a collection frequency of once per hour. Rainfall data are obtained using tipping bucket rain gauges or from nearby meteorological stations, and recorded in daily or hourly cumulative formats.

[0035] Acoustic emission signal data were collected using broadband accelerometers buried at different depths within the landslide body. In this embodiment, three boreholes were drilled at the leading, middle, and trailing edges of the landslide body. In each borehole, a broadband acoustic emission sensor (e.g., PAC R15α) was installed at the shallow (0–5 m), middle (5–15 m), and deep (15–30 m) layers. The sensor center frequency was 150 kHz, and the sampling frequency was no less than 100 kHz. The signal was continuously recorded by a data acquisition instrument after being preamplified (40 dB gain). The clock synchronization accuracy of all sensors was better than 1 millisecond.

[0036] Taking a landslide on the bank of a reservoir as an example: a total of 12 GNSS monitoring points, 5 inclinometer boreholes, and 3 acoustic emission boreholes were set up along the monitoring section. Data collection was conducted from June 2023 to October 2024, obtaining approximately 150,000 surface displacement data points, approximately 30,000 deep displacement data points, approximately 12,000 reservoir water level data points, approximately 500 rainfall data points, and approximately 2TB of continuous waveform data from acoustic emission signals.

[0037] S2. The sliding window local anomaly factor detection algorithm is used to extract candidate abnormal jump points in the time domain from the surface displacement data. At the same time, the sudden increase points of event rate and energy release rate are extracted from the acoustic emission signal data to obtain the candidate abnormal jump point set and the acoustic emission sudden increase point set. As one embodiment, step S2 specifically includes: (a) Detection of Local Loop Factor (LOF) in Sliding Window of Surface Displacement Data For the surface displacement time series of each monitoring point (Sampling interval Δt = 1 hour), first calculate the hourly displacement rate. (Unit: mm / h) forms a rate sequence. The sliding window length is set to L = 24 (i.e., 24 hours), and the sliding step size ΔL = 1 (i.e., sliding once per hour). Within each window, the Local Outlier Factor (LOF) of each data point relative to its k-neighborhood is calculated, where k is 5. The LOF value is defined as the average ratio of the local reachability density of point p to the local reachability density of all points in its neighborhood. When the LOF value is greater than k times the standard deviation of the mean LOF of all points within the window (k is 2.5), the point is marked as a candidate outlier jump point.

[0038] In this embodiment, LOF detection was performed on the surface displacement rate sequence of GNSS01 point, identifying 47 candidate anomalous jump points. Considering that transient instrument noise may produce isolated points, a temporal continuity check was performed: only clusters with at least two candidate anomalous jump points appearing within a consecutive 3-hour period were retained, while isolated single points were discarded. The final set of candidate anomalous jump points was obtained. It contains 32 jump events.

[0039] (ii) Extraction of sudden increases in event rate and energy release rate of acoustic emission signal data First, the raw acoustic emission signal is preprocessed: a fourth-order Butterworth bandpass filter is used to retain the 10kHz–50kHz frequency band (corresponding to the dominant frequency range of brittle rock fracture). The filter's passband attenuation is less than 1dB, and its stopband attenuation is greater than 40dB. After filtering, a threshold voltage (e.g., 3 times the background noise) is set to detect acoustic emission events. The event rate is defined as the number of events exceeding the threshold per second. The energy release rate is defined as the integral of the area enclosed by the waveform envelope of a single event or the square of the signal.

[0040] Calculate the moving average background value: Calculate the average event rate hourly using a 24-hour window. and average energy release rate The event rate E(t) at the current moment simultaneously satisfies: ( (The standard deviation of the event rate within 24 hours). And the energy release rate at the same time ; This moment is then marked as a point of sudden increase in the event rate. Similarly, the criteria for judging a point of sudden increase in the energy release rate are as follows: .

[0041] In this embodiment, data from a deep acoustic emission sensor (15-30m) were analyzed, and a total of 28 acoustic emission spike points were extracted, forming an acoustic emission spike point set. .

[0042] It should be noted that the sliding window local anomaly detection algorithm in step S2 specifically includes: S21. Set an analysis window with a length of L and a sliding step size of ΔL, slide it on the time series of the surface displacement data, and calculate the local outlier factor of the data points in each window. Specifically, the window length is set to L = 24 hours, and the sliding step size ΔL = 1 hour. For the surface displacement rate sequence... Take it starting from the 1st hour to This serves as the first window. The local reachability density of each data point within this window relative to its five nearest neighbors is calculated, thus obtaining the Local Outlier Factor (LOF) for each point. A higher LOF value indicates a more outlier point. The LOF values ​​of all points within the window form a set.

[0043] S22. Mark the points where the local outlier exceeds the mean plus k times the standard deviation within the window as candidate outlier jump points, where k is a preset sensitivity coefficient. Specifically, let the mean of the LOF within the window be . The standard deviation is The sensitivity coefficient k is taken as 2.5. Then, if at a certain point... These are marked as candidate abnormal jump points. For example, if the mean LOF within a certain window is 1.2 and the standard deviation is 0.3, then the threshold = 1.2 + 0.75 = 1.95, and a point with LOF = 2.1 is marked.

[0044] S23. Perform a time sequence continuity check on the candidate abnormal jump points, remove isolated noise points, and retain the cluster of continuously occurring abnormal jump points to form the candidate abnormal jump point set.

[0045] Specifically, a temporal continuity check is performed: because noise may generate isolated high LOF points, at least two marked points are required within a consecutive 3-hour period to be retained. For example, if only 14:00 is marked out of the three times 14:00, 14:30, and 15:00, while 15:30 and 16:00 are not marked, then the 14:00 point is discarded. If 14:00 and 14:30 are marked consecutively, then these two points are retained, forming a jump event. The final result is a set of candidate abnormal jump points. .

[0046] In this embodiment, a total of 32 candidate jump points were identified. After timing verification, 5 isolated points were removed, and 27 valid points were retained. The jump range of these points ranged from 3mm to 45mm.

[0047] It should be noted that the extraction process of sudden increases in the event rate and energy release rate of the acoustic emission signal data in step S2 specifically includes: S24. Perform short-time Fourier transform or wavelet packet transform on the acoustic emission signal to decompose it into at least 4 frequency bands; Specifically, the original acoustic emission signal is a continuous waveform with a sampling frequency of 100kHz. A short-time Fourier transform (STFT) with a window length of 1024 points (approximately 10ms), a Hanning window, and a 50% overlap rate is used to obtain the time-frequency spectrum. Alternatively, wavelet packet transform can be used, decomposing the signal into 4 layers to obtain 16 frequency band subspaces.

[0048] S25. Select a preset frequency band as the characteristic frequency band of brittle fracture, and calculate the acoustic emission event rate and energy release rate in the frequency band; Specifically, based on prior experiments, the dominant frequency range of brittle rock fracture is approximately 10kHz to 50kHz. A sub-band corresponding to this range (e.g., sub-bands 3 to 5) is selected, and waveforms within this band are extracted. The event count per second is calculated. An event is defined as a waveform pulse exceeding a threshold voltage (3 times the background noise RMS).

[0049] S26. Define the event rate spike point as: the event rate at the current moment exceeds N times the standard deviation of the moving average event rate of the previous T hours, and at the same time the energy release rate exceeds A times the background value. Specifically, the moving average background is calculated by using a 24-hour window and calculating the event rate every hour. (Unit: times / second). Calculate the average event rate for the previous 24 hours. and standard deviation If the event rate at the current moment satisfies And at the same time, the energy release rate If the sum of the event waveform envelope areas per second is greater than twice the average energy release rate of the previous 24 hours, it is marked as a point of sudden increase in event rate.

[0050] S27. Define the point of sudden increase in energy release rate as: the energy release rate at the current moment exceeds B times the standard deviation of the moving average energy release rate of the previous T hours, where T, N, A, and B are all preset values.

[0051] Specifically, the judgment of the point of sudden increase in energy release rate is slightly stricter: It is not required that the event rates increase simultaneously, because some high-energy, low-frequency events may not have high event rates. The two conditions can be used independently to obtain two sets of sudden increase points, which are then combined in step S3.

[0052] In this embodiment, during the period from 14:00 to 15:00 on August 15, 2023, the event rate of a certain landslide increased from an average of 5 times / second to 28 times / second (more than 3 times the standard deviation, where the standard deviation is 3), and the energy release rate increased from 0.2 J / s to 0.9 J / s (more than 2 times), triggering a sudden increase in the event rate. At 15:20, there was a separate sudden increase in the energy release rate to 1.2 J / s, but the event rate was only 8 times / second, which was also separately marked as a sudden increase in the energy release rate.

[0053] S3. Calculate the event matching degree between the candidate abnormal jump point set and the acoustic emission surge point set within the same spatiotemporal window. When the matching degree exceeds the first threshold, the corresponding spatiotemporal position is determined as a loosening deformation event with brittle loosening mechanical characteristics, and a loosening deformation preliminary judgment feature set is obtained. As one embodiment, step S3 specifically includes: For each candidate abnormal jump point According to the time of its occurrence Centered on the time window Every hour, collect all acoustic emission spikes within that window to form... Define the matching degree:

[0054] in For point and The absolute value of the time difference between the nearest acoustic emission spikes, if there are no acoustic emission spikes within the window. .coefficient We set it to 0.2 (so that for every 5-hour increase in time difference, the matching accuracy decreases to approximately 1 / e). When The event is determined to be a loosening deformation event. In this embodiment, Take 0.5.

[0055] right The above calculation is performed for each candidate jump point. For example, GNSS01 experienced a displacement jump at 14:00 on August 15, 2023, and within 6 hours of that point, the deep acoustic emission sensor showed a sudden increase in event rate at 15:20 on the same day. Hours, calculated The event was identified as a loosening deformation event. A preliminary feature set of loosening deformation events, comprising 18 events, was ultimately obtained.

[0056] For acoustic emission signals at different depths, the matching degree can be calculated layer by layer. If the matching degree of the middle and deep layers (5-30m) is significantly higher than that of the shallow layers (difference > 0.3), it is determined that the deep slip zone is loose.

[0057] It should be noted that the event matching degree in step S3 is calculated according to the following formula:

[0058] in, The set of candidate abnormal jump points, Centered on the candidate abnormal jump point time, the time window The set of acoustic emission abrupt increase points within, The time decay coefficient, Let be the absolute value of the time difference between the displacement jump point and the matched acoustic emission surge point; when The event was determined to be a loosening deformation event. The threshold is the first threshold.

[0059] As one embodiment, the matching degree calculation in step S3 is as follows: Let there be a set of candidate anomaly jump points. Acoustic emission sudden increase point set For each Search All that happened in Points within the range are denoted as In this embodiment Hour.

[0060] The matching degree formula is:

[0061] because It is either 1 (if there is at least one acoustic emission spike within the window) or 0 (if there is none), therefore it is actually If there is an acoustic emission point within the window, then 0; otherwise, 0.

[0062] Take 0.2 (per hour) ). for The absolute value of the time difference from the most recent acoustic emission spike. For example, if the time difference is 1.33 hours, then... First threshold Taking 0.5 corresponds to a time difference of approximately 3.47 hours. That is, if the time difference between the displacement jump point and the nearest acoustic emission surge point is greater than 3.47 hours, even if there is a match, it will not be recognized due to attenuation, thus ensuring the "simultaneity" requirement of brittle release.

[0063] For example, three candidate jump points, although they had sudden increases in acoustic emission within the window, had a time difference of more than 4 hours and a matching degree of less than 0.45, and were not judged as loosening events.

[0064] S4. Construct the deformation spatial gradient field of adjacent monitoring points on the same monitoring section, calculate the local deformation mutation index of each monitoring point, and determine the time domain anomaly identifier of each monitoring point based on the candidate anomaly jump point set. Combine the local deformation mutation index to calculate the spatiotemporal joint anomaly index. When the index exceeds the second threshold, the monitoring point is confirmed as the final loosening deformation event, and a loosening deformation confirmation feature set is obtained. As one embodiment, step S4 specifically includes: (a) Constructing the gradient field in the deformation space For all 12 GNSS monitoring points on the same monitoring section, at any given time Surface displacement rate at each point A spatial discrete point set is constructed. The spatial distribution map of displacement rate of the entire landslide surface is generated by inverse distance weighted interpolation (IDW), with an interpolation grid size of 5m×5m and a power exponent of 2.

[0065] (ii) Calculation of the local deformation abrupt change index For each monitoring point With its coordinates as the center, the radius The neighborhood is defined by a range of meters. The neighborhood must contain at least three other monitoring points. Calculate the range of all monitoring points within the neighborhood (including the point). The average displacement rate of itself and standard deviation The local deformation abrupt change index at that point is:

[0066] when When it is close to 0, take Then all monitoring points Linearly normalized to the [0,1] interval. In this embodiment, the spatial abrupt change threshold... A value of 2.0 (i.e., more than 2 standard deviations) is considered a significant spatial anomaly.

[0067] (III) Determining Time-Domain Anomaly Identifiers Based on the candidate abnormal jump point set obtained in step S2 For the current moment, if the monitoring point If this moment is marked as a candidate anomalous jump point, then the time-domain anomalous identifier... Otherwise, it is 0.

[0068] (iv) Calculation of the spatiotemporal joint anomaly index Set weight coefficients , (satisfy Indicator functions when The value is 1 if the time condition is met, and 0 otherwise. Spatiotemporal Joint Anomaly Index:

[0069] Second threshold Take 0.6. That is, only when... Only then is the monitoring point confirmed as the final loosening deformation event. In this embodiment, spatial verification is performed on the 18 preliminary events obtained in step S3. Although three of these events show anomalies in the time domain, the surrounding neighboring monitoring points also deform at a similar rate ( ),lead to The value just reached the threshold and was retained; two other events had significantly different deformation rates in their surrounding neighborhoods. ),but One event was confirmed as a strong loosening event; another event had an overall abnormality in its surrounding neighborhood but its own time domain abnormality was not significant. ), These were excluded. Ultimately, 15 loosening deformation events were confirmed, forming a loosening deformation confirmation feature set.

[0070] It should be noted that the construction of the deformation space gradient field and the calculation of the spatiotemporal joint anomaly index in step S4 specifically include: S41. Extract the surface displacement rate values ​​of all monitoring points on the same monitoring section at the same time, and use inverse distance weighted interpolation to generate a continuous spatial distribution map of displacement rate. For example, at a certain moment Surface displacement rate at all monitoring points A discrete point set is constructed using units of mm / h. A continuous grid surface is generated using inverse distance weighted interpolation (IDW). The exponent is set to 2, the search radius is 50 meters, and the minimum number of points is 3.

[0071] S42. For each monitoring point i, take all neighboring monitoring points within a radius R centered on i, and calculate the mean value of the neighborhood displacement rate. and standard deviation ; For example, for each monitoring point Take its radius All monitoring points within a certain range are considered as the neighborhood. In this embodiment, the neighborhood of GNSS01 includes three points: GNSS02, GNSS03, and GNSS04. The average displacement rate of these four points is calculated. and standard deviation For example, if the speeds at four points at a certain moment are 1.2, 1.3, 1.1, and 1.5 mm / h respectively, then the mean is 1.275 and the standard deviation is 0.147.

[0072] S43. Calculate the local deformation abruptness index. and will Normalize to the [0,1] interval; For example, monitoring points rate ,but All monitoring points Value normalization: The minimum in this embodiment The value is 0.1, with a maximum of 4.2. This is the normalized value of GNSS01. Spatial mutation threshold Set to 2.0 (corresponding to the original) ),Right now The time-indicative function is 1.

[0073] S44. Define the time-domain anomaly identifier. If monitoring point i belongs to the candidate abnormal jump point set, the value is 1; otherwise, it is 0. For example, if the current monitoring point At the current moment, it is marked as a candidate abnormal jump point (i.e., the set of candidate abnormal jump points in step S2). (in the middle), then Otherwise, 0.

[0074] S45. Calculate the spatiotemporal joint anomaly index using the following formula:

[0075] in, This is the spatial mutation threshold. For indicator functions, and For the weighting coefficients, satisfying And all are greater than 0; when This was confirmed as the final loosening deformation event. This is the second threshold.

[0076] Finally, the weighting coefficients , Spatiotemporal Joint Anomaly Index .when This is confirmed as the final loosening deformation event. For example: Situation A: ,but ,confirm.

[0077] Situation B: ,but It is exactly equal to the threshold, so it is confirmed (acceptable).

[0078] Case C: ,but Unconfirmed.

[0079] Situation D: ,but Unconfirmed.

[0080] The weighting coefficients can be adjusted according to the landslide type. For example, for homogeneous landslides, the spatial gradient may not be significant, so the weighting coefficients can be reduced. For landslides with well-developed fissures, spatial differences are significant, which can increase... In this embodiment, the optimal coefficient is obtained through grid search optimization. .

[0081] S5. Perform correlation analysis between the loosening deformation confirmation feature set and the synchronized reservoir water level data and rainfall data, and calculate the trigger response ratio; As one embodiment, step S5 specifically includes: For each confirmed loosening deformation event, extract the following parameters: Jump range (mm): The cumulative displacement increment from the start to the end of the event, which averages about 25mm in this embodiment.

[0082] Effective fluctuation of reservoir water level Take the absolute value of the daily change in reservoir water level over N=15 days before the event, and calculate using the decay memory model:

[0083] in For the first Daily reservoir water level change (current day's water level minus the previous day's water level, in meters), attenuation coefficient (That is, the influence weight of the previous day is 0.85, the weight of the two days before is 0.7225, and so on, decreasing sequentially). In this embodiment, an event occurred on August 15, 2023, and the weighted sum of the absolute values ​​of the daily fluctuations in reservoir water level over the previous 15 days was 2.35m.

[0084] cumulative rainfall Similarly, taking the first 15 days, the attenuation coefficient... :

[0085] in For the first Daily rainfall (in mm). In this embodiment, the weighted average rainfall for the 15 days prior to the event was 87.3 mm.

[0086] Lag Time Accumulation: Calculate the number of days lag between the occurrence of all loosening events in the history of this landslide and the triggering factors (rapid drop in reservoir water level or heavy rainfall), and sum them up. In this embodiment, the total historical lag days are 42 days, and the total observation duration is... Heaven, delayed penalty coefficient .

[0087] Trigger-response ratio calculation formula:

[0088] Pick (The contribution coefficient of the displacement for every 1m change in reservoir water level). (The contribution factor of each mm of rainfall to displacement). Substitute the values:

[0089] This value serves as the confirmation index for loosening and deformation.

[0090] It should be noted that the trigger response ratio in step S5 is calculated using the following formula:

[0091] in, The jump amplitude of the deformation jump event in the feature set of the loosened deformation confirmation is expressed in mm. The effective range of reservoir water level within N days prior to the event, calculated as follows: calculate, For the first Daily changes in reservoir water level The attenuation coefficient is the effect of water level. The cumulative rainfall within the same time period, calculated as follows: calculate, For the first Daily rainfall, This is the rainfall attenuation coefficient; These are the weighting coefficients for reservoir water level and rainfall, respectively. It is the sum of the lag times of all historical events of detachment. For the total observation time, This represents the lag penalty coefficient.

[0092] As one embodiment, the trigger response ratio calculation in step S5 is as follows: For a certain loosening deformation event, record the start time of the event. and end time Jump range ,in This represents the cumulative displacement of the surface.

[0093] Before the event occurred The weather serves as a window into influencing factors. For reservoir water levels, the absolute value of daily changes is calculated. Unit: m. Attenuation coefficient. Then the first The weighting coefficients from 10 days ago are .therefore:

[0094] Similarly, rainfall attenuation coefficient , ,Notice The rainfall on the day of the event is generally not included (because landslide deformation usually lags behind).

[0095] In this embodiment, mm / m, mm / mm. Historical lag time This is the sum of the number of days from the occurrence of the triggering factor (daily drop in reservoir water level > 0.5m or daily rainfall > 50mm) to the event itself. Total observation duration. The time (in days) from the first monitoring to the current time. Lag penalty coefficient. The correction factor is then... .

[0096] For example: A landslide experienced 5 loosening events in the past year, with lag times of 3, 4, 2, 5, and 4 days respectively, totaling 18 days. The correction factor = 1 + 0.2 × 18 / 365 = 1.0099, which is almost no correction. If the total lag time is large (such as the cumulative effect caused by long-term slow rainfall), the correction factor will be greater than 1, increasing the response ratio, that is, it is believed that long-term lag will exacerbate the risk.

[0097] S6. Based on the trigger response ratio and the adaptive multi-level early warning threshold matrix, a dynamic comparison is made to output the loosening landslide early warning level, wherein the adaptive multi-level early warning threshold matrix is ​​iteratively corrected according to the recovery damping coefficient determined by the ratio of residual displacement rate to background displacement rate after historical loosening events.

[0098] As one embodiment, step S6 specifically includes: (I) Initial multi-level early warning threshold matrix Based on the inversion analysis of historical landslide data, an initial threshold matrix is ​​established. The matrix rows correspond to reservoir water level ranges: low water level (<145m), medium water level (145~160m), and high water level (>160m); the columns correspond to warning levels: blue (caution), yellow (warning), orange (alert), and red (alarm). The initial thresholds are as follows (trigger response ratio R value): Table 1 Initial Threshold Settings

[0099] In this embodiment, when the event occurred, the reservoir water level was 152m (medium water level), R=1.587, which is within the yellow warning range (1.5~2.0), and an initial yellow warning was issued.

[0100] (II) Calculation of Restoration Damping Coefficient After each detachment event, the average displacement rate over the following 5 consecutive days is taken as the residual displacement rate. The average displacement rate during the pre-event stabilization period (at least 10 days) is taken as the background displacement rate. Calculate the restoring damping coefficient. In this embodiment, before the event mm / d, after the event mm / d, then .

[0101] (III) Adaptive Threshold Correction Assuming history has already occurred The damping coefficients for the subsequent loosening events were 1.1, 1.2, 1.3, 1.25, and 1.375, respectively, totaling 6.225, with an average damping coefficient of 1.245. Damping decay factor. .

[0102] Calculate the current early warning accuracy rate Where TP represents the number of correct alerts (the actual detachment event occurred and the alert level is ≥ blue), and FP represents the number of false alarms (an alert was issued but no detachment occurred). In this embodiment, among the last 10 alerts, TP=8 and FP=2, then... Target accuracy Accuracy feedback coefficient .

[0103] The yellow warning threshold (under medium water level conditions) is corrected as follows:

[0104] The corrected yellow alert threshold decreased from 1.5 to approximately 1.4. The event R=1.587 is still greater than 1.399, but closer to the threshold, indicating that landslides are more likely to trigger alerts after deterioration.

[0105] Meanwhile, since the residual displacement rate was greater than the background displacement rate after three consecutive loosening events ( This triggers the rule of mandating a higher level of warning, so an orange warning is actually issued (the original yellow warning is upgraded to orange).

[0106] The output warning information includes: warning level (orange), time of occurrence of the loosening event (14:00 on August 15, 2023), affected area (50 meters around GNSS01 point), and recommended measures (intensified monitoring and patrols).

[0107] It should be noted that the construction and dynamic correction process of the adaptive multi-level early warning threshold matrix in step S6 specifically includes: S61. Based on the inversion analysis of historical landslide events, construct an initial multi-level early warning threshold matrix. The rows correspond to different reservoir water level operating conditions, and the columns correspond to different loosening and deformation confirmation index threshold levels. Initial threshold matrix The system is based on statistical analysis of historical landslide events and engineering experience. The reservoir water level is divided into three operating ranges: low water level (<145m), medium water level (145~160m), and high water level (>160m). The warning levels are divided into four levels: blue (caution), yellow (warning), orange (alert), and red (alarm). Each warning level under each operating condition corresponds to a trigger response ratio. The lower limit value. The initial value is determined by trial and error or ROC curve optimization. In this embodiment, the values ​​in Table 1 are used.

[0108] S62. Record the residual displacement rate after each loosening deformation event ends. Compared with the background displacement rate before the event occurred Calculate the recovery damping coefficient And store the coefficient in the historical damping sequence; After each detachment event, the average displacement rate over the following 120 hours is taken as... The average displacement rate during the stable period (at least 240 hours, without significant jumps) prior to the event is taken as... .calculate For example, in this embodiment, the background velocity before the event is 0.8 mm / d, and the residual velocity after the event is 1.1 mm / d, then... Store this value in the database.

[0109] S63. Statistically calculate the average damping coefficient of the most recent M loosening events. And calculate the early warning accuracy of the current early warning system. TP represents the number of correct alerts, and FP represents the number of false alarms. Statistics recently Average damping coefficient of the secondary loosening event Calculate the accuracy of early warnings. In the past period (e.g., the last 30 days), after each warning (system output blue or higher), if loosening deformation (or displacement acceleration exceeding the threshold) actually occurs within the following 72 hours, it is recorded as a correct warning (TP); if no abnormality occurs within 72 hours after the warning, it is recorded as a false alarm (FP). In this embodiment... ,but Target accuracy It is set to 0.85, and can be adjusted according to management requirements.

[0110] S64. Correct the current warning threshold using the following formula:

[0111] in, The initial warning threshold, This is the damping decay factor, with a value ranging from 0.01 to 0.1. The number of loosening events that have occurred. This is the accuracy feedback coefficient, with a value ranging from 0.05 to 0.2. For target accuracy; When the sum of damping coefficients is large (indicating poor slope recovery capacity), the threshold is lowered and the warning is more sensitive; when the accuracy is lower than the target, the threshold is raised to reduce false alarms; when the accuracy is higher than the target, the threshold is lowered to increase the hit rate.

[0112] S65. When the mean of the historical damping sequence exceeds the preset upper limit. If the residual displacement rate is still greater than the background displacement rate after three consecutive loosening events, the subsequent warning level will be forcibly increased by one level.

[0113] If the historical damping sequence mean exceeds (i.e., the residual rate is more than 1.5 times the background rate), indicating that the landslide body has severely deteriorated. Even if the response ratio does not reach the next level threshold, the warning level should be forcibly raised by one level. Alternatively, if after three consecutive loosening events, the residual displacement rate is greater than the background displacement rate in each instance (i.e., Similarly, a forced upgrade of one level is imposed. In this embodiment, it is done three times consecutively. The values ​​are 1.2, 1.25, and 1.375 respectively, all greater than 1.0, therefore the yellow alert is forcibly upgraded to orange.

[0114] Example 2 To better understand the present invention, a specific application embodiment is given herein.

[0115] Case Background: A landslide on the bank of a reservoir had a volume of approximately 800,000 cubic meters and a slope of 25-35°, with the main lithology being sandstone interbedded with mudstone. Twelve GNSS monitoring points, five inclinometer boreholes, and three acoustic emission boreholes (three layers per borehole) were installed on the landslide. Monitoring was conducted from June 2023 to October 2024.

[0116] Example process: 1. Data Acquisition: On August 15, 2023, GNSS01 point displacement data showed an acceleration that began at 8:00 AM, with a cumulative displacement increase of 28 mm by 2:00 PM, at a rate of 4.2 mm / h, far exceeding the normal background value (0.1 mm / h). Simultaneously, the deep acoustic emission sensor recorded an increase in event rate from 5 times / second to 32 times / second and energy release rate from 0.2 J / s to 1.1 J / s between 2:30 PM and 3:00 PM. Reservoir water level data showed a rapid drop from 155 m to 148 m in the previous week, with a maximum daily drop of 0.6 m; the cumulative rainfall over the previous 15 days was 68 mm, but mainly concentrated between August 10-12 (with a maximum daily rainfall of 45 mm).

[0117] 2. Time Domain Anomaly Detection: Using the LOF algorithm (L=24h, k=5, threshold 2.5σ), the velocity point of GNSS01 at 14:00 on August 15th was marked as a candidate anomaly jump point, while no significant anomalies were found in the nearby GNSS02 and GNSS03.

[0118] 3. Acoustic emission matching: Time window ΔT=6h, time difference between the deep acoustic emission surge point and the displacement jump point is 0.5h, β=0.2, the matching degree is calculated as M=e^{-0.1}=0.905>0.5, which is judged as a preliminary loosening event.

[0119] 4. Spatial gradient field verification: At this moment, the velocity of GNSS01 is 4.2 mm / h, and the average velocity of the surrounding neighborhood (GNSS02, 03, 04) is 0.15 mm / h with a standard deviation of 0.05 mm / h. After normalization, it far exceeds the 2.0 threshold. Time domain identifier. Spacetime Joint Index This was confirmed as the final loosening event.

[0120] 5. Trigger-response ratio calculation: Jump amplitude Effective fluctuation of reservoir water level (Attenuation coefficient 0.85), cumulative rainfall (Attenuation coefficient 0.9), α=5, β=0.05, total historical lag 42 days. With γ=0.2, we get R=1.587.

[0121] 6. Adaptive Early Warning: The initial yellow threshold for the water level is 1.5. Due to the historical average damping coefficient of 1.245 and accuracy of 0.8, the corrected threshold is 1.399. R > threshold. Furthermore, because the residual velocity is greater than the background velocity three times consecutively, the warning is forcibly upgraded to orange. Output: "Orange warning, loose deformation, possible location is within 50 meters of GNSS01 point, depth 15-30m (deep slip zone). Increased monitoring and preparation of emergency measures are recommended." 7. Result Verification: On August 16, 2023, a field inspection revealed new cracks on the slope near GNSS01 point. The cracks were approximately 2 cm wide and extended in the same direction as the slip zone. In early September, inclinometer data showed a peak shear deformation at a depth of 22 m, confirming the prediction of deep slip zone loosening. This method successfully provided an early warning approximately one day in advance, preventing casualties.

[0122] Example 3 Please see Figure 2 , Figure 2 This is a schematic diagram of the hardware device in operation according to an embodiment of the present invention. The hardware device specifically includes: a loosening deformation identification and early warning device 401 for reservoir bank landslides, a processor 402, and a storage medium 403.

[0123] A loose deformation identification and early warning device 401 for reservoir bank landslides: The loose deformation identification and early warning device 401 for reservoir bank landslides implements the loose deformation identification and early warning method for reservoir bank landslides.

[0124] Processor 402: The processor 402 loads and executes the instructions and data in the storage medium 403 to implement the loosening deformation identification and early warning method for reservoir bank landslides.

[0125] Storage medium 403: The storage medium 403 stores instructions and data; the storage medium 403 is used to implement the loosening deformation identification and early warning method for reservoir bank landslides.

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying and warning of loosening deformation in reservoir bank landslides, characterized in that, Includes the following steps: S1. Collect multi-source monitoring data on the monitoring section of the reservoir bank landslide. The multi-source monitoring data includes surface displacement data, deep displacement data, reservoir water level data, rainfall data, and acoustic emission signal data inside the landslide body. S2. The sliding window local anomaly factor detection algorithm is used to extract candidate abnormal jump points in the time domain from the surface displacement data. At the same time, the sudden increase points of event rate and energy release rate are extracted from the acoustic emission signal data to obtain the candidate abnormal jump point set and the acoustic emission sudden increase point set. S3. Calculate the event matching degree between the candidate abnormal jump point set and the acoustic emission surge point set within the same spatiotemporal window. When the matching degree exceeds the first threshold, the corresponding spatiotemporal position is determined as a loosening deformation event with brittle loosening mechanical characteristics, and a loosening deformation preliminary judgment feature set is obtained. S4. Construct the deformation spatial gradient field of adjacent monitoring points on the same monitoring section, calculate the local deformation mutation index of each monitoring point, and determine the time domain anomaly identifier of each monitoring point based on the candidate anomaly jump point set. Combine the local deformation mutation index to calculate the spatiotemporal joint anomaly index. When the index exceeds the second threshold, the monitoring point is confirmed as the final loosening deformation event, and a loosening deformation confirmation feature set is obtained. S5. Perform correlation analysis between the loosening deformation confirmation feature set and the synchronized reservoir water level data and rainfall data, and calculate the trigger response ratio; S6. Based on the trigger response ratio and the adaptive multi-level early warning threshold matrix, a dynamic comparison is made to output the loosening landslide early warning level, wherein the adaptive multi-level early warning threshold matrix is ​​iteratively corrected according to the recovery damping coefficient determined by the ratio of residual displacement rate to background displacement rate after historical loosening events.

2. The method for identifying and warning of loosening deformation in reservoir bank landslides according to claim 1, characterized in that: The sliding window local anomaly detection algorithm in step S2 specifically includes: S21. Set an analysis window with a length of L and a sliding step size of ΔL, slide it on the time series of the surface displacement data, and calculate the local outlier factor of the data points in each window. S22. Mark the points where the local outlier exceeds the mean plus k times the standard deviation within the window as candidate outlier jump points, where k is a preset sensitivity coefficient. S23. Perform a time sequence continuity check on the candidate abnormal jump points, remove isolated noise points, and retain the cluster of continuously occurring abnormal jump points to form the candidate abnormal jump point set.

3. The method for identifying and warning of loosening deformation in reservoir bank landslides according to claim 1, characterized in that: The extraction process of sudden increases in the event rate and energy release rate of acoustic emission signal data in step S2 specifically includes: S24. Perform short-time Fourier transform or wavelet packet transform on the acoustic emission signal to decompose it into at least 4 frequency bands; S25. Select a preset frequency band as the characteristic frequency band of brittle fracture, and calculate the acoustic emission event rate and energy release rate in the frequency band; S26. Define the event rate spike point as: the event rate at the current moment exceeds N times the standard deviation of the moving average event rate of the previous T hours, and at the same time the energy release rate exceeds A times the background value. S27. Define the point of sudden increase in energy release rate as: the energy release rate at the current moment exceeds B times the standard deviation of the moving average energy release rate of the previous T hours, where T, N, A, and B are all preset values.

4. The method for identifying and warning of loosening deformation in reservoir bank landslides according to claim 1, characterized in that: The event matching degree in step S3 is calculated using the following formula: in, The set of candidate abnormal jump points, Centered on the candidate abnormal jump point time, the time window The set of acoustic emission abrupt increase points within, The time decay coefficient, Let be the absolute value of the time difference between the displacement jump point and the matched acoustic emission surge point; when The event was determined to be a loosening deformation event. The threshold is the first threshold.

5. The method for identifying and warning of loosening deformation in reservoir bank landslides according to claim 1, characterized in that: Step S4, specifically the construction of the deformation space gradient field and the calculation of the spatiotemporal joint anomaly index, includes: S41. Extract the surface displacement rate values ​​of all monitoring points on the same monitoring section at the same time, and use inverse distance weighted interpolation to generate a continuous spatial distribution map of displacement rate. S42. For each monitoring point i Take all neighboring monitoring points within a radius R centered on it, and calculate the neighborhood displacement rate. mean and standard deviation ; S43. Calculate the local deformation abruptness index. and will Normalize to the [0,1] interval; S44. Define the time-domain anomaly identifier. If monitoring point i If the candidate abnormal jump point set is included, the value is 1; otherwise, it is 0. S45. Calculate the spatiotemporal joint anomaly index using the following formula: in, This is the spatial mutation threshold. For indicator functions, and For the weighting coefficients, satisfying And all are greater than 0; when This was confirmed as the final loosening deformation event. This is the second threshold.

6. The method for identifying and warning of loosening deformation in reservoir bank landslides according to claim 1, characterized in that: The trigger response ratio in step S5 is calculated using the following formula: in, The jump amplitude of the deformation jump event in the feature set of the loosened deformation confirmation; The effective range of reservoir water level within N days prior to the event, calculated as follows: calculate, For the first Daily changes in reservoir water level The attenuation coefficient is the effect of water level. The cumulative rainfall within the same time period, calculated as follows: calculate, For the first Daily rainfall, This is the rainfall attenuation coefficient; These are the weighting coefficients for reservoir water level and rainfall, respectively. It is the sum of the lag times of all historical events of detachment. For the total observation time, This represents the lag penalty coefficient.

7. The method for identifying and warning of loosening deformation in reservoir bank landslides according to claim 1, characterized in that: The construction and dynamic correction process of the adaptive multi-level early warning threshold matrix in step S6 specifically includes: S61. Based on the inversion analysis of historical landslide events, construct an initial multi-level early warning threshold matrix. The rows correspond to different reservoir water level operating conditions, and the columns correspond to different loosening and deformation confirmation index threshold levels. S62. Record the residual displacement rate after each loosening deformation event ends. Compared with the background displacement rate before the event occurred Calculate the recovery damping coefficient And store the coefficient in the historical damping sequence; S63. Statistically calculate the average damping coefficient of the most recent M loosening events. And calculate the early warning accuracy of the current early warning system. ,in TP To ensure the correct number of warnings, FP This represents the number of false alarms. S64. Correct the current warning threshold using the following formula: in, The initial warning threshold, As a damping factor, The number of loosening events that have occurred. For accuracy feedback coefficient, For target accuracy; S65. When the mean of the historical damping sequence exceeds the preset upper limit. If the residual displacement rate is still greater than the background displacement rate after three consecutive loosening events, the subsequent warning level will be forcibly increased by one level.

8. A method for identifying and warning of loosening deformation in reservoir bank landslides according to any one of claims 1, characterized in that: The acoustic emission signal data is collected by broadband accelerometers buried at different depths in the landslide body. The depths include shallow (0-5m), middle (5-15m), and deep (15-30m). The signals of each channel are sampled synchronously, and the sampling frequency is not lower than a preset value. In step S3, the matching degree at each depth is calculated. When the difference between the matching degree of the deep layer and the matching degree of the shallow layer is greater than the preset value, it is determined that the deep slip zone is loose, and the output warning level is accompanied by a depth indicator.

9. A storage medium, characterized in that: The storage medium stores instructions and data to implement the loosening deformation identification and early warning method for reservoir bank landslides as described in any one of claims 1 to 8.

10. A loosening deformation identification and early warning device for reservoir bank landslides, characterized in that: include: A processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the loosening deformation identification and early warning method for reservoir bank landslides as described in any one of claims 1 to 8.

Images