A Comprehensive Risk Assessment Method for Geological Hazards Integrating Environmental Factors

Abstract

This invention discloses a comprehensive geological hazard risk assessment method integrating environmental factors, relating to the field of geological hazard early warning technology. It includes setting multiple independent stress-sensitive units circumferentially on the outer wall of a displacement sensor to simultaneously acquire contact stress information at different circumferential locations, forming a circumferential contact stress distribution sequence. By performing time-series analysis on this distribution sequence, localized compression caused by plant growth is identified at only some circumferential locations, and the amount of plant compression displacement change is determined accordingly. This displacement change is then removed from the original displacement sequence to recover the actual displacement sequence that truly reflects geological disturbance. Based on this, a statistical feature vector is constructed to detect abrupt changes, identifying changes that significantly deviate from historical stable behavior and issuing warnings. This invention solves the problem of plant growth interference causing distortion of displacement monitoring data, thus affecting the accuracy of geological hazard identification.

Description

Technical Field

[0001] This invention relates to the field of geological disaster early warning technology, and more specifically, to a comprehensive risk assessment method for geological disasters that integrates environmental factors. Background Technology

[0002] The formation and evolution of geological hazards are often long-term and insidious. Therefore, displacement sensors are commonly deployed at engineering sites to continuously monitor key components such as slopes, retaining walls, and foundations. Traditional displacement monitoring often treats the displacement sequence output by the sensors as a direct reflection of actual soil deformation, combining this with environmental information such as rainfall and groundwater levels to assess the stability of the monitored area. However, in complex real-world environments, monitoring points are often surrounded by vegetation, especially shrubs and trees with well-developed root systems. Over many years of slow growth, the roots or upper foliage of these trees gradually exert directional pressure on the outer wall of the displacement sensor, causing additional displacement of the sensor housing.

[0003] This type of localized compression caused by plant growth is spatially localized and temporally slow and monotonous, with its rhythm of change significantly different from the sudden or phased deformation caused by geological disturbance. If the influence of plant compression cannot be isolated from the displacement sequence, the slow displacement caused by vegetation growth may be mistaken for overall soil creep or gradual structural instability, thus amplifying the risk assessment results and even frequently triggering false alarms. Conversely, when parts of the circumference are continuously blocked by vegetation, the sensor's response to small-amplitude real displacements in certain directions may be masked, leading to the submergence of potential catastrophic signals and the risk of missed detections.

[0004] Current monitoring methods primarily focus on overall trend analysis or simple filtering of time series data. They struggle to simultaneously utilize circumferential force distribution, gradual rate of change characteristics, and statistical abrupt change information to precisely distinguish between plant compression components and actual geological deformation components. Furthermore, they lack methods for systematically modeling and geometrically constraining the spatial distribution of plant roots, making it difficult to construct stable and reliable comprehensive geological hazard risk assessment results from long-term monitoring data. Therefore, a risk assessment method is needed that can identify and eliminate the influence of plant compression at the displacement sensor level, while simultaneously constructing sensitive statistical features and detecting abrupt changes using the remaining actual displacement sequences. This would allow for explicit modeling of environmental factors and their integration into the comprehensive geological hazard risk assessment process. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a comprehensive risk assessment method for geological disasters that integrates environmental factors, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The advantages of this invention over the prior art are:

[0008] A comprehensive geological hazard risk assessment method incorporating environmental factors includes:

[0009] Multiple independent stress-sensitive units are arranged circumferentially on the outer wall of the displacement sensor to obtain the contact stress at different circumferential positions and form a circumferential contact stress distribution sequence.

[0010] Time series analysis is performed on the circumferential contact stress distribution sequence. When a contact stress change pattern occurs only in some circumferential positions within a preset time window, and the contact stress continuously increases and shows monotonous growth, and the change in contact stress per unit time is lower than the contact stress change rate threshold, the corresponding contact stress change pattern is determined to be local compression caused by plant growth, and the change in plant compression displacement corresponding to the contact stress change pattern is determined.

[0011] The plant compression displacement change is removed from the original displacement sequence output by the displacement sensor to obtain the actual displacement sequence;

[0012] Based on the actual displacement sequence, a statistical feature vector is constructed within a sliding time window, and abrupt change point detection is performed on the statistical feature vector. When an abrupt change point is detected in the statistical feature vector within a preset time window, a warning is given that the monitoring area is in a state of potential geological disaster risk.

[0013] Preferably, the process of determining the change in plant compression displacement includes:

[0014] In the early stage of monitoring, a preset learning time window is selected where the working conditions are stable and no geological disturbance has occurred. The circumferential contact stress distribution sequence and the original displacement sequence output by the displacement sensor are recorded within the preset learning time window.

[0015] Within the preset learning time window, the contact stress change pattern determined to be local compression caused by plant growth is used as input. By filtering and decomposing the original displacement sequence, displacement components that are time-synchronous with the contact stress change pattern and whose displacement change per unit time is lower than the displacement change rate threshold are extracted as synchronous displacement components.

[0016] Preferably, the process of determining the change in plant compression displacement further includes:

[0017] Using the contact stress variation mode and the synchronous displacement component, a static response model of contact stress-displacement under plant compression conditions is established by the least squares method.

[0018] In subsequent monitoring, the currently detected plant compression contact stress change pattern is substituted into the contact stress-displacement static response model to calculate the corresponding plant compression displacement change.

[0019] Preferably, the process of eliminating the plant compression displacement change includes:

[0020] The time series curves of the plant extrusion displacement components were constructed using the plant extrusion displacement changes calculated by the contact stress-displacement static response model.

[0021] Align the time series curve with the original displacement sequence output by the displacement sensor point by point on the same time axis;

[0022] After time alignment is completed, component separation operation is performed on the original displacement sequence. The displacement components with a correlation coefficient greater than the correlation coefficient threshold and a consistent change pattern with the time series curve are identified as plant compression displacement components and removed, thereby obtaining the actual displacement sequence after removing the influence of plant compression.

[0023] Preferably, constructing a statistical feature vector based on the actual displacement sequence within a sliding time window includes: within each sliding time window, calculating the average displacement, standard deviation of displacement, difference between maximum and minimum displacement values, and slope of displacement change over time obtained by least-squares linear fitting for the actual displacement sequence within that sliding time window, and arranging the average displacement, standard deviation of displacement, difference between maximum and minimum displacement values, and slope of displacement change over time in a preset order to form the statistical feature vector.

[0024] Preferably, the reference value of the statistical feature vector is determined as follows: within the preset learning time window at the beginning of monitoring, on the premise that the monitoring area is in a stable working condition and no geological disturbance has occurred, the actual displacement sequence within the preset learning time window is segmented into segments of the same length as the sliding time window. The corresponding statistical feature vector is calculated for the actual displacement sequence of each segment, and the average value of the statistical feature vector of each segment is calculated for each component. The vector composed of the average values ​​of each component is determined as the reference statistical feature vector under historical stable working conditions.

[0025] Preferably, the abrupt change detection of the statistical feature vector includes: in the subsequent monitoring process, calculating the current statistical feature vector for the actual displacement sequence within each sliding time window, performing a difference operation between the current statistical feature vector and the reference statistical feature vector on the corresponding components, and when the absolute difference of any component is greater than a preset feature difference threshold, determining the time position corresponding to the sliding time window as the abrupt change point of the statistical feature vector, and issuing a warning that the monitoring area is in a state of potential geological disaster risk.

[0026] Preferably, the stress-sensitive unit is a thin-film capacitive pressure sensor unit. Each thin-film capacitive pressure sensor unit includes upper and lower electrodes and a dielectric layer disposed between the upper and lower electrodes. The compressive force exerted by the external soil on the outer wall of the displacement sensor causes a change in the distance between the upper and lower electrodes, thereby causing a change in the capacitance value. The change in capacitance value is used to characterize the contact stress at the corresponding circumferential position.

[0027] Preferably, in addition to determining the plant compression displacement change by multiple displacement sensors, the method further includes: representing the spatial position coordinates of each displacement sensor in the coordinate system of the monitoring area, determining the plant compression direction of each displacement sensor according to the circumferential position of the local compression caused by plant growth, and forming a plant compression semi-straight line by combining the spatial position coordinates of each displacement sensor with the corresponding plant compression direction.

[0028] All plants are squeezed into a semi-straight line to form a set of plant squeezed semi-straight lines. Then, a geometric optimization operation is performed on the set of plant squeezed semi-straight lines to find the point that minimizes the sum of the squared distances from each plant squeezed semi-straight line to that point. This point is used as the coordinates of the estimated plant action point.

[0029] All the estimated plant action point coordinates are combined into an estimated plant action point set. A clustering operation based on Euclidean distance is performed on the estimated plant action point set. When the Euclidean distance between any two estimated plant action point coordinates in a certain cluster is less than the plant action point distance threshold, the cluster is determined as a plant squeezing cluster under the action of the same plant root system.

[0030] Preferably, for displacement sensors not included in any plant compression cluster, the following additional features are also included:

[0031] In the coordinate system of the monitoring area, the Euclidean distance between the displacement sensor and the coordinates of the plant action point of each plant squeezing cluster is calculated, and the sensor is assigned to the plant squeezing cluster with the smallest Euclidean distance.

[0032] Based on the coordinates of the plant action point of the assigned plant compression cluster and the spatial position coordinates of the displacement sensor, the direction vector from the plant action point to the displacement sensor is calculated, and the circumferential position corresponding to the direction vector is determined as the expected plant stress action direction of the displacement sensor.

[0033] A secondary verification and identification step is performed on the circumferential contact stress distribution sequence corresponding to the expected plant stress direction to identify the plant squeezing contact stress change pattern based on the monotonic increase of contact stress and the contact stress change rate threshold. When the verification result is local squeezing caused by plant growth, the plant squeezing displacement change is calculated and eliminated for the displacement sensor.

[0034] This invention is based on a stress-sensitive unit circumferentially arranged on the outer wall of a displacement sensor. By identifying contact stress change patterns that show monotonous and slow increases only in certain circumferential locations, it distinguishes between local compression caused by plant growth and actual geological deformation. Based on this, a quantifiable mechanism for eliminating plant compression displacement changes is constructed. This fundamentally solves the problems of false alarms, missed alarms, and trend distortions caused by long-term, silent compression from plants, such as plant roots, making displacement monitoring results closer to the actual geological conditions.

[0035] Furthermore, in multi-sensor scenarios, by using geometric optimization and Euclidean distance-based clustering to identify plant action points, the overall action distribution of plant roots can be reconstructed at the regional scale, and secondary verification compensation can be performed on unassigned sensors, making the removal of plant squeezing more complete and consistent, thereby further improving the accuracy and spatial consistency of the overall risk assessment. Attached Figure Description

[0036] Figure 1 This is an overall schematic diagram of the present invention;

[0037] Figure 2 This is a schematic diagram of the present invention for finding missed sensors. Detailed Implementation

[0038] The specific embodiments of the present invention will now be described with reference to the accompanying drawings.

[0039] like Figure 1 As shown, this invention relates to a displacement monitoring device deployed in slopes, retaining walls, or other geological formations. Multiple independent stress-sensitive units are arranged circumferentially on the outer wall of the displacement sensor. Each unit corresponds to a circumferential position on the outer wall and is used to measure the contact stress exerted by the external soil on the sensor housing at that position. This stress is continuously recorded on a time axis, thus forming a sequence of contact stress distributions for each circumferential position. In this way, the displacement sensor is no longer simply viewed as a device that outputs a one-dimensional displacement, but rather it simultaneously obtains a circumferential force image, enabling the identification of which directions are being gradually resisted.

[0040] In a further embodiment, after obtaining the circumferential contact stress distribution sequence, a time series analysis is performed on the sequence. Within a preset time window, the focus is on finding a change pattern where the contact stress continuously increases and generally shows a monotonically increasing trend at only a few circumferential locations during the monitoring period, and the change per unit time is always lower than the contact stress change rate threshold.

[0041] This slow, monotonous, and localized compression pattern is significantly different from the abrupt or phased alternating stress induced by rainstorms or slip surface activity. It is more consistent with the process of plant roots growing year by year, gradually expanding the surrounding soil and pushing against the outer wall of the sensor. Therefore, this contact stress change pattern is identified as local compression caused by plant growth. Here, plant growth can be the growth of the root system or the growth of the leaf structure above, depending on the sensor installation location.

[0042] Based on this judgment, the pattern can be mapped to the corresponding plant compression displacement change according to the amplitude and time evolution of the contact stress, which can be used to characterize how much of the sensor displacement under this working condition comes from the plant rather than from the deformation of the geological body itself.

[0043] In a further embodiment, to ensure that this mapping relationship is not estimated empirically but rather based on a quantifiable response model, a preset learning time window is selected during the initial monitoring phase, where the conditions are stable and no geological disturbances have occurred. This phase can be selected during the initial stage of project operation, in a period without concentrated rainfall or excavation disturbances. During this period, the circumferential contact stress distribution sequence and the original displacement sequence output by the displacement sensor are recorded. Within this learning window, using the aforementioned judgment results, contact stress change patterns identified as local compression caused by plant growth are selected and used as input signals to filter and decompose the original displacement sequence. The filtering method can be bandpass filtering, wavelet decomposition, or other signal processing methods suitable for separating slow trends from rapid disturbances. The aim is to extract displacement components that are temporally synchronized with the contact stress change pattern and whose displacement change per unit time is below the displacement change rate threshold.

[0044] This displacement component primarily reflects the slow top pressure caused by plant growth, without including significant geological disturbance; therefore, it is considered as the synchronous displacement component. The synchronous displacement component and the contact stress change pattern are matched point-to-point in time, which is beneficial for establishing the static response relationship later.

[0045] After obtaining the contact stress variation pattern and synchronous displacement components, a static response model between contact stress and displacement under plant compression conditions is established using the least squares method. This can be understood as finding a function that best fits the relationship between the increase in contact stress at a certain circumferential position and the corresponding displacement of the sensor in that direction. The simplest case can use a linear model, but it can also be extended to piecewise linear or low-order polynomial forms as needed, as long as the sum of squared errors is minimized through least squares fitting. Once the model is fitted during the learning phase, it can be reused in subsequent monitoring processes.

[0046] Specifically, when a plant compression contact stress change pattern that matches the characteristics of monotonous and slow growth is detected again in subsequent monitoring, this pattern is substituted into the static response model to calculate the corresponding plant compression displacement change. This is equivalent to being able to predict at any time how much of the sensor displacement is due to the plant pushing and how much is due to the geological body moving.

[0047] After obtaining the changes in vegetation compression displacement, to separate it from the total displacement at the signal level, a time-series curve of the vegetation compression displacement component needs to be constructed. This curve provides the corresponding vegetation compression displacement at each time step. Then, this time-series curve is aligned point-by-point with the original displacement sequence output by the sensor on the same time axis. This step requires that the sensor sampling clock and stress sampling clock be consistent or interpolated to a unified time scale. After alignment, component separation is performed on the original displacement sequence. Mathematically, correlation analysis combined with decomposition algorithms can be used. For example, the original displacement sequence can be decomposed into several candidate components, and the correlation coefficient between each component and the vegetation compression displacement time-series curve is calculated. Components with correlation coefficients greater than the correlation coefficient threshold and whose change patterns are consistent with the vegetation compression displacement time-series are identified as vegetation compression displacement components and uniformly removed from the original sequence. In this way, noise that is highly consistent with vegetation compression in both time and morphology is removed, leaving the actual displacement sequence after removing the influence of vegetation compression, thus more accurately reflecting the true deformation of the geological body itself.

[0048] In order to facilitate the identification of mutations based on the actual displacement sequence after obtaining the actual displacement sequence, this invention introduces the process of constructing statistical feature vectors and determining reference statistical feature vectors in a specific embodiment.

[0049] Specifically, the actual displacement sequence output by the displacement sensor and removed by plant compression is segmented according to a preset sliding time window, with each sliding time window covering a fixed length of continuous time step.

[0050] Within each sliding time window, the average displacement of the actual displacement sequence within that time window is first calculated to reflect the approximate magnitude of the displacement level during that period. Then, the standard deviation of the displacement is calculated to reflect the degree of fluctuation in the displacement during that period. Next, the difference between the maximum and minimum displacement values ​​within that time window is calculated to reflect the magnitude of the displacement change. In addition, the displacement-time relationship within that time window is fitted with a straight line using the least squares method, and the slope of the fitted line is used as the slope of the displacement change over time to characterize whether the displacement is rising, falling, or basically stable during that period.

[0051] The four quantities mentioned above represent the average displacement, standard deviation of displacement, difference between maximum and minimum displacement, and slope of displacement change over time within the same time window, respectively. They are arranged in a pre-defined order to form the statistical feature vector corresponding to the sliding time window. In this way, without changing the original time series structure, the overall level, fluctuation intensity, variation amplitude, and trend of the actual displacement sequence within each sliding time window can be centrally expressed by a low-dimensional vector. This facilitates subsequent comparisons and avoids introducing an overly complex feature construction process.

[0052] In a further embodiment, in order to provide a stable reference for subsequent mutation point detection, the present invention selects a preset learning time window in the early stage of monitoring. During the time period corresponding to the preset learning time window, the monitoring area is in a stable working condition and no geological disturbance occurs.

[0053] The actual displacement sequence within the preset learning time window is segmented into segments of the same length as the sliding time window described above, so that each segment can be used for feature calculation in the same way as during normal monitoring. For each segment of the actual displacement sequence within the preset learning time window, the average displacement, standard deviation of displacement, difference between maximum and minimum displacement, and slope of displacement change over time are calculated sequentially, and the statistical feature vector corresponding to that segment is constructed in the same order.

[0054] After obtaining multiple statistical feature vectors, the average value of each component of these feature vectors is calculated. Specifically, the average displacement value of all segments, the average displacement standard deviation of all segments, the average difference between the maximum and minimum displacement values ​​of all segments, and the average slope of the displacement change over time for all segments are calculated. The final result is a vector composed of these four average values. This vector reflects the typical displacement level, displacement fluctuation, displacement variation amplitude, and displacement trend of the monitored area under historical stable operating conditions. It is determined as the reference statistical feature vector under historical stable operating conditions and stored in the system as a benchmark for subsequent abrupt change detection.

[0055] In a further implementation, the process of detecting abrupt changes in the statistical feature vector is completed during the subsequent continuous monitoring phase. As the actual displacement sequence is continuously updated, the actual displacement sequence near the current moment is segmented according to a sliding time window of the same length as the preset learning time window, and the current statistical feature vector is calculated within each sliding time window.

[0056] For each current statistical feature vector, a difference operation is performed between it and the reference statistical feature vector at the corresponding component. This yields the differences between the current average displacement and the reference average displacement, the current standard deviation and the reference standard deviation, the difference between the current maximum and minimum displacement values ​​and the difference under the reference condition, and the difference between the current displacement slope and the reference displacement slope. Then, the absolute value of each of these differences is taken and compared with a pre-set feature difference threshold. When the absolute difference of any component exceeds the corresponding feature difference threshold, it is considered that the actual displacement behavior within the sliding time window has deviated from the historical stable condition, and a sudden change point occurs in the statistical feature vector at the corresponding time position of the sliding time window.

[0057] The system can record this time location as the occurrence time of a potential anomaly and trigger an alert indicating that the monitored area is at potential geological hazard risk. By directly comparing this with the statistical feature vector of historical stable conditions and using a simple threshold judgment method, abrupt changes in actual displacement sequences can be detected without introducing complex algorithms, facilitating engineering deployment and parameter adjustment.

[0058] At the sensing level, this invention employs a thin-film capacitive pressure sensor unit as the stress-sensitive unit. Each unit consists of upper and lower electrodes and a dielectric layer disposed between them, and is installed around the outer wall of the displacement sensor. When external soil applies compressive force to the outer wall at this location, the distance between the upper and lower electrodes changes slightly, thereby causing a change in the capacitance value. By measuring the capacitance using a circuit, the change in capacitance value can be converted into contact stress at the corresponding circumferential position. The thin-film structure facilitates high-density deployment around the circumference and possesses good flexibility and sealing properties, adapting to complex soil environments.

[0059] like Figure 2 As shown, in a further embodiment, the principle that plant growth is radial from the center outward is utilized to find sensors that may be affected by the same plant. Specifically:

[0060] Multiple displacement sensors are typically deployed within the monitoring area. This invention, in addition to determining the change in plant compression displacement for each sensor, also constructs a geometric map of the plant root system's action in space. First, a unified coordinate system is defined within the monitoring area, representing the installation position of each displacement sensor as its spatial coordinates. Simultaneously, the plant compression direction of the sensor is determined based on its circumferential position caused by localized compression during plant growth. Extending along this direction from the sensor's position can be considered a semi-straight line. Combining the spatial coordinates of each displacement sensor with its corresponding plant compression direction yields a semi-straight line of plant compression. All these semi-straight lines constitute the set of plant compression semi-straight lines.

[0061] Considering that the actual root action point of a plant should be close to the intersection region of these semi-straight lines, this invention performs geometric optimization operations on the set of semi-straight lines to find the point that minimizes the sum of the squared distances from each plant to a certain point by squeezing the semi-straight lines. This point is then used as the coordinates of the presumed plant action point for the corresponding combination. This can be understood as finding the most likely location of the root center in the least squares sense.

[0062] The coordinates of all presumed plant impact points are collected to form a set of presumed plant impact points. A clustering operation based on Euclidean distance is then performed on this set. When the Euclidean distance between any two presumed plant impact point coordinates within a cluster is less than a plant impact point distance threshold, these impact points can be considered to actually originate from the same plant or the same root system. In an engineering sense, this cluster is defined as a plant compression cluster under the influence of the same plant root system. In this way, plant compression phenomena detected by numerous sensors can be grouped into several physically meaningful plant clusters at a regional scale, providing spatial constraints for subsequent elimination and correction.

[0063] In the above steps, some displacement sensors may not be included within any plant compression clusters due to data noise, installation location, or other reasons. To avoid missing these points, this invention introduces a second-layer spatial correlation and verification mechanism for these sensors.

[0064] The specific method involves calculating the Euclidean distance between the sensors and the corresponding plant action points of each plant compression cluster in the coordinate system of the monitoring area, and assigning the sensors to the plant compression clusters with the smallest distance. Then, based on the coordinates of the plant action points of the assigned plant compression clusters and the spatial position coordinates of the displacement sensor, the direction vector from the plant action point to the displacement sensor is calculated, and the corresponding position of this direction vector on the circumference is determined as the expected direction of plant stress action of the displacement sensor.

[0065] Subsequently, a secondary verification and identification step based on the monotonically increasing contact stress and the contact stress change rate threshold is performed on the circumferential contact stress distribution sequence corresponding to this expected direction to identify the plant compression contact stress change pattern. If the verification result confirms the existence of local compression caused by plant growth, the aforementioned calculation and elimination operation of plant compression displacement change is repeated for the displacement sensor. This ensures that even sensors that were not initially included in the cluster can complete the environmental factor removal with the help of spatial reasoning and secondary verification, guaranteeing the consistency of the actual displacement sequences of all sensors in the entire monitoring area in terms of eliminating the influence of plant compression. This provides more reliable basic data for subsequent comprehensive risk assessment based on statistical characteristics and mutation point detection.

[0066] The thresholds involved in this invention are preferably set using a combination of engineering experience and statistical analysis. Regarding the contact stress change rate, the threshold can be defined by the amount of contact stress change per unit time. Generally, the peak change rate of contact stress under typical geological disturbance conditions is first statistically analyzed, and then the threshold is set to 5%–20% of this peak change rate. For scenarios where monitoring is conducted on a kilopascal (kPa) scale and a daily timescale, the contact stress change rate threshold can be selected within the range of 0.5–5 kPa per day, ensuring that the monotonous, gradual changes caused by slow plant growth are significantly lower than the change rates of rapid disturbances such as rainstorms and landslides. Regarding the displacement change rate, the displacement change rate threshold can be determined according to the sensor resolution and the noise level under stable, disaster-free conditions, and is generally selected to be 1–3 times the noise level. For slope displacement monitoring recorded in millimeters on a daily timescale, the displacement change rate threshold can be set within the range of 0.1–2 mm per day.

[0067] In the mutation detection technology scheme, the feature difference threshold is preferably determined based on the distribution of statistical feature vector components under historical stable operating conditions in the early stage of monitoring. Independent thresholds are given for four components: displacement average, displacement standard deviation, the difference between the maximum and minimum displacement values, and the slope of displacement change over time. Specifically, within a preset learning time window, the actual displacement sequence is segmented into segments of the same length as the sliding time window. First, the statistical feature vector corresponding to each segment is calculated. Then, the variance or standard deviation of each component is calculated, and the feature difference threshold is set to 2 to 3 times the standard deviation of that component. In this way, if the current statistical feature vector deviates from the reference statistical feature vector by more than 2 to 3 times the normal fluctuation range in a certain component, it will be considered to be abnormal. In slope monitoring scenarios measured in millimeters and on a timescale of days, the characteristic difference thresholds for the average displacement and the standard deviation of displacement can be roughly set within the range of 0.5 to 2 mm, the characteristic difference threshold for the difference between the maximum and minimum displacement values ​​can be set within the range of 1 to 5 mm, and the characteristic difference threshold for the slope of displacement over time can be set within the range of 0.1 to 1 mm per day. In practical engineering applications, fine-tuning can be performed within the above ranges based on the stiffness of the monitored object, sensor resolution, and on-site noise levels, using trial operation data.

[0068] In the process of spatial clustering to identify plant squeeze clusters, the distance threshold for plant action points can be determined by combining the horizontal expansion radius of common plant roots within the monitoring area. Typically, an empirical radius of the main influence range of the root system is first given based on the plant species and root development in the field. Then, the distance threshold for plant action points is preferably set to 0.5 to 1 times this empirical radius. For slope conditions where the root systems of small and medium-sized shrubs and trees are mainly distributed within a range of 1 to 3 meters, the distance threshold for plant action points can be set between 0.5 and 3 meters. This avoids mistakenly clustering two plants that are far apart into one cluster, and also aggregates the inferred action points derived from multiple sensors under the influence of the same plant's root system into the same plant squeeze cluster, providing a reasonable spatial constraint for regional-scale plant squeeze identification and removal.

[0069] All of the above thresholds can be calibrated during the initial deployment phase through trial operation and historical data statistics, and can be appropriately adjusted in the long-term operation process based on new data to adapt to different monitoring environments and sensor configurations.

[0070] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A comprehensive risk assessment method for geological hazards that integrates environmental factors, characterized in that, include: Multiple independent stress-sensitive units are set around the outer wall of the displacement sensor to obtain the contact stress of the external soil at different circumferential positions, forming a circumferential contact stress distribution sequence. Time series analysis is performed on the circumferential contact stress distribution sequence. When a contact stress change pattern occurs only in some circumferential positions within a preset time window, and the contact stress continuously increases and shows monotonous growth, and the change in contact stress per unit time is lower than the contact stress change rate threshold, the corresponding contact stress change pattern is determined to be local compression caused by plant growth, and the change in plant compression displacement corresponding to the contact stress change pattern is determined. The plant compression displacement change is removed from the original displacement sequence output by the displacement sensor to obtain the actual displacement sequence; Based on the actual displacement sequence, a statistical feature vector is constructed within a sliding time window, and abrupt change point detection is performed on the statistical feature vector. When an abrupt change point is detected in the statistical feature vector within a preset time window, a warning is given that the monitoring area is in a state of potential geological disaster risk. Multiple displacement sensors are deployed within the monitoring area. Based on determining the plant compression displacement change of each displacement sensor, the method also includes: representing the spatial position coordinates of each displacement sensor in the coordinate system of the monitoring area, determining the plant compression direction of each displacement sensor according to the circumferential position of the local compression caused by plant growth, and combining the spatial position coordinates of each displacement sensor with the corresponding plant compression direction to obtain a plant compression half-line. All plants are squeezed into a semi-straight line to form a set of plant squeezed semi-straight lines. Then, a geometric optimization operation is performed on the set of plant squeezed semi-straight lines to find the point that minimizes the sum of the squared distances from each plant squeezed semi-straight line to a certain point. This point is used as the coordinates of the estimated plant action point. All the estimated plant action point coordinates are combined into an estimated plant action point set. A clustering operation based on Euclidean distance is performed on the estimated plant action point set. When the Euclidean distance between any two estimated plant action point coordinates in a certain cluster is less than the plant action point distance threshold, the cluster is determined as a plant squeezing cluster under the action of the same plant root system.

2. The comprehensive geological hazard risk assessment method integrating environmental factors according to claim 1, characterized in that, The process of determining the change in plant compression displacement includes: In the early stage of monitoring, a preset learning time window is selected where the working conditions are stable and no geological disturbance has occurred. The circumferential contact stress distribution sequence and the original displacement sequence output by the displacement sensor are recorded within the preset learning time window. Within the preset learning time window, the contact stress change pattern determined to be local compression caused by plant growth is used as input. By filtering and decomposing the original displacement sequence, displacement components that are time-synchronous with the contact stress change pattern and whose displacement change per unit time is lower than the displacement change rate threshold are extracted as synchronous displacement components.

3. The comprehensive geological hazard risk assessment method integrating environmental factors according to claim 2, characterized in that, The process of determining the change in plant compression displacement also includes: Using the contact stress variation mode and the synchronous displacement component, a static response model of contact stress-displacement under plant compression conditions is established by the least squares method. In subsequent monitoring, the currently detected plant compression contact stress change pattern is substituted into the contact stress-displacement static response model to calculate the corresponding plant compression displacement change.

4. The comprehensive geological hazard risk assessment method integrating environmental factors according to claim 3, characterized in that, The process of eliminating the plant compression displacement change includes: The time series curves of the plant extrusion displacement components were constructed using the plant extrusion displacement changes calculated by the contact stress-displacement static response model. Align the time series curve with the original displacement sequence output by the displacement sensor point by point on the same time axis; After time alignment is completed, component separation operation is performed on the original displacement sequence. The displacement components with a correlation coefficient greater than the correlation coefficient threshold and a consistent change pattern with the time series curve are identified as plant compression displacement components and removed, thereby obtaining the actual displacement sequence after removing the influence of plant compression.

5. The comprehensive geological hazard risk assessment method integrating environmental factors according to claim 1, characterized in that, Constructing a statistical feature vector within a sliding time window based on the actual displacement sequence includes: within each sliding time window, calculating the average displacement, standard deviation of displacement, difference between maximum and minimum displacement values, and slope of displacement change over time obtained by least-squares linear fitting for the actual displacement sequence within that sliding time window; and arranging the average displacement, standard deviation of displacement, difference between maximum and minimum displacement values, and slope of displacement change over time in a preset order to form the statistical feature vector.

6. The comprehensive geological hazard risk assessment method integrating environmental factors according to claim 5, characterized in that, The reference value of the statistical feature vector is determined as follows: within the preset learning time window at the beginning of monitoring, on the premise that the monitoring area is in a stable working condition and no geological disturbance has occurred, the actual displacement sequence within the preset learning time window is segmented into segments of the same length as the sliding time window. The corresponding statistical feature vector is calculated for the actual displacement sequence of each segment, and the average value of the statistical feature vector of each segment is calculated for each component. The vector composed of the average values ​​of each component is determined as the reference statistical feature vector under historical stable working conditions.

7. The comprehensive geological hazard risk assessment method integrating environmental factors according to claim 6, characterized in that, The abrupt change detection of the statistical feature vector includes: in the subsequent monitoring process, calculating the current statistical feature vector for the actual displacement sequence within each sliding time window, performing a difference operation between the current statistical feature vector and the reference statistical feature vector on the corresponding components, and when the absolute difference of any component is greater than a preset feature difference threshold, determining the time position corresponding to the sliding time window as the abrupt change point of the statistical feature vector, and issuing a warning that the monitoring area is in a state of potential geological disaster risk.

8. The comprehensive geological hazard risk assessment method integrating environmental factors according to claim 1, characterized in that, The stress-sensitive unit is a thin-film capacitive pressure sensor unit. Each thin-film capacitive pressure sensor unit includes upper and lower electrodes and a dielectric layer disposed between the upper and lower electrodes. The compressive force exerted by the external soil on the outer wall of the displacement sensor causes a change in the distance between the upper and lower electrodes, thereby causing a change in the capacitance value. The change in capacitance value is used to characterize the contact stress at the corresponding circumferential position.

9. The comprehensive geological hazard risk assessment method integrating environmental factors according to claim 1, characterized in that, For displacement sensors not included in any plant compression cluster, the following are also included: In the coordinate system of the monitoring area, the Euclidean distance between the displacement sensor not included in any plant squeezing cluster and the coordinates of the plant action point of each plant squeezing cluster is calculated, and the sensor is assigned to the plant squeezing cluster with the smallest Euclidean distance. Based on the coordinates of the plant action point of the assigned plant compression cluster and the spatial position coordinates of the displacement sensor, the direction vector from the plant action point to the displacement sensor is calculated, and the circumferential position corresponding to the direction vector is determined as the expected plant stress action direction of the displacement sensor. A secondary verification and identification step is performed on the circumferential contact stress distribution sequence corresponding to the expected plant stress direction to identify the plant squeezing contact stress change pattern based on the monotonic increase of contact stress and the contact stress change rate threshold. When the verification result is local squeezing caused by plant growth, the plant squeezing displacement change is calculated and eliminated for the displacement sensor.

Images