Collapse thermal infrared displacement cooperative early warning method suitable for red shui Danxia landform

Abstract

The invention, which relates to the technical field of geological disaster monitoring and early warning, discloses a collapse thermal infrared displacement cooperative early warning method suitable for a red-water Danxia landform, comprising the following steps: (1) constructing a signal space-time asynchronous quantitative model, (2) improving an early warning trigger mechanism, and (3) assisting early warning by using biological and geogas escape. According to the method, time lag and spatial dislocation of thermal infrared and displacement signals in the red shui Danxia landform are accurately quantified, so that the limitation of traditional synchronism hypothesis is broken through; multi-scale chaotic feature extraction and Voronoi diagram weighted matching are fused to realize asynchronous signal efficient fusion, and a biological-physical-chemical multi-source collaborative system is constructed in combination with specific moss physiological indexes and geogas Ca < 2 + > Raman feature peaks; the early warning threshold value is dynamically adjusted through the Q-learning algorithm and the ART2 neural network, false abnormal signals are automatically recognized, missing report and false report are effectively reduced, and the accuracy, reliability and environmental adaptability of collapse early warning are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of geological disaster monitoring and early warning, in particular to a collapse hot infrared displacement collaborative early warning method suitable for Chishui Danxia landform. BACKGROUND

[0002] In the field of rock mass collapse early warning, the collaborative application of hot infrared monitoring and displacement monitoring is an important technical means to improve the early warning accuracy. In the existing technology, hot infrared monitoring identifies potential instability signals by capturing the micro changes of rock surface temperature (such as heat exchange anomalies of water evaporation, temperature fluctuations of thermal expansion and cold contraction), and displacement monitoring judges the instability process by identifying rock deformation (such as crack expansion, shear sliding). The collaborative logic of the two has long been based on the default premise of "spatiotemporal synchronization of signals" - that is, hot infrared anomalies and displacement anomalies are considered as immediate correlation signals of the same instability process, and hot signals will inevitably be accompanied by displacement changes in the short term, and the spatial distribution of the two overlaps ("hot anomaly area is the displacement first area"). Based on this, the existing collaborative early warning technology usually takes "spatiotemporal matching of hot infrared anomalies and contemporaneous displacement anomalies" as the early warning trigger condition, relying on the synchronization of signals to realize risk judgment.

[0003] However, in the practice of collapse early warning of Chishui Danxia landform, the above-mentioned "synchronization default premise" is invalid due to the uniqueness of the landform, resulting in significant limitations of the existing collaborative early warning technology. As a typical red sandstone Danxia landform, Chishui Danxia is characterized by the dual particularity of lithology and topography, which directly leads to the "spatiotemporal asynchronous coupling" phenomenon of hot infrared signals and displacement signals, and this phenomenon has not been recognized and quantified by existing technology, forming an early warning blind area.

[0004] From the perspective of lithology, the red sandstone of Chishui Danxia has a high porosity of 15%-25%, and a high content of clay minerals, resulting in strong water absorption and water retention capacity. After rainwater infiltration, the water on the rock surface causes heat exchange anomalies through evaporation, and hot infrared sensors can capture temperature micro-decrease or fluctuation anomalies within a few hours to one day, forming hot infrared anomaly signals; but water needs to penetrate to deep fissures through pore filling, overcoming capillary resistance, and then triggering displacement through the triple action of "pore water pressure rise", "rock mass strength softening" and "increased self-weight". This infiltration process is significantly delayed by the connectivity of pores and the thickness of the rock mass, resulting in a delay of displacement signals relative to hot infrared signals, and the delay time varies depending on the inducement - the greater the rainfall intensity and the longer the duration, the deeper the water penetration, and the delay time can be up to 0.5-3 days; while the thermal expansion and cold contraction caused by the diurnal temperature difference only affects the shallow rock mass, and the delay time is shortened to 0.5-1 days.

[0005] From the perspective of topographic features, Chishui Danxia is mainly in the form of steep cliff (slope often > 60°), and a large number of concave rock cavities and convex rock ridges are developed, which leads to the spatial distribution of thermal infrared and displacement signals. Thermal infrared anomalies depend on the heat exchange of surface water evaporation, and are more likely to concentrate in areas where "water is easy to stay and expose", such as convex rock ridges and upper cliff. Displacement anomalies are caused by deep fracture instability, and their spatial position depends on "stress concentration area" and "deep water permeable area", such as the lower part of concave rock cavity (rainwater is easy to penetrate into the deep part) and the lower part of cliff (stress concentration area bearing the weight of the upper part), which eventually leads to a 2-5 meter misalignment distance between the spatial distribution centers of thermal infrared anomalies and displacement anomalies in the same instability process.

[0006] The core defect of the prior art is that it has never realized the above-mentioned "spatio-temporal asynchrony" of thermal infrared and displacement signals in Danxia landform, and has not established a quantitative model for its lithology and topographic features. On the one hand, the time lag length of the two under different inducements (rainfall / temperature difference) is not clear, and the problem of "signal asynchrony" cannot be solved. On the other hand, the spatial misalignment distance under different micro-terrain units (concave rock cavity / convex rock ridge) is not defined, and the problem of "signal position mismatch" cannot be solved. Based on the "synchronization assumption", in the scenario where thermal signals appear while displacement has not occurred (time lag), displacement appears while thermal signals have weakened or spatial misalignment (spatial misalignment), false negatives are likely to occur due to signal matching deviation. When relying solely on thermal signals for early warning, false positives are likely to occur due to "thermal anomalies not accompanied by immediate displacement".

[0007] Further, traditional geological monitoring techniques focus on the trend correlation of a single signal, without deeply exploring the spatio-temporal fine coupling relationship of multiple signals. Existing multi-source signal fusion techniques take "spatio-temporal alignment of signals" as a prerequisite, lack quantitative models for asynchronous signals, and existing research has not paid attention to the synergistic effect of high porosity-strong water absorption characteristics of Danxia sandstone and steep topography, mistakenly transferring the signal rules of other landforms, resulting in the above-mentioned "spatio-temporal asynchronous coupling blind area" cannot be solved by traditional knowledge system, which seriously restricts the reliability of Chishui Danxia landform collapse warning. Therefore, for the lithology and topographic features of Chishui Danxia landform, breaking through the quantitative blind area of thermal infrared-displacement signal spatio-temporal asynchrony is a key requirement to improve the accuracy of its collapse collaborative warning.

[0008] Therefore, a collapse thermal infrared-displacement collaborative warning method suitable for Chishui Danxia landform is provided to overcome the above problems. SUMMARY

[0009] The purpose of the present application is to provide a collapse thermal infrared-displacement collaborative warning method suitable for Chishui Danxia landform to solve the problems raised in the above background art.

[0010] To solve the above technical problems, the present application provides a kind of collapse hot infrared displacement collaborative early warning method suitable for Chishui Danxia landform, comprising the following steps:

[0011] (1) construct signal space-time asynchronous quantization model: the time lag quantization relationship between thermal infrared signal and displacement signal is established through experiment, the interval length from thermal infrared signal to displacement signal under different inducements is quantified;A micro-terrain stress-thermal radiation coupling simulation method is established, and the spatial misplacement distance of thermal infrared anomaly and displacement anomaly is quantified;

[0012] (2) improve early warning trigger mechanism: based on space-time asynchronous quantization model, multi-scale chaotic feature extraction method is used to fuse and analyze thermal infrared signal and displacement signal, and asynchronous signal weighted matching is realized through Voronoi diagram space-time topological search window;Design multi-stage early warning process combined with biological indicators, set graded early warning threshold combining biological indicators and physical signals;

[0013] (3) use biological and soil gas emission to assist early warning: monitor the physiological indicators of Chishui Danxia unique moss to build a water stress correlation model, and analyze the correlation between the characteristic peak intensity of soil gas and the crack propagation rate through surface enhanced Raman spectroscopy, which is used for multi-source signal collaborative early warning.

[0014] Further, the establishment of time lag quantization relationship in step (1) includes: using micro fiber grating sensor array to collect layered expansion data of simulated sandstone structure under rainfall or temperature change, and establishing layered expansion amount-moisture penetration depth-time dynamic correlation matrix through gradient moisture content calibration method;Pore connectivity correction coefficient is introduced, and time lag length is quantified by formula.

[0015] Further, the quantification of spatial misplacement distance in step (1) includes: using 1:50 scale micro-terrain model, stress and thermal radiation signals are synchronously collected by thermal infrared sensor and atomic force microscope principle displacement sensor;Introduce topographic curvature correction factor, and quantify spatial misplacement distance by formula.

[0016] Further, the multi-scale chaotic feature extraction method in step (2) includes: the maximum Lyapunov exponent is calculated by variable window size for thermal infrared signal and displacement signal, and the correlation dimension is extracted by sliding embedding dimension optimization G-P algorithm;When thermal infrared temperature anomaly fluctuation exceeds ±0.5℃ and duration is greater than 1h, start Voronoi diagram space-time topological search window, set weight coefficient according to distance from thermal anomaly center, and execute weighted signal matching with maximum Lyapunov exponent greater than 0.05 / h and correlation dimension greater than 2.5 as threshold.

[0017] Further, the multi-stage early warning process of the biological indicator in step (2) comprises: selecting Polytrichum formosum as the indicator organism, monitoring its chlorophyll fluorescence parameter Fv / Fm, and calculating the water stress index; setting the early warning logic: issuing a blue warning when the thermal infrared anomaly is greater than 0.15; upgrading the yellow warning when the displacement is greater than 0.05 mm and greater than 0.3 within the prediction time lag range; issuing a red warning when the displacement is greater than 0.5 mm and greater than 0.4.

[0018] Further, step (2) further comprises a feedback adjustment mechanism: a Q-learning algorithm is used to construct a parameter optimization model, the early warning threshold is used as the action space of the agent, and the reward function considers both the early warning accuracy and the early warning lead time; combining ART2 neural network for clustering analysis of monitoring data, automatically identifying pseudo-anomaly signal patterns and dynamically adjusting the early warning threshold and additional conditions.

[0019] Further, the biological indicator monitoring in step (3) comprises: setting up a Polytrichum formosum sample plot in a cliff with a 5m x 5m grid, collecting fluorescence spectrum curves of 650-750nm wavelength by a fluorescence spectrometer, calculating the characteristic peak intensity ratio and Fv / Fm value, and establishing a three-level transfer model of fluorescence parameters-moss water content-rock mass water content; when the fluorescence intensity ratio decreases by more than 20% and Fv / Fm is less than 0.7, combined with a slight decrease in thermal infrared temperature of more than 0.3℃ and a slight change in displacement of more than 0.02mm, the rock mass water saturation critical state is determined.

[0020] Further, the soil gas emission monitoring in step (3) comprises: setting up sampling probes at the bottom of the cliff with a 10m interval, collecting soil gas samples by Au nanoparticle modified silicon SERS substrate, analyzing Ca 2 + characteristic peak intensity at 10132 nm; using Tucker decomposition and deep belief network to fuse thermal infrared, displacement and soil gas signals, when the SERS characteristic peak intensity increases by more than 50% within 24h and lasts for more than 6h, combined with thermal infrared temperature fluctuation of ±0.4℃ and displacement of 0.03mm, the rock mass unstable state is determined.

[0021] Compared with the prior art, the beneficial effects of the present application are:

[0022] I. Precise quantification of signal spatiotemporal asynchrony, reducing false positives and false negatives:

[0023] Precise quantification of time lag: Through the micro fiber grating sensor array to collect layered expansion data, combined with gradient moisture content calibration method and pore connectivity correction coefficient, the time lag quantitative model of thermal infrared signal and displacement signal under different incentives (rainfall, temperature difference) is established, which can accurately calculate the interval time length between signals (error control within ±1.5h). The problem of missing report of thermal signal appearing while displacement does not occur (missing report) or simply relying on thermal signal false report caused by default "signal synchronization" in the prior art is solved, which provides a precise basis in time dimension for asynchronous signal matching.

[0024] Precise quantification of spatial dislocation: Based on a 1:50 scale micro-topographic model, data are synchronously collected by thermal infrared sensor and atomic force microscope principle displacement sensor, and a topographic curvature correction factor is introduced to quantify the spatial dislocation distance (boundary recognition accuracy of ±0.3m). The spatial dislocation rule (2-5m range) of convex rock ridge thermal anomaly area and concave rock cavity lower displacement concentration area is clarified, which solves the early warning deviation caused by signal space distribution mismatch in the prior art, and provides a precise basis in space dimension for asynchronous signal matching.

[0025] II. Multi-source signal collaborative fusion to improve early warning accuracy and reliability

[0026] Efficient fusion of asynchronous signals: Multi-scale chaotic feature extraction method (variable window calculation of maximum Lyapunov exponent, optimization of G-P algorithm to extract correlation dimension) is adopted, combined with Voronoi diagram space-time topology search window to realize weighted matching of asynchronous signals. Through dynamic adjustment of distance weight coefficient to match priority, thermal infrared and displacement signals with time and space dislocation can be effectively fused, which solves the limitation of traditional multi-source fusion technology relying on "signal space alignment", and improves the robustness of signal fusion.

[0027] Biological-physical signal coupling early warning: A multi-stage early warning process integrating the physiological indicators of the unique moss (Polytrichum formosum) in Chishui Danxia is designed, and a water stress correlation model is established by monitoring the chlorophyll fluorescence parameter Fv / Fm, which combines biological response with physical signals (thermal infrared, displacement) to set graded thresholds (blue, yellow, red early warning). The introduction of biological indicators provides an early indication of rock mass water saturation critical state, which prolongs the early warning lead time (up to 6-12h).

[0028] Ground gas emission auxiliary verification: The surface enhanced Raman spectrum (SERS) is used to analyze the Ca 2 + characteristic peak intensity at 10132nm, and a correlation model between the characteristic peak intensity and the crack propagation rate is established. Through Tucker decomposition and deep belief network fusion of thermal infrared, displacement and ground gas signals, the classification accuracy is improved. The change of ground gas characteristic peak intensity provides a chemical level evidence for the unstable state of rock mass, which further improves the reliability of early warning.

[0029] III. Dynamic adaptive early warning mechanism, enhancing environmental adaptability:

[0030] Intelligent threshold dynamic adjustment: Introduce Q-learning reinforcement learning algorithm to build parameter optimization model, take early warning threshold as agent action space, combine ART2 neural network for clustering analysis of monitoring data, can automatically identify pseudo abnormal signals (such as thermal infrared fluctuation caused by short-term strong sunshine) and dynamically adjust early warning threshold and additional conditions (such as extending the duration requirement). Solving the problem that traditional fixed threshold is difficult to adapt to complex environmental changes, reducing false positives caused by environmental interference.

[0031] Multi-scene precise adaptation: Through zoning and time sampling strategy (setting sampling frequency according to lithology difference) and quantum computing optimized multivariate regression model, it can adapt to the monitoring needs of different lithology areas (pure sandstone area, argillaceous sandstone area, interlayer area) in Chishui Danxia. Combined with micro-topographic stress-thermal radiation coupling simulation, it realizes precise early warning of special micro-topography such as concave rock cavity and convex rock ridge, and improves the applicability of the method in complex topography.

[0032] IV. Expand early warning dimensions:

[0033] Break through the limitations of traditional single physical signal monitoring, and build a "physical-biological-chemical" multi-source collaborative early warning system.

[0034] Data reliability guarantee: Use environmental data processing method combined with quantum key distribution (QKD) technology and block chain storage technology to ensure the security and tamper resistance of monitoring data; train multivariate regression model through quantum annealing algorithm, improve data processing efficiency and prediction accuracy, and provide high-quality data support for early warning decision.

[0035] In summary, the present application significantly improves the accuracy, reliability and advance of Chishui Danxia landform collapse early warning by quantifying signal spatiotemporal asynchrony, fusing multi-source monitoring data, and building dynamic adaptive mechanism, providing effective technical support for geological disaster prevention in the region. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 A schematic diagram of a collapse thermal infrared displacement collaborative early warning method suitable for Chishui Danxia landform. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0038] Referring to Figure 1 , the present application provides a technical solution:

[0039] Referring to Figure 1 An embodiment of a collapse hot infrared displacement collaborative early warning method suitable for the Danxia landform in Chishui:

[0040] Embodiment one: constructing a signal space asynchronous quantization model:

[0041] 1. Time lag quantization:

[0042] 1.1. In view of the high porosity (15%-25%) and strong water absorption of Chishui Danxia red sandstone, a time lag calculation method based on "moisture absorption expansion effect tracking" is established. Select artificial sintered sandstone simulation structures (reproduce the particle size distribution and pore distribution of Chishui Danxia sandstone) with porosity of 18%-22%, and arrange a micro fiber grating sensor array with accuracy of ±0.1με inside the structure at an interval of 5cm×5cm. Design a "gradient water content calibration method": divide the simulation structure into 5 layers (2cm thick each) along the depth direction, and preset different initial water contents (5%-25% gradient) for each layer. When the simulation rainfall (rainfall amount controlled at 5-50mm / h) or temperature change (temperature control accuracy ±0.5℃) occurs, the sensor array synchronously collects 0.01%-0.1% linear expansion data and fiber grating wavelength shift value (resolution 0.1pm) of each layer, and establishes a "layered expansion amount-water permeation depth-time" dynamic correlation matrix.

[0043] 1.2. Based on more than 100 sets of experimental data, a three-dimensional calibration model of moisture absorption expansion amount-time-water permeation depth is constructed, an "equivalent permeation coefficient correction method" is proposed, and the pore connectivity parameter (determined by CT scanning, range 0.3-0.7) of the rock mass is introduced to correct the time lag formula, and the quantization formula is:

[0044]

[0045] Where λ is the pore connectivity correction coefficient (0.3-0.7), which solves the defect that the traditional model does not consider the difference in pore structure. The parameters are explained as follows:

[0046] T is the time lag duration (unit: h), which refers to the interval time from the appearance of the hot infrared signal to the generation of the displacement signal;

[0047] ΔL is the moisture absorption expansion amount of the structure (unit: mm), which is obtained by converting the wavelength shift value monitored by the fiber grating sensor;

[0048] L0 is the initial length of the structure (unit: mm), which is the reference length of the simulation sandstone structure;

[0049] 4.2 is the experimental calibration coefficient, determined by multiple control experiments under the rainfall intensity of 20 mm / h;

[0050] λ is the pore connectivity correction coefficient, obtained by CT scanning to determine the pore connectivity of the rock mass.

[0051] 1.3, the environmental data processing method using quantum key distribution (QKD) technology, design "partitioned time sampling strategy": the monitoring area is divided into three types of sampling area (pure sandstone area, argillaceous sandstone area, interlayer area) according to the lithology difference, and independent sampling frequency is set for each type of area (pure sandstone area 1 time / 10 min, argillaceous sandstone area 1 time / 5 min). Quantum annealing algorithm is used to train more than 100,000 groups of data, and a multiple regression model is established:

[0052] T = a × P + b × D + c × ΔT + d × K,

[0053] Wherein:

[0054] P is the rainfall intensity (unit: mm / h);

[0055] D is the thickness of rock mass (unit: m);

[0056] ΔT is the day and night temperature difference (unit: ℃);

[0057] K is the content of clay minerals (unit: %), determined by X-ray diffraction analysis;

[0058] a, b, c, d are the coefficients optimized by quantum calculation (0.32, 0.56, 0.18, 0.08 respectively);

[0059] The prediction error of the model is controlled within ±1.5h, which is improved compared with the traditional model, and the data is ensured to be tamper-proof through blockchain.

[0060] 2, spatial dislocation quantization:

[0061] 2.1, For the influence of the micro-landform (concave rock cavity depth 1-10 m, convex rock ridge height 2-8 m) on the signal distribution in the Danxia landform of Chishui, a "micro-landform stress-thermal radiation coupling simulation" method was established. A 1:50 scale micro-landform model (made of zirconium titanate piezoelectric ceramic and lithium niobate pyroelectric material) was used to design a "double field synchronous acquisition system". Thermal infrared sensors based on CdSe / ZnS quantum dots (particle size 5 nm) (temperature measurement range -20-80℃, resolution 0.02℃) and atomic force microscope principle displacement sensors (range 0-100μm, resolution 0.1nm) were arranged on the surface and inside of the model. The electrical signal changes (sampling rate 1kHz) of piezoelectric materials (0.1-10mV) and pyroelectric materials (0.01-0.1mV) were synchronously collected. The dynamic correlation between the two was established through the "stress-temperature field coupling coefficient" (calibrated by finite element simulation, range 0.02-0.08).

[0062] 2.2, The spatial dislocation quantification formula introduces a "terrain curvature correction factor", which is:

[0063] L = 0.12 x h x sinα x β ，

[0064] Where β is the terrain curvature correction factor (calculated by unmanned aerial vehicle laser radar data, convex rock ridge 1.2-1.5, concave rock cavity 0.8-1.0), and the parameters are explained as follows:

[0065] L is the horizontal dislocation distance between the displacement anomaly center and the thermal infrared anomaly center (unit: m);

[0066] h is the depth of the concave rock cavity (unit: m), which is obtained by unmanned aerial vehicle laser radar measurement;

[0067] α is the cliff slope (unit: °), which is calculated from the terrain data;

[0068] β is the terrain curvature correction factor, which reflects the influence of micro-landform curvature on signal distribution;

[0069] 0.12 is the experimental calibration coefficient, which is verified by 20 different micro-landform models.

[0070] 2.3, The "micro-landform unit intelligent division algorithm" was developed using the GIS spatial analysis method based on alliance chain technology. A distributed node network was built through the Hyperledger Fabric architecture. The point cloud density of the unmanned aerial vehicle laser radar (50 points / m 2Hash value check on collected terrain data, automatic identification of "stress-terrain sensitive areas" (special areas such as the lower part of concave rock cavity and the root of convex rock ridge) by smart contract, boundary recognition accuracy improved to ±0.3m, spatial dislocation prediction value of each unit updated at a frequency of 1 time / h, forming a time-stamped spatial position reference system.

[0071] Example two: improve early warning trigger mechanism:

[0072] 1. Asynchronous signal fusion trigger:

[0073] Establish a fusion algorithm based on "signal chaotic characteristics and spatiotemporal topology correlation", design "multi-scale chaotic feature extraction method": when analyzing the chaotic characteristics of thermal infrared signals (sampling frequency 1 Hz) and displacement signals (sampling frequency 10 Hz), use variable window size (1000-3000 data points) to calculate the maximum Lyapunov exponent, and use sliding embedding dimension (2-10) to optimize the G-P algorithm to extract the correlation dimension, solving the feature distortion problem caused by traditional fixed window analysis. According to the time lag quantitative model, generate the probability distribution map of displacement signal chaotic characteristics within 0.5-3 days after the appearance of thermal infrared signal.

[0074] When the thermal infrared temperature anomaly fluctuation exceeds ±0.5℃ (duration >1h), start the spatiotemporal topology search window based on Voronoi diagram, introduce "risk weight decay factor": in the predicted displacement anomaly area (radius 5-10m), set the weight coefficient according to the distance from the thermal anomaly center (0.1-1.0 linear decay), and use the maximum Lyapunov exponent >0.05 / h and the correlation dimension >2.5 as the threshold for weighted signal matching. For example, when the thermal infrared anomaly appears on a 5m high convex rock ridge, the matching weight of the displacement signal in the 3-8m concave rock cavity below the ridge is higher, and the matching is successful, which triggers the early warning.

[0075] Construct a fuzzy reasoning method based on quantum logic gates, design "dynamic fuzzy rule base": use 2-qubit logic gate circuit to realize fuzzy operation, divide the thermal infrared signal temperature change amplitude into five levels: 0-0.2℃ (extremely small), 0.2-0.5℃ (small), 0.5-1℃ (medium), 1-2℃ (large), >2℃ (extremely large); displacement is divided into six levels: 0-0.01mm (extremely small), 0.01-0.05mm (small), 0.05-0.2mm (small), 0.2-0.5mm (large), 0.5-1mm (large), >1mm (extremely large). The rule base introduces a "time decay coefficient" (dynamically adjusted from 0.8 to 1.2 as the time of thermal signal appearance increases), and outputs 5 levels of risk level (1-5 levels) through 20 preset fuzzy rules (such as "thermal infrared extremely large and displacement large → risk extremely high").

[0076] 2、Multi-stage early warning:

[0077] Design multi-stage early warning process with biological indicators, establish "biological-physical signal coupling criteria": select drought-tolerant moss (Polytrichum commune) unique to Danxia in Chishui as an indicator organism, monitor its chlorophyll fluorescence parameter Fv / Fm (range 0.6-0.8), and propose "water stress index":

[0078]

[0079] When I > 0.15, it is determined that the water stress state is entered. The early warning logic is optimized as follows: when thermal infrared anomaly (temperature fluctuation > 0.5℃) and I > 0.15 (last > 6h), blue warning (risk level 1) is issued; when displacement > 0.05mm (daily change rate > 0.02mm / d) and I > 0.3 within the prediction time lag, yellow warning (risk level 2) is upgraded; when displacement > 0.5mm (hourly change rate > 0.05mm / h) and I > 0.4, red warning (risk level 3) is issued.

[0080] Establish a feedback adjustment mechanism based on reinforcement learning and adaptive resonance theory, and design a "double-target reward function": use Q-learning algorithm to build a parameter optimization model, take early warning threshold as agent action space, and reward function considers both early warning accuracy (target > 90%) and early warning lead time (target > 6h), solving the problem of insufficient lead time caused by traditional single-target optimization. Combined with ART2 neural network (vigilance parameter 0.7-0.9) for clustering analysis of hourly updated monitoring data, "pseudo-anomaly signal pattern" (such as thermal infrared fluctuation caused by short-term strong sunlight) is automatically identified. For example, when thermal infrared anomaly does not collapse for 3 consecutive times in a certain area, the system not only raises the thermal infrared early warning threshold from 0.5℃ to 0.7℃, but also automatically adds the additional condition of "duration > 2h", and updates the ART2 network clustering center simultaneously.

[0081] Example Three: Use natural phenomena to assist early warning:

[0082] 1、Biological indicator monitoring:

[0083] The moss physiological index monitoring method is used to design the "moss-rock moisture transfer model": the multi-form tower moss sample plots (10 cm x 10 cm) are arranged on the cliff according to the 5 m x 5 m grid, the portable fluorescence spectrometer (detection wavelength 650-750 nm, resolution 1 nm) is used to collect the fluorescence spectrum curve every 2 hours, the characteristic peak intensity ratio (690 nm / 730 nm) and Fv / Fm value are calculated, and the "fluorescence parameter-moss water content-rock water content" three-level transfer model is established. When the fluorescence intensity ratio decreases by more than 20% and Fv / Fm is less than 0.7, combined with the slight decrease (more than 0.3°C) of the thermal infrared temperature and the slight change (more than 0.02 mm) of the displacement, it is determined that the rock mass is in the critical state of moisture saturation, and the early warning is given 6-12 hours in advance.

[0084] The surface temperature and humidity gradient analysis method is used to propose the "temperature and humidity gradient anomaly degree index":

[0085]

[0086] ΔH is the humidity difference, ΔT is the temperature difference, t is the duration, and 10 / 100 m 2 The sensor array (temperature ±0.2°C, humidity ±2%RH) is densely arranged to monitor the temperature and humidity difference within 5 m vertically. When G>5 (i.e. humidity difference>15%, temperature difference>2°C and duration>6h), the temperature and humidity gradient-stress state correlation model is constructed combined with the thermal infrared and displacement signals, for example, 3 days after the rainfall, the humidity at the lower part of the concave rock cavity is 20% higher, the temperature is 1.5°C lower, and the daily displacement increases by 0.05 mm, it is determined that the deep water causes stress change, and multi-dimensional reference data is provided.

[0087] 2. Ground gas escape monitoring:

[0088] The surface enhanced Raman spectroscopy (SERS) based ground gas analysis method is established, and the "characteristic peak intensity-fracture propagation rate calibration curve" is developed: the sampling probe (air inlet height 1.5 m) is arranged at the bottom of the cliff with a 10 m interval, the Au nanoparticle modified silicon-based SERS substrate (enhancement factor 10 6 ), micro-pump (100 mL / min) and spectrometer (400-2000 cm -1 , resolution 4 cm -1 ) are used to collect the ground gas sample in real time, and it is found that the characteristic peak intensity of Ca 2 + at 10132 nm is linearly correlated with the fracture propagation rate When the concentration increases by more than 50% within 24 hours and lasts for more than 6 hours, the rock mass fracture propagation rate is more than 0.01 mm / h, combined with the thermal infrared temperature fluctuation ±0.4°C and the displacement 0.03 mm data, it is determined that the rock mass enters the unstable stage, and the early warning is given 1-2 days in advance.

[0089] A multi-source data fusion method based on Tucker decomposition and deep belief network(DBN) was proposed, and an "asynchronous signal time alignment layer" was introduced. The dynamic time warping(DTW) algorithm layer was added to the input stage of the model to align the time axis of thermal infrared, displacement, and ground-air three types of asynchronous signals, solving the defect that the traditional fusion model did not consider the time sequence difference of signals. The SERS characteristic peak intensity, thermal infrared temperature change, and displacement value three types of high-dimensional data(100-500 dimensions) were compressed into 20-dimensional feature tensors by Tucker decomposition, and input into the DBN model composed of three layers of restricted Boltzmann machine. The classification accuracy of the model on rock mass state(stable / potentially dangerous / dangerous) was improved. When the fusion data shows that the SERS peak intensity increases by 50%, the temperature decreases by 0.6 ℃, and the displacement is 0.08 mm, the output "potentially dangerous" state triggers the intermediate warning.

Claims

1. A collapse hot infrared displacement cooperative early warning method suitable for Chishui Danxia landform, characterized in that, Comprise the following steps: (1) Constructing the time-space asynchronous quantification model of signals: Establish the time lag quantification relationship between thermal infrared signals and displacement signals through experiments, and quantify the interval time length from the occurrence of thermal infrared signals to the generation of displacement signals under different inducements; Establish a micro-topography stress-thermal radiation coupling simulation method to quantify the spatial misplacement distance between thermal infrared anomalies and displacement anomalies; (2) Improving the early warning trigger mechanism: Based on the time-space asynchronous quantification model, use multi-scale chaotic feature extraction method to fuse and analyze thermal infrared signals and displacement signals, and realize asynchronous signal weighted matching through Voronoi diagram time-space topology search window; Design a multi-stage early warning process that integrates biological indicators, set graded early warning thresholds combining biological physiological indicators and physical signals; (3) Using biological and geogas emission to assist early warning: Monitor the physiological indicators of the unique moss in the Danxia of Chishui to build a water stress correlation model, and analyze the correlation between the intensity of geogas characteristic peaks and the crack propagation rate through surface enhanced Raman spectroscopy, which is used for multi-source signal collaborative warning.

2. The collapse hot infrared displacement collaborative early warning method suitable for Chishui Danxia landform according to claim 1, characterized in that: The establishment of time lag quantification relationship in step (1) includes: Using a micro fiber grating sensor array to collect layered expansion data of simulated sandstone structures under rainfall or temperature changes, and establishing a dynamic correlation matrix of layered expansion volume, water infiltration depth and time through gradient water content calibration method; Introduce the pore connectivity correction coefficient and quantify the time lag length through the formula.

3. The collapse hot infrared displacement collaborative early warning method suitable for Chishui Danxia landform according to claim 1, characterized in that: The quantification of spatial misplacement distance in step (1) includes: Using a 1:50 scale micro-topography model, synchronously collecting stress and thermal radiation signals through thermal infrared sensors and atomic force microscope principle displacement sensors; Introduce the terrain curvature correction factor and quantify the spatial misplacement distance through the formula.

4. The collapse hot infrared displacement collaborative early warning method suitable for Chishui Danxia landform according to claim 1, characterized in that: The multi-scale chaotic feature extraction method in step (2) includes: Using variable window size to calculate the maximum Lyapunov exponent for thermal infrared signals and displacement signals, and extracting the correlation dimension through sliding embedding dimension optimization G-P algorithm; When the thermal infrared temperature anomaly fluctuation exceeds ±0.5℃ and the duration is >1h, start the Voronoi diagram time-space topology search window, set the weight coefficient according to the distance from the thermal anomaly center, and execute weighted signal matching with the threshold of maximum Lyapunov exponent >0.05 / h and correlation dimension >2.

5.

5. The collapse hot infrared displacement collaborative early warning method suitable for Chishui Danxia landform according to claim 1, characterized in that: The multi-stage early warning process integrating biological indicators in step (2) includes: Selecting Polytrichum formosanum as the indicator organism, monitoring its chlorophyll fluorescence parameter Fv / Fm, and calculating the water stress index; Set the early warning logic: issue blue warning when thermal infrared anomaly is >0.15; upgrade to yellow warning when displacement is >0.05mm and >0.3 within the predicted time lag range; issue red warning when displacement is >0.5mm and >0.

4.

6. The collapse hot infrared displacement collaborative early warning method suitable for Chishui Danxia landform according to claim 1, characterized in that: Step (2) also includes a feedback adjustment mechanism: Use Q-learning algorithm to build a parameter optimization model, take the early warning threshold as the action space of the agent, and consider the early warning accuracy and early warning lead time in the reward function; Combine ART2 neural network for clustering analysis of monitoring data, automatically identify pseudo-anomaly signal patterns and dynamically adjust the early warning threshold and additional conditions.

7. The collapse hot infrared displacement collaborative early warning method suitable for Chishui Danxia landform according to claim 1, characterized in that: The biological indication monitoring in step (3) comprises: arranging a plurality of Diphyscium squarrosum sample plots in a 5m×5m grid on the cliff wall, collecting a fluorescence spectrum curve of 650-750nm wavelength by a fluorescence spectrometer, calculating a characteristic peak intensity ratio and a Fv / Fm value, and establishing a three-level transfer model of fluorescence parameters-moss water content-rock mass water content; when the fluorescence intensity ratio decreases by more than 20% and the Fv / Fm is less than 0.7, the rock mass water saturation critical state is determined in combination with a thermal infrared temperature slight decrease of more than 0.3 DEG C and a slight displacement change of more than 0.02mm.

8. The collapse hot infrared displacement collaborative early warning method suitable for Chishui Danxia landform according to claim 1, characterized in that: The ground gas escape monitoring in step (3) comprises: arranging sampling probes at 10 m intervals at the bottom of the cliff wall, collecting ground gas samples through Au nanoparticle modified silicon-based SERS substrates, and analyzing Ca 2 + characteristic peak intensity at 10132 nm; Tucker decomposition and deep belief network are used for fusing thermal infrared, displacement and ground gas signals, and when the SERS characteristic peak intensity growth is greater than 50% within 24 h and lasts for more than 6 h, combined with thermal infrared temperature fluctuation of ±0.4℃ and displacement of 0.03mm, the rock mass unstable state is determined.

Images