Cooperative early warning system and method for geological disasters

Abstract

The invention relates to the technical field of geological disaster early warning, in particular to a geological disaster collaborative early warning system and method. The defects that traditional animal behavior observation is high in subjectivity, cannot be quantized and is difficult to standardize are overcome. Through a standardized multi-mode biosensing terminal, an objective baseline modeling algorithm and an automatic anomaly recognition process, the concept that a fuzzy animal can predict disasters is converted into a set of engineering technical scheme which is driven by data, clear in process and reproducible in result. Through a reverse verification mechanism of a physical signal to a biological signal, the technical bottleneck of high false alarm rate of a traditional biological observation method is fundamentally solved. The whole process from data acquisition to early warning release of the whole system is automatic, continuous manual intervention is not needed, the monitoring efficiency is greatly improved, and the system has practical feasibility of deployment and popularization in a large-scale geological disaster high-incidence area.

Description

Technical Field

[0001] This disclosure relates to the field of geological disaster early warning technology, and in particular to a collaborative geological disaster early warning system and method. Background Technology

[0002] The development of geological disaster monitoring and early warning technology has evolved from macroscopic qualitative observation to microscopic precise quantitative monitoring. Early methods relied on manual inspections and simple equipment, resulting in limited early warning capabilities. With the development of modern sensing technology, spatial information technology, and computer technology, an integrated "air-space-ground" physical monitoring system has been formed, represented by Synthetic Aperture Radar Interferometry (InSAR), Global Navigation Satellite System (GNSS), UAV remote sensing, and ground sensors (crack gauges, rain gauges, infrasound sensors, etc.). In recent years, with the integration of big data and artificial intelligence technologies, research focus in this field has shifted to the intelligent fusion of multi-source data, dynamic assessment of disaster risks, and the construction of pre-disaster early warning models, aiming to improve the accuracy and timeliness of early warnings.

[0003] (1) Existing technology 1: Early warning system and data fusion method based on physical monitoring.

[0004] This type of system represents the current mainstream technology. Its system architecture typically involves deploying sensors such as GNSS and crack gauges at disaster-prone locations and then aggregating the data to a central server for analysis. At the data fusion level, there are two main technical approaches: Linear weighted fusion: As disclosed in Chinese patent CN112363251A, this method assigns fixed weights to different physical quantities such as displacement and rainfall through expert scoring or the analytic hierarchy process (AHP), and then performs a weighted summation to obtain a comprehensive risk index. The advantage of this method is its simplicity and intuitiveness, but its linear assumption ignores the complex nonlinear causal relationships between the various physical quantities.

[0005] "Black Box" Intelligent Fusion: Some cutting-edge research has begun to employ machine learning models such as neural networks (NN) and support vector machines (SVM) to train all physical monitoring data as input features in an end-to-end manner. These models can fit nonlinear relationships, but their "black box" nature makes the decision-making process uninterpretable, making it difficult to determine whether it is based on real physical laws or spurious statistical correlations. This is a major flaw in high-risk decision-making fields such as geological safety.

[0006] (2) Existing technology 2: Quantitative analysis technology of animal behavior.

[0007] There have been some preliminary explorations into using animal behavior in disaster research. The main technical methods include: GPS trajectory analysis: By fitting animals with GPS collars, spatial behavioral patterns such as their activity range and migration routes can be analyzed. For example, analyzing changes in the activity radius of cattle and sheep can help explore their correlation with geological disasters.

[0008] Video image analysis: Infrared cameras are set up at fixed locations, and computer vision technology is used to identify and count the frequency and number of animal appearances, as well as specific behaviors such as running and gathering. These technologies quantify single behavioral dimensions, but they are usually isolated and lack the ability to comprehensively analyze multiple behavioral modalities such as animal posture and vocalizations, and to establish a scientific model that can describe its normal state as a "multidimensional behavioral benchmark".

[0009] (3) Existing technology three: a preliminary attempt to combine biological and physical monitoring.

[0010] Currently, publicly available technical solutions combining biosensing and physical monitoring are extremely rare, and most remain at the conceptual or very rudimentary stage. For example, some solutions propose adding a "human observer" position to the physical monitoring system to receive reports of abnormal animal behavior from the public, which are then manually interpreted by experts. This approach is highly subjective, suffers from severe response delays, and lacks a systematic approach, making it impossible to automate or implement in engineering. Currently, there is no collaborative early warning system in this field capable of automatically, in real-time, and deeply integrating multimodal biosensor data with multi-source physical monitoring data.

[0011] Based on the above analysis, the inventors discovered the following fundamental technical bottlenecks in the process of realizing this disclosure: The contradiction between the timeliness of early warning and the completeness of precursor information: Although physical monitoring is accurate, it often responds to macroscopic phenomena in the middle and late stages of a disaster process, resulting in a short warning window; while microscopic precursor information that animals can perceive cannot be captured and utilized scientifically and reliably by current technology, leading to a "perception blind spot".

[0012] The dilemma of data fusion: traditional weighted fusion methods are too simplistic to characterize nonlinear relationships; while "black box" machine learning models are not trusted in high-risk decision-making domains due to their lack of interpretability. There is an urgent need in this field for a fusion model that can handle nonlinear causal relationships while possessing "white box" interpretability.

[0013] The “application gap” of biosensing: Although the potential of animals has been recognized in academia, there is a huge technological gap in three key areas: how to comprehensively quantify various animal behaviors, how to establish objective criteria for judging anomalies, and how to seamlessly integrate them as a reliable data source into automated early warning systems.

[0014] The lack of system-level synergy: Existing technologies, whether physical monitoring or biological observation, are independent and fragmented systems. There is a lack of a top-level design that can organically integrate the two into a complete early warning system that works collaboratively, cross-validates, and makes unified decisions.

[0015] This disclosure aims to solve the following technical problems: Transforming animals' sensitive biological perception capabilities into quantifiable and analyzable scientific data can fill the "perception blind spots" of physical monitoring.

[0016] Establish a collaborative early warning model that integrates biological behavioral data and traditional physical monitoring data to achieve complementary advantages and improve the timeliness and accuracy of early warning.

[0017] Transform unstructured, subjective observations of animal behavior into an automated, objective, and engineerable monitoring and early warning system.

[0018] Furthermore, this disclosure aims to solve a long-standing technical problem in the field: how to identify and suppress abnormal animal behavior "false signals" caused by non-geological disaster factors such as the appearance of natural enemies and human activity interference, and at the same time, how to use these potentially noisy biological signals to guide physical monitoring systems to work more efficiently, so as to maximize the timeliness advantage of biological early warning without sacrificing reliability. Summary of the Invention

[0019] To address the aforementioned issues, this disclosure provides a collaborative early warning system and method for geological disasters, which solves the shortcomings of existing geological disaster monitoring technologies, such as insufficient early warning timeliness, the existence of perception blind spots, and shallow data fusion levels.

[0020] Firstly, a collaborative early warning system for geological disasters includes: The system comprises a biological behavior sensing subsystem, an environmental physical monitoring subsystem, a data transmission and processing center, and a collaborative early warning and decision-making platform. The biological behavior sensing subsystem is used to acquire animal behavior data within the monitoring area; The environmental physical monitoring subsystem is used to acquire ground physical sensor information and spatial remote sensing data within the monitoring area; The data transmission and processing center is used to transmit data collected by the biological behavior sensing subsystem and the environmental physical monitoring subsystem to the center and to establish a database with a unified spatiotemporal reference. The collaborative early warning and decision-making platform, based on a unified spatiotemporal benchmark database, is used to establish a behavioral baseline model and identify abnormal animal behavior; based on abnormal animal behavior and environmental physical data, it outputs the probability of geological disasters and issues early warnings.

[0021] Furthermore, the biological behavior perception subsystem includes: a multimodal biosensing terminal; A multimodal biosensor terminal, worn on a selected target animal, is used to acquire physiological stress indicators of the animal; Multimodal biosensing terminals specifically include: The positioning module is used to obtain the animal's real-time geographical location; The posture and activity monitoring module is used to collect data on the animal's activity intensity and posture. Bioacoustic acquisition module, used to record animal calls; Data transceiver and control unit, used for local caching and wireless transmission of data after acquisition; Furthermore, the environmental physical monitoring subsystem includes: The ground-based physical sensor network specifically includes GNSS receivers, crack gauges, rain gauges, and infrasound sensors deployed in the monitoring area, used for environmental physical monitoring through ground-based physical sensors; Spatial remote sensing data source, used to acquire spatial remote sensing data of the monitoring area.

[0022] Furthermore, the data transmission and processing center includes: A wireless communication network is used to transmit data collected by the biological behavior sensing subsystem and the environmental physical monitoring subsystem to the center, using LoRa, NB-IoT or satellite communication methods; A data fusion and storage server is used to receive, clean, align, and store heterogeneous data to establish a database with a unified spatiotemporal benchmark.

[0023] Furthermore, the collaborative early warning decision-making platform includes: The animal behavior baseline modeling module is used to run machine learning algorithms to learn from the collected animal behavior data and establish a baseline of normal behavior patterns that describes the range of activity, activity rhythm and vocalization patterns of animals under normal behavior conditions. The abnormal behavior identification module is used to compare real-time collected animal behavior data with the baseline of normal behavior patterns, and uses the isolated forest anomaly detection algorithm to identify individual or group abnormal behaviors that deviate from the normal behavior baseline and to perform quantitative rating. The collaborative risk assessment and dynamic verification module is used to dynamically adjust the judgment threshold of the normal behavior baseline; based on abnormal animal behavior and environmental physical data, it outputs the probability of geological disasters through a Bayesian fusion engine.

[0024] Furthermore, the animal behavior baseline modeling module is specifically used for: The collected animal GPS locations, the mean and variance of accelerometer readings, and the Mel frequency cepstral coefficients of vocalizations are used to form a multidimensional behavioral feature vector. This multidimensional behavioral feature vector is then input into a Gaussian mixture model for cluster training to obtain several Gaussian distribution centers and covariance matrices that characterize normal behavior.

[0025] Furthermore, the abnormal behavior recognition module is specifically used for: For real-time collected animal behavior data to form a multidimensional behavioral feature vector, calculate its log-likelihood probability under the established Gaussian mixture model; When the probability value is lower than the preset statistical threshold, it is judged as abnormal behavior, and an abnormality index between 0 and 1 is calculated based on the probability value.

[0026] Furthermore, the threshold for judging normal behavior baseline is dynamically adjusted, including: When the abnormal animal behavior index output by the abnormal behavior recognition module exceeds the preset judgment threshold, the data acquisition frequency of the physical sensor is automatically increased or the judgment threshold is decreased. When abnormal animal behavior is detected, the physical signal co-verification module simultaneously evaluates the status of physical evidence nodes. If no co-verification anomalies are found in the physical monitoring data within a certain time window, a weighting coefficient is generated and fed back to the Bayesian fusion engine to reduce the weight of the abnormal animal behavior signals in order to correct the probability of geological disasters.

[0027] Furthermore, the Bayesian fusion engine employs a dynamic Bayesian network, the structure and parameters of which are obtained through machine learning training using historical disaster data, to represent the uncertain causal relationship between various elements and monitoring indicators.

[0028] Furthermore, dynamic Bayesian networks include: Nodes include physical evidence nodes, biological evidence nodes, intermediate nodes, and target nodes; Physical evidence nodes include rainfall, historical deformation, and real-time deformation; Biological evidence nodes, including abnormal animal behavior; Intermediate nodes include soil saturation, slope stability, and precursory signals of disasters that serve as a physical-biological bridge; The target node outputs the probability of geological disasters. Directed edges between nodes represent causal dependencies; Real-time monitoring data is input into the network, and the probability distribution of the entire dynamic Bayesian network is updated using the confidence propagation inference algorithm. The posterior probability of the target node is then calculated as the probability of geological disaster.

[0029] Secondly, a collaborative early warning method for geological disasters. A collaborative early warning system for geological disasters based on the above; The biological behavior sensing subsystem and the environmental physical monitoring subsystem are launched in parallel to collect animal behavior data streams and environmental physical data streams, and are spatiotemporally aligned through the data transmission and processing center. A baseline of normal behavioral patterns for the target animal population is generated based on historical data; The real-time collected animal behavior data is compared with the baseline of normal behavior patterns to determine whether there are any abnormalities; if an abnormality is detected, the abnormal event is quantified to obtain the animal behavior abnormality index. The probability of geological disasters is obtained by combining the abnormal animal behavior index with the synchronously acquired physical monitoring data as input. The probability of geological disasters is compared with the preset multi-level early warning thresholds to obtain the corresponding early warning level, and then an early warning is issued.

[0030] Furthermore, the abnormal animal behavior index is fused with simultaneously acquired physical monitoring data as input to calculate the probability of geological disasters, specifically including: If the animal's abnormal behavior index exceeds the preset judgment threshold, the collection frequency of the physical sensor will be increased. Animal behavior abnormality index and physical monitoring data were used as evidence and input into a dynamic Bayesian network. To determine whether there is a synergy between physical monitoring data and animal abnormal behavior indices; When physical monitoring data and abnormal animal behavior indices show synergy, it is identified as a high-confidence event, and the probability of geological disaster calculated by dynamic Bayesian network is output. When there is no correlation between physical monitoring data and abnormal animal behavior index, it is judged as a low-confidence event. The abnormal animal behavior signal is downweighted, and the corrected probability of geological disaster is output.

[0031] This disclosure includes at least the following beneficial effects: This disclosure utilizes the natural sensitivity of specific animals to precursory geological hazards such as infrasound, electromagnetic field changes, and micro-vibrations. These precursory signals often appear before macroscopic surface deformation occurs. Through the bio-behavioral perception subsystem and anomaly recognition module of this disclosure, it is possible to capture the stress responses of animal groups triggered by these weak precursors, thereby obtaining earlier warning signals than relying solely on deformation monitoring methods such as GNSS or InSAR. This has irreplaceable value for preventing sudden landslides, collapses, and other geological hazards that have subtle precursors and develop rapidly, providing a valuable window of opportunity for disaster prevention and mitigation.

[0032] Existing physical monitoring networks primarily target limited physical quantities such as displacement, deformation, and rainfall, exhibiting "perception blind spots" for complex geophysical and geochemical precursor information. This disclosure innovatively utilizes animal populations as a distributed, natural "biosensor network," effectively compensating for this deficiency. Animals possess multi-dimensional perception capabilities, comprehensively reflecting environmental anomalies that are difficult for human instruments to fully capture. Therefore, this disclosure constructs a complementary "bio-physical" monitoring system, enabling more comprehensive and multi-dimensional acquisition of disaster precursor information and significantly reducing the risk of missed reports due to incomplete information.

[0033] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the disclosure. The objects and other advantages of this disclosure may be realized and obtained by means of the structures pointed out in the description and the accompanying drawings. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a schematic diagram of the overall architecture of the collaborative early warning system according to an embodiment of this disclosure; Figure 2 This is a schematic diagram of the dynamic Bayesian network structure of the collaborative risk assessment and dynamic verification module in an embodiment of this disclosure; Figure 3 This is a flowchart illustrating the collaborative early warning method according to an embodiment of the present disclosure; Figure 4 This is a schematic diagram illustrating the principle of the collaborative verification and dynamic adjustment mechanism in the embodiments of this disclosure. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0037] like Figure 1 As shown, a collaborative early warning system for geological disasters includes: The system comprises a biological behavior sensing subsystem, an environmental physical monitoring subsystem, a data transmission and processing center, and a collaborative early warning and decision-making platform. The biological behavior sensing subsystem is used to acquire animal behavior data within the monitoring area; The environmental physical monitoring subsystem is used to acquire ground physical sensor information and spatial remote sensing data within the monitoring area; The data transmission and processing center is used to transmit data collected by the biological behavior sensing subsystem and the environmental physical monitoring subsystem to the center and to establish a database with a unified spatiotemporal reference. The collaborative early warning and decision-making platform, based on a unified spatiotemporal benchmark database, is used to establish a behavioral baseline model and identify abnormal animal behavior; based on abnormal animal behavior and environmental physical data, it outputs the probability of geological disasters and issues early warnings.

[0038] The specific implementation details are as follows: It includes at least: a biological behavior sensing subsystem, an environmental physical monitoring subsystem, a data transmission and processing center, and a collaborative early warning decision-making platform.

[0039] (I) System Composition (in conjunction with Appendix) Figure 1 illustrate): Reference Figure 1 The collaborative early warning system provided in this disclosure has an overall architecture that includes at least: a biological behavior sensing subsystem S11, an environmental physical monitoring subsystem S12, a data transmission and processing center S13, and a collaborative early warning decision platform S14. Biological Behavior Perception Subsystem (S11): Multimodal biosensing terminal (S111): A miniaturized, low-power device worn on a selected target animal (such as the marmot from the northern slope of the Tianshan Mountains selected in this embodiment). In some embodiments, the terminal may also integrate a body temperature sensor or a heart rate sensor to obtain physiological stress indicators of the animal as a supplementary dimension for anomaly detection. Its internal integration includes: Positioning module (GPS / BeiDou) (S111.1): Acquires the animal's real-time geographical location for trajectory analysis.

[0040] Posture and Activity Monitoring Module (Triaxial Accelerometer, Magnetometer) (S111.2): Collects data on the animal's activity intensity and posture (such as running, digging, and standing still).

[0041] Bioacoustic acquisition module (miniature microphone) (S111.3): Records animal calls for voiceprint feature analysis.

[0042] Data transceiver and control unit (S111.4): responsible for data acquisition, local buffering and wireless transmission.

[0043] Target animal population (S112): Select representative native animal populations that are sensitive to environmental changes within the monitoring area.

[0044] Environmental physical monitoring subsystem (S12): Ground physical sensor network (S121): includes GNSS receivers, crack gauges, rain gauges, infrasound sensors, etc. deployed in the monitoring area.

[0045] Space remote sensing data source (S122): Regularly accessed InSAR, high-resolution optical satellite, and other data.

[0046] Data transmission and processing center (S13): Wireless communication network (S131): Responsible for transmitting the data collected by subsystems S11 and S12 to the center, and can use LoRa, NB-IoT or satellite communication.

[0047] Data fusion and storage server (S132): Receives, cleans, aligns and stores heterogeneous data from all front ends, and establishes a database with a unified spatiotemporal reference.

[0048] Collaborative Early Warning Decision-Making Platform (S14): Animal behavior baseline modeling module (S141): Runs machine learning algorithms to learn from long-term collected animal behavior data and establishes a "behavioral baseline model" describing the animal's activity range, activity rhythm, and vocalization patterns under normal conditions. It constructs a multi-dimensional feature vector from the collected animal GPS location, the mean and variance of accelerometer readings, and the Mel-frequency cepstral coefficients of vocalizations. This feature vector is then input into a Gaussian mixture model for cluster training, thereby obtaining multiple Gaussian distribution centers and covariance matrices representing normal behavior.

[0049] The abnormal behavior identification module (S142) compares new animal behavior data with the baseline model in real time, using anomaly detection algorithms such as isolated forest to identify individual or group abnormal behaviors that significantly deviate from the baseline, and performs quantitative rating. For new behavioral feature vectors, it calculates their log-likelihood probability under the established Gaussian mixture model. When this probability value is lower than a preset statistical threshold (e.g., lower than the 1% quantile of the probability of all training samples), it is judged as abnormal behavior, and an anomaly index between 0 and 1 is calculated based on the probability value.

[0050] Collaborative Risk Assessment and Dynamic Verification Module (S143): Please refer to Figure 4This module is the core decision-making module of this disclosure. It not only includes a Bayesian fusion engine 45 for deep integration of biological and physical monitoring data, but also embeds a two-way verification and dynamic adjustment mechanism. In a preferred embodiment, the core of this module is a dynamic Bayesian network, the structure and parameters of which can be obtained through machine learning training using historical disaster data. This network can scientifically express the uncertain causal relationships between various geological elements and monitoring indicators.

[0051] by Figure 2 Taking the network topology shown as an example, its specific definition is as follows: Node Definitions: The network contains three types of nodes. Physical evidence nodes include rainfall (S221), historical deformation (S222), and real-time deformation (S223); biological evidence nodes are abnormal animal behavior (S26), whose confidence value is calculated by module S142; intermediate nodes include soil saturation (S23), slope stability (S24), and disaster precursor signals that serve as a physical-biological bridge (S25); the target node is the final output probability of geological disasters (S27).

[0052] Logical Relationship: Directed edges between nodes define causal dependencies. The core logical chain is as follows: physical factors jointly determine slope stability (S24); slope instability triggers precursory signals of disaster (S25); precursory signals lead to abnormal animal behavior (S26). Ultimately, slope stability (S24), representing the physical state, and abnormal animal behavior (S26), representing biological evidence, jointly determine the probability of geological disasters (S27), achieving cross-validation of two independent chains of evidence at the decision-making level.

[0053] Reasoning process: During system operation, real-time monitoring data is input into the network as evidence. The module employs reasoning algorithms such as confidence propagation to update the probability distribution of the entire network, and finally calculates the posterior probability of the target node as a comprehensive risk value.

[0054] Reference Figure 4 The aforementioned "two-way verification and dynamic adjustment mechanism" dynamically adjusts the physical monitoring mode (positive triggering) based on the intensity of the biological signal, and verifies and suppresses "false signals" of the biological signal (reverse verification) based on the response of the physical signal, thereby achieving true intelligent collaborative decision-making. Figure 4As shown, when the input animal behavior abnormality index (S41) exceeds the observation threshold (S42), an enhanced physical monitoring mode is positively triggered (S43), acting on physical monitoring (S44). Simultaneously, biological and physical data are jointly input into the Bayesian fusion engine (S45). The engine's output undergoes reverse verification by the physical signal co-verification module (S46). If verification fails (i.e., physical signal silence), the dynamic weighting controller (S47) generates a weighting coefficient, which is fed back to the fusion engine (S45), thereby outputting a corrected, more reliable risk probability, ultimately yielding the final risk probability (S48). This mechanism constitutes the core intelligent decision-making closed loop of this disclosure. The specific implementation is as follows: (1) "Positive triggering" of biological signals for physical monitoring: When the abnormal animal behavior index (S41) output by the abnormal behavior recognition module (S142) exceeds a preset "observation threshold" (S42), the system automatically triggers the enhanced physical monitoring mode (S43) to increase the data acquisition frequency of key physical sensors (S44) or reduce their judgment threshold.

[0055] (2) "Reverse verification" of physical signals to biological signals: When the system detects abnormal animal behavior, the physical signal co-verification module (S46) will simultaneously evaluate the status of physical evidence nodes. If no co-verification abnormality is found in the physical monitoring data within a certain time window, the dynamic weighting controller (S47) will generate a weighting coefficient and feed it back to the Bayesian fusion engine (S45) to perform weighting processing on the abnormal biological signals, thereby outputting a corrected and more reliable final risk probability (S48), effectively avoiding false alarms.

[0056] A collaborative early warning method for geological disasters, A collaborative early warning system for geological disasters based on the above; The biological behavior sensing subsystem and the environmental physical monitoring subsystem are launched in parallel to collect animal behavior data streams and environmental physical data streams, and are spatiotemporally aligned through the data transmission and processing center. A baseline of normal behavioral patterns for the target animal population is generated based on historical data; The real-time collected animal behavior data is compared with the baseline of normal behavior patterns to determine whether there are any abnormalities; if an abnormality is detected, the abnormal event is quantified to obtain the animal behavior abnormality index. The probability of geological disasters is obtained by combining the abnormal animal behavior index with the synchronously acquired physical monitoring data as input. The probability of geological disasters is compared with the preset multi-level early warning thresholds to obtain the corresponding early warning level, and then an early warning is issued.

[0057] The specific implementation details are as follows: S31: Multi-source data acquisition and synchronization. The system starts the biological behavior perception subsystem (S11) and the environmental physical monitoring subsystem (S12) in parallel to acquire animal behavior data streams and environmental physical data streams, and performs spatiotemporal alignment through the data transmission and processing center (S13).

[0058] S32: Animal behavior baseline modeling. In the early stages of system operation or during non-disaster periods, the animal behavior baseline modeling module (S141) is invoked to train and generate a baseline of normal behavioral patterns for the target animal population based on historical data.

[0059] S33: Real-time Abnormal Behavior Recognition. The abnormal behavior recognition module (S142) is invoked to compare the real-time collected animal behavior data with the baseline model to determine if any abnormalities exist. If no abnormalities are found, the process returns to step S31 to continue monitoring; if an abnormality is detected, the abnormal event is quantified, and the process proceeds to the next step.

[0060] S34: Collaborative Risk Assessment and Dynamic Verification. Trigger the collaborative risk assessment and dynamic verification module (S143), which takes the abnormal animal behavior index output by S33 as input and the synchronously acquired physical monitoring data such as deformation rate and rainfall as input, and performs fusion calculation to obtain the comprehensive disaster risk probability P.

[0061] S34.1: Trigger Enhanced Physical Monitoring. If the animal behavior abnormality index I_bio output by S33 is greater than the threshold_A, the system enters the "Enhanced Physical Monitoring" mode to increase the sampling rate of the physical sensors.

[0062] S34.2: Coordinated Risk Assessment. Using the bioanomaly index I_bio and (possibly enhanced) physical monitoring data as evidence, input them into a dynamic Bayesian network to calculate the comprehensive disaster risk probability P.

[0063] S34.3: Physical Signal Co-verification. During the evaluation process, determine whether there is a synergy between physical monitoring data and abnormal biological signals. If yes, proceed to step S34.5; otherwise, proceed to step S34.4.

[0064] S34.4: Dynamic weight reduction processing. If the event is determined to be of low confidence, the biological abnormal signal is weighted and the corrected risk probability P' is output. Then proceed to step S35.

[0065] S34.5: Output high-confidence risk. If an event is determined to be of high confidence, the risk probability P calculated by the Bayesian network is directly output.

[0066] S35: Early Warning Decision and Issuance. The final risk probability (P or P') is compared with the preset four-level early warning threshold. If P < P1, no early warning is issued; if P1 ≤ P < P2, a blue early warning is issued; and so on. The early warning issuance module (S144) is invoked to push the decision results to relevant departments via SMS, the internet, etc.

[0067] Compared with existing technologies, this disclosure, by deeply integrating animal behavior science with modern physical monitoring technology and constructing an intelligent collaborative decision-making model, brings the following significant beneficial effects: (1) Significantly improve the timeliness of early warning and enhance the ability to warn of sudden disasters.

[0068] This disclosure utilizes the natural sensitivity of specific animals to precursory geological hazards such as infrasound, electromagnetic field changes, and micro-vibrations. These precursory signals often appear before macroscopic surface deformation occurs. Through the bio-behavioral perception subsystem and anomaly recognition module of this disclosure, it is possible to capture the stress responses of animal groups triggered by these weak precursors, thereby obtaining earlier warning signals than relying solely on deformation monitoring methods such as GNSS or InSAR. This has irreplaceable value for preventing sudden landslides, collapses, and other geological hazards that have subtle precursors and develop rapidly, providing a valuable window of opportunity for disaster prevention and mitigation.

[0069] (2) Eliminate the “blind spots” in monitoring and build a more comprehensive disaster precursor information network.

[0070] Existing physical monitoring networks primarily target limited physical quantities such as displacement, deformation, and rainfall, exhibiting "perception blind spots" for complex geophysical and geochemical precursor information. This disclosure innovatively utilizes animal populations as a distributed, natural "biosensor network," effectively compensating for this deficiency. Animals possess multi-dimensional perception capabilities, comprehensively reflecting environmental anomalies that are difficult for human instruments to fully capture. Therefore, this disclosure constructs a complementary "bio-physical" monitoring system, enabling more comprehensive and multi-dimensional acquisition of disaster precursor information and significantly reducing the risk of missed reports due to incomplete information.

[0071] (3) Through the two-way verification mechanism, the problem of “false alarm” in biological early warning has been solved in a revolutionary way, which greatly improves the reliability of decision-making.

[0072] The core innovation of this disclosure lies in constructing a collaborative verification and dynamic noise reduction closed loop for biological and physical evidence chains. Through the mechanism of "reverse verification of biological signals by physical signals," the system can intelligently identify and suppress "false signals" of abnormal animal behavior caused by non-geological factors such as predators and human activities, fundamentally solving the technical bottleneck of high false alarm rates in traditional biological observation methods. Conversely, through the mechanism of "positive triggering of physical monitoring by biological signals," it ensures that no early biological anomalies triggered by genuine, weak precursors are easily overlooked. This intelligent collaborative working mode makes the robustness and accuracy of early warning decisions far exceed that of any single monitoring system or simple information overlay system.

[0073] (4) It realizes the scientific, engineering and automation of biological sensing methods, and has high operability and promotion value.

[0074] This disclosure overcomes the shortcomings of traditional animal behavior observation, which is highly subjective, unquantifiable, and difficult to standardize. Through standardized multimodal biosensor terminals, objective baseline modeling algorithms, and automated anomaly identification processes, this disclosure transforms the vague concept of "animals predicting disasters" into a data-driven, clearly structured, and reproducible engineering solution. The entire system is fully automated from data acquisition to early warning dissemination, requiring no continuous human intervention, greatly improving monitoring efficiency and making it practically feasible for deployment and promotion in large-scale geological disaster-prone areas.

[0075] To enable those skilled in the art to better understand this disclosure, the principles of this disclosure are explained below in conjunction with the accompanying drawings: Example 1: This example uses a landslide-prone area as an example to illustrate the specific application of this disclosure.

[0076] 1. System Deployment: Target animal selection: Marmots, which are widely distributed in the area and sensitive to vibration, were selected as the target animal population. Fifty healthy adult marmots were selected.

[0077] Deployment of the biosensor terminal: Each marmot is equipped with the multimodal biosensor terminal (S111) described in this disclosure. The terminal weighs approximately 30 grams, is powered by solar energy, and has a battery life of more than one year.

[0078] Physical monitoring deployment: Three high-precision GNSS receivers were deployed at key deformation points in the region, and five crack gauges were deployed in potential crack zones. Simultaneously, regional rainfall data from the National Meteorological Administration and regularly published Sentinel-1 satellite InSAR deformation data were integrated.

[0079] Data center construction: Deploy a server at the regional management station as a data transmission and processing center (S13) to receive front-end data via LoRa network.

[0080] 2. Method execution: The method execution process in this embodiment strictly follows... Figure 3 The process is shown below.

[0081] like Figure 3 As shown, after identifying significant abnormal animal behavior (S33), the system first determines whether it exceeds the observation threshold (S34). If so, an enhanced physical monitoring mode is triggered (S34.1) to capture physical precursors with higher sensitivity. Subsequently, in the collaborative risk assessment (S34.2), the system performs reverse verification (S34.3) to determine whether the physical signals are coordinated. If the physical signals are silent, the biological signals are dynamically downweighted, and finally, an early warning decision is made based on the corrected risk probability (S36), thereby effectively avoiding false alarms. Details are as follows: S31-S32 (Baseline Modeling Phase - First 3 Months): The system continuously collects GPS tracks, activity intensity, and vocalization data of 50 marmots. The animal behavior baseline modeling module (S141) uses a Gaussian mixture model to model the marmots' daily activity range (normal activity radius is 500 meters), activity intensity rhythm (active during the day and resting at night), and vocalization spectrum characteristics (distinguishing between normal social calls and high-frequency alarm calls), forming the "normal behavior baseline" for this population.

[0082] S33 (Real-time Anomaly Detection): On a certain day, the abnormal behavior detection module (S142) detected the following composite anomaly: Spatial anomaly: Within 2 hours, GPS tracks of 35 marmots (70% of the total) showed that they migrated rapidly in a northeast direction, with an average migration distance of more than 1,500 meters, far exceeding their normal activity radius.

[0083] Abnormal activity: Accelerometer data shows that these marmots have consistently high activity levels, exhibiting prolonged running.

[0084] Acoustic Anomaly: The microphone picked up a large number of long-lasting, high-frequency alarm calls. The module classifies this as a "Level 1 Group Anomaly" and outputs an anomaly index of 0.85 (range 0-1).

[0085] S34 (Collaborative Risk Assessment): Because the detected anomaly index of 0.85 far exceeds the observation threshold (e.g., 0.6), the system first triggers enhanced physical monitoring (corresponding to...). Figure 3 (Step S34.1). Subsequently, in the collaborative risk assessment (corresponding to...) Figure 3 In step S34.2), the system discovered a high degree of synergy between physical signals such as GNSS displacement acceleration and heavy rainfall and biological signals. Therefore, it verified the synergy through physical signals (corresponding to...). Figure 3In step S34.3), the final output is a high-confidence risk probability P=0.92 (corresponding to...). Figure 3 (Step S34.5).

[0086] S35 (Early Warning Decision): Because the risk probability P=0.92 exceeds the preset red early warning threshold (P4=0.9), the system executes an early warning decision and issues an alert (corresponding to...). Figure 3 In step S35), the warning release module (S144) is invoked to immediately and automatically generate a "red warning" and send the warning information to the local emergency management department and the person in charge of the threatened area through a dedicated SMS channel. The content includes: "The risk of landslide in XX area is extremely high. The highest level emergency plan should be activated immediately and personnel should be evacuated!"

[0087] 3. Effect Verification: Six hours after the warning was issued, a medium-sized landslide occurred in the area. Thanks to the timely warning, all threatened personnel were safely evacuated, avoiding any casualties.

[0088] Example 2: Application of Intelligent Biological Monitoring Terminal Based on Edge Computing This embodiment describes a distributed, low-latency implementation of the present disclosure, which is particularly suitable for scenarios with limited network bandwidth or extremely high real-time response requirements. Its core is to move computational tasks such as identifying abnormal biological behavior to the edge devices at the monitoring front end.

[0089] 1. System Deployment: Core hardware change: The "multimodal biosensing terminal (S111)" in Example 1 is upgraded to an "edge intelligent processing unit". This edge intelligent processing unit not only includes sensors such as GPS, accelerometer, and microphone, but also has a built-in low-power AI processor with local computing capabilities.

[0090] Physical and biological monitoring deployment: Consistent with Example 1.

[0091] Data Center Role Change: The role of the server (S13) in the regional management station changes from "data transmission and processing center" to "risk aggregation and decision-making center". It no longer needs to process massive amounts of raw biosensor data, but instead receives structured "abnormal event" data from the front-end edge intelligent processing unit after local analysis.

[0092] 2. Method execution: S31-S32 (Baseline Modeling and Model Deployment): Animal behavior baseline modeling (S141) is still done on the central server because it requires a global historical dataset for training.

[0093] Key difference: After training, the generated "normal behavior baseline model" is compiled and optimized, and then distributed and deployed to each front-end edge intelligent processing unit via a wireless network (such as 4G or during regular maintenance).

[0094] S33 (Edge real-time anomaly recognition): Computing location: The entire computing process of anomaly behavior recognition (S142) runs in real-time entirely on the device worn by the marmot.

[0095] Local processing: The AI processor on the edge intelligent processing unit analyzes the raw data streams from the GPS, accelerometer, and microphone in real-time and compares them with the baseline model stored locally.

[0096] Data transmission: When the edge intelligent processing unit detects a compound anomaly, it no longer uploads a large amount of raw data traces or audio. Instead, it generates a highly condensed and standardized JSON-format "anomaly alert" data packet and sends it to the central server via the LoRa network.

[0097] Example of data packet: {"terminal_id": "EIT_035", "timestamp": "2025-10-27T14:30:00Z", "anomaly_type": ["spatial_migration", "acoustic_alarm"], "anomaly_index": 0.85} Data volume comparison: The size of this data packet is only a few bytes, while the original GPS traces and audio data can be up to several megabytes.

[0098] S34 - S35 (Central collaborative risk assessment and early warning): After the central server receives the "anomaly alerts" from one or more intelligent processing units, it triggers the collaborative risk assessment and dynamic verification module (S143).

[0099] The subsequent processes are the same as those in Embodiment 1: Integrate physical monitoring data, calculate the comprehensive risk probability, and issue an early warning.

[0100] 3. Beneficial effects of this embodiment: Greatly reduces the network bandwidth requirements, making it possible to carry out large-scale deployments in remote mountainous areas without fiber optic or 5G coverage. Achieves lower early warning latency because anomaly recognition is completed immediately at the scene of the event. Enhances the robustness of the system. Even if the central network is briefly interrupted, the front-end intelligent processing unit can still work independently and even achieve preliminary early warning through local linkage.

[0101] Example 3: Application of a wide-area adaptive early warning system with cloud-edge collaboration This embodiment describes a state-of-the-art, adaptively optimized system architecture for deployment on a larger scale (such as a county on the northern slope of the Tianshan Mountains), embodying a collaborative working mode between the cloud brain and edge neurons.

[0102] 1. System Deployment: Architecture: It adopts a two-level architecture of "cloud platform + edge node".

[0103] Cloud-based brain: Deployed on a public or private cloud, it possesses the capabilities for massive data storage, large-scale parallel computing, and complex model training. It is responsible for global situational awareness, iterative model optimization, and cross-regional early warning decision-making.

[0104] Edge nodes: A wide-area monitoring network consisting of hundreds of edge intelligent processing units as described in Example 2, covering multiple different geological hazard points.

[0105] Data flow: Edge nodes are responsible for real-time perception and preliminary analysis, while the cloud-based brain is responsible for in-depth analysis and global optimization.

[0106] 2. Method execution: Initial phase: The cloud platform uses historical data to train the first generation of "general baseline model" and "risk assessment model" and distributes them to all edge nodes.

[0107] Collaborative workflow: Edge perception and reporting: Each edge node independently performs real-time anomaly identification and uploads "anomaly alarms" and a small number of "typical normal" data samples to the cloud brain.

[0108] Cloud-based global situational analysis: The collaborative risk assessment and dynamic verification module (S143) of the cloud-based brain not only integrates physical and biological data from a single site, but also analyzes the spatiotemporal correlation of anomalous events reported by multiple edge nodes. For example, it can identify a macroscopic pattern of "abnormal behavior spreading from southwest to northeast along a fault zone," which is imperceptible to a single edge node.

[0109] It integrates larger-scale physical data, such as regional weather radar images and wide-area InSAR subsidence field data, to conduct comprehensive regional risk assessment.

[0110] Cloud-based adaptive model optimization: The animal behavior baseline modeling module (S141) of the cloud-based brain continuously gathers new data uploaded from all edge nodes. It retrains and optimizes the "behavioral baseline model" and "risk assessment model" regularly (e.g., monthly) or after major events (e.g., a successful alert or a missed alert) using the new data.

[0111] For example, it found that marmots in region A are more sensitive to rainfall, while marmots in region B are more sensitive to micro-seismic events, thus generating personalized, customized models adapted to the local environment for edge intelligent processing units in different regions.

[0112] Online model updates: The optimized new model is wirelessly pushed to the corresponding edge nodes via over-the-air download technology, enabling continuous and automatic iterative upgrades of the intelligence level of the entire early warning system.

[0113] S35 (Tiered Early Warning Decision): Based on the risk assessment results, the cloud platform can make more refined early warning decisions, such as issuing a red alert for the core area and an orange alert for the potentially affected area.

[0114] 3. The beneficial effects of this embodiment: This represents a leap from "single-point early warning" to "regional collaborative prevention," significantly improving the accuracy and coverage of early warning systems. It endows the system with self-learning and evolutionary capabilities, enabling it to continuously adapt to environmental changes and shifts in animal behavior, thus preventing model rigidity and failure. It provides a complete technical solution for large-scale, operational geological disaster monitoring networks, possessing extremely high commercial value and widespread application potential.

[0115] Comparative Example 1: Comparison with Simple Weighted Fusion Method To highlight the superiority of the collaborative early warning model disclosed herein, a comparative scheme is set up. This scheme eliminates the dynamic Bayesian network and two-way verification mechanism of this disclosure, and adopts the linear weighted fusion method commonly used in existing technologies. That is, the quantified animal abnormality index and the abnormality index of each physical monitoring indicator are weighted and summed according to fixed weights preset by experts (e.g., animal abnormality accounts for 30%, displacement accounts for 40%, and rainfall accounts for 30%) to obtain a comprehensive risk score.

[0116] Scenario simulation and result comparison: A simulated real-world scenario: The presence of a large predator (snow leopard) in a certain area causes a large-scale, prolonged running and alarm calls from the marmot population, with an animal anomaly index as high as 0.9. However, at this time, all physical monitoring indicators, such as GNSS and crack gauges, show stability.

[0117] Results of the comparative scheme: Due to the use of fixed weights, the high index of animal abnormalities (0.9 * 30%) directly caused the comprehensive risk score to exceed the warning threshold, and the system erroneously issued a blue warning.

[0118] The results of Embodiment 1 of this disclosure: The system detected an animal anomaly index of 0.9, but the "reverse verification" mechanism found that the physical evidence chain was completely silent. Therefore, the system judged this event as a "high-probability event of non-geological origin", dynamically reduced the weight of the biological anomaly index, and the final comprehensive risk probability did not exceed the warning threshold. No warning was issued, and a false alarm was successfully avoided.

[0119] This comparative example clearly demonstrates that the collaborative verification mechanism disclosed herein can effectively distinguish between disaster precursors and environmental interference. Its technical effects in reducing false alarm rates and improving early warning reliability are unmatched by simple weighted fusion methods, and it has significant inventiveness.

[0120] Quantification and baseline modeling techniques for animal behavior: How to transform unstructured animal behaviors (position, posture, sound) into analyzable structured data and establish mathematical models that can accurately describe their "normal" state.

[0121] The spatiotemporal alignment and deep fusion algorithm for biological-physical heterogeneous data: The core is the decision model in the collaborative risk assessment and dynamic verification module. It must be able to scientifically process two completely different sources, scales, and response characteristics of data, and provide reliable fusion results.

[0122] Low-power, miniaturized multimodal biosensor terminal design: Hardware implementation is the prerequisite for the application of the entire system, and it is necessary to ensure minimal impact on animals, long battery life, and reliable data acquisition.

[0123] A collaborative early warning system for geological disasters includes: a biological behavior sensing subsystem for acquiring quantitative animal behavior data; an environmental physical monitoring subsystem for acquiring surface physical parameters; and a collaborative early warning decision platform configured with a collaborative decision model, the platform being able to integrate data from the two subsystems to assess disaster risk.

[0124] A collaborative early warning method for geological disasters includes: collecting biological behavior data and physical monitoring data in parallel; establishing an animal behavior baseline model based on historical data; identifying abnormal behaviors that deviate from the baseline in real time; fusing the abnormal behaviors and physical monitoring data to calculate a comprehensive risk probability; and making early warning decisions based on the probability.

[0125] The collaborative early warning decision-making platform uses a Bayesian network model for data fusion. This network includes physical evidence nodes, biological evidence nodes, and intermediate nodes that connect the two.

[0126] An edge computing architecture is adopted, in which the identification of abnormal animal behavior is performed on the front-end sensing terminal.

[0127] A cloud-edge collaborative architecture is adopted, in which the cloud platform is responsible for training and optimizing the model, and can distribute the updated model to the edge nodes.

[0128] The multimodal sensing terminal for animal behavior perception integrates a positioning module, an activity monitoring module, and an acoustic acquisition module, and can optionally be configured with a local AI processing unit.

[0129] Although the present disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. A collaborative early warning system for geological disasters, characterized in that, include: The system comprises a biological behavior sensing subsystem, an environmental physical monitoring subsystem, a data transmission and processing center, and a collaborative early warning and decision-making platform. The biological behavior sensing subsystem is used to acquire animal behavior data within the monitoring area; The environmental physical monitoring subsystem is used to acquire ground physical sensor information and spatial remote sensing data within the monitoring area; The data transmission and processing center is used to transmit data collected by the biological behavior sensing subsystem and the environmental physical monitoring subsystem to the center and to establish a database with a unified spatiotemporal reference. The collaborative early warning and decision-making platform, based on a database with a unified spatiotemporal reference, is used to establish a behavioral baseline model and identify abnormal animal behavior. Based on abnormal animal behavior and environmental physical data, the probability of geological disasters is output and early warnings are issued.

2. The geological disaster collaborative early warning system according to claim 1, characterized in that, The biological behavior perception subsystem includes: a multimodal biological sensing terminal; A multimodal biosensor terminal, worn on a selected target animal, is used to acquire physiological stress indicators of the animal; Multimodal biosensing terminals specifically include: The positioning module is used to obtain the animal's real-time geographical location; The posture and activity monitoring module is used to collect data on the animal's activity intensity and posture. Bioacoustic acquisition module, used to record animal calls; The data transceiver and control unit is used for local caching and wireless transmission of data after acquisition.

3. The geological disaster collaborative early warning system according to claim 1, characterized in that, The environmental physical monitoring subsystem includes: The ground-based physical sensor network specifically includes GNSS receivers, crack gauges, rain gauges, and infrasound sensors deployed in the monitoring area, used for environmental physical monitoring through ground-based physical sensors; Spatial remote sensing data source, used to acquire spatial remote sensing data of the monitoring area.

4. The geological disaster collaborative early warning system according to claim 1, characterized in that, Data transmission and processing center, including: A wireless communication network is used to transmit data collected by the biological behavior sensing subsystem and the environmental physical monitoring subsystem to the center, using LoRa, NB-IoT or satellite communication methods; A data fusion and storage server is used to receive, clean, align, and store heterogeneous data to establish a database with a unified spatiotemporal benchmark.

5. A collaborative early warning system for geological disasters according to claim 1, characterized in that, The collaborative early warning and decision-making platform includes: The animal behavior baseline modeling module is used to run machine learning algorithms to learn from the collected animal behavior data and establish a baseline of normal behavior patterns that describes the range of activity, activity rhythm and vocalization patterns of animals under normal behavior conditions. The abnormal behavior identification module is used to compare real-time collected animal behavior data with the baseline of normal behavior patterns, and uses the isolated forest anomaly detection algorithm to identify individual or group abnormal behaviors that deviate from the normal behavior baseline and to perform quantitative rating. The collaborative risk assessment and dynamic verification module is used to dynamically adjust the judgment threshold of the normal behavior baseline; based on abnormal animal behavior and environmental physical data, it outputs the probability of geological disasters through a Bayesian fusion engine.

6. A collaborative early warning system for geological disasters according to claim 5, characterized in that, The animal behavior baseline modeling module is specifically used for: The collected animal GPS locations, the mean and variance of accelerometer readings, and the Mel frequency cepstral coefficients of vocalizations are used to form a multidimensional behavioral feature vector. This multidimensional behavioral feature vector is then input into a Gaussian mixture model for cluster training to obtain several Gaussian distribution centers and covariance matrices that characterize normal behavior.

7. A collaborative early warning system for geological disasters according to claim 5, characterized in that, The abnormal behavior recognition module is specifically used for: For real-time collected animal behavior data to form a multidimensional behavioral feature vector, calculate its log-likelihood probability under the established Gaussian mixture model; When the probability value is lower than the preset statistical threshold, it is judged as abnormal behavior, and an abnormality index between 0 and 1 is calculated based on the probability value.

8. A collaborative early warning system for geological disasters according to claim 5, characterized in that, The threshold for judging normal behavior baseline is dynamically adjusted, including: When the abnormal animal behavior index output by the abnormal behavior recognition module exceeds the preset judgment threshold, the data acquisition frequency of the physical sensor is automatically increased or the judgment threshold is decreased. When abnormal animal behavior is detected, the physical signal co-verification module simultaneously evaluates the status of physical evidence nodes. If no co-verification anomalies are found in the physical monitoring data within the time window, a weighting coefficient is generated and fed back to the Bayesian fusion engine to reduce the weight of the abnormal animal behavior signals in order to correct the probability of geological disasters.

9. A collaborative early warning system for geological disasters according to claim 5, characterized in that, The Bayesian fusion engine employs a dynamic Bayesian network, whose structure and parameters are obtained through machine learning training using historical disaster data. This network is used to represent the uncertain causal relationships between various elements and monitoring indicators.

10. A collaborative early warning system for geological disasters according to claim 9, characterized in that, Dynamic Bayesian networks include: Nodes include physical evidence nodes, biological evidence nodes, intermediate nodes, and target nodes; Physical evidence nodes include rainfall, historical deformation, and real-time deformation; Biological evidence nodes, including abnormal animal behavior; Intermediate nodes include soil saturation, slope stability, and precursory signals of disasters that serve as a physical-biological bridge; The target node outputs the probability of geological disasters. Directed edges between nodes represent causal dependencies; Real-time monitoring data is input into the network, and the probability distribution of the entire dynamic Bayesian network is updated using the confidence propagation inference algorithm. The posterior probability of the target node is then calculated as the probability of geological disaster.

11. A collaborative early warning method for geological disasters, characterized in that, A collaborative early warning system for geological disasters based on any one of claims 1-10; The biological behavior sensing subsystem and the environmental physical monitoring subsystem are launched in parallel to collect animal behavior data streams and environmental physical data streams, and are spatiotemporally aligned through the data transmission and processing center. A baseline of normal behavioral patterns for the target animal population is generated based on historical data; The real-time collected animal behavior data is compared with the baseline of normal behavior patterns to determine whether there are any abnormalities. If an anomaly is detected, the abnormal event is quantified to obtain an animal behavior anomaly index; The probability of geological disasters is obtained by combining the abnormal animal behavior index with the synchronously acquired physical monitoring data as input. The probability of geological disasters is compared with the preset multi-level early warning thresholds to obtain the corresponding early warning level, and then an early warning is issued.

12. The method for collaborative early warning of geological disasters according to claim 11, characterized in that, The probability of geological disasters is calculated by fusing animal behavior abnormality index with synchronously acquired physical monitoring data, specifically including: If the animal's abnormal behavior index exceeds the preset judgment threshold, the collection frequency of the physical sensor will be increased. Animal behavior abnormality index and physical monitoring data were used as evidence and input into a dynamic Bayesian network. To determine whether there is a synergy between physical monitoring data and animal abnormal behavior indices; When physical monitoring data and abnormal animal behavior indices show synergy, it is identified as a high-confidence event, and the probability of geological disaster calculated by dynamic Bayesian network is output. When there is no correlation between physical monitoring data and abnormal animal behavior index, it is judged as a low-confidence event. The abnormal animal behavior signal is downweighted, and the corrected probability of geological disaster is output.

Images