A health preventive monitoring system for ancient trees

Abstract

This invention relates to the field of ancient tree protection technology and discloses an ancient tree health preventive monitoring system, including a data input module, a monitoring module, an identification module, and an output module. The data input module collects multi-dimensional heterogeneous data, performs normalization processing to generate a normalized feature matrix; the monitoring module receives the normalized feature matrix and real-time monitoring vectors, performs inference calculations, and generates a risk outbreak probability and a composite risk intensity; the identification module integrates the risk outbreak probability, composite risk intensity, and the normalized real-time monitoring vectors to output a comprehensive health and safety score, compares the comprehensive health and safety score with a four-level safety threshold range to determine a preliminary risk level, extracts historical risk data of the same tree species in the same area, runs a misjudgment correction algorithm to process the preliminary risk level, and outputs a final risk level; the output module triggers different response commands to execute automated disposal operations based on the final risk level. This invention can reduce long-term risk prediction errors.

Description

Technical Field

[0001] This invention relates to the field of ancient tree protection technology, specifically to an ancient tree health preventive monitoring system. Background Technology

[0002] As important natural and historical cultural heritage, the health monitoring and preventive protection of ancient trees are crucial. Traditional ancient tree health monitoring systems deploy sensors to collect physiological and environmental data, and combine this data with models to provide early warnings of potential risks to ancient trees.

[0003] However, existing ancient tree health monitoring systems have significant shortcomings in practical applications. Currently, monitoring methods for the health of ancient trees mainly rely on the following approaches: Traditional manual inspection: This relies on forestry experts to conduct regular on-site inspections and make judgments based on experience. This method is inefficient, highly subjective, cannot achieve continuous monitoring, and is difficult to detect early minor anomalies in a timely manner, often only being discovered when the risk has deteriorated to an irreversible stage. Single-point sensor monitoring: This involves deploying a few environmental sensors (such as thermometers and hygrometers) around the ancient tree, triggering an alarm only when the data exceeds a fixed threshold. This type of solution has a single monitoring dimension, does not consider the coupling effect between multiple risk factors, lacks effective model extrapolation capabilities, and has high false alarm and false negative rates; moreover, the system is open-loop and cannot self-optimize and correct based on feedback. Existing technologies still have significant shortcomings in terms of the comprehensiveness of monitoring dimensions, quantification of multi-risk factor coupling, dynamic adaptive updating of models, and automated hierarchical handling.

[0004] Chinese patent application CN202510183047 discloses a multi-factor integrated monitoring method and system for the growth environment of ancient trees. This scheme utilizes sensors to collect meteorological, soil, and pest and disease data, and uses a chaotic theory model (including phase space reconstruction, chaotic attractor extraction, and Lyapunov exponent calculation) to predict future trends in the growth environment of ancient trees. While this scheme can perform dynamic trend analysis of environmental factors, it primarily focuses on predicting changes in external environmental factors and does not integrate the monitoring of the ancient tree's own structural characteristics (such as trunk tilt, internal rot rate, and root exposure) with its real-time physiological state (such as stress wave transmission and chlorophyll fluorescence efficiency). Furthermore, this scheme lacks quantitative analysis of the coupling effects between multiple risk factors, does not establish a closed-loop linkage mechanism between monitoring results and graded early warning and automated response, and lacks the model self-optimization capability based on manual verification and feedback.

[0005] For example, Chinese patent application CN202410910703 discloses a health monitoring system and method for the maintenance of ancient trees. This scheme monitors the runoff (transpiration) flow of ancient trees and combines this with temperature and light intensity parameters to establish a health assessment model (three-dimensional surface fitting) to determine the health status of the ancient trees. This scheme, starting from the physiological perspective of trees and reflecting health problems through changes in transpiration, has a certain degree of innovation. However, its monitoring dimensions are relatively singular, relying solely on runoff parameters as the core judgment indicator, and failing to fully consider multi-source heterogeneous risk factors such as trunk rot, root exposure, pest and disease vibration, and photosynthetic efficiency. Furthermore, its health assessment model is based on static surface fitting of historical data, lacking the ability to adaptively adjust to fluctuations in sensor data, and failing to achieve multi-level risk classification response and dynamic iterative optimization of model parameters.

[0006] Therefore, this invention proposes a preventive monitoring system for the health of ancient trees to address the shortcomings of existing technologies. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a preventative health monitoring system for ancient trees.

[0008] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a preventive health monitoring system for ancient trees, which includes a data input module, a monitoring module, an identification module, and an output module.

[0009] The data input module is used to collect multi-dimensional heterogeneous data and perform normalization processing to generate a normalized feature matrix; The monitoring module is communicatively connected to the data input module. The monitoring module includes: a real-time monitoring unit deployed at the ancient tree site, used to collect real-time monitoring vectors at a set sampling frequency; and a preventive monitoring model, used to receive the normalized feature matrix and the real-time monitoring vectors, calculate the comprehensive weight vector through a hierarchical-entropy weight composite algorithm, and calculate the probability of risk outbreak and the intensity of composite risk based on a spatiotemporal evolution inference algorithm. The identification module, which is communicatively connected to the monitoring module, is used to receive the risk outbreak probability, the composite risk intensity, and the normalized real-time monitoring vector, perform data fusion calculation, output a comprehensive health and safety score, compare the comprehensive health and safety score with a preset four-level safety threshold range, calculate and output a preliminary risk level, extract historical risk data of ancient trees of the same tree species in the same area to construct a priori knowledge input set, run a misjudgment correction algorithm to process the preliminary risk level, and output the final risk level. The output module, which is communicatively connected to the identification module, is used to receive the final risk level and automatically trigger response instructions at different levels to perform automated handling operations for the final risk level with different range thresholds.

[0010] The formula used by the data input module to perform positive index standardization is: ; Defined as the normalized output value of a positive indicator; Defined as the original value of the positive indicator input; Defined as the historical maximum value of the input indicator; Defined as the historical statistical minimum value of the input indicator; The formula for the inverse normalization of negative indicators is: ; Defined as the normalized output value of a negative indicator; Defined as the original value of the negative indicator input.

[0011] The real-time monitoring unit includes a trunk stress wave sensor, a biological attack monitoring sensor, a millimeter-wave radar, and a chlorophyll fluorescence sensor, which are used to collect stress wave transmission time, pest vibration signal intensity, canopy micro-displacement, and photosynthetic efficiency index, respectively, and to construct a real-time monitoring vector. , Defined as the stress wave propagation time measured by the tree trunk stress wave sensor; Defined as the intensity of the insect vibration signal measured by the biological infestation monitoring sensor; Defined as the micro-displacement of the canopy measured by millimeter-wave radar; Defined as the photosynthetic efficiency index measured by a chlorophyll fluorescence sensor.

[0012] The monitoring module calculates the comprehensive weight vector using a hierarchical-entropy weighted composite algorithm. : ; Defined as the first obtained after fusion The combined weight of each factor; Defined as the fusion coefficient, the system sets the fusion coefficient. The value is 0.6; Defined as data-driven weights based on the coefficient of variation; Defined as the objective weight calculated using the entropy weight method; the monitoring module combines and arranges the comprehensive weights of all factors to form a comprehensive weight vector. The overall weight vector W = [W1, W2, ..., W...] m ] T , where m is a variable representing the number of factors.

[0013] The monitoring module dynamically adjusts the fusion coefficient λ based on the data fluctuation dispersion and enforces numerical out-of-bounds limits, keeping it within the range of [0.3, 0.8]. , Defined as the composite weight fusion coefficient; Defined as the basic fusion coefficient value, set by the system. The value is 0.6; Defined as a penalty adjustment coefficient, set by the system. The value is 0.15; Defined as volatility dispersion.

[0014] The monitoring module defines meteorological factors, biological factors, and tree factors as primary factors, and soil factors, topographic factors, and genetic and maintenance factors as secondary factors, and calculates the coupling strength coefficient between primary and secondary factors. The following calculation formula is satisfied: ; W i and W j The first-order factor F i and second-order factor F j The corresponding weights; Defined as the Pearson correlation coefficient between the time series of the primary factor and the secondary factor.

[0015] The comprehensive loss function constructed by the monitoring module is: ; In the above formula, Defined as a comprehensive loss function; Defined as the total sample size; Defined as sample sequence number; Defined as the first Predicted risk value for each sample; Defined as the first The actual observed risk value for each sample; β is defined as the regularization coefficient, and the system sets the value of β to 0.01; Defined as the total number of weight parameters; Defined as parameter sequence number; Defined as the first One weight parameter.

[0016] The identification module calculates a comprehensive health and safety score. The risk level is determined by comparing the data with the four-level safety threshold ranges [0.8, 1.0], [0.6, 0.8], [0.4, 0.6], and [0, 0.4]. The calculation formula satisfies: ; In the above formula, Defined as a comprehensive health and safety score; Defined as a weighting coefficient for the probability of risk occurrence; Defined as the probability of a risk occurring; Defined as the weighting coefficient of the composite risk intensity after normalization; Defined as the composite risk intensity after normalization; Defined as the weight coefficient of the real-time monitoring vector after normalization; Defined as the real-time monitoring vector after normalization.

[0017] The Bayesian inference misjudgment correction formula executed by the recognition module is as follows: In the above formula, Defined as posterior probability; Defined as the first A preliminary risk level; Defined as a sequence number for the initial risk level. The set of possible values ​​for is 1, 2, 3, and 4. Defined as a comprehensive health and safety score based on observational data; Defined as in the first Observational data appeared at a preliminary risk level. Class conditional probability; Defined as the first The prior probability of a preliminary risk level; Defined as the accumulating variable index that iterates through the four levels of security threshold intervals; Defined as the first A preliminary risk level; Defined as in the first Observational data appeared at a preliminary risk level. Class conditional probability; Defined as the first The prior probability of a preliminary risk level.

[0018] The automated processing operations of the output module include: Low-risk zone: Generate routine maintenance reminders and store them in the ancient tree electronic archive database; Medium-risk zone: Send preventative maintenance recommendations to the management platform and send a frequency modification instruction to the monitoring module, changing the simulation frequency from 24 hours / time to 12 hours / time; High-risk zone: Send early warning execution plans to mobile terminals and management platforms; Extremely high risk zone: Trigger on-site audible and visual alarm devices and report emergency warning information to the forestry authority system.

[0019] This invention provides a preventative health monitoring system for ancient trees. It has the following beneficial effects: 1. This invention extracts predicted risk values ​​and actual observed risk values ​​through a monitoring module to construct a comprehensive loss function with a regularization penalty term. It then executes a truncation operation logic when the absolute value of the initial gradient value exceeds a gradient truncation threshold, outputting a corrected gradient value for weight iteration calculation. This mechanism limits the single update amplitude of weight parameters under conditions of abrupt changes in sensor input data, avoids gradient divergence during gradient descent model iteration, ensures the stability of the preventative monitoring model's weight matrix optimization, and thus improves the accuracy of model inference calculations.

[0020] 2. This invention extracts historical risk data of ancient trees of the same species in the same region through an identification module, runs a Bayesian inference-based error correction algorithm to process the initial risk level and select the maximum posterior probability value, and periodically executes an exponential smoothing time-weighted algorithm to update the prior probability. This method uses regional historical prior data to correct erroneous classification results caused by instantaneous environmental noise or data spikes, and incorporates data offset features caused by seasonal environmental changes into the Bayesian inference model, reducing the frequency of false alarms and narrowing the risk prediction error over long-term operation.

[0021] 3. This invention, through its output module, activates corresponding response branches to perform automated actions such as routine maintenance reminders, modifying model simulation frequencies, issuing terminal pop-up warnings, and activating on-site audible and visual alarms for final risk levels falling into low-risk, medium-risk, high-risk, and extremely high-risk ranges. This hierarchical response control logic can automatically allocate matching hardware and software control actions according to different risk levels, solving the problem that a unified alarm mode cannot match diverse risk characteristics, eliminating the intermediate step of manual judgment and resource allocation, and shortening the system's action command response time. Attached Figure Description

[0022] Figure 1 This is a system block diagram of the present invention; Figure 2 This is a flowchart of the process of the present invention; Figure 3 This is one of the comparative diagrams of the effects of the ancient tree health preventive monitoring system of the present invention, showing the dynamic change curve of risk identification accuracy over 12 months; Figure 4 This is the second comparative chart showing the effects of the ancient tree health preventive monitoring system of the present invention, which is a bar chart of the core indicators of the system's overall false alarm rate and final grading accuracy. Detailed Implementation

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] See attached document Figure 1 , Figure 2 This invention provides a preventive monitoring system for the health of ancient trees, including a data input module, a monitoring module, an identification module, and an output module. The modules work together to form a closed loop of data acquisition, model deduction, risk identification, early warning, and response.

[0025] The data input module establishes a communication connection with the monitoring module. The data input module transmits the normalized feature matrix, generated after normalization processing, as the main output data to the monitoring module. Communication methods: internal bus communication (when modules are deployed on the same physical server or embedded device) or network communication based on the TCP / IP protocol (when modules are distributed).

[0026] The main hardware of the monitoring module is a real-time monitoring unit deployed at the ancient tree site. This unit includes a trunk stress wave sensor, a biological attack monitoring sensor, a millimeter-wave radar, and a chlorophyll fluorescence sensor. It collects physiological data of the ancient tree and data on subtle environmental changes at a set sampling frequency (e.g., 15 minutes / sample). The monitoring module can send commands to the real-time monitoring unit based on system strategies (e.g., identifying medium-risk responses triggered by the identification module) to dynamically adjust the sensor's sampling frequency or operating mode. Wireless communication technologies such as 4G / 5G, LoRa, Wi-Fi, or ZigBee are typically used to adapt to the dispersed deployment environment of the ancient trees in the field. The communication protocol must support low power consumption, long-distance transmission, and breakpoint resume functionality.

[0027] The monitoring module establishes a communication connection with the identification module, employing efficient internal process communication or network communication, such as lightweight data serialization formats like JSON and Protocol Buffers, to ensure transmission efficiency. The monitoring module transmits its core calculation results—the probability of risk outbreak and the intensity of compound risk—to the identification module. Simultaneously, the monitoring module can also forward normalized real-time monitoring vectors. Similar to inter-module communication, the monitoring module's calculation results provide the identification module with key risk characteristic inputs, which the identification module relies on for subsequent risk level determination. This connection serves as a bridge between the two core functions of "model inference" and "risk identification."

[0028] The identification module receives real-time monitoring vectors from the real-time monitoring unit, integrates the risk outbreak probability, composite risk intensity, and real-time monitoring vectors to calculate a comprehensive health and safety score, classifies risk levels, and uses historical risk data of ancient trees of the same species in the same area for misjudgment correction, outputting the final risk level. The identification module establishes a communication connection with the output module. The identification module sends the final risk level, generated after Bayesian misjudgment correction, as a decision instruction to the output module, which serves as the basis for the output module to execute tiered responses. This connection realizes the instruction issuance from "intelligent analysis" to "automated handling." The output module establishes connections with various external execution devices and systems through multiple output interfaces, forming a multi-terminal linkage early warning and handling network. Diverse communication protocols are used depending on the execution terminal. For example, connections to databases use JDBC / ODBC; connections to mobile apps use HTTP / HTTPS push gateways; connections to on-site alarm devices can use RS485 wired control or LoRa wireless control; and connections to higher-level management systems use secure Web Services or API interfaces. The output module transforms abstract risk levels into concrete, executable physical actions and informational instructions, which is a key link in the system's automated closed-loop handling.

[0029] In addition to the data flow mentioned above, an important two-way feedback path has been established within the system to enable continuous optimization of the model.

[0030] Manual verification feedback loop: The management platform's built-in manual verification feedback interface, upon receiving the on-site inspection and risk assessment results, quantifies and maps them into actual observed risk values. This value is then transmitted back to the monitoring module via a reverse communication connection between the management platform and the monitoring module.

[0031] Model update trigger: Upon receiving a real value, the monitoring module triggers a dynamic model update procedure, iteratively optimizing the internal weight matrix using the loss function and gradient descent algorithm. The optimized weight matrix will be used for subsequent inference calculations, thus forming a closed-loop learning process of "monitoring-inference-verification-optimization," enabling the system's prediction accuracy to continuously improve over time.

[0032] In summary, the four core modules of this invention form an organic whole that is both clearly defined in terms of division of labor and closely collaborative through clearly defined interfaces and communication protocols. Through the data input module, monitoring module, identification module, and output module, they jointly execute a closed-loop workflow of data collection, model inference, risk identification, and early warning response, thus ensuring the intelligent, automated, and precise realization of the ancient tree health preventive monitoring task.

[0033] The working principles and processes of each module are explained in detail below: I. Data Input Module: The data input module connects to image acquisition equipment, sensors, lidar scanning equipment, and external databases to obtain basic tree data. External databases include forestry historical archives, a GIS (Geographic Information System), a meteorological platform, and a platform for managing ancient and famous trees. The data input module obtains historical damage and disaster data by accessing the forestry historical archives, acquires basic environmental data by connecting to the GIS and meteorological platforms, and obtains genetic and maintenance data by connecting to the platform for managing ancient and famous trees. The operation flow of the data input module is as follows: S1. Data Acquisition: S101. Obtain basic tree data and construct the tree state vector. Tree State Vector Set as: ;(Formula 1) In the above formula, Defined as trunk tilt, it refers to the angle of deviation of the central axis of the tree trunk from the vertical ground. Defined as the decay rate within the cross-section of the tree trunk, it can be measured using stress wave tomography. Defined as the percentage of the area with exposed roots.

[0034] S102. Obtain historical data on damage and disasters, and construct a historical risk vector: Historical Risk Vector for: ;(Formula 2) Defined as the pest outbreak index over the past five years, it combines the hazard level of invasive species with the frequency of outbreaks; Defined as the natural disaster impact index over the past three years, it integrates the degree of damage caused by windstorms, lightning strikes, etc. Both indices are retrieved from the forestry historical archives database.

[0035] S103. Environmental Basic Data Collection: Connect with the GIS geographic information system and meteorological platform to obtain quantitative values ​​of soil permeability. Topographic slope Meteorological background parameters And construct an environment parameter vector. Environment parameter vector Set as: ;(Formula 3) in , All data are numerical and can be directly retrieved after being measured and stored in a GIS geographic information system and meteorological platform. The discrete feature quantization mapping algorithm is used for transformation, assuming the soil texture set... (such as sandy soil, loam, clay, etc.), through mapping functions Convert it into a breathability quantification value The transformation formula satisfies: ;(Formula 4) Defined for soil types The mapping function; Defined as the first in the set of soil textures Soil type; Defined as the first Standard air permeability coefficient of soil type; Defined as the reference value for the maximum soil permeability coefficient set by the system.

[0036] S104. Data Collection and Maintenance: The data input module connects to the ancient and famous tree management platform to extract textual data on stress resistance levels. , compared with historical maintenance effect feedback values Numerical transformation is performed using fuzzy membership functions, and the following definition is provided: The higher the value, the stronger the self-repair ability at the gene level.

[0037] S2. Normalize the multi-dimensional heterogeneous data collected in step S1: To eliminate the dimensional differences between the data, all raw data are divided into positive indicators (higher values ​​indicate higher risk, such as internal decay rate and tilt) and negative indicators (higher values ​​indicate lower risk, such as stress resistance and maintenance effect). All heterogeneous data collected in the above steps are then uniformly mapped to... The interval eliminates the influence of dimensions, completes data standardization, and the positive index calculation formula satisfies: ;(Formula 5) In the above formula, Defined as the normalized output value of a positive indicator; Defined as the original value of the positive indicator input; Defined as the historical maximum value of the input indicator; Defined as the historical minimum value of the input indicator.

[0038] The negative indicator is calculated using an inverse normalization formula, which satisfies the following formula: ;(Formula 6) In the above formula, Defined as the normalized output value of a negative indicator; Defined as the original value of the negative indicator input; and The meaning is the same as above.

[0039] All normalized output values ​​are combined to construct a normalized feature matrix. The data is then transmitted to the monitoring module.

[0040] II. Monitoring Module: The monitoring module includes an external real-time monitoring unit and an embedded preventative monitoring model. The real-time monitoring unit is deployed on the trunk, root zone, and periphery of the ancient tree's canopy, and includes a trunk stress wave sensor, a biological attack monitoring sensor, a millimeter-wave radar, and a chlorophyll fluorescence sensor. Each sensor collects physiological data and environmental micro-change data of the ancient tree at a sampling frequency of 15 minutes / sample, and then transmits the data to the preventative monitoring model. The preventative monitoring model is a multi-algorithm fusion computing architecture that extrapolates and calculates the normalized feature matrix transmitted from the data input module and the real-time monitoring vectors detected by the real-time monitoring unit, generating a risk outbreak probability and recording it as... Calculate and generate the composite risk intensity and denot it as The probability of risk outbreak and the intensity of the combined risk are then transmitted to the identification module.

[0041] J1. Real-time monitoring vector construction: Real-time monitoring vector Satisfying the formula: ;(Formula 7) In the above formula, Defined as the stress wave propagation time measured by the tree trunk stress wave sensor; Defined as the intensity of the insect vibration signal measured by the biological infestation monitoring sensor; Defined as the micro-displacement of the canopy measured by millimeter-wave radar; Defined as the photosynthetic efficiency index measured by a chlorophyll fluorescence sensor.

[0042] J2. Weight Calculation: The comprehensive weight vector W is calculated using the hierarchical-entropy weight composite algorithm. The specific steps are as follows: J21. Data-Driven Weights Based on Coefficient of Variation ; J211. Calculate the coefficient of variation of the index: Based on the real-time monitoring vector data collected in step J1, calculate the coefficient of variation (CV) of the j-th index. j : (Formula 8); in Standard deviation, The mean is the average value, and the coefficient of variation reflects the relative volatility of the indicator. The larger the value, the more sensitive the indicator is to changes in risk.

[0043] J212. Constructing the Judgment Matrix: Using the coefficient of variation as the quantification of the importance of the indicator, define the elements a of the judgment matrix. ij for: (Formula 9) J213. Calculate the data-driven weights: Normalize each column of the judgment matrix A, then calculate the arithmetic mean of the rows to obtain the data-driven weights. : (Formula 10) m is the total number of indicators (or factors) involved in the weight calculation; J22. Calculate the objective weight vector using the entropy weight method. The formula for calculating the information entropy contained in the objective weight vector satisfies: ;(Formula 11) In the above formula, Defined as the first Information entropy of each indicator; Defined as the number of samples; Defined as sample sequence number; Defined as the first The first indicator The proportion of features in each sample.

[0044] Objective weight ;(Formula 12) The objective weights satisfy: 0 ≤ ≤1, =1; J23. Based on the principle of minimum discriminative information, the data-driven weight vector and the objective weight vector are fused to calculate the comprehensive weight of a single factor. The formula for calculating the comprehensive weight of a single factor satisfies: ;(Formula 13) In the above formula, Defined as the first obtained after fusion The combined weight of each factor; Defined as the fusion coefficient, the system sets the fusion coefficient. The value is 0.6; Defined as data-driven weights based on the coefficient of variation; Defined as the objective weight calculated using the entropy weight method; the monitoring module combines and arranges the comprehensive weights of all factors to form a comprehensive weight vector. The overall weight vector W = [W1, W2, ..., W...] m ] T , where m is a variable representing the number of factors.

[0045] J3. Risk Factor Coupling Analysis: This involves analyzing the meteorological background parameters derived from step S103. Defined as a meteorological factor, the pest outbreak index derived from step S102 is... Defined as a biological factor, the trunk tilt derived from step S101 Corrosion rate within the cross-section of the tree trunk Percentage of exposed root area Defined as tree factors, the aforementioned meteorological factors, biological factors, and tree factors play a direct and dominant role in the health of ancient trees and are all primary factors. The soil type value derived from step S103 is quantified. Topographic slope Defined as environmental factors, environmental factors are secondary factors. Other unmentioned variables, such as the gene data and maintenance data collected in step S104, are also considered secondary factors. Secondary factors play an indirect regulatory or auxiliary role in the health of ancient trees. The coupling strength coefficient between primary and secondary factors is also considered. The following calculation formula is satisfied: ;(Formula 14) In the above formula, Defined as the coupling strength coefficient; W i and W j The first-order factor F i and second-order factor F j The corresponding weights; Defined as the Pearson correlation coefficient between the time series of the primary and secondary factors, this formula is used to characterize the coupling effect of a sharp amplification of risk when "strong winds" and "tree trunk rot" occur simultaneously. It should be noted that the coupling strength is... Defined only between primary and secondary factors. For factor pairs that belong to the same primary or secondary factors, the system sets their coupling strength coefficient. =0, meaning its coupling effect is not calculated.

[0046] J4. Spatiotemporal Evolution Deduction: Time Dimension: A Markov chain model is used to predict the probability of risk outbreaks at future moments. First, a risk state space K = {k1, k2, k3, k4} is defined, corresponding to four levels: low risk, medium risk, high risk, and extremely high risk, respectively. Each state corresponds to a preset risk intensity value V. k By statistically analyzing the frequency of risk state transitions in historical monitoring data, a state transition probability matrix P is constructed. Let the initial state distribution vector at time t=0 be S0, then the state distribution at time t is S... t =S0.P t The formula for calculating the probability of a risk outbreak is: ;(Formula 15) In the above formula, Defined as the future The probability of a risk outbreak at any given moment; Defined as a specific sequence number of the risk status; Defined as At the moment in the first The probability of a certain risk state; Defined as the first The risk intensity value of a certain risk state.

[0047] The above process quantifies the health risk of ancient trees from two dimensions: the time evolution trend (Markov chain) and the current multi-factor spatial coupling (weighted superposition). Together, they constitute the system's risk prediction capability.

[0048] J5. Composite Risk Intensity Calculation: The monitoring module extracts the normalized risk value of each individual factor, based on the comprehensive weight vector W and the coupling strength coefficient. The normalized risk value is then used to generate a composite risk intensity that includes coupling effects using a weighted superposition algorithm. Its calculation formula satisfies: ;(Formula 16) In the above formula, Defined as the total number of risk factors involved in the calculation; Defined as the sequence number of a single risk factor; Defined as the first Normalized risk value of each single factor; Defined as the first The combined weight of each single factor; Defined as not equal to in the system The ordinal numbers of the other single factors; Defined as the first The single factor and the first Coupling strength coefficient between individual factors; Defined as the first The normalized risk value of each individual factor. Formula 17 reflects the situation when multiple risk factors are strongly coupled ( (If the risk is high), the composite risk value output by the system will grow exponentially in a non-linear manner.

[0049] The monitoring module uses a Markov chain model to predict the probability P of future risk outbreaks. risk The current composite risk intensity R is quantified through a weighted superposition coupling model. composite Together, these two elements constitute a two-dimensional quantitative description of the health risks of ancient trees, providing decision-making basis for the identification module from the two dimensions of time and degree.

[0050] As a preferred implementation, step J6 is also included: dynamic optimization of the comprehensive weight fusion coefficient. In Formula 13, which calculates the comprehensive weight of a single factor, if the fusion coefficient λ remains constant, when the sensor data fluctuates drastically due to sudden environmental disturbances (such as strong winds, rainstorms, or animal collisions), the objective weight (entropy weight method) will mistakenly amplify the importance of those instantaneous noise factors, leading to distortion of the comprehensive weight and ultimately causing a large deviation in risk prediction.

[0051] Therefore, λ needs to be dynamically adjusted based on the overall volatility and dispersion of the data. Greater data volatility indicates a more unstable environment and decreased reliability of sensor data. In this case, the proportion of data-driven weights should be reduced (λ decreases), relying more on objective weights. Conversely, when the data is stable, data-driven weights dominate, and the monitoring module performs dynamic optimization calculation of the composite weight fusion coefficients according to the following steps: Step J61: Calculation of fluctuation dispersion: ;(Formula 17) In the above formula, Defined as the volatility dispersion; Defined as the total number of feature dimensions (i.e., the number of risk factors involved in the calculation, such as trunk tilt, internal rot rate, pest vibration signal, etc.). Defined as the ordinal number of the feature dimension; Defined as the number of samples within the sampling time window (e.g., 96 samples were collected in the last 24 hours). Defined as the sample number; Defined as the first The th feature dimension Normalized values ​​of each sample; Defined as the first The sample mean of each feature dimension within the sampling time window; for each feature dimension u, calculate its sample standard deviation, which measures the degree of fluctuation of that dimension within the time window. Sum the standard deviations of all dimensions and divide by the total number of dimensions D to obtain the average standard deviation. The volatility dispersion reflects the overall volatility of all monitored indicators in the entire system over a recent period. A larger value indicates more unstable data from each sensor (e.g., simultaneous large jumps in multiple indicators); a smaller value indicates a stable system.

[0052] J62: Calculation of Composite Weight Fusion Coefficient: The monitoring module compares the fluctuation dispersion with a preset dispersion benchmark threshold and calculates the composite weight fusion coefficient using the fluctuation dispersion. The formula for calculating the composite weight fusion coefficient satisfies: ;(Formula 18) In the above formula, Defined as the composite weight fusion coefficient; Defined as the basic fusion coefficient value, set by the system. The value is 0.6; Defined as a penalty adjustment coefficient, set by the system. The value is 0.15; Defined as volatility dispersion.

[0053] J63: Numerical out-of-bounds restrictions: The monitoring module performs a numerical out-of-bounds limit operation on the composite weight fusion coefficient to ensure that it is within a reasonable range: when If the value is less than 0.3, force the assignment of λ = 0.3. when When the value is greater than 0.8, the value λ is forcibly assigned to 0.8; J64: Weight Update After the monitoring module completes the operation to prevent the composite weight fusion coefficient from exceeding the numerical limit, it will update the value. Enter the comprehensive weight calculation formula 13 and re-execute the calculation.

[0054] Through the above dynamic optimization mechanism, the fluctuation dispersion V dispersion Within the range of drastic fluctuations exceeding 0.4 (i.e., average standard deviation exceeding 0.4), the weighting of objective data is increased, thereby narrowing the range of error in the overall risk rating.

[0055] As a preferred implementation, step J7, dynamic model update, is also included: The management platform has a built-in manual verification and feedback interface. This interface receives the on-site inspection and grading results from forestry experts and quantifies and maps these results to actual observed risk values ​​between 0 and 1. The management platform then sends these actual observed risk values ​​back to the monitoring module. The monitoring module performs a dynamic model update, triggering a weight matrix update command every 24 hours. The monitoring module extracts the generated predicted risk values ​​and the real-time acquired actual risk values ​​to construct a loss function.

[0056] J71. The monitoring module extracts the predicted risk value and the actual observed risk value obtained through the manual verification feedback interface of the management platform. It also extracts the weight parameters for the current iteration period and constructs a comprehensive loss function with a regularization penalty term. ;(Formula 19) In the above formula, Defined as a comprehensive loss function; Defined as the total sample size; Defined as sample sequence number; Defined as the first Predicted risk value for each sample; Defined as the first The actual observed risk value for each sample; β is defined as the regularization coefficient, and the system sets the value of β to 0.01; Defined as the total number of weight parameters; Defined as parameter sequence number; Defined as the first One weight parameter.

[0057] J72. Gradient Calculation: After the monitoring module receives the actual observed risk values ​​from the management platform, it initiates an iterative optimization of the internal weight matrix using the gradient descent algorithm based on the loss function result in Formula 19. The newly optimized weight matrix is ​​then used to replace the initial comprehensive weights for subsequent model derivation. The monitoring module performs partial derivative calculations on the comprehensive loss function to calculate the initial gradient values ​​corresponding to the weight parameters. The formula for calculating the initial gradient value satisfies: ;(Formula 20) In the above formula, Defined as the first The initial gradient values ​​corresponding to each weight parameter; Defined as a comprehensive loss function for the first The partial derivatives of each weight parameter.

[0058] J73. Gradient Truncation: The monitoring module executes an anti-divergence mechanism by extracting the absolute value of the initial gradient and comparing it with the system's preset gradient truncation threshold. If the absolute value of the initial gradient is greater than the system's preset gradient truncation threshold, the monitoring module runs the truncation operation logic and calculates and outputs a corrected gradient value. The calculation formula for the truncation operation logic satisfies: ;(Formula 21) In the above formula, Defined as correcting the gradient value; Defined as the system-preset gradient cutoff threshold, the system sets... The value is 5.0. Based on the statistical distribution of the weight parameters, 99% of the absolute values ​​of the gradient are less than 5.0. If the sensor data noise increases significantly (such as during thunderstorms), it can be temporarily increased to 8.0 to prevent excessive truncation. Defined as the first The initial gradient values ​​corresponding to each weight parameter; Defined as the absolute value of the initial gradient.

[0059] If the absolute value of the initial gradient value is less than or equal to the system's preset gradient cutoff threshold, the monitoring module sets the corrected gradient value to be equal to the initial gradient value.

[0060] J74. Weight Iterative Update: The monitoring module extracts the corrected gradient value from Formula 21. The weights are then input into the weight update program to perform one weight iteration calculation. The calculation formula of the weight update program satisfies: ;(Formula 22) In the above formula, Defined as the 1st iteration after the completion of the iteration One weight parameter; Defined as the first iteration before execution One weight parameter; Defined as the learning rate built into the system.

[0061] The monitoring module utilizes a comprehensive loss function with regularization penalty term combined with truncation operation logic to limit the single update amplitude of weight parameters under the condition of sudden changes in sensor input data, thereby maintaining the iterative convergence success rate of the gradient descent model in a numerical range greater than 99%.

[0062] ;(Formula 23) In the above formula, Defined as the new weight matrix after one iteration; Defined as the historical weight matrix before performing the iterative operation; learning rate When the value is greater than 0.10, the loss function oscillates and does not converge; when it is less than 0.01, the convergence time exceeds 30 days. Therefore, it is recommended to initially set it to 0.03. If oscillation of the loss function is observed, it should be reduced to below 0.02. It is defined as the partial derivative of the loss function with respect to the historical weight matrix.

[0063] To address the technical problem that existing monitoring models often lack clear mathematical support, resulting in low risk identification accuracy and a lack of dynamic correction capabilities, the monitoring module acquires high-frequency data vectors by deploying physical sensors. It then combines preventive monitoring model preprocessing, spatiotemporal evolution extrapolation algorithms, and gradient descent iterative logic to redistribute weights and perform nonlinear weighted summation calculations on the risk data stream. This achieves the technical effect of automatically optimizing prediction parameters as monitoring data accumulates and quantifying the comprehensive composite risk values ​​faced by the tree structure.

[0064] Steps J6 and J7 represent two parameter optimization strategies within the system. These strategies do not conflict with each other and work together. The first level (short-term adaptive, implemented by J6): Based on the fluctuation dispersion of sensor data over the past 24 hours, the composite weight fusion coefficient is dynamically adjusted. This coefficient is used to calculate the comprehensive weight in the current forward inference and does not change the historical stored values ​​of the weight matrix. This level responds to instantaneous data fluctuations and requires no manual feedback.

[0065] The second level (long-term learning, implemented by J7): Every 24 hours, based on the actual observed risk values ​​verified by manual feedback, the weight values ​​of each factor in the weight matrix are permanently updated through the loss function (Equation 19) and gradient descent (Equation 22). This level corrects the systematic bias of the model.

[0066] Collaborative process: Within each simulation cycle (e.g., 15 minutes), the system first calculates the current λ according to J6. F And in conjunction with the current weight matrix (determined by the last update result of J7), calculate the comprehensive weight W used in this simulation. j When J7's timed update is triggered, the system independently performs iterative optimization of the weight matrix. The optimization result overwrites the old weight matrix and is used by J6 in the next cycle. There is no numerical conflict between the two layers because J6 only modifies the fusion coefficients, while J7 modifies the weight matrix itself.

[0067] III. Recognition Module: After the monitoring module outputs the comprehensive and compound risk values ​​faced by the tree, the identification module receives the risk outbreak probability and compound risk intensity transmitted by the monitoring module, and executes the specific process of comprehensive risk identification and adaptive level determination.

[0068] T1. Comprehensive Health and Safety Score Calculation: Risk Outbreak Probability Extracted from Formula 15 by the Identification Module The composite risk intensity obtained from Formula 16 The normalized real-time monitoring vectors are then subjected to weighted fusion calculations to output a comprehensive health and safety score. The calculation formula satisfies: ;(Formula 24) In the above formula, Defined as a comprehensive health and safety score; Defined as a weighting coefficient for the probability of risk occurrence; Defined as the weighting coefficient of the composite risk intensity after normalization; Defined as the composite risk intensity after normalization; Defined as the weight coefficient of the real-time monitoring vector after normalization; Defined as the real-time monitoring vector after normalization.

[0069] T2. Four-level security threshold interval division. The identification module internally stores four-level security threshold intervals, which are divided into low-risk interval, medium-risk interval, high-risk interval and extremely high-risk interval.

[0070] Low-risk zone: The ancient tree is in a healthy and stable state, ∈[0.8,1.0]. Medium risk range: The ancient tree has potential health risks (∈[0.6,0.8)). High-risk area: The health condition of the ancient trees has significantly deteriorated (∈[0.4,0.6)). Extremely high risk zone: ∈[0,0.4), the ancient tree faces an urgent survival crisis; The identification module compares the comprehensive health and safety score with the four-level safety threshold range to determine and output a preliminary risk level. T3. Misjudgment Correction Based on Bayesian Inference: The identification module extracts historical risk data of ancient trees of the same species in the same region and calculates the prior probability and class conditional probability. For continuous comprehensive health and safety scores, a scoring statistical binning interval with a step size of 0.05 is set. The identification module maps and classifies the comprehensive health and safety scores used as observation data into the corresponding scoring statistical binning interval. By statistically analyzing the historical sample frequency of each preliminary risk level falling into this scoring statistical binning interval in the historical risk data of ancient trees of the same species in the same region, the class conditional probability is calculated. The identification module runs a misjudgment correction algorithm based on Bayesian inference to process the preliminary risk level, and the calculation formula satisfies: ;(Formula 25) In the above formula, Defined as posterior probability; Defined as the first A preliminary risk level; Defined as a sequence number for the initial risk level. The set of possible values ​​for is 1, 2, 3, and 4. Defined as a comprehensive health and safety score based on observational data; Defined as in the first Observational data appeared at a preliminary risk level. Class conditional probability; Defined as the first The prior probability of a preliminary risk level; Defined as the accumulating variable index that iterates through the four levels of security threshold intervals; Defined as the first A preliminary risk level; Defined as in the first Observational data appeared at a preliminary risk level. Class conditional probability; Defined as the first The prior probability of a preliminary risk level.

[0071] The identification module calculates and generates four posterior probability values ​​corresponding to the four risk levels, selects the maximum posterior probability value, marks the preliminary risk level corresponding to the maximum posterior probability value as the final risk level, and transmits it to the output module.

[0072] To address the technical issue that the false alarm rate of a single threshold judgment method exceeds 20% due to interference from sensor environmental noise, the identification module executes a comprehensive risk identification and adaptive level judgment process. It adopts a hard interval division combined with a Bayesian inference-based error correction algorithm, and uses regional historical prior data to correct erroneous classification results caused by instantaneous data spikes, thereby achieving a technical effect of improving the system's level judgment accuracy to over 95%.

[0073] IV. Output Module: The output module receives the final risk level transmitted by the identification module. Internally, it has four response branches: low risk, medium risk, high risk, and extremely high risk. Each branch connects to the regular alert interface, the simulation frequency control logic, the dual early warning triggering logic, and the emergency plan reporting logic. The output module activates the corresponding connection branch to perform the output operation based on the final risk level value.

[0074] F1. Low-Risk Response: Activate the low-risk response branch when the final risk level falls into the low-risk range. The output module calls the text generation logic to generate routine maintenance reminders. The output module connects to the ancient tree electronic archive database, combines the routine maintenance reminders with the current timestamp to generate an archive update data package, and sends the archive update data package to the ancient tree electronic archive database to perform a storage append write operation.

[0075] F2. Medium-Risk Response: Activate the medium-risk response branch when the final risk level falls within the medium-risk range. The output module retrieves preventative maintenance recommendations from the built-in maintenance database and sends them to the management platform. The output module sends a frequency modification command to the monitoring module, forcibly changing the simulation frequency parameter of the preventative monitoring model within the monitoring module from once every 24 hours to once every 12 hours.

[0076] F3. High-Risk Response: Activate the high-risk response branch when the final risk level falls into the high-risk range. The output module extracts the abnormal indicator values ​​and location coordinates that triggered the warning to construct the warning execution plan. Establish network communication connections with the mobile app and management platform. The output module sends a terminal push command to the mobile app, which receives the command and displays the warning execution plan. Simultaneously, the output module sends a front-end pop-up command to the management platform. The management platform receives the pop-up command and displays the warning execution plan on the top-level interface of the display.

[0077] F4. Extremely High Risk Response: Activate the extremely high risk response branch when the final risk level falls into the extremely high risk range. The output module connects to the control terminal of the on-site audible and visual alarm device and sends a high-level trigger signal for 5 seconds to activate the sound and light-emitting components. The output module extracts the tree state vector, environmental parameter vector, and comprehensive health and safety score to generate emergency warning information, which is then reported to the forestry authority's system via a network interface.

[0078] To address the technical issue that the unified alarm mode cannot match diverse risk characteristics, resulting in a manual resource allocation response time exceeding 30 minutes, the output module implements a hierarchical response control and automated handling output process. The output module establishes a hard mapping logic from four levels of risk to multiple devices and external systems, achieving the technical effect of automatically allocating matching software and hardware control actions according to different risk levels, and limiting the operation delay of the output module from receiving the final risk level to completing the action command transmission to within 50 milliseconds.

[0079] Specific application examples: The above describes the complete preferred embodiment of the ancient tree health preventive monitoring system and its working mechanism of the present invention. To further verify the effectiveness and technical advantages of the system in different application scenarios, two application examples and experimental data of a comparative example are given below in conjunction with specific application scenarios.

[0080] The experimental subjects and experimental group settings are as follows: This experiment selected 15 ancient trees from the same climate and soil environment area as monitoring subjects, and divided them into one comparative example group and two implementation example groups: Comparative Example 1: Five ancient camphor trees under Level III protection (aged 150-200 years) were selected, and a conventional scheme of "traditional manual inspection and single-point temperature and humidity sensor monitoring" was adopted. The frequency of manual inspection was set to once a month, and the sensor system only triggered an alarm when the temperature and humidity data exceeded the set hard threshold. There was no closed-loop prediction model or multi-dimensional data fusion mechanism.

[0081] Example 1: Five ancient camphor trees from the same region, age range, and protection level as Comparative Example 1 were selected, and the ancient tree health preventive monitoring system provided by this invention was deployed comprehensively. The system's data input module collects data on the tree structure, history, environment, and genes across all dimensions and performs normalization preprocessing; the real-time monitoring unit collects physiological and environmental micro-change data such as stress waves, micro-displacement, and photosynthetic efficiency at a sampling frequency of 15 minutes / time; the preventive monitoring model built into the monitoring module triggers gradient descent iterative updates of weights at a frequency of 24 hours / time; the output module automatically outputs multi-terminal collaborative early warning and response plans based on the comprehensively identified risk classification.

[0082] Example 2: Five Class I protected Masson Pinus trees (age ≥ 500 years) were selected and the ancient tree health preventive monitoring system of this invention was deployed. Considering the physiological characteristics of Masson Pinus trees, which are highly susceptible to pine wilt nematodes and termites, the management personnel adjusted the initial weights in the preventive monitoring model through the system interface, increasing the calculation weight of the biological attack risk factor by 20% to adapt to the susceptibility of Masson Pinus trees to biological attacks. All other system parameters remained at their default settings.

[0083] Experimental process and effect comparison data: The experimental period was set to run continuously for 12 months, during which the region experienced complex external environmental changes such as the peak season for pests and diseases in spring and typhoons and rainstorms in summer. The management platform synchronously recorded the number of alarms, verified false alarms / missed alarms, risk identification accuracy, and system automation response time for the three schemes.

[0084] See attached document Figure 3 , attached Figure 3 The system performance comparison results of the embodiments of the present invention and the comparative examples over a 12-month operating period are presented. (Appendix) Figure 3 The curve showing the dynamic change in risk identification accuracy over 12 months is attached. Figure 4 This is a bar chart showing the core indicators of the system's overall false alarm rate and final classification accuracy. (Combined with the appendix...) Figure 3 and attached Figure 4 A comparison of monitoring and verification results shows that: Comparative Example 1 running data: Refer to Appendix Figure 3 The accuracy of the circle-marked curve, lacking a comprehensive evaluation and dynamic correction mechanism, shows significant fluctuations and declines in the accuracy of Comparative Example 1 during the typhoon-prone summer and the spring when pests and diseases are prevalent. Single-point sensors are severely affected by environmental noise, generating invalid alarms. (See attached diagram.) Figure 4 After comprehensive calculation, the risk identification accuracy rate of Comparative Example 1 was only 65%, and due to the reliance on manual review, the average emergency response time exceeded 24 hours.

[0085] Example 1 Operational Data: Refer to Appendix Figure 3 The dashed square markers, thanks to the fusion of multi-source heterogeneous data and Bayesian inference algorithms, successfully filtered out transient data interference. Over a 12-month period, as the monitoring module continuously optimized the weight matrix based on the loss function using gradient descent algorithms, the system's risk identification accuracy climbed from 88% in the cold start phase to a stable level of around 96.5%, exhibiting self-learning and error convergence characteristics. (See attached diagram) Figure 4 Its false alarm rate has been reduced to 4%, and the automated response time for high-risk events has been shortened to within 5 minutes.

[0086] Example 2 Operational Data: Refer to Appendix Figure 3The triangular markers and dashed lines, along with the customized weighting scheme for specific tree species, demonstrate exceptionally high early warning capabilities. In the early stages of termite activity in late spring, the system detected minute anomalies and issued a medium-risk warning seven days in advance, successfully containing the disease at its nascent stage. Example 2, under year-round high-frequency monitoring, achieved a risk classification accuracy of 97.1% (with a false alarm rate of only 2.5%), with the curve consistently at its highest and most stable position, proving the system's adaptive adjustment capabilities specific to tree species.

[0087] Experimental results show that Examples 1 and 2 outperform Comparative Example 1 in terms of risk identification accuracy, resistance to environmental interference, and emergency response speed. This invention systematically solves the technical bottlenecks of traditional monitoring methods, such as one-way communication without feedback, limited monitoring dimensions, and high false alarm rates, verifying the predictive accuracy and practical engineering application value of this invention.

Claims

1. A preventative health monitoring system for ancient trees, characterized in that, include: The data input module is used to collect multi-dimensional heterogeneous data and perform normalization processing to generate a normalized feature matrix; The monitoring module is communicatively connected to the data input module. The monitoring module includes: a real-time monitoring unit deployed at the ancient tree site, used to collect real-time monitoring vectors at a set sampling frequency; and a preventive monitoring model, used to receive the normalized feature matrix and the real-time monitoring vectors, calculate the comprehensive weight vector through a hierarchical-entropy weight composite algorithm, and calculate the probability of risk outbreak and the intensity of composite risk based on a spatiotemporal evolution inference algorithm. The identification module, which is communicatively connected to the monitoring module, is used to receive the risk outbreak probability, the composite risk intensity, and the normalized real-time monitoring vector, perform data fusion calculation, output a comprehensive health and safety score, compare the comprehensive health and safety score with a preset four-level safety threshold range, calculate and output a preliminary risk level, extract historical risk data of ancient trees of the same tree species in the same area to construct a priori knowledge input set, run a misjudgment correction algorithm to process the preliminary risk level, and output the final risk level. The output module, which is communicatively connected to the identification module, is used to receive the final risk level and automatically trigger response instructions at different levels to perform automated handling operations for the final risk level with different range thresholds.

2. The ancient tree health preventive monitoring system according to claim 1, characterized in that, The formula used by the data input module to perform positive index standardization is: ; Defined as the normalized output value of a positive indicator; Defined as the original value of the positive indicator input; Defined as the historical maximum value of the input indicator; Defined as the historical statistical minimum value of the input indicator; The formula for the inverse normalization of negative indicators is: ; Defined as the normalized output value of a negative indicator; Defined as the original value of the negative indicator input.

3. The ancient tree health preventive monitoring system according to claim 1, characterized in that, The real-time monitoring unit includes a trunk stress wave sensor, a biological attack monitoring sensor, a millimeter-wave radar, and a chlorophyll fluorescence sensor, which are used to collect stress wave transmission time, pest vibration signal intensity, canopy micro-displacement, and photosynthetic efficiency index, respectively, and to construct a real-time monitoring vector. , Defined as the stress wave propagation time measured by the tree trunk stress wave sensor; Defined as the intensity of the insect vibration signal measured by the biological infestation monitoring sensor; Defined as the micro-displacement of the canopy measured by millimeter-wave radar; Defined as the photosynthetic efficiency index measured by a chlorophyll fluorescence sensor.

4. The ancient tree health preventive monitoring system according to claim 1, characterized in that, The monitoring module calculates the comprehensive weight vector using a hierarchical-entropy weighted composite algorithm. : ； Defined as the first obtained after fusion The combined weight of each factor; Defined as the fusion coefficient; Defined as data-driven weights; Defined as the objective weight obtained by the entropy weight method; The monitoring module combines and arranges the comprehensive weights of all factors to form a comprehensive weight vector. The overall weight vector W = [W1, W2, ..., W...] m ] T , where m is a variable representing the number of factors.

5. The ancient tree health preventive monitoring system according to claim 4, characterized in that, The monitoring module dynamically adjusts the fusion coefficient λ based on the data fluctuation dispersion and enforces numerical out-of-bounds limits, keeping it within the range of [0.3, 0.8]. , Defined as the composite weight fusion coefficient; Defined as the basic fusion coefficient value; Defined as the penalty adjustment coefficient; Defined as volatility dispersion.

6. The ancient tree health preventive monitoring system according to claim 1, characterized in that, The monitoring module defines meteorological factors, biological factors, and tree factors as primary factors, and soil factors, topographic factors, and genetic and maintenance factors as secondary factors, and calculates the coupling strength coefficient between primary and secondary factors. The following calculation formula is satisfied: ； W i and W j The first-order factor F i and second-order factor F j The corresponding weights; Defined as the Pearson correlation coefficient between the time series of the primary factor and the secondary factor.

7. The ancient tree health preventive monitoring system according to claim 1, characterized in that, The comprehensive loss function constructed by the monitoring module is: ； In the above formula, Defined as a comprehensive loss function; Defined as the total sample size; Defined as sample sequence number; Defined as the first Predicted risk value for each sample; Defined as the first The actual observed risk value for each sample; Defined as the regularization coefficient; Defined as the total number of weight parameters; Defined as parameter sequence number; Defined as the first Each weight parameter.

8. The ancient tree health preventive monitoring system according to claim 1, characterized in that, The identification module calculates a comprehensive health and safety score. The risk level is determined by comparing the data with the four-level safety threshold ranges [0.8, 1.0], [0.6, 0.8], [0.4, 0.6], and [0, 0.4]. The calculation formula satisfies: ； In the above formula, Defined as a comprehensive health and safety score; Defined as a weighting coefficient for the probability of risk occurrence; Defined as the probability of a risk occurring; Defined as the weighting coefficient of the composite risk intensity after normalization; Defined as the composite risk intensity after normalization; Defined as the weight coefficient of the real-time monitoring vector after normalization; Defined as the real-time monitoring vector after normalization.

9. The ancient tree health preventive monitoring system according to claim 1, characterized in that, The Bayesian inference misjudgment correction formula executed by the recognition module is as follows: In the above formula, Defined as posterior probability; Defined as the first A preliminary risk level; Defined as a sequence number for the initial risk level. The possible values ​​for are 1, 2, 3, and 4; Defined as a comprehensive health and safety score based on observational data; Defined as in the first Observational data appeared at a preliminary risk level. Class conditional probability; Defined as the first The prior probability of a preliminary risk level; Defined as the accumulating variable index that iterates through the four levels of security threshold intervals; Defined as the first A preliminary risk level; Defined as in the first Observational data appeared at a preliminary risk level. Class conditional probability; Defined as the first The prior probability of a preliminary risk level.

10. The ancient tree health preventive monitoring system according to claim 1, characterized in that, The automated processing operations of the output module include: Low-risk zone: Generate routine maintenance reminders and store them in the ancient tree electronic archive database; Medium-risk zone: Send preventative maintenance recommendations to the management platform and send a frequency modification instruction to the monitoring module, changing the simulation frequency from 24 hours / time to 12 hours / time; High-risk zone: Send early warning execution plans to mobile terminals and management platforms; Extremely high risk zone: Trigger on-site audible and visual alarm devices and report emergency warning information to the forestry authority system.

Images