A forest pest control method based on big data mining

Abstract

The application discloses a forestry disease and pest control method based on big data mining, which comprises the following steps: collecting and preprocessing multi-source heterogeneous data of a target forest area; extracting multi-source features and performing fusion processing according to the correlation of ecological factors; performing spatio-temporal correlation analysis and constraint fusion processing based on the multi-source fusion feature set; performing spatio-temporal prediction modeling and result analysis on the comprehensive representation sequence of diseases and pests based on an improved AGCRN model, and generating disease and pest prediction results; performing risk zoning and control decision determination processing on different monitoring areas in the target forest area; constructing a control decision mapping relationship and generating a control strategy set; constructing a strategy evaluation index and performing comprehensive evaluation and sorting optimization to determine the target control strategy. The application adopts a multi-source data fusion and spatio-temporal correlation modeling method, realizes accurate prediction of diseases and pests and optimization of control strategies, and has the advantages of high prediction accuracy, strong decision scientificity and high resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of forestry disaster prevention and control, and in particular to a method for the prevention and control of forestry pests and diseases based on big data mining. Background Technology

[0002] With the continuous expansion of forestry resources, the occurrence and spread of pests and diseases are showing a trend of complex spatial and temporal distribution and diversified influencing factors. Existing technologies mostly rely on single data sources or simple statistical analysis methods for monitoring and predicting pests and diseases, typically making empirical judgments based on meteorological data or local pest data, lacking the ability to fuse and process multi-source heterogeneous data. Furthermore, while some methods incorporate remote sensing or IoT data, the data processing often remains at the feature splicing or shallow analysis stage, failing to fully explore the coupling relationships between different ecological factors. This makes it difficult to accurately characterize the intrinsic correlation between the occurrence, spread, and damage to trees caused by pests and diseases, resulting in insufficient accuracy and stability of prediction results.

[0003] In terms of prevention and control decision-making, existing technologies mostly adopt fixed rules or simple threshold division methods for risk zoning and prevention and control strategy formulation, lacking comprehensive consideration of spatiotemporal propagation relationships, resource constraints, and ecological protection requirements, making it difficult to achieve refined and dynamic prevention and control scheduling. Furthermore, existing solutions rely heavily on single evaluation indicators or empirical weights in the evaluation and selection of prevention and control strategies, failing to establish a multi-dimensional evaluation indicator system. This results in shortcomings in resource utilization efficiency, operational timeliness, and ecological adaptability of the selected strategies, making it difficult to meet the actual needs of high-precision pest and disease control in complex forestry scenarios.

[0004] Therefore, how to provide a forestry pest and disease control method based on big data mining is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a forestry pest and disease control method based on big data mining. This invention adopts a multi-source data fusion and spatiotemporal correlation modeling method to achieve accurate prediction of pests and diseases and optimization of control strategies. It has the advantages of high prediction accuracy, strong scientific decision-making and high resource utilization efficiency.

[0006] A forestry pest and disease control method based on big data mining according to an embodiment of the present invention includes the following steps:

[0007] Collect multi-source heterogeneous data from the target forest area during the target monitoring period and preprocess them to generate a standardized forestry monitoring dataset;

[0008] Multi-source features are extracted from standardized forestry monitoring datasets and fused according to the correlation of ecological factors to generate a multi-source fused feature set.

[0009] Spatiotemporal correlation analysis and constraint fusion processing are performed based on multi-source fusion feature sets to generate integrated pest and disease characterization sequences.

[0010] Based on the improved AGCRN model, spatiotemporal prediction modeling and result analysis of the integrated pest and disease characterization sequence are performed to generate pest and disease prediction results for each monitoring area.

[0011] Based on the pest and disease prediction results, risk zoning and prevention and control decision-making are carried out for different monitoring areas in the target forest area to generate zoning prevention and control needs results.

[0012] Based on the results of regional prevention and control needs and preset prevention and control constraints, a prevention and control decision mapping relationship is constructed and a set of prevention and control strategies is generated.

[0013] Based on the set of prevention and control strategies, strategy evaluation indicators are constructed, and comprehensive evaluation and ranking are carried out to determine the target prevention and control strategies.

[0014] Optionally, the multi-source heterogeneous data includes meteorological environmental data, insect monitoring data, remote sensing image data, and forest growth status data. The preprocessing includes timestamp unification, spatial coordinate unification, outlier removal, missing value completion, noise suppression, dimension normalization, and association and alignment according to monitoring area identifier, monitoring time period identifier, and forest object identifier.

[0015] Optionally, the generation of the multi-source fusion feature set specifically includes:

[0016] Based on meteorological and environmental data in the standardized forestry monitoring dataset, meteorological driving factor analysis and coupled modeling were performed to obtain a set of meteorological driving features.

[0017] Based on the insect monitoring data in the standardized forestry monitoring dataset, insect activity behavior characterization and population dynamic analysis were performed to obtain a set of insect activity characteristics.

[0018] Spatial representation extraction and anomaly response quantification were performed on remote sensing image data in the standardized forestry monitoring dataset to obtain a set of remote sensing representation features;

[0019] By combining the tree growth status data in the standardized forestry monitoring dataset, a set of tree health features is obtained through health status analysis and damage characterization.

[0020] Based on the meteorological driving feature set, insect activity feature set, remote sensing characterization feature set and forest health feature set, cross-source association matching and spatiotemporal registration processing are performed to generate a registration feature group under a unified spatiotemporal benchmark.

[0021] Semantic association encoding and fusion processing are performed on the registered feature groups according to the correlation of ecological factors to generate a multi-source fusion feature set.

[0022] Optionally, the generation of the integrated pest and disease characterization sequence specifically includes:

[0023] Based on the multi-source fusion feature set, regional aggregation and temporal unfolding are performed according to the monitoring area identifier, monitoring time period identifier, and forest object identifier to construct the regional temporal feature group corresponding to each monitoring area;

[0024] The environmental impact correlations were determined based on the regional temporal characteristic groups and the spatial relationships between each monitoring area.

[0025] Based on the inter-regional change trajectories of regional temporal characteristic groups and insect activity characteristics, the association between insect pest transmission was determined;

[0026] The host distribution association was determined based on the forest health characteristics in the regional time series feature group and the forest type distribution information obtained from the analysis of remote sensing image data.

[0027] Spatiotemporal correlation modeling was performed based on environmental impact correlations, pest transmission correlations, and host distribution correlations to construct a multi-relationship spatiotemporal correlation structure among monitoring areas;

[0028] Based on the multi-relation spatiotemporal association structure, the multi-source fusion features of the current monitoring period in each monitoring area are subjected to association constraint fusion processing with the multi-source fusion features of the adjacent monitoring periods in the associated monitoring areas to generate a comprehensive pest and disease characterization sequence.

[0029] Optionally, the generation of the pest and disease prediction results specifically includes:

[0030] The integrated pest and disease representation sequence is input into the improved AGCRN model, and node representation encoding is performed in the node embedding generation structure to generate node embedding representation results for each monitoring area.

[0031] The node embedding representation results are processed by a dual-graph generation structure to generate an environment association graph and a propagation association graph;

[0032] Based on the node association relationships represented by the environmental association graph and the propagation association graph, cross-regional feature propagation and weighted aggregation processing are performed on the node embedding representation results through a node adaptive graph convolution structure to generate graph association feature representation results corresponding to each monitoring area;

[0033] Based on the graph association feature representation results, the temporal state update processing is performed through a cyclic temporal modeling structure, and the temporal information is dynamically adjusted by combining a time delay-aware gating structure to generate spatiotemporal prediction characterization sequences of pests and diseases corresponding to each monitoring area.

[0034] Based on the spatiotemporal prediction characterization sequence of pests and diseases, pest and disease category discrimination processing and pest and disease occurrence risk analysis processing are performed to generate pest and disease category identifiers, occurrence probability values ​​and risk level identifiers.

[0035] Based on the spatiotemporal prediction characterization sequence of pests and diseases, diffusion path analysis and trend evolution analysis are performed to generate diffusion direction identifiers, diffusion range identifiers, and diffusion intensity identifiers.

[0036] The pest and disease category identifiers, occurrence probability values, risk level identifiers, diffusion direction identifiers, diffusion range identifiers, and diffusion intensity identifiers corresponding to each monitoring area are collected and processed to generate pest and disease prediction results for each monitoring area.

[0037] Optionally, the generation of the zoning prevention and control requirement results specifically includes:

[0038] Based on the occurrence probability value, the intensity value corresponding to the diffusion intensity indicator, the range value corresponding to the diffusion range indicator, and the risk level indicator in the pest and disease prediction results, risk quantification is performed on each monitoring area to determine the comprehensive regional risk value corresponding to each monitoring area.

[0039] Adjacent monitoring areas with the same pest and disease category identification and a regional risk comprehensive value difference lower than the preset zoning threshold are aggregated and divided. Based on the aggregated monitoring area set, the risk zoning identifier is determined, and the pest and disease risk zoning result is generated.

[0040] The risk zone, the comprehensive risk value of the region, and the intensity value corresponding to the diffusion intensity indicator of each monitoring area in the results of the pest and disease risk zoning are weighted and calculated to generate a priority judgment value. The prevention and control priority indicator is determined according to the result of the priority judgment value falling into the preset priority range.

[0041] Based on the occurrence probability value, diffusion intensity change trend, diffusion range expansion trend and control priority indicator in the pest and disease prediction results, a weighted calculation is performed on each monitoring area to generate a time window judgment value, and the control time window indicator is determined according to the result of the time window judgment value falling into the preset time window interval.

[0042] The comprehensive regional risk value, the priority level value corresponding to the prevention and control priority indicator, and the time urgency value corresponding to the prevention and control time window indicator are weighted and integrated to generate a prevention and control intensity judgment value. The prevention and control intensity level indicator is determined based on the result of the prevention and control intensity judgment value falling into the preset intensity level range.

[0043] The pest and disease category identifiers, prevention and control priority identifiers, prevention and control time window identifiers, and prevention and control intensity level identifiers corresponding to each monitoring area are collected and processed to generate the regional prevention and control requirements results.

[0044] Optionally, the generation of the prevention and control strategy set specifically includes:

[0045] Based on the pest and disease category identifiers, prevention and control priority identifiers, prevention and control time window identifiers, and prevention and control intensity level identifiers in the results of regional prevention and control needs, and combined with preset prevention and control constraints, the monitoring areas are correlated and matched to construct a prevention and control decision mapping relationship. The prevention and control constraints include forest area information, forest type information, prevention and control resource information, pesticide use restriction information, ecological protection constraint information, and operation period constraint information.

[0046] Based on the prevention and control decision mapping relationship, forest objects, operation areas and prevention and control units that are compatible with the current pest and disease categories and meet the requirements of ecological protection are screened in each monitoring area to determine the target prevention and control object identification;

[0047] Based on the prevention and control decision mapping relationship, the time interval corresponding to the prevention and control time window identifier is matched and filtered with the allowable work interval corresponding to the work period constraint information to determine the target work period identifier for each monitoring area.

[0048] Based on the prevention and control decision mapping relationship, the target prevention and control object identifier, target operation period identifier, prevention and control priority identifier, prevention and control intensity level identifier and prevention and control resource information are associated and allocated to determine the corresponding pesticide demand, equipment demand and personnel demand for each monitoring area, and the resource allocation result is formed by combining the available resources.

[0049] Based on the prevention and control decision mapping relationship, the operational coverage area corresponding to the target prevention and control object identification, resource allocation results, pesticide use restriction information, ecological protection constraint information, and forest area information is constrained, screened, and matched to determine the measure judgment value and prevention and control measure identification for each monitoring area.

[0050] Based on the prevention and control decision mapping relationship, the measure judgment value, prevention and control measure identifier, target operation period identifier and resource allocation result are combined and configured to determine the corresponding pesticide application parameters, equipment scheduling parameters and operation time parameters for each monitoring area, and are collected with the prevention and control measure identifier to generate a set of prevention and control strategies.

[0051] Optionally, the determination of the target prevention and control strategy specifically includes:

[0052] Based on the control measures identifiers, pesticide application parameters, equipment scheduling parameters, and operation time parameters in the control strategy set, and combined with the pest and disease prediction results and regional control needs results for each monitoring area, a set of strategy evaluation indicators corresponding to each candidate control strategy is extracted.

[0053] Based on the strategy evaluation index set, the matching between each candidate prevention and control strategy and the current prevention and control needs of the monitoring area is weighted and calculated to determine the appropriate evaluation value for each candidate prevention and control strategy.

[0054] Based on the strategy evaluation index set, resource requirements are calculated for the pesticide delivery parameters, equipment scheduling parameters and operation time parameters of each candidate prevention and control strategy. The pesticide consumption, equipment occupancy and personnel input are determined and compared with the current available resources to obtain the resource occupancy ratio. The resource occupancy ratios are weighted and summed according to the preset resource weights to generate resource evaluation values.

[0055] Based on the strategy evaluation index set, the timeliness matching between the operation time parameters and the allowable operation range and the ecological constraint satisfaction of each candidate prevention and control strategy are jointly calculated to determine the timeliness evaluation value and ecological evaluation value of each candidate prevention and control strategy.

[0056] The adaptation evaluation value, timeliness evaluation value, and ecological evaluation value are used as positive evaluation components, and the resource evaluation value is used as negative evaluation component. After multiplying each of them by the corresponding evaluation weight, the weighted sum is obtained to obtain the comprehensive evaluation value of each candidate prevention and control strategy.

[0057] The candidate prevention and control strategies are ranked from highest to lowest according to their comprehensive evaluation values. The candidate prevention and control strategy with the best comprehensive evaluation value is selected as the target prevention and control strategy. The target prevention and control strategy includes the target prevention and control area identifier, the target pest and disease object identifier, the prevention and control measure identifier, the pesticide application parameters, the equipment scheduling parameters, the operation time parameters, and the operation path parameters.

[0058] The beneficial effects of this invention are:

[0059] This invention constructs a unified processing and fusion analysis framework based on multi-source heterogeneous data, enabling the collaborative utilization of multi-dimensional data such as meteorological environment, insect pest monitoring, remote sensing imagery, and forest growth status. It establishes correlations between multi-source features under the same spatiotemporal benchmark, providing a more comprehensive characterization of the coupling mechanism between pest occurrence, spread, and forest damage. Compared to existing methods relying on single data sources or shallow feature fusion, this application effectively enhances the comprehensive characterization capability of pests and diseases through semantic fusion processing driven by ecological factor correlations and a spatiotemporal correlation modeling mechanism. This significantly improves the accuracy and stability of subsequent prediction results, providing more reliable data support for the early identification and dynamic monitoring of forest pests and diseases.

[0060] At the prediction and analysis level, this application introduces an improved AGCRN model. While retaining the original advantages of spatiotemporal modeling, it constructs a dual-graph generation structure and a time-delay-aware gating mechanism. This enables the model to simultaneously characterize environmental influence relationships, pest transmission relationships, and host distribution relationships. Furthermore, by incorporating transmission lag characteristics, it dynamically adjusts the temporal state, thereby significantly enhancing its ability to characterize pest and disease spread paths, evolutionary trends, and risk levels. The pest and disease prediction results generated by this improved model not only include category and probability information but also output key indicators such as spread direction, range, and intensity, making the prediction results more comprehensive and providing multi-dimensional support for subsequent prevention and control decisions.

[0061] In terms of prevention and control decision-making and strategy optimization, this application constructs a risk zoning and zoning prevention and control demand system based on prediction results, and establishes a prevention and control decision mapping relationship by combining prevention and control resources, ecological protection, and operational constraints, thereby achieving coordinated determination of prevention and control targets, operational timing, and resource allocation. Simultaneously, by constructing a multi-dimensional strategy evaluation index system, candidate prevention and control strategies are comprehensively evaluated and ranked from multiple dimensions such as adaptability, resource consumption, timeliness, and ecological constraint satisfaction, effectively improving the scientific nature and feasibility of prevention and control strategies. Compared with existing strategy selection methods that rely on experience or single indicators, this application can improve resource utilization efficiency and prevention and control effectiveness while ensuring ecological security, achieving refined, dynamic, and intelligent forestry pest and disease control. Attached Figure Description

[0062] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0063] Figure 1 This is a flowchart of a forestry pest and disease control method based on big data mining proposed in this invention;

[0064] Figure 2 This is a flowchart illustrating the generation of pest and disease prediction results for a forestry pest and disease control method based on big data mining proposed in this invention.

[0065] Figure 3 This is a flowchart illustrating the generation of zonal control requirements results for a forestry pest and disease control method based on big data mining proposed in this invention. Detailed Implementation

[0066] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0067] refer to Figures 1-3A method for controlling forest pests and diseases based on big data mining includes the following steps:

[0068] Collect multi-source heterogeneous data from the target forest area during the target monitoring period and preprocess them to generate a standardized forestry monitoring dataset;

[0069] Multi-source features are extracted from standardized forestry monitoring datasets and fused according to the correlation of ecological factors to generate a multi-source fused feature set.

[0070] Spatiotemporal correlation analysis and constraint fusion processing are performed based on multi-source fusion feature sets to generate integrated pest and disease characterization sequences.

[0071] Based on the improved AGCRN model, spatiotemporal prediction modeling and result analysis of the integrated pest and disease characterization sequence are performed to generate pest and disease prediction results for each monitoring area.

[0072] Based on the pest and disease prediction results, risk zoning and prevention and control decision-making are carried out for different monitoring areas in the target forest area to generate zoning prevention and control needs results.

[0073] Based on the results of regional prevention and control needs and preset prevention and control constraints, a prevention and control decision mapping relationship is constructed and a set of prevention and control strategies is generated.

[0074] Based on the set of prevention and control strategies, strategy evaluation indicators are constructed, and comprehensive evaluation and ranking are carried out to determine the target prevention and control strategies.

[0075] In this embodiment, the multi-source heterogeneous data includes meteorological environment data, insect monitoring data, remote sensing image data, and forest growth status data. The preprocessing includes unifying timestamps, unifying spatial coordinates, removing outliers, completing missing values, suppressing noise, normalizing dimensions, and associating and aligning data according to monitoring area identifiers, monitoring time period identifiers, and forest object identifiers.

[0076] In this embodiment, the generation of the multi-source fusion feature set specifically includes:

[0077] Meteorological driving factor analysis and coupled modeling were performed on meteorological environmental data in the standardized forestry monitoring dataset to obtain a set of meteorological driving features. The meteorological environmental data included temperature, rainfall, air humidity, wind speed, wind direction, sunshine duration, and soil moisture content. The meteorological driving features included temperature-humidity coupled driving strength, rainfall-induced activation strength, wind migration driving strength, and comprehensive meteorological suitability strength. Among them, the temperature-humidity coupled driving strength was determined by multiplying the air humidity value and the temperature value and combining the deviation of the historical average value for the same period; the rainfall-induced activation strength was determined by weighted summation of the current rainfall value and the cumulative rainfall values ​​of multiple previous monitoring periods; the wind migration driving strength was jointly determined by the consistency of the wind speed value and the wind direction with the historical spread direction of the target pests and diseases; and the comprehensive meteorological suitability strength was obtained by multiplying the temperature-humidity coupled driving strength, rainfall-induced activation strength, and wind migration driving strength by the corresponding ecological sensitivity weights and summing them.

[0078] Based on the insect monitoring data in the standardized forestry monitoring dataset, insect activity behavior characterization and population dynamic analysis were performed to obtain a set of insect activity characteristics. The insect monitoring data includes the number of insects trapped, insect density, adult proportion, larval proportion, insect age structure, insect activity frequency, and diurnal variation data. The insect activity characteristics include insect activity intensity, insect age diffusion potential, population growth rate, and intensity of pre-spreading signs. Among them, insect activity intensity is determined by normalizing the product of the number of insects trapped and the insect activity frequency, combined with the forest coverage area; insect age diffusion potential is determined by a weighted combination of the adult proportion, larval proportion, and high migration risk insect age proportion; population growth rate is determined by normalizing the difference between the insect density in the current monitoring period and the previous monitoring period, combined with the insect density in the previous monitoring period; and intensity of pre-spreading signs is determined by the degree of matching between insect activity intensity, insect age diffusion potential, and population growth rate and the threshold range of historical outbreak samples.

[0079] Spatial representation extraction and anomaly response quantification were performed on remote sensing image data in the standardized forestry monitoring dataset to obtain a set of remote sensing representation features. The remote sensing image data included multispectral images, thermal infrared images, and texture images. Spatial cropping, regional grid division, and pixel aggregation were performed. The remote sensing representation features included canopy anomaly index, thermal stress intensity, patch expansion degree, and texture fragmentation degree. The canopy anomaly index was determined by normalizing the difference between the near-infrared band reflectance value and the red band reflectance value, combined with the sum of the two. The thermal stress intensity was determined by the difference between the average canopy temperature of the target monitoring area and the average canopy temperature of the healthy control area. The patch expansion degree was determined by the ratio of the number of anomalous pixels to the total number of pixels, combined with the growth rate of the connected region area. The texture fragmentation degree was determined by jointly using the contrast value, entropy value, and homogeneity value in the texture statistical features.

[0080] By combining the forest growth status data in the standardized forestry monitoring dataset, a set of forest health characteristics was obtained through health status analysis and damage characterization. The forest growth status data includes tree height, diameter at breast height (DBH), crown width, leaf area index (LAI), leaf chlorosis degree, branch dieback ratio, and growth rate. The forest health characteristics include growth decline intensity, crown damage intensity, water stress degree, and comprehensive health decline degree. Among them, growth decline intensity is determined by the difference between the current growth rate and the historical average normal growth rate, and normalized by combining the historical average normal growth rate. Crown damage intensity is determined by the weighted sum of leaf chlorosis degree, branch dieback ratio, and crown shrinkage ratio. Water stress degree is determined by the degree of coupling deviation between the leaf area index change and the soil moisture content change. The comprehensive health decline degree is obtained by multiplying the growth decline intensity, crown damage intensity, and water stress degree by their respective health evaluation weights and then summing them.

[0081] Based on meteorological-driven feature sets, insect activity feature sets, remote sensing characterization feature sets, and forest health feature sets, cross-source association matching and spatiotemporal registration processing are performed to generate a registration feature group under a unified spatiotemporal reference. Cross-source feature matching establishes a correspondence according to the monitoring area identifier, monitoring time period identifier, and forest object identifier, and jointly determines the association matching strength based on the absolute value of the feature numerical difference and the consistency of the change direction. Spatiotemporal registration determines the spatiotemporal registration coefficient according to the spatial distance of the monitoring area, the time interval of the monitoring period, and the consistency of the insect pest transmission trajectory, and performs unified spatiotemporal reference correction on various features based on the spatiotemporal registration coefficient.

[0082] Semantic association coding and fusion processing are performed on the registered feature groups according to the ecological factor correlation to generate a multi-source fusion feature set. The ecological factor correlation refers to the multidimensional coupling relationship formed between meteorological environmental conditions, insect activity status, forest health status and remote sensing representation response in the same monitoring area and the same monitoring period, around the occurrence, spread and damage process of pests and diseases, including occurrence-driven correlation, propagation-driven correlation and damage response correlation. The semantic association coding divides the registered feature groups into occurrence-driven semantic components, propagation-driven semantic components and damage response semantic components according to the ecological factor correlation, and generates semantic association coding values ​​based on the sum of the products of each semantic component and the corresponding semantic contribution weight.

[0083] In this embodiment, the generation of the integrated pest and disease characterization sequence specifically includes:

[0084] Based on the multi-source fusion feature set, regional aggregation and temporal unfolding are performed according to the monitoring area identifier, monitoring time period identifier, and forest object identifier to construct the regional temporal feature group corresponding to each monitoring area;

[0085] The environmental impact correlation is determined based on the regional temporal characteristic groups and the spatial location relationship between each monitoring area. The environmental impact correlation analysis includes extracting the temperature difference, air humidity difference, rainfall difference, wind speed difference, and wind direction consistency between any two monitoring areas, and combining them with the spatial distance between the monitoring areas and the degree of synchronization of environmental changes for joint calculation. The degree of synchronization of environmental changes is determined by the ratio of the number of times the meteorological driving characteristic change direction of the two monitoring areas is consistent in multiple consecutive monitoring periods to the total number of comparisons. The environmental impact correlation strength is obtained by multiplying the degree of wind direction consistency, the degree of synchronization of environmental changes, and the reciprocal of the spatial distance.

[0086] The correlation of pest transmission is determined based on the inter-regional change trajectory of regional temporal characteristic groups and pest activity characteristics. The determination of the correlation of pest transmission includes extracting the difference in pest population density, the difference in the intensity of pre-transmission indicators, the difference in the diffusion potential of insect age, and the response interval between monitoring periods between any two monitoring areas. It is also combined with the prevailing wind direction, regional connectivity conditions, and the degree of overlap of historical migration paths for joint calculation. Among them, the response interval between monitoring periods is determined by the time difference between the monitoring period when the pest population activity intensity in the upstream monitoring area first exceeds the preset activity threshold and the monitoring period when the intensity of pre-transmission indicators in the downstream monitoring area first exceeds the preset warning threshold. The degree of overlap of historical migration paths is determined by the ratio of the number of times the line direction between the target monitoring areas is consistent with the historical pest migration direction to the total number of historical migration records. The pest transmission correlation strength is obtained by weighted summation of the degree of consistency of insect age diffusion potential changes, the reciprocal of the response interval between monitoring periods, and the degree of overlap of historical migration paths.

[0087] The host distribution correlation is determined based on the forest health characteristics in the regional time series feature group and the forest type distribution information obtained from the analysis of remote sensing image data. The determination of the host distribution correlation includes extracting the consistency of forest types, the consistency of comprehensive health decline, and the consistency of changes in canopy damage intensity between any two monitoring areas, and combining them with the continuous distribution range of the host and the contact ratio of the regional boundary. Among them, the consistency of forest types is determined by the proportion of the area corresponding to the same forest type to the union of the total forest area of ​​the two monitoring areas. The consistency of comprehensive health decline is determined by the ratio of the number of times the change direction of comprehensive health decline in the two monitoring areas is consistent in multiple consecutive monitoring periods to the total number of comparisons. The host distribution correlation strength is obtained by weighted summation of the consistency of forest types, the consistency of comprehensive health decline, the consistency of changes in canopy damage intensity, the continuous distribution range of the host, and the contact ratio of the regional boundary.

[0088] Spatiotemporal correlation modeling is performed based on environmental impact correlation, pest transmission correlation, and host distribution correlation to construct a multi-relationship spatiotemporal correlation structure among monitoring areas. The spatiotemporal correlation modeling process includes establishing inter-regional correlation units with each monitoring area as the correlation starting area and correlation ending area, and mapping the environmental impact correlation strength, pest transmission correlation strength, and host distribution correlation strength to the environmental impact weight, transmission correlation weight, and host distribution weight of the corresponding correlation unit, respectively. The comprehensive correlation strength of the multi-relationship is obtained by multiplying the environmental impact weight, transmission correlation weight, and host distribution weight by the corresponding correlation contribution weight and summing them.

[0089] Based on a multi-relation spatiotemporal association structure, the multi-source fusion features of each monitoring area during the current monitoring period are combined with the multi-source fusion features of adjacent monitoring periods of related monitoring areas to generate a comprehensive pest and disease characterization sequence. The association constraint fusion process includes using the multi-source fusion features of each monitoring area during the current monitoring period as the current area input, extracting the multi-source fusion features of related monitoring areas during adjacent monitoring periods as the related area input, weighting and aggregating the related area inputs according to the multi-relation comprehensive association strength, and combining them with the current area input for constraint fusion to obtain the comprehensive pest and disease characterization results of each monitoring area for each monitoring period. The comprehensive pest and disease characterization results of each monitoring area for each monitoring period are then arranged in chronological order to form a comprehensive pest and disease characterization sequence.

[0090] In this embodiment, the generation of pest and disease prediction results specifically includes:

[0091] The integrated pest management (IPM) sequence is input into the improved AGCRN model. Node representation encoding is performed in the node embedding generation structure to generate node embedding representations for each monitoring area. The node representation encoding process includes: splitting the multidimensional representation components corresponding to each monitoring area in the IPM sequence for each monitoring period and arranging them in a preset order to construct initial node feature vectors; calculating the temporal variation characteristics based on the value changes of the initial node feature vectors over multiple consecutive monitoring periods; determining the variation amplitude according to the difference between the initial node feature vector corresponding to the current monitoring period and the initial node feature vector corresponding to the previous monitoring period; and determining the variation amplitude based on the number of times the change direction is consistent over multiple consecutive monitoring periods. The ratio of the total number of changes determines the stability of the change; the initial node feature vector and the corresponding temporal change feature are jointly encoded, and the initial node feature vector and the temporal change feature are weighted and combined with the corresponding encoding weights to generate a temporal enhanced node feature representation; based on the temporal enhanced node feature representation and the correlation strength information between monitoring areas, cross-regional feature fusion processing is performed, and the temporal enhanced node feature representation corresponding to the current monitoring area and the temporal enhanced node feature representation corresponding to the associated monitoring area are weighted and converged to generate a relation enhanced node feature representation; normalization and scale alignment processing are performed on the relation enhanced node feature representation to generate the node embedding representation results corresponding to each monitoring area;

[0092] Compared with the existing AGCRN model, the improved AGCRN model has the following improvements: First, in terms of graph structure modeling, the single adaptive graph generation structure in the original model is transformed into a dual graph generation structure. While retaining the node embedding representation results, an environmental association graph and a propagation association graph are constructed respectively. The environmental association graph is used to represent the environmental impact association relationship, and the propagation association graph is used to represent the pest propagation association relationship and the host distribution association relationship, thereby improving the ability to express multiple types of spatial associations. Second, in terms of temporal modeling, the gating computation method based on graph convolution in the original model is transformed into a time delay-aware gating structure. In the cyclic temporal modeling process, the monitoring time interval, the propagation response time difference, and the regional association transmission order are introduced to adjust the state update, realizing dynamic weighting of historical states and candidate states, thereby enhancing the model's ability to characterize propagation lag and temporal evolution characteristics, and thus improving the accuracy and stability of pest and disease prediction results.

[0093] The improved AGCRN model adopts a multi-objective joint loss function, which is composed of a weighted loss of classification, risk prediction, and diffusion intensity regression. The total loss value is generated by weighted summation of the three types of losses. The weights of classification loss, risk loss, and diffusion loss are set in a ratio of 3:2:1. The model parameters are iteratively updated based on the total loss value. When the change of the total loss value is less than a preset threshold in multiple consecutive iterations, the model is considered to have reached the convergence state and the parameter update is stopped.

[0094] The improved AGCRN model employs a gradient-based iterative parameter update method during training. The training parameters are set as follows: the initial learning rate is set to the range of 0.0001 to 0.001, and is gradually reduced during training according to a preset decay strategy; the batch sample size is set to an integer between 16 and 64; the number of training rounds is set to an integer between 50 and 200; the node embedding dimension is set to between 16 and 128; the time-series input window length is set to 5 to 20 monitoring periods; and the weight coefficient of the latency impact factor is set to between 0.2 and 0.5.

[0095] The node embedding representation results are processed using a dual-graph generation structure to generate an environmental correlation graph and a propagation correlation graph. The generation of the environmental correlation graph includes: arranging the node embedding representation results according to monitoring area identifiers within the dual-graph generation structure to construct a regional node embedding sequence; for any two monitoring areas, extracting the corresponding node embedding representation results from the regional node embedding sequence, performing difference calculations and absolute value processing on each dimension of the feature components at the corresponding locations to obtain a feature difference component sequence, and then performing a weighted summation of the feature difference component sequences to generate inter-regional feature difference values; based on the node embedding representation results, applying the results to multiple consecutive monitoring periods... The changes within the region are analyzed by comparing the direction of change of each corresponding feature component, counting the number of times the direction of change is consistent, and calculating the ratio of the number of times the direction of change is consistent to the total number of comparisons to generate the degree of consistency of changes between regions. The feature difference values ​​between regions are weighted and combined with the environmental impact correlation strength in the environmental impact correlation relationship, and the consistency adjustment of the weighted combination result is performed based on the degree of consistency of changes between regions to generate environmental correlation weight values ​​between regions. The environmental correlation weight values ​​between all monitoring regions are arranged in rows and columns according to the monitoring region identifier to construct an environmental correlation weight matrix, and an environmental correlation diagram is generated based on the environmental correlation weight matrix.

[0096] The generation of the propagation correlation graph includes: arranging the node embedding representation results according to the monitoring area identifier in the dual-graph generation structure to construct a regional node embedding sequence; for any two monitoring areas, extracting the corresponding node embedding representation results from the regional node embedding sequence, calculating the change amplitude of each dimension feature component at the corresponding position and performing weighted summation to generate the inter-regional propagation change intensity; based on the changes in the node embedding representation results over multiple consecutive monitoring periods, determining the monitoring period when each monitoring area reaches a preset change threshold, and calculating the time difference between the corresponding monitoring periods of any two monitoring areas to generate the inter-regional propagation time difference; weighting and fusing the inter-regional propagation change intensity with the pest propagation correlation intensity in the pest propagation correlation relationship and the host distribution correlation intensity in the host distribution correlation relationship, respectively, and adjusting the fusion result based on the inter-regional propagation time difference to generate inter-regional propagation correlation weight values; arranging the propagation correlation weight values ​​between all monitoring areas according to the monitoring area identifier in rows and columns to construct a propagation correlation weight matrix, and generating the propagation correlation graph based on the propagation correlation weight matrix;

[0097] Based on the node relationships represented by the environmental association graph and the propagation association graph, a node adaptive graph convolutional structure is used to perform cross-regional feature propagation and weighted aggregation processing on the node embedding representation results, generating graph association feature representation results corresponding to each monitoring region. The generation of graph association feature representation results includes: determining the set of associated monitoring regions and their association weight values ​​corresponding to each monitoring region based on the environmental association graph and the propagation association graph; extracting corresponding feature vectors from each associated monitoring region according to the node embedding representation results, and weighting the extracted feature vectors according to the environmental association weight value and the propagation association weight value to generate association feature contribution values; weighting and aggregating the association feature contribution values ​​corresponding to each associated monitoring region, and fusing them with the node embedding representation results corresponding to the current monitoring region to generate graph association feature representation results.

[0098] Based on the graph association feature representation results, a cyclic temporal modeling structure is used to perform temporal state update processing, and a time delay-aware gating structure is combined to dynamically adjust the temporal information, generating a spatiotemporal prediction representation sequence of pests and diseases corresponding to each monitoring area. The temporal state update processing includes: in the cyclic temporal modeling structure, a temporal input sequence corresponding to each monitoring area is constructed based on the graph association feature representation results; the graph association feature representation results corresponding to the current monitoring period and the temporal state representation results corresponding to the previous monitoring period are extracted; the temporal state representation results of the previous monitoring period and the graph association feature representation results of the current monitoring period are weighted and combined to generate candidate state representation results of the current monitoring period; and the candidate state representation results are fused and updated with the temporal state representation results of the previous monitoring period to generate the temporal state representation results of the current monitoring period.

[0099] Dynamic adjustment includes: In the time delay perception gating structure, a time delay influence factor is calculated based on the monitoring time interval, propagation response time difference, and regional association transmission order between each monitoring area. The monitoring time interval is determined according to the absolute value of the difference between the corresponding monitoring time interval numbers of the current monitoring area and the associated monitoring area. The propagation response time difference is determined according to the difference between the monitoring time interval when the temporal state representation result of the associated monitoring area first exceeds the preset response threshold and the corresponding monitoring time interval of the current monitoring area. The regional association transmission order is determined according to the shortest path hierarchy order from the associated monitoring area to the current monitoring area in the environmental association diagram and the propagation association diagram. The monitoring time interval, propagation response time difference, and regional association transmission order are normalized and then weighted and summed with the corresponding time delay weights to generate the time delay influence factor. Based on the time delay influence factor, the temporal state representation result of the previous monitoring period and the candidate state representation result of the current monitoring period are weighted and adjusted to generate historical state adjustment results and candidate state adjustment results. Finally, the historical state adjustment results and candidate state adjustment results are weighted and fused to generate the spatiotemporal prediction representation sequence of pests and diseases corresponding to each monitoring area.

[0100] Based on the spatiotemporal prediction representation sequence of pests and diseases, pest and disease category discrimination processing and pest and disease occurrence risk analysis processing are performed to generate pest and disease category identifiers, occurrence probability values ​​and risk level identifiers. The pest and disease category discrimination processing includes extracting the spatiotemporal prediction representation vector of pests and diseases corresponding to each monitoring area according to the monitoring area identifier based on the spatiotemporal prediction representation sequence of pests and diseases, performing a dimension-by-dimensional matching calculation on the spatiotemporal prediction representation vector of pests and diseases corresponding to each monitoring area with the preset pest and disease category feature pattern, performing a weighted summation according to the absolute value of the difference between the corresponding feature components to generate the matching difference value corresponding to each category, sorting each category according to the size of the matching difference value, selecting the category with the smallest matching difference value as the target pest and disease category and generating the pest and disease category identifier.

[0101] The risk analysis and processing of pest and disease occurrence includes extracting the spatiotemporal prediction intensity value of pest and disease corresponding to each monitoring area based on the spatiotemporal prediction intensity value of pest and disease, comparing and mapping the spatiotemporal prediction intensity value of pest and disease with the risk distribution interval in the historical pest and disease occurrence samples, calculating the occurrence probability value according to the relative position of the current intensity value in the historical sample intensity distribution, and determining the risk level label according to the range in which the occurrence probability value falls within the preset risk level interval.

[0102] Based on the spatiotemporal prediction characterization sequence of pests and diseases, diffusion path analysis and trend evolution analysis are performed to generate diffusion direction identifiers, diffusion range identifiers, and diffusion intensity identifiers. The diffusion path analysis process includes constructing a regional temporal intensity change sequence based on the spatiotemporal prediction characterization sequence of pests and diseases according to the monitoring area identifier and the monitoring time period identifier. For any two monitoring areas, the changes in the spatiotemporal prediction characterization intensity values ​​of pests and diseases within the corresponding monitoring time period are extracted. The difference in intensity growth rate between adjacent monitoring areas is calculated and the order of growth is determined. The diffusion relationship between regions is screened by combining the diffusion association weight value in the diffusion association diagram. The region connection relationship with the large intensity growth rate and the high diffusion association weight value is selected to construct a diffusion path candidate set. The diffusion path candidate set is screened according to the path connection and continuity constraints according to the monitoring time period order to generate the main diffusion path. Then, the diffusion direction identifier is determined according to the spatial connection direction from the starting monitoring area to the target monitoring area in the main diffusion path. The diffusion range identifier is determined according to the number of monitoring areas covered by the main diffusion path and the spatial boundary range of the corresponding monitoring areas.

[0103] The trend evolution analysis process includes extracting the spatiotemporal prediction intensity value sequence of pests and diseases in each monitoring area within multiple consecutive monitoring periods based on the spatiotemporal prediction characterization sequence of pests and diseases, calculating the intensity change difference between adjacent monitoring periods and counting the number of consecutive increases and consecutive decreases, determining the degree of growth trend based on the ratio of the number of consecutive increases to the total number of monitoring periods, and weighting the trend change in combination with the changes in the propagation association weight in the propagation association diagram of each monitoring area to generate a trend evolution intensity value. The trend evolution intensity value is then normalized across all monitoring areas, and finally, the diffusion intensity indicator is determined based on the normalized trend evolution intensity value.

[0104] The pest and disease category identifiers, occurrence probability values, risk level identifiers, diffusion direction identifiers, diffusion range identifiers, and diffusion intensity identifiers corresponding to each monitoring area are collected and processed to generate pest and disease prediction results for each monitoring area.

[0105] In this embodiment, the generation of the zoning prevention and control demand results specifically includes:

[0106] Based on the occurrence probability value, the intensity value corresponding to the diffusion intensity indicator, the range value corresponding to the diffusion range indicator, and the risk level indicator in the pest and disease prediction results, risk quantification is performed on each monitoring area to determine the comprehensive regional risk value for each monitoring area. The determination of the comprehensive regional risk value includes normalizing the occurrence probability value, diffusion intensity value, and diffusion range value to generate probability risk component, intensity risk component, and range risk component, and then weighting and summing the probability risk component, intensity risk component, and range risk component with the level value corresponding to the risk level indicator to obtain the comprehensive regional risk value for each monitoring area.

[0107] Adjacent monitoring areas with the same pest and disease category identification and a regional risk comprehensive value difference lower than the preset zoning threshold are aggregated and divided. Based on the aggregated monitoring area set, the risk zoning identifier is determined, and the pest and disease risk zoning result is generated.

[0108] The risk zone, the comprehensive risk value of the region, and the intensity value corresponding to the diffusion intensity indicator of each monitoring area in the results of the pest and disease risk zoning are weighted and calculated to generate a priority judgment value. The prevention and control priority indicator is determined according to the result of the priority judgment value falling into the preset priority range.

[0109] Based on the occurrence probability value, diffusion intensity change trend, diffusion range expansion trend and control priority indicator in the pest and disease prediction results, a weighted calculation is performed on each monitoring area to generate a time window judgment value, and the control time window indicator is determined according to the result of the time window judgment value falling into the preset time window interval.

[0110] The comprehensive regional risk value, the priority level value corresponding to the prevention and control priority indicator, and the time urgency value corresponding to the prevention and control time window indicator are weighted and integrated to generate a prevention and control intensity judgment value. The prevention and control intensity level indicator is determined based on the result of the prevention and control intensity judgment value falling into the preset intensity level range.

[0111] The pest and disease category identifiers, prevention and control priority identifiers, prevention and control time window identifiers, and prevention and control intensity level identifiers corresponding to each monitoring area are collected and processed to generate the regional prevention and control requirements results.

[0112] In this embodiment, the generation of the prevention and control strategy set specifically includes:

[0113] Based on the pest and disease category identifiers, prevention and control priority identifiers, prevention and control time window identifiers, and prevention and control intensity level identifiers in the results of regional prevention and control needs, and combined with preset prevention and control constraints, the monitoring areas are correlated and matched to construct a prevention and control decision mapping relationship. The prevention and control constraints include forest area information, tree type information, prevention and control resource information, pesticide use restriction information, ecological protection constraint information, and operation period constraint information. The correlation and matching process includes: matching the prevention and control time window identifier with the operation period constraint information in terms of time interval; matching the prevention and control intensity level identifier with the resource supply capacity in the prevention and control resource information in terms of intensity; matching the pest and disease category identifier with the tree type information and pesticide use restriction information in terms of applicability; matching the forest area information with the operation coverage requirements corresponding to the prevention and control intensity level identifier in terms of scope; and filtering out matching results that do not meet the ecological constraints in combination with the ecological protection constraint information.

[0114] Based on the prevention and control decision mapping relationship, forest objects, operation areas and prevention and control units that are compatible with the current pest and disease categories and meet the requirements of ecological protection are screened in each monitoring area to determine the target prevention and control object identification;

[0115] Based on the prevention and control decision mapping relationship, the time interval corresponding to the prevention and control time window identifier is matched and filtered with the allowable work interval corresponding to the work period constraint information to determine the target work period identifier for each monitoring area.

[0116] Based on the prevention and control decision mapping relationship, the target prevention and control object identifier, target operation period identifier, prevention and control priority identifier, prevention and control intensity level identifier and prevention and control resource information are associated and allocated to determine the corresponding pesticide demand, equipment demand and personnel demand for each monitoring area, and the resource allocation result is formed by combining the available resources.

[0117] Based on the prevention and control decision mapping relationship, the operational coverage corresponding to the target prevention and control object identification, resource allocation results, pesticide use restriction information, ecological protection constraint information, and forest area information is constrained, screened, and compared. The measure judgment value and prevention and control measure identification corresponding to each monitoring area are determined. The constraint screening and matching comparison includes eliminating measures that do not meet the use conditions from the candidate prevention and control measures based on pesticide use restriction information, screening measures whose impact on the ecological environment exceeds the allowable range based on ecological protection constraint information, filtering measures that exceed the feasible range based on the operational coverage corresponding to the forest area information, comparing the resource matching degree of the remaining candidate prevention and control measures based on the resource allocation results, performing matching calculations based on the consistency between the target prevention and control object identification and the applicable scope of the candidate prevention and control measures, generating a measure judgment value by comprehensively considering the constraint satisfaction and matching degree of each candidate prevention and control measure, and determining the prevention and control measure identification based on the magnitude of the measure judgment value.

[0118] Based on the prevention and control decision mapping relationship, the measure judgment value, prevention and control measure identifier, target operation period identifier and resource allocation result are combined and configured to determine the corresponding pesticide application parameters, equipment scheduling parameters and operation time parameters for each monitoring area, and are collected with the prevention and control measure identifier to generate a set of prevention and control strategies.

[0119] In this embodiment, the determination of the target prevention and control strategy specifically includes:

[0120] Based on the control measures identifiers, pesticide application parameters, equipment scheduling parameters, and operation time parameters in the control strategy set, and combined with the pest and disease prediction results and regional control needs results for each monitoring area, a set of strategy evaluation indicators is extracted for each candidate control strategy. The set of strategy evaluation indicators includes the degree of control suitability, the degree of resource consumption, the degree of operation timeliness, and the degree of ecological constraint satisfaction.

[0121] The matching evaluation value of each candidate prevention and control strategy is determined by weighted calculation based on the strategy evaluation index set to assess the matching degree between each candidate prevention and control strategy and the current prevention and control needs of the monitored area. The matching evaluation value is obtained by weighted summation of the matching degree between the prevention and control measure identifier and the pest and disease category identifier, the matching degree between the pesticide application parameters and the prevention and control intensity level identifier, the matching degree between the equipment scheduling parameters and the operational needs corresponding to the target prevention and control area identifier, and the matching degree between the operational time parameters and the prevention and control time window identifier.

[0122] Based on the strategy evaluation index set, resource requirements are calculated for the pesticide delivery parameters, equipment scheduling parameters and operation time parameters of each candidate prevention and control strategy. The pesticide consumption, equipment occupancy and personnel input are determined and compared with the current available resources to obtain the resource occupancy ratio. The resource occupancy ratios are weighted and summed according to the preset resource weights to generate resource evaluation values.

[0123] Based on the strategy evaluation index set, the timeliness matching between the operation time parameters and the allowable operation intervals, as well as the ecological constraint satisfaction, of each candidate prevention and control strategy are jointly calculated to determine the timeliness evaluation value and ecological evaluation value of each candidate prevention and control strategy. The timeliness evaluation value is obtained by calculating the degree of overlap between the operation time parameters and the allowable time intervals corresponding to the prevention and control time window identifiers, and by weighted summing the results in combination with the urgency corresponding to the prevention and control priority identifiers. The ecological evaluation value is obtained by matching the degree of conformity between the pesticide application parameters, equipment scheduling parameters, and target prevention and control area range and the operation coverage range corresponding to the pesticide use restriction information, ecological protection constraint information, and forest area information.

[0124] The adaptation evaluation value, timeliness evaluation value, and ecological evaluation value are used as positive evaluation components, and the resource evaluation value is used as negative evaluation component. After multiplying each of them by the corresponding evaluation weight, the weighted sum is obtained to obtain the comprehensive evaluation value of each candidate prevention and control strategy.

[0125] The candidate control strategies are ranked from highest to lowest according to their comprehensive evaluation values. The candidate control strategy with the best comprehensive evaluation value is selected as the target control strategy. The target control strategy includes the target control area identifier, the target pest and disease identifier, the control measure identifier, the pesticide application parameters, the equipment scheduling parameters, the operation time parameters, and the operation path parameters.

[0126] Example 1: To verify the feasibility of this invention in practice, it was applied to a pine forest resource protection scenario in a hilly forest area in southern my country. This area has significant topographic relief, complex forest stand structure, and is significantly affected by the monsoon climate. Under the high temperature and humidity conditions of spring and summer, pests and diseases are prone to occur and spread rapidly. Traditional methods relying on manual patrols and single pest monitoring are insufficient to promptly grasp the development trend of pests and diseases, resulting in problems such as delayed detection, difficulty in determining spread paths, and unreasonable allocation of control resources. This leads to repeated outbreaks of pests in local areas, causing a continuous impact on tree growth.

[0127] In this scenario, meteorological monitoring equipment, insect trapping devices, and remote sensing data acquisition methods are first deployed to continuously collect data on the meteorological environment, insect activity, and forest growth status in different monitoring areas within the forest region. The collected data undergoes unified temporal and spatial alignment processing. Based on this, meteorological driving characteristics, insect activity characteristics, remote sensing representation characteristics, and forest health characteristics are correlated and fused to form a comprehensive feature sequence reflecting the occurrence and spread of pests and diseases. Subsequently, combining the spatial relationships between different monitoring areas and the patterns of pest transmission, joint modeling is performed on the environmental impact relationships, transmission relationships, and host distribution relationships between areas, thereby obtaining a comprehensive pest and disease representation sequence with spatiotemporal correlation characteristics. Furthermore, by introducing a prediction structure with spatiotemporal modeling capabilities, the development trend of pests and diseases in each monitoring area is continuously predicted, yielding prediction results including information such as pest and disease types, occurrence probabilities, and spread directions.

[0128] During the prevention and control decision-making stage, risk quantification and zoning of each monitoring area are conducted based on the prediction results. Areas with similar risk levels and spatial adjacency are aggregated to form a clear risk zoning structure. Based on this, prevention and control priorities and time windows are determined by considering the risk level, diffusion trend, and operational time constraints of each area. Furthermore, the prevention and control intensity is rationally determined by comprehensively considering resource supply capacity and ecological protection constraints. Subsequently, through a comprehensive evaluation of different prevention and control measures in terms of resource consumption, operational timeliness, and the degree to which ecological constraints are met, multiple candidate prevention and control strategies are ranked and screened, ultimately determining the target prevention and control strategy suitable for each monitoring area.

[0129] In actual operation, analysis and comparison based on data from multiple consecutive monitoring periods revealed that after applying the method of this invention, the spread trend of pests and diseases could be identified at an early stage, the risk zoning results showed a high degree of consistency with the actual distribution of affected areas, the priority of prevention and control matched the needs of on-site prevention and control, and the allocation of prevention and control resources was more concentrated and reasonable. Operational records from different time periods within the same forest area showed that the prevention and control strategies generated by the method of this invention effectively reduced repetitive operations and improved the timeliness and targeting of prevention and control responses, thereby improving the overall effectiveness of forestry pest and disease control while ensuring ecological and environmental constraints.

[0130] Table 1. Performance Comparison of the Invention and Traditional Forestry Pest and Disease Control Methods

[0131] Indicator Categories Traditional methods Method of the present invention Prediction accuracy (%) 78.6 84.2 Risk partitioning accuracy (%) 76.3 82.1 Prevention and control priority matching rate (%) 74.5 80.4 Prevention and control time window matching rate (%) 71.2 77.6 Resource utilization rate (%) 81.0 87.3 Job coverage (%) 83.5 88.9 Average response time (h) 11.8 8.9

[0132] As can be clearly seen from Table 1, the method of the present invention is superior to the traditional method in many indicators.

[0133] Regarding prediction accuracy, the traditional method achieves 78.6%, while the method of this invention improves it to 84.2%. The main reason for this improvement is that this invention uses multi-source data fusion and spatiotemporal correlation modeling to form a unified expression of meteorological, insect, and forest status information, reducing the bias caused by a single data source, thereby improving stability and accuracy in the prediction process.

[0134] The accuracy of risk zoning has been improved from 76.3% to 82.1%. This invention calculates the comprehensive risk value of the region and divides adjacent monitoring areas by aggregation. This makes risk zoning no longer rely on a single threshold judgment, but is based on a comprehensive analysis of multi-dimensional factors. Therefore, it can better reflect the actual distribution of pests.

[0135] The matching rate of prevention and control priorities increased from 74.5% to 80.4%. This improvement is due to the introduction of regional risk comprehensive value and diffusion intensity value in the priority determination process, so that the priority is no longer based solely on experience or single indicator judgment, thus better reflecting the actual distribution of prevention and control needs.

[0136] The matching rate of prevention and control time windows increased from 71.2% to 77.6%. This invention combines occurrence probability values, spread trends, and priority indicators for comprehensive calculation, making the determination of time windows more consistent with the development rhythm of pests and diseases, thereby improving the rationality of the timing of operations.

[0137] Resource utilization increased from 81.0% to 87.3%, and operational coverage increased from 83.5% to 88.9%. This is mainly due to the introduction of resource constraints and decision mapping relationships in the generation of prevention and control strategies in this invention, which makes resource allocation more centralized and avoids redundant investment, thereby improving overall utilization efficiency and coverage.

[0138] The average response time has been reduced from 11.8 hours to 8.9 hours. This invention reduces the manual analysis and decision-making process by linking the prediction results with prevention and control decisions, enabling prevention and control strategies to be generated and executed more quickly, thereby improving response efficiency.

[0139] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A forest pest control method based on big data mining, characterized in that, Includes the following steps: Collect multi-source heterogeneous data from the target forest area during the target monitoring period and preprocess them to generate a standardized forestry monitoring dataset; Multi-source features are extracted from standardized forestry monitoring datasets and fused according to the correlation of ecological factors to generate a multi-source fused feature set. Spatiotemporal correlation analysis and constraint fusion processing are performed based on multi-source fusion feature sets to generate integrated pest and disease characterization sequences. Based on the improved AGCRN model, spatiotemporal prediction modeling and result analysis of the integrated pest and disease characterization sequence are performed to generate pest and disease prediction results for each monitoring area. Based on the pest and disease prediction results, risk zoning and prevention and control decision-making are carried out for different monitoring areas in the target forest area to generate zoning prevention and control needs results. Based on the results of regional prevention and control needs and preset prevention and control constraints, a prevention and control decision mapping relationship is constructed and a set of prevention and control strategies is generated. Based on the set of prevention and control strategies, strategy evaluation indicators are constructed, and comprehensive evaluation and ranking are carried out to determine the target prevention and control strategies. 2.The forestry pest control method based on big data mining of claim 1, wherein, The multi-source heterogeneous data includes meteorological environmental data, insect monitoring data, remote sensing image data, and forest growth status data. The preprocessing includes unifying timestamps, unifying spatial coordinates, removing outliers, completing missing values, suppressing noise, normalizing dimensions, and associating and aligning data according to monitoring area identifiers, monitoring time period identifiers, and forest object identifiers. 3.The forestry disease and pest control method based on big data mining of claim 1, wherein, The generation of the multi-source fusion feature set specifically includes: Based on meteorological and environmental data in the standardized forestry monitoring dataset, meteorological driving factor analysis and coupled modeling were performed to obtain a set of meteorological driving features. Based on the insect monitoring data in the standardized forestry monitoring dataset, insect activity behavior characterization and population dynamic analysis were performed to obtain a set of insect activity characteristics. Spatial representation extraction and anomaly response quantification were performed on remote sensing image data in the standardized forestry monitoring dataset to obtain a set of remote sensing representation features; By combining the tree growth status data in the standardized forestry monitoring dataset, a set of tree health features is obtained through health status analysis and damage characterization. Based on the meteorological driving feature set, insect activity feature set, remote sensing characterization feature set and forest health feature set, cross-source association matching and spatiotemporal registration processing are performed to generate a registration feature group under a unified spatiotemporal benchmark. Semantic association encoding and fusion processing are performed on the registered feature groups according to the correlation of ecological factors to generate a multi-source fusion feature set.

4. The forestry pest and disease control method based on big data mining according to claim 1, characterized in that, The generation of the integrated pest and disease characterization sequence specifically includes: Based on the multi-source fusion feature set, regional aggregation and temporal unfolding are performed according to the monitoring area identifier, monitoring time period identifier, and forest object identifier to construct the regional temporal feature group corresponding to each monitoring area; The environmental impact correlations were determined based on the regional temporal characteristic groups and the spatial relationships between each monitoring area. Based on the inter-regional change trajectories of regional temporal characteristic groups and insect activity characteristics, the association between insect pest transmission was determined; The host distribution association was determined based on the forest health characteristics in the regional time series feature group and the forest type distribution information obtained from the analysis of remote sensing image data. Spatiotemporal correlation modeling was performed based on environmental impact correlations, pest transmission correlations, and host distribution correlations to construct a multi-relationship spatiotemporal correlation structure among monitoring areas; Based on the multi-relation spatiotemporal association structure, the multi-source fusion features of the current monitoring period in each monitoring area are subjected to association constraint fusion processing with the multi-source fusion features of the adjacent monitoring periods in the associated monitoring areas to generate a comprehensive pest and disease characterization sequence.

5. A method for forestry pest and disease control based on big data mining according to claim 1, characterized in that, The generation of the pest and disease prediction results specifically includes: The integrated pest and disease representation sequence is input into the improved AGCRN model, and node representation encoding is performed in the node embedding generation structure to generate node embedding representation results for each monitoring area. The node embedding representation results are processed by a dual-graph generation structure to generate an environment association graph and a propagation association graph; Based on the node association relationships represented by the environmental association graph and the propagation association graph, cross-regional feature propagation and weighted aggregation processing are performed on the node embedding representation results through a node adaptive graph convolution structure to generate graph association feature representation results corresponding to each monitoring area; Based on the graph association feature representation results, the temporal state update processing is performed through a cyclic temporal modeling structure, and the temporal information is dynamically adjusted by combining a time delay-aware gating structure to generate spatiotemporal prediction characterization sequences of pests and diseases corresponding to each monitoring area. Based on the spatiotemporal prediction characterization sequence of pests and diseases, pest and disease category discrimination processing and pest and disease occurrence risk analysis processing are performed to generate pest and disease category identifiers, occurrence probability values ​​and risk level identifiers. Based on the spatiotemporal prediction characterization sequence of pests and diseases, diffusion path analysis and trend evolution analysis are performed to generate diffusion direction identifiers, diffusion range identifiers, and diffusion intensity identifiers. The pest and disease category identifiers, occurrence probability values, risk level identifiers, diffusion direction identifiers, diffusion range identifiers, and diffusion intensity identifiers corresponding to each monitoring area are collected and processed to generate pest and disease prediction results for each monitoring area.

6. The forestry pest and disease control method based on big data mining according to claim 1, characterized in that, The generation of the zoning prevention and control requirements results specifically includes: Based on the occurrence probability value, the intensity value corresponding to the diffusion intensity indicator, the range value corresponding to the diffusion range indicator, and the risk level indicator in the pest and disease prediction results, risk quantification is performed on each monitoring area to determine the comprehensive regional risk value corresponding to each monitoring area. Adjacent monitoring areas with the same pest and disease category identification and a regional risk comprehensive value difference lower than the preset zoning threshold are aggregated and divided. Based on the aggregated monitoring area set, the risk zoning identifier is determined, and the pest and disease risk zoning result is generated. The risk zone, the comprehensive risk value of the region, and the intensity value corresponding to the diffusion intensity indicator of each monitoring area in the results of the pest and disease risk zoning are weighted and calculated to generate a priority judgment value. The prevention and control priority indicator is determined according to the result of the priority judgment value falling into the preset priority range. Based on the occurrence probability value, diffusion intensity change trend, diffusion range expansion trend and control priority indicator in the pest and disease prediction results, a weighted calculation is performed on each monitoring area to generate a time window judgment value, and the control time window indicator is determined according to the result of the time window judgment value falling into the preset time window interval. The comprehensive regional risk value, the priority level value corresponding to the prevention and control priority indicator, and the time urgency value corresponding to the prevention and control time window indicator are weighted and integrated to generate a prevention and control intensity judgment value. The prevention and control intensity level indicator is determined based on the result of the prevention and control intensity judgment value falling into the preset intensity level range. The pest and disease category identifiers, prevention and control priority identifiers, prevention and control time window identifiers, and prevention and control intensity level identifiers corresponding to each monitoring area are collected and processed to generate the regional prevention and control requirements results.

7. A method for forestry pest and disease control based on big data mining according to claim 1, characterized in that, The generation of the prevention and control strategy set specifically includes: Based on the pest and disease category identifiers, prevention and control priority identifiers, prevention and control time window identifiers, and prevention and control intensity level identifiers in the results of regional prevention and control needs, and combined with preset prevention and control constraints, the monitoring areas are correlated and matched to construct a prevention and control decision mapping relationship. The prevention and control constraints include forest area information, forest type information, prevention and control resource information, pesticide use restriction information, ecological protection constraint information, and operation period constraint information. Based on the prevention and control decision mapping relationship, forest objects, operation areas and prevention and control units that are compatible with the current pest and disease categories and meet the requirements of ecological protection are screened in each monitoring area to determine the target prevention and control object identification; Based on the prevention and control decision mapping relationship, the time interval corresponding to the prevention and control time window identifier is matched and filtered with the allowable work interval corresponding to the work period constraint information to determine the target work period identifier for each monitoring area. Based on the prevention and control decision mapping relationship, the target prevention and control object identifier, target operation period identifier, prevention and control priority identifier, prevention and control intensity level identifier and prevention and control resource information are associated and allocated to determine the corresponding pesticide demand, equipment demand and personnel demand for each monitoring area, and the resource allocation result is formed by combining the available resources. Based on the prevention and control decision mapping relationship, the operational coverage area corresponding to the target prevention and control object identification, resource allocation results, pesticide use restriction information, ecological protection constraint information, and forest area information is constrained, screened, and matched to determine the measure judgment value and prevention and control measure identification for each monitoring area. Based on the prevention and control decision mapping relationship, the measure judgment value, prevention and control measure identifier, target operation period identifier and resource allocation result are combined and configured to determine the corresponding pesticide application parameters, equipment scheduling parameters and operation time parameters for each monitoring area, and are collected with the prevention and control measure identifier to generate a set of prevention and control strategies.

8. A method for forestry pest and disease control based on big data mining according to claim 1, characterized in that, The determination of the target prevention and control strategy specifically includes: Based on the control measures identifiers, pesticide application parameters, equipment scheduling parameters, and operation time parameters in the control strategy set, and combined with the pest and disease prediction results and regional control needs results for each monitoring area, a set of strategy evaluation indicators corresponding to each candidate control strategy is extracted. Based on the strategy evaluation index set, the matching between each candidate prevention and control strategy and the current prevention and control needs of the monitoring area is weighted and calculated to determine the appropriate evaluation value for each candidate prevention and control strategy. Based on the strategy evaluation index set, resource requirements are calculated for the pesticide delivery parameters, equipment scheduling parameters and operation time parameters of each candidate prevention and control strategy. The pesticide consumption, equipment occupancy and personnel input are determined and compared with the current available resources to obtain the resource occupancy ratio. The resource occupancy ratios are weighted and summed according to the preset resource weights to generate resource evaluation values. Based on the strategy evaluation index set, the timeliness matching between the operation time parameters and the allowable operation range and the ecological constraint satisfaction of each candidate prevention and control strategy are jointly calculated to determine the timeliness evaluation value and ecological evaluation value of each candidate prevention and control strategy. The adaptation evaluation value, timeliness evaluation value, and ecological evaluation value are used as positive evaluation components, and the resource evaluation value is used as negative evaluation component. After multiplying each of them by the corresponding evaluation weight, the weighted sum is obtained to obtain the comprehensive evaluation value of each candidate prevention and control strategy. The candidate prevention and control strategies are ranked from highest to lowest according to their comprehensive evaluation values. The candidate prevention and control strategy with the best comprehensive evaluation value is selected as the target prevention and control strategy. The target prevention and control strategy includes the target prevention and control area identifier, the target pest and disease object identifier, the prevention and control measure identifier, the pesticide application parameters, the equipment scheduling parameters, the operation time parameters, and the operation path parameters.

Images