Orchard pest early warning and prevention resource management decision support method based on multi-source data

Abstract

The invention relates to the technical field of orchard disease and insect pest early warning and prevention resource management, in particular to an orchard disease and insect pest early warning and prevention resource management decision support method based on multi-source data, and the method comprises the steps: firstly obtaining orchard multi-source data, and constructing a space-time fusion data set with a geographic grid as a unit through space-time alignment and standardization processing; inputting the data set into a machine learning multi-classification model, and outputting a risk quantification index of a future specific pest and disease damage; triggering a prevention and treatment decision engine based on an index, and combining a resource knowledge base to generate an optimal prevention and treatment prescription through multi-objective optimization; then, matching prevention and control resource real-time inventory with agricultural machinery space positions, and generating a resource scheduling scheme and agricultural machinery operation path planning; and finally, the integrated information is visualized on an electronic map, a control instruction is issued, and feedback is received to form closed-loop management. The method improves the accuracy of disease and pest early warning, achieves the optimal prevention and control scheme, improves the utilization efficiency of resources and agricultural machinery, and provides support for scientific management of orchard diseases and pests.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of orchard disease and pest early warning and prevention and control resource management, in particular to an orchard disease and pest early warning and prevention and control resource management decision support method based on multi-source data. BACKGROUND

[0002] Traditional orchard disease and pest early warning relies on a single data source, such as only referring to macro meteorological data or local historical records. The data lacks spatio-temporal consistency and completeness, making it difficult to accurately quantify the risk of future disease and pest occurrence, and prone to early warning lag or misjudgment. Prevention and control decision-making often relies on manual experience, without systematically integrating prevention and control effectiveness, economic cost and environmental impact, resulting in either insufficient prevention and control effect or serious resource waste or additional burden on the ecological environment. At the same time, prevention and control resource management and agricultural machinery operation scheduling are mutually exclusive, with low matching degree between resource inventory and actual demand of the orchard, and the agricultural machinery operation path is not optimized, resulting in unreasonable resource allocation and low agricultural machinery operation efficiency. Moreover, the entire disease and pest management process lacks a closed-loop mechanism, making it difficult to monitor the operation progress and resource status in real time, and difficult to adjust in time when problems occur, affecting the overall prevention and control effect. Therefore, the present application provides an orchard disease and pest early warning and prevention and control resource management decision support method based on multi-source data to solve the problems mentioned in the background. SUMMARY

[0003] The present application aims to provide an orchard disease and pest early warning and prevention and control resource management decision support method based on multi-source data to solve the problems mentioned in the background.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical solutions: The orchard disease and pest early warning and prevention and control resource management decision support method based on multi-source data comprises the following steps: Step S1, data perception and fusion: acquiring various source data of the orchard, performing spatio-temporal alignment and standardization processing on the various source data of the orchard, and constructing a spatio-temporal fusion data set with the orchard geographic grid as the unit; Step S2, disease and pest risk quantification: inputting the spatio-temporal fusion data set into a disease and pest prediction model, the disease and pest prediction model being a multi-classification model based on machine learning algorithm and characterized by environmental factor time series data and historical disease and pest occurrence regularity; the output result of the disease and pest prediction model being a risk quantification index for a specific disease and pest in a future preset time period, the risk quantification index being a numerical value integrating occurrence probability and potential harm degree; Step S3, prevention and treatment prescription generation: triggering a prevention and treatment decision engine based on the risk quantification index; the prevention and treatment decision engine is built-in with a prevention and treatment resource knowledge base, which records the prevention and treatment efficiency coefficient, economic cost and environmental impact factor of different prevention and treatment resources under different risk levels; the prevention and treatment decision engine takes maximizing comprehensive benefits as the goal, and obtains the optimal prevention and treatment prescription matched with the current risk level by solving a multi-objective optimization algorithm, the optimal prevention and treatment prescription at least includes the recommended prevention and treatment resource type, application dose and operation time window; Step S4, resource-space matching and scheduling: matching the optimal prevention and treatment prescription with the real-time inventory data of prevention and treatment resources in the orchard and the spatial position information of operation equipment to generate a specific resource scheduling scheme and agricultural machinery operation path planning; the resource scheduling scheme includes the allocation quantity of the required resources from the warehouse to the specific orchard grid, and the agricultural machinery operation path planning is based on the orchard electronic map and the operation point generated by the prescription for path optimization calculation; Step S5, decision visualization and instruction issuing: integrating the risk quantification index, the optimal prevention and treatment prescription, the resource scheduling scheme and the agricultural machinery operation path planning, and performing superposition rendering and visual display on the electronic map of the management terminal, and generating a set of controllable control instructions for driving the intelligent agricultural machinery to execute the prevention and treatment prescription.

[0005] As a preferred scheme, the multiple source data of the orchard includes real-time environmental data collected through Internet of Things devices, fruit tree canopy image data obtained through image collection devices, and macro weather data and historical pest data accessed from external systems.

[0006] As a preferred scheme, the multiple source data of the orchard is acquired, and the multiple source data of the orchard is subjected to spatio-temporal alignment and standardized processing to construct a spatio-temporal fusion data set with the orchard geographic grid as a unit, including: S1-1, converting the acquired real-time environmental data, fruit tree canopy image data, macro weather data and historical pest data into standardized data sequences with unified time stamps and geographic coordinate system identifiers respectively; S1-2, based on the orchard electronic map, the standardized data sequences subjected to unified preprocessing are spatially divided and data correlated according to the preset geographic grid rules to generate a multi-source initial data set in each orchard geographic grid unit; S1-3, performing integrity check on the multi-source initial data set in each orchard geographic grid unit in the time dimension, filling in the data of the data points with missing values by using time series interpolation method, and performing smoothing filter processing on the data sequences with abnormal fluctuations to generate regular time series data at the grid unit level; S1-4, the environment factor data in each grid cell after filling and smoothing processing, the vegetation index and pest and disease visual feature data extracted from the canopy image, and the gridded meteorological elements and historical pest and disease occurrence intensity data are subjected to feature layer superposition and association, to form a time-space fusion data set with the orchard geographical grid as the basic unit.

[0007] As a preferred solution, the time-space fusion data set is input into a pest and disease prediction model, the pest and disease prediction model is a multi-classification model based on a machine learning algorithm, and the environment factor time series data and the historical pest and disease occurrence regularity are characteristics; the output result of the pest and disease prediction model is a risk quantization index for a specific pest and disease in a future preset time period, the risk quantization index is a numerical value that comprehensively considers the occurrence probability and the potential harm degree, including: S2-1, sample division is performed on the time-space fusion data set to generate an input sample set for model training and prediction, wherein each sample corresponds to environment factor time series data of an orchard geographical grid at a historical time point and actual pest and disease occurrence records in a preset time period after the time point; S2-2, multi-dimensional features related to pest and disease occurrence are extracted from the environment factor time series data of each input sample, the multi-dimensional features include statistical features, periodic features, and change trend features of the environment factors, and are combined with the historical pest and disease occurrence regularity features corresponding to the orchard geographical grid to form a complete feature vector; S2-3, the feature vector is input into a multi-classification model based on a machine learning algorithm, the multi-classification model is trained by historical data and is used to learn the complex nonlinear relationship between environment features and multiple pest and disease occurrence conditions; S2-4, the trained multi-classification model is used to predict the time-space fusion data at the current time, and outputs a risk quantization index for a specific pest and disease in a future preset time period, the risk quantization index is obtained by weighting the occurrence probability output by the model and the preset harm weight coefficient of the pest and disease.

[0008] As a preferred solution, step S3 includes: S3-1, according to the risk quantization index, a preset risk level division threshold is combined to determine the pest and disease risk level of the current orchard geographical grid, and a control decision engine is triggered based on the pest and disease risk level; S3-2, the control decision engine retrieves a candidate control resource set from a control resource knowledge base according to the pest and disease risk level, the control resource knowledge base records the control efficiency coefficient, economic cost and environmental impact factor of different control resources under different risk levels; S3-3, based on the candidate prevention and treatment resource set, a multi-objective optimization problem is constructed with the maximum comprehensive benefit as the target, wherein the comprehensive benefit target function at least includes the maximum prevention and treatment effectiveness, the minimum economic cost and the minimum environmental impact; S3-4, a multi-objective optimization algorithm is used to solve the multi-objective optimization problem, and a Pareto optimal solution set is generated, each solution in the Pareto optimal solution set corresponds to a potential prevention and treatment prescription; S3-5, an optimal solution is selected from the Pareto optimal solution set according to a preset decision rule, and an optimal prevention and treatment prescription matching the current risk level is generated, the optimal prevention and treatment prescription at least includes a recommended prevention and treatment resource type, a dose, and a operation time window.

[0009] As a preferred solution, step S4 includes: S4-1, the optimal prevention and treatment prescription is obtained, and real-time inventory data of prevention and treatment resources in the orchard and spatial position information of operation equipment are accessed; S4-2, based on the recommended prevention and treatment resource type and the dose in the optimal prevention and treatment prescription, the area of the orchard geographic grid and the pest risk distribution, the required number of prevention and treatment resources for each orchard geographic grid is calculated; S4-3, according to the real-time inventory data of the prevention and treatment resources, the required number of prevention and treatment resources for each orchard geographic grid is matched with the inventory and verified for availability, if the inventory is insufficient, a resource replenishment warning is triggered, and a resource allocation plan is generated, the resource allocation plan includes the allocation number of the required resources from the warehouse to the specific orchard grid; S4-4, based on the operation points generated by the optimal prevention and treatment prescription and the electronic map of the orchard, combined with the spatial position information of the operation equipment, a path optimization algorithm is used for agricultural machinery operation path planning, the path optimization algorithm takes the minimum total distance or the minimum operation time as the target, and calculates the optimal order and path of the agricultural machinery to traverse all operation points; S4-5, the resource allocation plan and the agricultural machinery operation path planning are integrated to generate a specific resource scheduling scheme and agricultural machinery operation instruction, the resource scheduling scheme includes resource allocation details and scheduling timetable, and the agricultural machinery operation instruction includes path navigation information and operation parameter setting.

[0010] As a preferred solution, step S5 includes: S5-1, the risk quantification index, the optimal prevention and treatment prescription, the resource scheduling scheme and the agricultural machinery operation path planning are integrated to generate prevention and treatment decision integrated data; S5-2, call the electronic map service engine of the management terminal, associate the spatial elements and attribute information in the prevention and control decision integrated data, and perform overlay rendering and visual display on the electronic map, wherein the risk quantization index is displayed on the corresponding orchard geographic grid in the form of a heat map, the optimal prevention and control prescription is associated and displayed in the form of a pop-up window, the resource scheduling scheme is displayed in the form of a material flow chart, and the agricultural machine operation path planning is dynamically displayed in the form of a preset icon and a path line; S5-3, based on the visual display result, generate a controllable instruction set according to a user confirmation instruction or a system preset rule, the controllable instruction set including a navigation control instruction for the intelligent agricultural machine, an operation parameter setting instruction, and a warehouse scheduling instruction for the resource storage system; S5-4, deliver the controllable instruction set to the corresponding intelligent agricultural machine terminal and resource storage control system through an Internet of Things communication network; S5-5, real-time receive operation state information returned by the intelligent agricultural machine and inventory update information fed back by the resource storage system, and dynamically update and display on the electronic map of the management terminal to form a closed-loop management of decision-making, execution, and feedback.

[0011] As can be seen from the technical solutions provided by the above-mentioned application, the fruit orchard pest warning and prevention resource management decision support method based on multi-source data provided by the application has the beneficial effects that: The application integrates real-time environmental data, crown image data, macro-weather data, and historical pest data through multi-source data spatio-temporal fusion, constructs a standardized spatio-temporal fusion data set with an orchard geographic grid as a unit, solves the problem of one-sidedness of single-source data information and spatio-temporal inconsistency, provides a high-quality data basis for subsequent warning, extracts environmental factor multi-dimensional features and historical pest rules through a machine learning multi-classification model, outputs a risk quantization index of comprehensive occurrence probability and harm degree, realizes accurate quantitative warning of specific pests in a preset time period in the future instead of fuzzy judgment, effectively avoids missed or mistaken judgment, and provides a scientific basis for early deployment of prevention measures; The prevention and control decision engine combines an internal knowledge base, constructs a multi-objective optimization problem with the goal of maximizing comprehensive benefits, considers the maximization of prevention and control effectiveness, minimization of economic cost, and minimization of environmental impact, generates a Pareto optimal solution set through an algorithm and selects an optimal prevention and control prescription, and clearly defines the resource type, application dose, and operation time window; this process avoids the situation of single pursuit of prevention and control effect while ignoring cost waste or environmental damage, ensures effective pest control, reduces orchard operation cost, reduces the use of high-pollution resources, meets the needs of green agriculture and sustainable development, and meets the needs of green agriculture and sustainable development; The optimal prevention and control prescription is precisely matched with real-time inventory and the spatial location of agricultural machinery. The resource demand of the grid is calculated to achieve on-demand allocation. When the inventory is insufficient, an early warning is triggered and the demand of high-risk grids is prioritized to avoid resource idleness or shortage. At the same time, the optimal operation path of agricultural machinery is generated based on the orchard electronic map and operation points. The path is planned with the goal of minimizing the operation distance or time, reducing the ineffective driving of agricultural machinery, reducing energy consumption and time costs, improving the overall efficiency of resource scheduling and agricultural machinery operation, and avoiding the arbitrariness of traditional manual allocation and operation planning. By visually displaying risk indices, prescriptions, scheduling plans, and agricultural machinery routes through electronic maps, managers can intuitively grasp the overall situation. The generated control command set is sent to the intelligent agricultural machinery and warehousing system through the Internet of Things to ensure precise implementation in the execution process. At the same time, it receives real-time feedback on operation status and inventory updates, dynamically updates and displays the data to form a closed-loop management system. If problems such as agricultural machinery failure or inventory abnormalities occur, the plan can be adjusted in time and instructions can be reissued to avoid the continuous impact of execution deviations on the prevention and control effect, thereby improving the controllability and flexibility of the entire orchard pest and disease management process. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the steps in the orchard pest and disease early warning and control resource management decision support method based on multi-source data, as described in this invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0014] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0015] like Figure 1 As shown, this embodiment of the invention provides a decision support method for orchard pest and disease early warning and control resource management based on multi-source data, including the following steps: Step S1, Data Perception and Fusion: Acquire multiple source data from the orchard, perform spatiotemporal alignment and standardization on the multiple source data from the orchard, and construct a spatiotemporal fusion dataset with the orchard geographic grid as the unit; Step S2, Pest and Disease Risk Quantification: Input the spatiotemporal fusion dataset into the pest and disease prediction model. The pest and disease prediction model is a multi-classification model built based on machine learning algorithms and characterized by environmental factor time series data and historical pest and disease occurrence patterns. The output of the pest and disease prediction model is a risk quantification index for a specific pest and disease within a preset future time period. The risk quantification index is a value that combines the probability of occurrence and the degree of potential harm. Step S3, prevention and treatment prescription generation: trigger the prevention and treatment decision engine based on the risk quantification index; the prevention and treatment decision engine is built-in with a prevention and treatment resource knowledge base, which records the prevention and treatment efficiency coefficient, economic cost and environmental impact factor of different prevention and treatment resources under different risk levels; the prevention and treatment decision engine takes the maximization of comprehensive benefits as the goal, and obtains the optimal prevention and treatment prescription matched with the current risk level by solving the multi-objective optimization algorithm, the optimal prevention and treatment prescription at least includes the recommended prevention and treatment resource type, application dose, operation time window; Step S4, resource-space matching and scheduling: matching the optimal prevention and treatment prescription with the real-time inventory data of prevention and treatment resources in the orchard and the spatial position information of operation equipment to generate specific resource scheduling scheme and agricultural machinery operation path planning; the resource scheduling scheme includes the allocation quantity of the required resources from the warehouse to the specific orchard grid, and the agricultural machinery operation path planning is based on the orchard electronic map and the operation point generated by the prescription for path optimization calculation; Step S5, decision visualization and instruction issuing: integrating the risk quantification index, the optimal prevention and treatment prescription, the resource scheduling scheme and the agricultural machinery operation path planning, and superimposing and rendering them on the electronic map of the management terminal for visual display, and generating a set of controllable control instructions for driving the intelligent agricultural machinery to execute the prevention and treatment prescription.

[0016] In this embodiment, the various source data of the orchard includes real-time environmental data collected through Internet of Things devices, fruit tree canopy image data obtained through image collection devices, and macro weather data and historical pest data accessed from external systems; The core of step S1 is to obtain multi-source data of the orchard and perform spatio-temporal alignment and standardization processing, and finally build a spatio-temporal fusion data set with the orchard geographic grid as the unit, to provide spatio-temporally consistent and high-quality basic data support for subsequent pest risk quantification, which specifically includes the following sub-steps: S1-1, multi-source data standardization conversion: Obtain the real-time environmental data of the orchard collected through the Internet of Things devices, which includes temperature, humidity, light intensity, soil moisture content and other indicators, extract data points at fixed time intervals, assign a unified format timestamp to each data point, and associate the geographic coordinates of the collection device within the orchard range to form a real-time environmental standardized data sequence; Obtain the fruit tree canopy image data obtained through the image collection device, perform geometric correction on the image to accurately correspond the image pixel position to the actual geographic coordinates of the orchard, add a shooting timestamp to each image, and convert the image to a universal data format to form a fruit tree canopy image standardized data sequence; Acquire macro-meteorological data from external systems, which includes information such as regional rainfall, wind speed, wind direction, and temperature. Filter the corresponding data according to the orchard's location, unify the data time granularity to be consistent with real-time environmental data, add a unified timestamp to the data, and associate it with the geographical coordinate range of the orchard's location to form a standardized macro-meteorological data sequence. Acquire historical pest and disease data of the orchard, which includes information such as the types of pests and diseases, the time of occurrence, the location of occurrence, the degree of damage, and control records. Organize the time information of each record and convert it into a unified format timestamp. Clarify the specific location of the orchard corresponding to each record and associate it with geographical coordinates to form a standardized data sequence of historical pests and diseases. S1-2, Geographic Grid Division and Data Linkage: By calling up the electronic map of the orchard, and combining the actual area of ​​the orchard, terrain features, and fruit tree planting distribution, the geographic grid rules are preset to generate a geographic grid system covering the entire orchard on the electronic map, and each geographic grid is assigned a unique identifier. The standardized real-time environmental data sequence, the standardized fruit tree canopy image data sequence, the standardized macro-meteorological data sequence, and the standardized historical pest and disease data sequence, which have undergone unified preprocessing, are spatially matched with the geographic grid system to determine the geographic grid to which the geographic range corresponding to each data point belongs. According to the geographic grid identifier, various standardized data that match the same geographic grid are integrated and associated to form a multi-source initial dataset within each orchard geographic grid unit. Each dataset contains real-time environmental data, fruit tree canopy image data, macro meteorological data, and historical pest and disease data corresponding to that grid unit. S1-3, Temporal normalization of multi-source initial datasets: For the multi-source initial dataset within each orchard geographic grid unit, the integrity of each data category is checked according to the time dimension. The data missingness of each type of data within the preset time range is statistically analyzed, and the time location and corresponding data type of the missing data points are determined. For data points with missing values, time-series interpolation methods are used to impute the data: if the missing data are a few discontinuities in a continuous time series, linear interpolation is used to calculate the value of the missing points; if the missing data are continuous missing over a long period of time, spline interpolation is used to construct a smooth curve to fit the existing data points and estimate the data within the missing time period. Anomaly fluctuation detection is performed on standardized data sequences of various types of data. The mean and standard deviation of the data sequence are calculated, and an outlier judgment threshold is set. Data that exceeds the range of mean plus or minus 3 times the standard deviation are judged as outliers. For data sequences with abnormal fluctuations, smoothing filtering is used: for continuous numerical data such as real-time environmental data and macro weather data, the moving average filtering method is used to replace the current data point with the average of adjacent data points; for non-continuous data such as orchard canopy image data and historical pest data, the median filtering method is used to replace the abnormal data point with the median of adjacent data points, and finally the regular time series data at the grid cell level is generated; S1-4, feature layer superposition and spatio-temporal fusion data set construction: From the regular time series data of each grid cell after filling and smoothing, multi-dimensional features are extracted: from environmental factor data, original time series features of temperature, humidity, and light intensity are extracted; from the orchard canopy image data, vegetation index and pest visual features are extracted through image recognition algorithm, where the vegetation index includes normalized vegetation index and green leaf area index, and the pest visual feature includes pest area ratio and pest number estimation; from macro weather data, grid-based rainfall, wind speed, wind direction, and temperature time series features are extracted; from historical pest data, occurrence frequency, occurrence intensity, and occurrence cycle of various pests in the grid cell are extracted; Align the environmental factor features, vegetation index features, pest visual features, macro weather element features, and historical pest occurrence regularity features extracted in the same grid cell according to the time dimension to ensure that different types of features are correlated at the same timestamp; Integrate the multi-dimensional features after time alignment, combine the multi-source features of each grid cell at different time points into a spatio-temporal feature set of the grid cell, and finally form a spatio-temporal fusion data set with orchard geographical grid as the basic unit.

[0017] In this embodiment, the function of step S2 is to perform deep processing on the spatio-temporal fusion data set through a multi-classification model based on machine learning algorithm, to realize risk quantification of specific pests in a future preset time period, and to accurately output the risk quantification index of comprehensive occurrence probability and potential harm degree, to provide scientific and accurate risk basis for subsequent prevention and control decision engine triggering and optimal prevention and treatment prescription generation, to ensure the pertinence and foresight of prevention and control measures; the following are the detailed steps: S2-1, sample division and input sample set construction of spatio-temporal fusion data set: Determine the sample time and space reference, first determine the sample time granularity, such as setting the historical time points by day or by week, and then take the orchard geographical grid as the spatial unit to ensure the clear spatio-temporal range of each sample; Divide the sample type, divide the spatio-temporal fusion data set into training set, validation set and test set in the ratio of 7:2:1, and keep the distribution of different orchard geographical grids and different historical time period data balanced in each set during the division process to avoid sample bias; Constructing a single sample structure, each sample needs to contain two parts of core content, one is the environmental factor time series data corresponding to the orchard geographic grid at a certain historical time point, covering the temperature, humidity, light intensity, rainfall and other environmental index time series sequence for a continuous preset length (such as 7 days) before and after the time point; The second is the record of the actual occurrence of diseases and pests in the orchard geographic grid within a preset period of time (such as 14 days) after the historical time point, including disease and pest species, occurrence intensity, damage range and other information, forming a labeled supervised learning sample; Completing sample labeling and storage, standardizing the record of the occurrence of diseases and pests for each sample, such as using the value 0 to represent no occurrence, 1 to represent mild occurrence, 2 to represent moderate occurrence, and 3 to represent severe occurrence, and then storing the labeled samples according to the training set, validation set and test set to provide structured data support for model training and validation; S2-2, multi-dimensional feature extraction and feature vector construction: Extracting multi-dimensional features of environmental factor time series data, extracting three types of core features from the environmental factor time series data of each sample; statistical features include mean, variance, maximum, minimum, and median of environmental factors within a preset length, such as daily average temperature and temperature variance within 7 days; Periodic features are extracted by the STL time series decomposition method, such as the weekly fluctuation of temperature; Trend features are obtained by calculating the linear regression slope or the growth rate of the sliding window, such as the humidity growth rate for 3 consecutive days; Extracting historical disease and pest occurrence regularity features, for each orchard geographic grid, retrieve its historical disease and pest data for the past 3-5 years, extract the disease and pest occurrence frequency, average occurrence intensity, and adjacent year occurrence probability of the grid in the same time period (such as June to July each year) as the current sample, forming a historical regularity feature set; Constructing a complete feature vector, integrating the multi-dimensional features of the environmental factor of the same sample with the historical disease and pest occurrence regularity features of the corresponding orchard geographic grid, removing redundant features (such as selecting features with a variance greater than a preset threshold through variance analysis), and finally forming a fixed-dimensional feature vector, ensuring that each sample corresponds to a structurally unified feature vector, meeting the model input requirements; S2-3, multi-classification model training based on machine learning: Selecting a model algorithm, selecting a machine learning algorithm suitable for handling complex nonlinear relationships to construct a multi-classification model, such as gradient boosting tree (XGBoost), random forest or support vector machine, and preferentially selecting XGBoost algorithm to improve the model's ability to handle high-dimensional features and prediction accuracy; Model training is carried out, the feature vector of the training set is input into the multi-classification model, the occurrence of the sample labeled pest and disease is taken as the supervision label, the cross-entropy loss function is taken as the model optimization objective, and the model parameters are adjusted through iterative training, such as setting the decision tree depth to 3-10, setting the learning rate to 0.01-0.1, and setting the regularization coefficient to 0.1-1, minimizing the error between the model prediction value and the actual label; Model hyperparameter optimization is performed, and the hyperparameters of the model are optimized using the validation set. Grid search or random search method is used to traverse the preset hyperparameter combination, and the F1 score of the validation set is taken as the evaluation index to select the hyperparameter combination with the highest F1 score to ensure the optimal performance of the model. The generalization ability of the model is verified, the feature vector of the test set is input into the optimized model, and the accuracy, recall rate and F1 score of the model on the test set are calculated. If all three indicators are greater than the preset threshold (such as accuracy ≥ 0.85, recall rate ≥ 0.8, F1 score ≥ 0.82), the generalization ability of the model meets the standard, and the training is completed. If it does not meet the standard, adjust the feature extraction method or hyperparameter combination and retrain until it meets the standard. S2-4, current data prediction and risk quantification index calculation: The current spatio-temporal fusion data is processed to obtain the spatio-temporal fusion data of each geographical grid in the orchard at the current time. According to the feature extraction method of step S2-2, the same dimension of multi-dimensional features and historical pest and disease occurrence regularity features are extracted to construct the feature vector of the current data, ensuring that the feature vector structure is consistent with the model training. Output the probability of pest and disease occurrence. The feature vector of the current data is input into the trained multi-classification model, and the model outputs the probability of occurrence of a specific pest and disease in each geographical grid of the orchard within a preset time period (such as 14 days). The probability value ranges from 0 to 1, and the larger the value, the higher the probability of occurrence. Calculate the risk quantification index. The risk quantification index is calculated using the formula RI = P x W (where RI is the risk quantification index, the value range is 0 to 10; P is the probability of occurrence of a specific pest and disease output by the model, the value range is 0 to 1; W is the preset damage weight coefficient of the pest and disease, which is set according to the influence of the pest and disease on the yield and quality of the fruit tree, such as setting the damage weight coefficient of the borer W to 10 for severe damage and setting the damage weight coefficient of the leaf spot disease W to 5 for light damage). Output the grid risk data. The risk quantification index of each geographical grid of the orchard is sorted to form grid risk distribution data, and the risk quantification results of each grid are determined to provide accurate numerical basis for the risk level determination and prevention decision triggering in subsequent step S3.

[0018] In this embodiment, the role of step S3 is to trigger the prevention and control decision engine based on the pest risk quantification index output in step S2, combine the built-in prevention and control resource knowledge base, and solve the optimal solution considering the prevention and control effectiveness, economic cost and environmental impact through a multi-objective optimization algorithm to generate the optimal prevention and control prescription matching the current risk level, clearly recommend the type of recommended prevention and control resources, application dose and operation time window, provide specific execution basis for subsequent orchard resource scheduling and agricultural machinery operation, and ensure that the prevention and control measures are scientific, efficient and meet the requirement of maximizing comprehensive benefits. The following are the detailed steps: S3-1, risk level determination and prevention and control decision engine triggering: Retrieve the preset risk level threshold. The system is built-in with the risk level threshold developed based on the historical prevention and control experience of orchards and industry standards, for example, risk quantification index RI < 3 is low risk, 3 ≤ RI < 7 is medium risk, and RI ≥ 7 is high risk. Different pest types can correspond to different thresholds (for example, the high risk threshold for borer is set to RI ≥ 6, and the high risk threshold for leaf spot disease is set to RI ≥ 8). Match the risk quantification index with the threshold. Compare the risk quantification index of each orchard geographic grid generated in step S2 with the risk level threshold of the corresponding pest one by one to determine the specific risk level of each grid. For example, if the RI of a certain grid for borer is 7.2, it is determined as high risk. Trigger the prevention and control decision engine of the corresponding level. The low risk level triggers the lightweight decision process (only filters low environmental impact and low cost prevention and control resources), and the medium and high risk levels trigger the complete decision process (comprehensively calls the knowledge base and optimization algorithm) to ensure that the response intensity of the decision engine is adapted to the risk level of the pest, avoiding resource waste or insufficient prevention and control. S3-2, candidate prevention and control resource set retrieval: Retrieve the core data of the prevention and control resource knowledge base. The knowledge base stores the key attributes of various prevention and control resources, including resource type (biological pesticides, chemical pesticides, physical prevention and control equipment, natural enemy insects, etc.), prevention and control effectiveness coefficient under different risk levels (for example, the effectiveness coefficient of a certain biological pesticide for aphids under medium risk is 0.85, and the larger the value, the better the prevention and control effect), unit area economic cost (for example, the application cost of a certain chemical pesticide per hectare is 300 yuan), and environmental impact factor (for example, the environmental impact factor of a certain physical prevention and control equipment is 0.1, and the smaller the value, the smaller the disturbance to the ecological environment). Filter the adaptive resources according to the risk level. The prevention and control decision engine filters the adaptive prevention and control resources in the knowledge base according to the risk level of the current grid. For example, under the low risk level, only biological pesticides, physical traps and other low pollution and low cost resources are retrieved. Under the high risk level, comprehensive prevention and control resources including high efficiency chemical pesticides and biological pesticide combinations, natural enemy insect release schemes, etc. are retrieved. Sort the candidate prevention and control resource set, classify the screened prevention and control resources by type, record the performance coefficient, economic cost and environmental impact factor of each resource under the current risk level, form a candidate prevention and control resource set, for example, the candidate set of medium-risk aphids includes biological pesticides A, chemical pesticides B, physical traps C and corresponding attribute data; S3-3, multi-objective optimization problem construction: Determine the comprehensive benefit objective function, take the maximization of comprehensive benefit as the core, and construct three interrelated sub-objective functions: Maximize prevention and control performance: Let the number of candidate resources be n, the performance coefficient of the kth resource be E k , and the resource usage ratio be x k . The performance objective function is MaxF1=Σ(E k ×x k ) (where F1 is the total prevention and control performance, E k is the performance coefficient of the kth resource, x k is the usage ratio of the kth resource, and n is the number of candidate resources). Minimize economic cost: Let the unit area cost of the kth resource be C k . The cost objective function is MinF2=Σ(C k ×x k ) (where F2 is the total economic cost per unit area, C k is the unit area cost of the kth resource, x k is the usage ratio of the kth resource, and n is the number of candidate resources). Minimize environmental impact: Let the environmental impact factor of the kth resource be En k . The environmental objective function is MinF3=Σ(En k ×x k ) (where F3 is the total environmental impact factor, En k is the environmental impact factor of the kth resource, x k is the usage ratio of the kth resource, and n is the number of candidate resources). Set optimization constraints, including resource usage ratio constraints (x k ≥ 0 and Σx k = 1 to ensure complete resource combination and no negative proportion usage), resource usage limits (such as x k ≤ 0.3 for high-toxicity chemical pesticides under low risk to avoid excessive use), and performance bottom line constraints (F1 ≥ 0.6 to ensure that the prevention and control effect does not fall below the basic requirement). Integrate to form a multi-objective optimization problem, combine the three sub-objective functions and constraints, and clearly define the solution direction of the problem as finding the resource usage ratio combination that maximizes F1, minimizes F2, and minimizes F3 under the premise of meeting the constraints. S3-4, Multi-objective optimization algorithm solving and Pareto optimal solution set generation: Select a multi-objective optimization algorithm, use the non-dominated sorting genetic algorithm (NSGA-II) for solving. This algorithm can efficiently handle multi-objective conflict problems, distinguish the quality of the solution through fast non-dominated sorting, and maintain the diversity of the solution set through congestion calculation, making it suitable for the optimization scenario of orchard prevention resource combination; Configure algorithm parameters, set the population size to 60-100 (ensure that the search range covers enough resource combinations), the number of iterations to 40-60 (balance solution accuracy and efficiency), the crossover probability to 0.8 (promote high-quality gene combinations), and the mutation probability to 0.02 (avoid the algorithm falling into local optimum); Execute algorithm iteration solving, input the constructed multi-objective optimization problem into the NSGA-II algorithm, generate random resource usage ratio combinations through population initialization, and gradually filter out non-dominated solutions (i.e., solutions that cannot improve one objective without degrading another) through iteration operations such as selection, crossover, and mutation; Generate Pareto optimal solution set, aggregate all non-dominated solutions in the iteration process, and form the Pareto optimal solution set after removing duplicate solutions. Each solution corresponds to a specific resource usage ratio combination, such as solution 1 with biological pesticide A usage ratio 0.7 and physical trap C usage ratio 0.3, and solution 2 with chemical pesticide B usage ratio 0.5 and biological pesticide A usage ratio 0.5. Each solution is labeled with corresponding F1, F2, and F3 values; S3-5, Optimal prevention prescription selection and generation: Execute the preset decision rule to filter the optimal solution. The system has built-in decision rules to balance actual application needs, such as the rule to prioritize solutions with F1 ≥ 0.7, F2 ≤ 1.1 times the average cost of all solutions, and F3 ≤ 0.4. If there are multiple solutions that meet the conditions, select the solution with the largest F1. Traverse the Pareto optimal solution set and exclude solutions that do not meet the conditions, such as a solution with F1 = 0.68 (lower than 0.7) is excluded. Finally, determine the unique optimal solution; Calculate specific prevention resource parameters. According to the resource usage ratio of the optimal solution, calculate the resource usage based on the area of the orchard geographic grid (e.g., a grid area of 1.2 hectares). For example, the optimal solution is biological pesticide A usage ratio 0.6 and chemical pesticide B usage ratio 0.4. The recommended usage of biological pesticide A is 1500 ml / ha, and the recommended usage of chemical pesticide B is 800 ml / ha. Therefore, the grid needs 1.2 × 1500 × 0.6 = 1080 ml of biological pesticide A and 1.2 × 800 × 0.4 = 384 ml of chemical pesticide B; Determine the operation time window, retrieve the optimal application period of the resources contained in the optimal solution from the prevention and control resource knowledge base (such as the optimal application period of biological pesticide A is 6-9 am to avoid the volatilization of drug efficacy caused by high temperature; chemical pesticide B needs to avoid rain to avoid liquid loss), and combine the future 72-hour weather data accessed by the external system (such as predicting no rain in the morning of the next day, and the temperature is 18-22°C) to determine the specific operation time range (such as 6:30-8:30 the next day); Integrate to generate the optimal prevention and control prescription, integrate the screened prevention and control resource types (biological pesticide A, chemical pesticide B), the calculated application dose (1080 ml, 384 ml), and the determined operation time window (6:30-8:30 the next day) to form a structured optimal prevention and control prescription, ensuring that each parameter is clear and can be directly used for subsequent resource scheduling and agricultural machine operation.

[0019] In this embodiment, the role of step S4 is to link the optimal prevention and control prescription generated in step S3 with the decision visualization and instruction issuing of step S5. By accurately matching the optimal prevention and control prescription with the real-time inventory of orchard prevention and control resources and the spatial position of operation equipment, a feasible resource scheduling scheme and agricultural machine operation path planning are generated, ensuring that prevention and control resources are allocated as needed, agricultural machines operate efficiently, and resources are not wasted or operation is not delayed, providing specific operation basis for subsequent intelligent execution links. The following are the detailed steps: S4-1, Core data acquisition and integration: Retrieve optimal prevention and control prescription data, extract the optimal prevention and control prescription for all orchard geographic grids generated in step S3 from the system database, clearly define the required prevention and control resource types (such as biological pesticide A, chemical pesticide B, physical trap C), unit area application dose, operation time window, and other key information for each grid, and classify and organize the prescription list by grid number; Access real-time inventory data of prevention and control resources, and through the interface with the orchard resource warehouse management system, obtain the current inventory of various types of prevention and control resources in the warehouse, the storage partition position (such as biological pesticides are stored in warehouse area 1 and chemical pesticides are stored in area 2), resource shelf life (to avoid allocation of expired resources), and other data to form a real-time inventory list; Collect operation equipment spatial position and state data, and through the Internet of Things positioning module (such as GPS or Beidou positioning) installed on agricultural machines, obtain the real-time spatial coordinates, working state (such as idle, operating, fault), equipment type (matching different prevention and operation needs), and operable time length (based on fuel or power) of all operating agricultural machines (such as spraying machines and trap laying machines) to form a list of agricultural machine state and position; Integrate the three types of core data, associate the prescription list, real-time inventory list, and agricultural machine state and position list by orchard geographic grid number to ensure that the resource demand, inventory supply, and agricultural machine matching data of each grid correspond one by one, laying a foundation for subsequent matching calculation; S4-2, Calculation of Orchard Geographic Grid Resource Demand: Determine the basic grid parameters and extract the actual planting area S of each geographic grid from the orchard electronic map, in hectares; based on the grid pest and disease risk level determined in step S2, set the risk distribution coefficient K (K is 1.0 for high-risk grids, 0.9 for medium-risk grids, and 0.8 for low-risk grids to ensure that resource allocation is appropriate for the risk level). Calculate the single-resource requirement for a single grid. For each type of required control resource in each grid, based on the unit area application dose D (in ml / ha or kg / ha) in the optimal control prescription, calculate the demand Q of that grid for that type of resource using the formula Q=S×D×K (where Q is the demand of a single orchard geographic grid for a certain type of control resource, S is the actual planting area of ​​a single orchard geographic grid, D is the unit area application dose of a certain type of control resource in the optimal control prescription, and K is the pest and disease risk distribution coefficient of a single orchard geographic grid). Generate a complete list of resource requirements for the entire grid. Perform the above calculations on each of the orchard geographic grids and calculate the total demand for each type of prevention and control resource (the sum of Q values ​​for all grids). This will form a complete demand list containing grid number, resource type, and demand quantity, which will serve as the basis for inventory matching. S4-3, Inventory Matching Verification and Resource Allocation Plan Generation: Perform inventory and demand matching verification, compare the total list of resource demands across the entire grid with the real-time inventory list one by one, and for each type of prevention and control resource, determine whether its total inventory meets the total demand; if the total inventory is greater than or equal to the total demand, it is determined that the inventory is sufficient; if the total inventory is less than the total demand, it is determined that the inventory is insufficient, and a resource replenishment warning is immediately triggered and pushed to the management personnel terminal through the system message, specifying the type of warning resource, the total demand gap, and the gap ratio of high-risk grids; Establish resource allocation rules. When inventory is sufficient, allocate resources according to demand. The allocation amount of a certain type of resource to a single grid is Q_point = Q × (total inventory / total demand) (where Q_point is the allocation amount of a certain type of prevention and control resource in a single orchard grid, Q is the demand for a certain type of prevention and control resource in a single orchard grid, the total inventory is the current inventory of that type of prevention and control resource in the warehouse, and the total demand is the sum of the demand for that type of prevention and control resource from all orchard grids). This ensures that each grid receives resources according to its demand ratio. When inventory is insufficient, allocate existing inventory to high-risk grids (allocation amount not less than 90% of their demand) according to the risk priority principle, then allocate it to medium-risk grids (allocation amount not less than 70% of their demand), and postpone allocation to low-risk grids until resources are replenished. Generate a resource allocation plan, specify the allocation details of each type of resource from the warehouse to each grid (grid number, allocation amount, corresponding warehouse storage partition), transportation scheduling arrangement (transport vehicle number, departure time, estimated arrival time of the grid, transportation route), and the replenishment requirements of insufficient resources (the type of resources needed to be replenished, the gap amount, the recommended procurement channel, and the estimated replenishment time to the warehouse), forming a structured resource allocation plan document; S4-4, agricultural machinery operation path planning: Mark the operation points and match the agricultural machinery. On the orchard electronic map, mark the operation points (such as the center of the planting row of fruit trees in each grid, the center of the high-incidence area of pests and diseases, and the number of operation points in each grid is determined according to the planting density, usually one operation point per 0.1 hectares) according to the operation requirements of each grid in the optimal control prescription. According to the type of agricultural machinery and the operation requirements, screen the idle and suitable agricultural machinery, and divide the responsible area of each agricultural machinery according to the principle of geographical proximity (such as No. 1 spray machine responsible for 5 adjacent grids in the east of the orchard, No. 2 spray machine responsible for 4 adjacent grids in the west), to avoid increasing the travel distance of cross-regional operation; Select path optimization algorithm and target. Adopt path optimization algorithm (such as ant colony algorithm, which is good at solving the optimal path problem of multi-node traversal, suitable for complex orchard terrain), set the optimization target as the shortest total distance or the minimum total time of operation (if the operation time window is tight, the time target is preferred, otherwise the distance target is preferred); The input parameters of the algorithm include the real-time position coordinates of the agricultural machinery, the coordinates of all operation points in the responsible area, the coordinates of obstacles in the orchard (such as irrigation channels, roads, buildings, which need to be avoided), and the travel speed of the agricultural machinery (according to the orchard terrain, 5 kilometers per hour in flat areas, 3 kilometers per hour in slope areas); Calculate the optimal operation path. The algorithm initializes at the starting position of the agricultural machinery, and traverses all operation points according to the rule of closer distance, higher priority selection probability. In the iteration process, the path selection is optimized (superior path is preferentially retained, inferior path is gradually eliminated), until the iteration times reach the preset value (usually set to 50 times), and the optimal order and path coordinate sequence of each agricultural machinery traversing all operation points in its responsible area are output. For example, No. 1 spray machine starts from position X1Y1, and then passes through operation point AX2Y2, operation point CX3Y3, and operation point BX4Y4, and finally returns to the starting position, forming a complete path planning, and marking the distance and estimated travel time of each path segment; S4-5, integration of resource scheduling scheme and agricultural machinery operation instruction: Integrate resource scheduling and path planning data, associate the resource allocation plan generated in step S4-3 with the agricultural machinery operation path planning generated in step S4-4, ensure that the operation path of each agricultural machine covers the grid of the allocated resource in the responsible area, and the resource transportation time and agricultural machine operation time are matched (such as the agricultural machine needs to start operation within 1 hour after the resource arrives at the grid); Generate a resource scheduling scheme, which includes resource allocation details (resource type, allocation amount, warehouse delivery time by grid number), transportation scheduling table (transportation vehicle number, transportation route, grid resource arrival time), and inventory dynamic update record (update inventory data in real time after each batch of resources is delivered), to ensure that the management personnel clearly understand the whole process of resource flow; Generate agricultural machinery operation instructions, which include path navigation information (agricultural machinery number, optimal path coordinate sequence, turning prompt, obstacle avoidance point), operation parameter setting (such as the spray pressure, spray volume, and operation height of the spraying agricultural machinery, which are adjusted according to the application dose in the optimal control prescription, for example, 1500 ml / ha of application dose corresponds to 300 ml / min of spray volume), and operation time requirement (need to be completed within the operation time window of the optimal control prescription, with the latest start time and the earliest end time marked), to ensure that the agricultural machinery operator can directly execute the operation according to the instructions; Store and pre-check, store the integrated resource scheduling scheme and agricultural machinery operation instructions in the system core database, and at the same time, check whether the resource allocation amount is consistent with the grid demand, whether the agricultural machinery path covers all operation points, and whether the operation time is within the window through the built-in verification module of the system, if there is inconsistency, return to the corresponding step for correction, to ensure the accuracy of the scheme and the instructions.

[0020] In this embodiment, the function of step S5 is to integrate the risk quantification index of step S2, the optimal control prescription of step S3, the resource scheduling scheme of step S4, and the agricultural machinery operation path planning, and to visually display the decision information through electronic map visualization to make it intuitive and searchable, and to generate and issue control instructions to drive intelligent agricultural machinery and warehouse systems to execute operations, and to dynamically update the status through real-time feedback to form a closed-loop management of decision-making, execution, and feedback, to ensure that the control decision is efficient, controllable, and adjustable; the following are the detailed steps: S5-1, decision-making integrated data construction: Determine the integrated data range, and specify the data to be integrated, including the pest risk quantification index of each orchard geographic grid generated in step S2 (including grid number, pest type, and risk value), the optimal control prescription output in step S3 (including grid number, control resource type, application dose, and operation time window), the resource scheduling scheme formed in step S4 (including grid number, resource allocation amount, transportation information, and inventory status), and the agricultural machinery operation path planning (including agricultural machinery number, responsible grid, operation point coordinates, path sequence, and expected operation time length); Establish data association rules, and use the orchard geographic grid number as the core association key to bind the risk quantification index, the optimal control prescription, and the resource scheduling information of the same grid. Use the agricultural machinery number as the auxiliary association key to associate the agricultural machinery operation path planning with the resource scheduling and control prescription information of the corresponding responsible grid, so as to ensure that the data from different sources are consistent in the time and space dimensions. Generate integrated control decision data, and organize the data in a grid-resource-machinery three-layer structure. Each layer contains attribute information and associated data (for example, the grid layer contains risk index and prescription, the resource layer contains allocation amount and transportation, and the machinery layer contains path and operation requirements), forming a structured data set to provide a unified data basis for subsequent visualization and instruction generation. S5-2, visualization of decision information electronic map: Call the electronic map service engine to start the electronic map service of the management terminal, load the high-precision electronic map of the orchard (including grid boundaries, fruit tree planting areas, warehouse locations, roads, and obstacle areas), set map zooming, panning, layer control, and other interactive functions, and support management personnel to view details of different areas. Implement multi-type data overlay rendering, and use differentiated display methods according to data types: Risk quantification index display: use heat map rendering to map risk values to color gradients (for example, low risk with green, medium risk with yellow, and high risk with red, with darker colors indicating higher risk), and overlay them on the corresponding orchard geographic grid. Mouse-over the grid to display the specific risk value and pest type. Optimal control prescription display: use a pop-up window association method to label the prescription icon at the center of the corresponding grid, click the icon to pop up a pop-up window, and display the control resource type, application dose, operation time window, and control efficiency coefficient of the resource. Resource scheduling scheme display: use material flow chart rendering to connect the warehouse and the target grid with lines with arrows. The line thickness corresponds to the resource allocation amount (the larger the allocation amount, the thicker the line). The line is labeled with the transportation vehicle number and the expected arrival time. The warehouse location is labeled with the current inventory status (sufficient in blue and insufficient in orange). The agricultural machine operation path planning display: dynamic icons and path lines are used for rendering, different color icons are used to distinguish agricultural machine types on the map (such as blue triangles for spraying machines and green circles for trap setter machines), the icon position is synchronized with the current position of the agricultural machine in real time, the planned path (including the sequence of operation points, which are marked with dots) is marked with a solid line, the expected operation time and the remaining operation time are marked on the path, and the path animation simulation operation process is supported; S5-3, control instruction set generation can be issued: Determine the instruction generation trigger condition. If the system is in automatic mode, when the integrated data of the control decision passes the pre-verification (such as sufficient resource inventory, idle agricultural machine, and non-expired operation time window), the instruction generation is automatically triggered. If it is in manual mode, the generation process is started after receiving the management personnel's confirmation instruction (clicking the instruction generation button of the management terminal); Build the instruction content framework. The control instruction set is divided into two categories: Intelligent agricultural machine control instruction: including agricultural machine number, navigation control instruction (operation point coordinate sequence, path planning data, turning prompt, obstacle avoidance point), operation parameter setting instruction (application dose, spraying pressure, operation speed, operation start / end time); Resource warehouse system instruction: including warehouse number, warehouse dispatching instruction (resource type, warehouse quantity, corresponding grid number, transportation vehicle matching information, warehouse time window), inventory updating instruction (warehouse data record requirement after warehouse dispatching); Execute instruction verification. The system built-in verification module checks the completeness of the instruction content (such as whether the agricultural machine number and resource type are missing) and the logical consistency (such as whether the operation time is within the prescription window and whether the warehouse quantity is less than the inventory), and returns to the previous step for correction if there is an exception. The final control instruction set is generated after the verification is passed; S5-4, control instruction set Internet of Things delivery: Establish the Internet of Things communication connection, start the Internet of Things communication link (using LoRa or 4G / 5G communication protocol to ensure low delay and high reliability) of the management terminal, intelligent agricultural machine terminal, and resource warehouse control system, and verify the online state of the equipment (display the online icons of agricultural machines and warehouse systems, and mark the offline equipment with a red warning); Execute instruction batch delivery, sort by operation priority (high-risk grid corresponds to priority of agricultural machine and resource instruction), and transmit the control instruction set to the corresponding device terminal: the intelligent agricultural machine terminal receives the navigation and operation parameter instruction, stores it in the local instruction library, and returns the instruction reception success confirmation message; the resource warehouse control system receives the warehouse and inventory instruction, triggers the warehouse management module to prepare the warehouse operation, and returns the warehouse preparation completion message; The record instruction is issued, and the log is stored in the system database. The instruction issuing time, receiving device number, instruction content, confirmation message receiving time and other information are stored to form a traceable instruction issuing record, which is convenient for subsequent problem troubleshooting; S5-5, job and inventory state feedback and closed-loop update: Real-time feedback information is received, and the job state information (including current job position, completed job point number, job parameter execution situation, and device fault alarm) returned by the intelligent agricultural machine terminal and the inventory update information (including the type / quantity of resources out of the warehouse, the out-of-warehouse time, the current remaining inventory, and the information of the replenished resources into the warehouse) fed back by the resource storage system are continuously received through the Internet of Things communication network. Dynamic update and display of feedback information are realized, and the intelligent agricultural machine icon is moved to the current job position on the management terminal electronic map in real time. The completed job points are marked as green check marks, and the uncompleted points are marked as gray dots. The fault agricultural machine is marked with a red fault icon and pops up the fault type. The inventory state of the warehouse position is updated in real time (the remaining inventory value is updated, and the insufficient state is converted to sufficient if replenished into the warehouse), and the completed transportation line is marked as delivered in the resource flow diagram. A closed-loop management is formed, and the management personnel judge the job progress and resource state based on real-time feedback information: if the job is normally completed and the inventory is updated accurately, the current prevention and control decision-making process is closed loop; if there is an agricultural machine fault or inventory anomaly, the system automatically triggers the adjustment mechanism (such as dispatching a standby agricultural machine to replace the faulty device, and extending the replenishment warning of the insufficient resource), updates the resource scheduling scheme and agricultural machine job instruction, and reissues until the job is completed, ensuring that the entire process is continuously controllable.

[0021] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A decision support method for orchard pest and disease early warning and control resource management based on multi-source data, characterized by: Includes the following steps: Step S1, Data Perception and Fusion: Acquire multiple source data from the orchard, perform spatiotemporal alignment and standardization on the multiple source data from the orchard, and construct a spatiotemporal fusion dataset with the orchard geographic grid as the unit; Step S2, Pest and Disease Risk Quantification: Input the spatiotemporal fusion dataset into the pest and disease prediction model. The pest and disease prediction model is a multi-classification model built based on machine learning algorithms and characterized by environmental factor time series data and historical pest and disease occurrence patterns. The output of the pest and disease prediction model is a risk quantification index for a specific pest and disease within a preset future time period. The risk quantification index is a value that combines the probability of occurrence and the degree of potential harm. Step S3, Prevention and Treatment Prescription Generation: Trigger the prevention and treatment decision engine based on the risk quantification index; The prevention and control decision engine has a built-in prevention and control resource knowledge base, which records the prevention and control effectiveness coefficients, economic costs and environmental impact factors of different prevention and control resources under different risk levels. The prevention and control decision engine aims to maximize comprehensive benefits and uses a multi-objective optimization algorithm to obtain the optimal prevention and control prescription that matches the current risk level. The optimal prevention and control prescription includes at least the recommended prevention and control resource type, application dosage and operation time window. Step S4, Resource-Space Matching and Scheduling: Match the optimal prevention prescription with the real-time inventory data of prevention resources in the orchard and the spatial location information of the operating equipment to generate a specific resource scheduling plan and agricultural machinery operation path planning; the resource scheduling plan includes the allocation quantity of required resources from the warehouse to the specific orchard grid, and the agricultural machinery operation path planning is based on the orchard electronic map and the operation points generated by the prescription to perform path optimization calculations; Step S5, Decision Visualization and Command Issuance: Integrate the risk quantification index, optimal prevention and control prescription, resource scheduling plan and agricultural machinery operation path planning, overlay and render and visualize them on the electronic map of the management terminal, and generate a set of control commands that can be issued. The control command set is used to drive the intelligent agricultural machinery to execute the prevention and control prescription.

2. The decision support method for orchard pest and disease early warning and control resource management based on multi-source data as described in claim 1, characterized in that: The orchard's various data sources include real-time environmental data collected through IoT devices, fruit tree canopy image data acquired through image acquisition devices, and macro-meteorological data and historical pest and disease data accessed from external systems.

3. The orchard pest and disease early warning and control resource management decision support method based on multi-source data according to claim 2, characterized in that: Multiple source data points from the orchard were acquired, and spatiotemporal alignment and standardization were performed on the data to construct a spatiotemporal fusion dataset based on orchard geographic grids, including: S1-1. The acquired real-time environmental data, fruit tree canopy image data, macro meteorological data and historical pest and disease data are converted into standardized data sequences with unified timestamps and geographic coordinate system identifiers. S1-2. Based on the orchard electronic map, the standardized data sequence that has undergone unified preprocessing is spatially divided and associated according to the preset geographic grid rules to generate a multi-source initial dataset within each orchard geographic grid unit. S1-3. Perform a time-dimensional integrity check on the multi-source initial dataset within each orchard geographic grid unit, use temporal interpolation to fill in missing data points, and perform smoothing filtering on data sequences with abnormal fluctuations to generate regular temporal data at the grid unit level. S1-4. The environmental factor data after filling and smoothing in each grid cell, the vegetation index and visual feature data of pests and diseases extracted from the canopy image, and the gridded meteorological elements and historical pest and disease occurrence intensity data are superimposed and correlated in the feature layer to form a spatiotemporal fusion dataset with the orchard geographical grid as the basic unit.

4. The decision support method for orchard pest and disease early warning and control resource management based on multi-source data as described in claim 1, characterized in that: The spatiotemporal fusion dataset is input into the pest and disease prediction model, which is a multi-classification model built based on machine learning algorithms and characterized by environmental factor time-series data and historical pest and disease occurrence patterns. The output of the pest and disease prediction model is a risk quantification index for a specific pest or disease within a preset future time period. The risk quantification index is a value that combines the probability of occurrence and the degree of potential harm, including: S2-1. Divide the spatiotemporal fusion dataset into samples to generate an input sample set for model training and prediction. Each sample corresponds to the environmental factor time series data of an orchard geographic grid at a historical time point and the actual occurrence of pests and diseases within a preset time period after that time point. S2-2. Extract multi-dimensional features related to the occurrence of pests and diseases from the time series data of environmental factors of each input sample. The multi-dimensional features include the statistical features, periodic features, and trend features of environmental factors. Combine these features with the historical pest and disease occurrence patterns corresponding to the geographical grid of the orchard to form a complete feature vector. S2-3. Input the feature vector into a multi-classification model built on machine learning algorithms. The multi-classification model is trained with historical data and is used to learn the complex nonlinear relationship between environmental features and the occurrence of various pests and diseases. S2-4. Use the trained multi-classification model to predict the spatiotemporal fusion data at the current moment and output the risk quantification index for a specific pest or disease within a preset time period in the future. The risk quantification index is obtained by weighting the probability of occurrence output by the model with the preset harm weight coefficient of the pest or disease.

5. The decision support method for orchard pest and disease early warning and control resource management based on multi-source data according to claim 1, characterized in that: Step S3 includes: S3-1. Based on the risk quantification index and the preset risk level classification threshold, determine the pest and disease risk level of the current orchard geographic grid, and trigger the prevention and control decision engine based on the pest and disease risk level. S3-2, The prevention and control decision engine retrieves a set of candidate prevention and control resources from the prevention and control resource knowledge base based on the risk level of pests and diseases. The prevention and control resource knowledge base records the prevention and control effectiveness coefficient, economic cost and environmental impact factor of different prevention and control resources under different risk levels. S3-3. Based on the candidate set of prevention and control resources, construct a multi-objective optimization problem with the goal of maximizing comprehensive benefits. The comprehensive benefit objective function includes at least maximizing prevention and control effectiveness, minimizing economic costs, and minimizing environmental impact. S3-4. A multi-objective optimization algorithm is used to solve the multi-objective optimization problem and generate a Pareto optimal solution set. Each solution in the Pareto optimal solution set corresponds to a potential prevention and treatment prescription. S3-5. Select the optimal solution from the Pareto optimal solution set according to the preset decision rules, and generate the optimal prevention and control prescription that matches the current risk level. The optimal prevention and control prescription shall include at least the recommended prevention and control resource type, application dosage, and operation time window.

6. The decision support method for orchard pest and disease early warning and control resource management based on multi-source data according to claim 1, characterized in that: Step S4 includes: S4-1. Obtain the optimal prevention and control prescription, and at the same time access the real-time inventory data of prevention and control resources in the orchard and the spatial location information of the operating equipment; S4-2. Based on the recommended control resource types and application dosages in the optimal control prescription, and combined with the area of ​​the orchard geographic grid and the distribution of pest and disease risks, calculate the amount of control resources required for each orchard geographic grid. S4-3. Based on the real-time inventory data of prevention and control resources, perform inventory matching and availability verification on the quantity of prevention and control resources required for each orchard geographic grid. If the inventory is insufficient, trigger a resource replenishment warning and generate a resource allocation plan. The resource allocation plan includes the allocation quantity of the required resources from the warehouse to the specific orchard grid. S4-4. Based on the orchard electronic map and the optimal prevention and control prescription, the operation points are generated. Combined with the spatial location information of the operation equipment, the path optimization algorithm is used to plan the operation path of agricultural machinery. The path optimization algorithm aims to minimize the total operation distance or the operation time and calculates the optimal order and path for agricultural machinery to traverse all operation points. S4-5. Integrate the resource allocation plan with the agricultural machinery operation path planning to generate specific resource scheduling schemes and agricultural machinery operation instructions. The resource scheduling scheme includes resource allocation details and scheduling timetable, and the agricultural machinery operation instructions include path navigation information and operation parameter settings.

7. The decision support method for orchard pest and disease early warning and control resource management based on multi-source data according to claim 1, characterized in that: Step S5 includes: S5-1. Integrate the risk quantification index, optimal prevention and control prescription, and resource scheduling plan with agricultural machinery operation path planning to generate integrated prevention and control decision data. S5-2. Call the electronic map service engine of the management terminal to associate the spatial elements and attribute information in the integrated data of prevention and control decision, and overlay and render and visualize them on the electronic map. Among them, the risk quantification index is displayed in the form of a heat map on the corresponding orchard geographical grid, the optimal prevention and control prescription is displayed in the form of a pop-up window, the resource scheduling plan is overlaid in the form of a material flow map, and the agricultural machinery operation path planning is dynamically displayed in the form of preset icons and path lines. S5-3. Based on the visualization results, generate a set of control instructions that can be issued according to user confirmation instructions or system preset rules. The set of control instructions includes navigation control instructions for intelligent agricultural machinery, operation parameter setting instructions, and outbound scheduling instructions for resource storage systems. S5-4. The control command set is sent to the corresponding intelligent agricultural machinery terminal and resource storage control system through the Internet of Things communication network; S5-5: Receives real-time operational status information from intelligent agricultural machinery and inventory update information from the resource storage system, and dynamically updates and displays it on the electronic map of the management terminal, forming a closed-loop management system of decision-making, execution, and feedback.

Images