Big data driven green pest control decision-making method

Abstract

The present application relates to the field of agricultural pest control technology, and discloses a green pest control decision method driven by big data. The method comprises collecting multi-source heterogeneous agricultural environment data containing meteorological elements, crop physiological indexes and field biological activity information, generating dynamic weight coefficients according to the spatial and temporal resolution and measurement dimension of the data source; constructing a pest risk dynamic assessment model, calculating the pest occurrence probability index of different regions and different growth stages based on the coefficient fused data; establishing a green control measure knowledge base, matching the candidate control scheme set according to the crop type and growth cycle; combining the pest occurrence probability index and the candidate scheme set, weighing the control effect, environmental impact and economic cost to generate the optimal control strategy sequence; implementing a dynamic adjustment mechanism, updating the pest occurrence probability index according to the real-time farmland feedback data, and iteratively correcting the optimal control strategy sequence to help sustainable green agricultural production.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agricultural pest control technology, in particular to a big data driven green pest control decision method. BACKGROUND

[0002] In the process of agricultural production, the occurrence and spread of pests are always important factors affecting crop yield and quality. Traditional pest control work mostly relies on the experience of the grower, which is greatly influenced by subjective factors and is difficult to accurately grasp the rules and trends of pest occurrence. With the development of agricultural scale and intensification, the single experience-based prevention and control mode cannot meet the needs of modern agricultural production.

[0003] At present, some agricultural production has begun to introduce data monitoring means to assist in pest control, but the existing technology still has many limitations. The data collection dimension is relatively single, often only focusing on meteorological data or crop appearance data, ignoring crop physiological indicators, field biological activity information and other factors that have important influence on pest occurrence, resulting in insufficient comprehensive data support and difficulty in accurately reflecting the potential risk of pest occurrence; in data processing and utilization, there is a lack of effective dynamic weight distribution mechanism, which cannot reasonably fuse data according to the differences in spatial resolution and measurement dimension of different data sources, so that the pest assessment model based on these data has low accuracy and is difficult to accurately calculate the pest occurrence probability of different regions and different crop growth stages.

[0004] The selection of existing control measures mostly relies on fixed scheme library and fails to fully match crop types, growth cycles and specific pest risk situations dynamically. In the decision-making process, only the prevention and control effect is focused on, while the influence of the prevention and control measures on the environment and the economic cost factors are ignored, resulting in the lack of comprehensiveness and practicality of the generated prevention and control strategy. In addition, the existing technology lacks effective dynamic adjustment mechanism and cannot update the pest assessment results and correct the prevention and control strategy in time according to the real-time feedback data of the farmland, which cannot quickly respond when the pest occurrence situation changes, easily causing the problems of untimely or excessive prevention and control, affecting the prevention and control effect and possibly causing environmental burden and economic loss. SUMMARY

[0005] The purpose of the present application is to provide a big data driven green pest control decision method to solve the problems raised in the background technology.

[0006] To achieve the above purpose, the present application provides a big data driven green pest control decision method, which comprises:

[0007] Collecting multi-source heterogeneous agricultural environment data, the multi-source heterogeneous agricultural environment data at least including meteorological elements, crop physiological indicators and field biological activity information, generating dynamic weight coefficients according to the spatial and temporal resolution and measurement dimension of the data source;

[0008] Constructing a disease and pest risk dynamic evaluation model, fusing multi-source heterogeneous agricultural environment data based on the dynamic weight coefficients, and calculating disease and pest occurrence probability indexes of different regions at different growth stages;

[0009] Establishing a green prevention and control measure knowledge base, the green prevention and control measure knowledge base including the applicable conditions and action intensity parameters of biological control, ecological regulation and physical control technology, and matching a candidate prevention and control scheme set according to the crop type and growth cycle;

[0010] Generating a multi-objective optimization decision scheme, based on the disease and pest occurrence probability indexes and the candidate prevention and control scheme set, and generating an optimal prevention and control strategy sequence by weighing the prevention and control effect, environmental impact and economic cost;

[0011] Implementing a prevention and control strategy dynamic adjustment mechanism, updating the disease and pest occurrence probability indexes according to the real-time collected farmland feedback data, and triggering the iteration correction of the optimal prevention and control strategy sequence.

[0012] Preferably, the constructing a disease and pest risk dynamic evaluation model, fusing multi-source heterogeneous agricultural environment data based on the dynamic weight coefficients, and calculating disease and pest occurrence probability indexes of different regions at different growth stages, comprises:

[0013] Dividing the characteristic dimensions of agricultural environment data, the characteristic dimensions including meteorological factors, soil factors and biological factors, and calculating the correlation contribution degrees of each characteristic dimension according to the historical disease and pest occurrence records;

[0014] Generating feature fusion coefficients based on the correlation contribution degrees, analyzing the temporal fluctuation and spatial heterogeneity of multi-source heterogeneous agricultural environment data through a sliding time window, and dynamically adjusting the distribution proportion of the feature fusion coefficients;

[0015] Integrating multi-source heterogeneous agricultural environment data by using a nonlinear weighted aggregation algorithm to generate a regional disease and pest risk map, the regional disease and pest risk map including probability indexes and confidence evaluation values;

[0016] Layering and calibrating the regional disease and pest risk map according to the crop growth stage and climate zone characteristics, and outputting a standardized disease and pest occurrence probability index.

[0017] Preferably, the integrating multi-source heterogeneous agricultural environment data by using a nonlinear weighted aggregation algorithm to generate a regional disease and pest risk map, comprises:

[0018] Extracting abnormal fluctuation features in multi-source heterogeneous agricultural environment data, calculating the deviation and mutation frequency of meteorological elements and biological activity information;

[0019] Based on the deviation and mutation frequency, a risk sensitive factor is constructed, and the monitoring data in the high deviation and high mutation frequency areas are marked as priority processing data stream;

[0020] Noise interference in the priority processing data stream is eliminated by an adaptive filtering algorithm to generate a purified environmental feature vector;

[0021] The purified environmental feature vector is input into a risk prediction network to output a pest risk probability distribution map with time and space labels.

[0022] Preferably, the noise interference in the priority processing data stream is eliminated by an adaptive filtering algorithm to generate a purified environmental feature vector, comprising:

[0023] Identify periodic interference patterns in the priority processing data stream, and generate a filtering threshold range according to the intensity of the interference patterns;

[0024] Segmented smoothing processing is performed on the priority processing data stream using dynamic filtering coefficients, which are associated with data acquisition frequency and sensor accuracy;

[0025] Calculate the stability index of the smoothed data segment, eliminate data segments below the stability index threshold, and retain high-confidence data segments;

[0026] High-confidence data segments are reorganized according to time and space dimensions to generate a purified environmental feature vector.

[0027] Preferably, the green prevention and control measure knowledge base is established, and the green prevention and control measure knowledge base contains the applicable conditions and action strength parameters of biological control, ecological regulation and physical control technology, and matches the candidate control scheme set according to crop type and growth cycle, including:

[0028] Analyzing the technical specifications of green prevention and control measures, extracting environmental parameter thresholds and crop growth stage constraints required for measure implementation;

[0029] Constructing a prevention and control measure effect evaluation matrix, the effect evaluation matrix contains prevention efficiency, environmental compatibility and resource consumption level indicators;

[0030] Based on the crop type and real-time growth state, match the constraint conditions of the preliminary control measure set;

[0031] Sort the preliminary control measure set by a multi-dimensional utility function to generate a candidate control scheme set.

[0032] Preferably, the sorting of the initial set of prevention and control measures by the multi-dimensional utility function generates a candidate set of prevention and control schemes, including:

[0033] Obtaining current farmland environment state data, calculating the matching degree of each measure in the initial set of prevention and control measures and the environment state;

[0034] Evaluating the ratio of the implementation cost of the measure to the expected prevention and treatment efficiency, generating a cost-benefit coefficient;

[0035] Fusing the matching degree and the cost-benefit coefficient to calculate a comprehensive priority score, and arranging the prevention and control measures in descending order of the score;

[0036] Selecting measures with a comprehensive priority score higher than a preset threshold to form the candidate set of prevention and control schemes.

[0037] Preferably, the multi-objective optimization decision scheme is generated based on the pest occurrence probability index and the candidate set of prevention and control schemes, and the optimal prevention and control strategy sequence is generated by balancing prevention and control effect, environmental impact and economic cost, including:

[0038] Establishing a multi-objective decision model, the multi-objective decision model taking maximizing prevention and control effect, minimizing environmental impact and optimizing economic cost as parallel targets;

[0039] Inputting the pest occurrence probability index and the candidate set of prevention and control schemes into the multi-objective decision model to generate a Pareto optimal solution set;

[0040] Selecting the optimal prevention and control strategy from the Pareto optimal solution set based on decision preference weights, the decision preference weights being dynamically configured according to policy requirements and farmer needs;

[0041] Arranging the optimal prevention and control strategy into a prevention and control strategy sequence in the order of implementation time and regional priority.

[0042] Preferably, the selection of the optimal prevention and control strategy from the Pareto optimal solution set based on the decision preference weights includes:

[0043] Analyzing the current agricultural policy orientation and the farmer's business objectives to generate a policy compliance coefficient and an economic preference coefficient;

[0044] Calculating the strategy bias index of each solution in the Pareto optimal solution set, the strategy bias index reflecting the degree of satisfaction of the solution for different targets;

[0045] Integrating the policy compliance coefficient and the economic preference coefficient into a decision preference weight vector;

[0046] Calculating the degree of fit of each solution and the decision preference weight vector by a weighted projection algorithm, and selecting the solution with the highest degree of fit as the optimal prevention and control strategy.

[0047] Preferably, the implementation of the prevention and control strategy dynamic adjustment mechanism updates the disease and pest occurrence probability index according to the real-time collected farmland feedback data, and triggers the iteration correction of the optimal prevention and control strategy sequence, including:

[0048] Monitoring the farmland response data after the implementation of the prevention and control strategy, the farmland response data including the growth and decline of the disease and pest and the change of the crop physiological state;

[0049] Calculating the deviation degree of the implementation effect and the expected target, and if the deviation degree exceeds the allowable range, the strategy correction process is started;

[0050] Adjusting the sensitive parameters in the disease and pest occurrence probability index according to the deviation direction and amplitude;

[0051] Re-running the multi-objective optimization decision scheme to generate an updated prevention and control strategy sequence.

[0052] Preferably, the calculation of the deviation degree of the implementation effect and the expected target, and if the deviation degree exceeds the allowable range, the strategy correction process is started, including:

[0053] Setting the deviation degree allowable range, which is dynamically adjusted according to the crop sensitivity and disaster level;

[0054] Comparing the difference between the real-time farmland response data and the expected target value, calculating the absolute deviation and the relative deviation ratio;

[0055] If either the absolute deviation or the relative deviation ratio exceeds the allowable range, the strategy correction instruction is triggered;

[0056] According to the deviation type, the parameter recalibration or model reconstruction method is selected to execute the strategy correction process.

[0057] Compared with the prior art, the beneficial effects of the present application are:

[0058] Through the collection and fusion of multi-source heterogeneous agricultural environment data, the problem of single data dimension and insufficient support in the prior art is effectively solved. The collected data covers multiple aspects such as meteorological elements, crop physiological indicators and field biological activity information, which can comprehensively reflect various factors affecting the occurrence of diseases and pests, and provide a more rich and comprehensive data basis for subsequent risk assessment and decision-making. At the same time, according to the dynamic weight coefficient generated according to the spatial and temporal resolution and measurement dimension of the data source, the reasonable weighted fusion of multi-source data is realized, the data processing deviation caused by the neglect of data importance difference is avoided, and the data fusion result can more accurately reflect the actual situation.

[0059] In the disease and pest risk assessment aspect, the disease and pest risk dynamic assessment model constructed based on the multi-source data fused based on the dynamic weight coefficient can accurately calculate the disease and pest occurrence probability index of different regions in different growth stages. This dynamic assessment method breaks the limitations of the traditional assessment model fixation and static, fully considers the influence of regional differences and crop growth cycle changes on disease and pest occurrence, makes the assessment result more targeted and accurate, can provide accurate risk basis for prevention and control decision, and helps users to master the disease and pest occurrence trend in advance.

[0060] The establishment of the green prevention and control measure knowledge base integrates the applicable conditions and action intensity parameters of biological control, ecological regulation and physical control technology, and matches the candidate prevention and control scheme set according to the crop type and growth cycle, which changes the fixed selection of the prevention and control scheme in the prior art and the lack of pertinence. Through accurate matching, it is ensured that the candidate prevention and control scheme can adapt to the specific crop situation and growth stage, improves the applicability of the prevention and control scheme, and lays a good foundation for subsequent generation of the optimal prevention and control strategy.

[0061] When generating the multi-objective optimization decision scheme, the limitations of single attention to the prevention and control effect in the existing decision process are broken through by weighing the prevention and control effect, environmental impact and economic cost. This multi-objective optimization method can minimize the negative impact of the prevention and control measures on the environment while ensuring the prevention and control effect, and control the economic cost, so that the generated optimal prevention and control strategy sequence is more comprehensive and practical, meets the demand of green and sustainable development of modern agriculture, and can also better consider the economic interests of the growers.

[0062] The implementation of the dynamic adjustment mechanism of the prevention and control strategy updates the disease and pest occurrence probability index according to the real-time collected farmland feedback data, and triggers the iteration correction of the optimal prevention and control strategy sequence, effectively solving the problem of lack of dynamic response capability in the prior art. This dynamic adjustment method can timely capture the changes of the disease and pest occurrence in the farmland, quickly update the evaluation result and correct the prevention and control strategy, ensure that the prevention and control measures always match the actual disease and pest occurrence, avoid the problems of untimely prevention and control or excessive prevention and control, ensure the prevention and control effect, reduce unnecessary resource waste and environmental burden, and improve the flexibility and effectiveness of the entire prevention and control decision process. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 The working principle diagram of the big data driven disease and pest green prevention and control decision method is described in the present application;

[0064] Figure 2 The flowchart for constructing the disease and pest risk dynamic assessment model is described in the present application;

[0065] Figure 3 The flowchart for generating the self-adaptive filtering and environmental feature vector is described in the present application. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0067] Please refer to Figure 1 The present application provides a big data driven green pest control decision-making method, which comprises the following steps: integrating multi-source data acquisition, dynamic model evaluation, knowledge base matching, optimized decision generation and feedback adjustment mechanism. Deploy Internet of Things sensor network and remote sensing equipment to continuously collect multi-source heterogeneous agricultural environment data in farmland environment. These data cover meteorological elements such as temperature, humidity and precipitation, crop physiological indicators such as leaf area index and photosynthetic rate, and field biological activity information such as pest density and natural enemy population dynamics. During the data acquisition process, dynamic weight coefficients are calculated according to the differences in spatial and temporal resolution and measurement dimension of each data source; for example, high spatial and temporal resolution sensor data is given a higher weight, and the weight of low frequency observation data is adjusted accordingly to ensure the rationality of data fusion. When constructing the dynamic evaluation model of plant diseases and insect pests, multi-source data are integrated based on dynamic weight coefficients, and a weighted fusion algorithm is used to calculate the plant disease and insect pest occurrence probability index of different regions at each growth stage of the crop. This index reflects the potential risk of disease and insect pest outbreak and is dynamically updated with environmental changes. A green control measure knowledge base is established, which systematically records the applicable conditions and action strength parameters of biological control such as natural enemy release, ecological regulation such as habitat optimization, and physical control such as trapping technology. The knowledge base automatically matches a set of candidate control schemes according to the crop type such as rice and wheat and the growth cycle such as the seedling stage and the flowering stage. In the multi-objective optimization decision scheme generation stage, the disease and insect pest occurrence probability index and the candidate control scheme set are combined to generate an optimal control strategy sequence by using a multi-objective decision algorithm, with prevention and control effect, environmental impact and economic cost as optimization objectives. The sequence is arranged according to the implementation priority and time window. The dynamic adjustment mechanism of the implementation of the control strategy is used to update the risk index and trigger strategy iteration correction through real-time monitoring of farmland feedback data such as the growth and decline of plant diseases and insect pests, so as to ensure the adaptability of the control measures.

[0068] Embodiment 1: Please refer to Figure 2The division of the characteristic dimensions of agricultural environmental data is based on the internal logic of the agricultural ecosystem. Meteorological factors include temperature, humidity, light intensity, precipitation, wind speed, and other climate factors that directly affect the occurrence and development of pests and diseases. Soil factors include soil temperature, humidity, pH, organic matter content, nitrogen, phosphorus, and potassium nutrient concentrations, which reflect the basic conditions for crop growth. Biological factors include pest population density, natural enemy number, pathogen spore capture amount, and crop leaf humidity, which directly represent biological interactions. Each characteristic dimension is set with several sub-features, such as temperature indicators in meteorological factors, which can be further divided into daily average temperature, diurnal temperature difference, and accumulated temperature. This multi-dimensional division can comprehensively capture environmental driving factors that affect the occurrence of pests and diseases. To calculate the correlation contribution of each characteristic dimension based on historical pest and disease occurrence records, at least three complete observation data of the growth season are required. The feature importance evaluation algorithm is used to analyze the correlation strength between each factor and the occurrence of pests and diseases. For continuous variables, the Pearson correlation coefficient is calculated using correlation analysis. For categorical variables, variance analysis or chi-square test is used to evaluate their discriminant ability. Finally, the contribution of each characteristic dimension to the occurrence of pests and diseases is ranked. Pearson correlation coefficient is used as a basic statistical tool to preliminarily quantify the strength and direction of the linear relationship between continuous environmental variables and the occurrence of pests and diseases. It calculates the ratio of the covariance between two variables to the product of their standard deviations, with a result range of -1 to 1. A positive value indicates a positive correlation, i.e., as the environmental variable value increases, the risk of pest and disease occurrence tends to increase; a negative value indicates a negative correlation. The closer the absolute value is to 1, the stronger the linear relationship. For example, it can be used to quickly evaluate the linear correlation strength between continuous meteorological indicators such as average temperature and cumulative precipitation and insect population density. However, it is important to note that Pearson correlation coefficient can only measure linear relationships and may not effectively identify complex nonlinear relationships, so it is usually combined with other nonlinear analysis methods as a component of comprehensive feature correlation evaluation.

[0069] The process of generating feature fusion coefficients based on correlation contribution needs to consider the spatiotemporal variability of features. High-contribution features obtain higher weight coefficients in data fusion, but the allocation of weights needs to be dynamically adjusted in combination with the availability and reliability of data. The temporal window size is determined according to the crop growth cycle and the occurrence of pests and diseases. For crops with short growth cycles, a shorter time window is used to quickly capture the impact of environmental changes on pests and diseases. The analysis of spatial heterogeneity is realized through geographic information system technology, which divides the monitoring area into grid cells and calculates the coefficient of variation of environmental features in each grid. For areas with large spatial variation, the weight of local features is appropriately increased. When dynamically adjusting the allocation ratio of feature fusion coefficients, an adaptive weighting mechanism is introduced. When a specific environmental factor in a certain area shows abnormal fluctuations, the weight of this factor in feature fusion is automatically increased to ensure the model's response to sudden environmental changes. When integrating multi-source heterogeneous agricultural environmental data using nonlinear weighted aggregation algorithms, a mathematical model for multi-source data fusion needs to be designed, which can handle environmental data of different scales and units. The core of nonlinear weighting is to establish a nonlinear mapping relationship between feature values and pest and disease occurrence probabilities, and the optimal weight combination is obtained through machine learning algorithms. The process of generating regional pest and disease risk maps includes spatial interpolation and time series prediction. Spatial interpolation uses Kriging or inverse distance weighting to extend the risk values of discrete points to continuous areas. Time series prediction uses historical data to establish a trend model of risk changes. The probability index in the risk map represents the likelihood of pest and disease occurrence, which is quantified as a value between 0 and 1. The confidence evaluation value reflects the reliability of the probability estimate, which is calculated based on data quality, model fitting goodness, and sample size. Nonlinear weighted aggregation algorithm is a core data fusion technology designed for the nonlinear relationship between factors and pest and disease risk in agricultural environmental systems. Its core idea is not simply multiplying each feature value by a fixed weight and then adding them up, but using a more complex function mapping to simulate the real effects of the combined action of these factors. A common implementation is to use machine learning models as aggregators. For example, artificial neural networks can automatically learn the complex nonlinear combination of input features through their hidden layers and activation functions (such as Sigmoid, ReLU) and output a comprehensive risk score. Support vector machines use kernel functions to map data to high-dimensional spaces for nonlinear classification or regression. These algorithms can simulate complex interaction effects and threshold effects such as "only when temperature and humidity exceed a certain threshold at the same time, the risk of disease increases significantly." Through this nonlinear aggregation, the actual risk level under the synergistic action of multiple environmental factors can be more realistically reflected.

[0070] Layered calibration of regional pest and disease risk maps based on crop growth stages and climate zone characteristics is a crucial step in improving model accuracy. Crop growth stages are categorized into seedling, tillering, jointing, heading, flowering, grain-filling, and maturity stages, each with significant differences in crop sensitivity and resistance to pests and diseases. Layered calibration requires establishing growth stage-specific correction functions to adjust parameter settings in the risk calculation model for each growth stage. Climate zone calibration considers the differences in pest and disease occurrence patterns across different climate regions; the occurrence thresholds and prevalence conditions for the same pest or disease differ in temperate, subtropical, and tropical regions. The output of the standardized pest and disease occurrence probability index needs to be normalized to ensure comparability of risk values ​​across different regions and periods. Normalization methods include minimum-maximum scaling or Z-score standardization, resulting in an index that is readily usable by the decision-making system. During implementation, real-time acquisition and transmission of multi-source data requires reliable IoT infrastructure support, and sensor nodes deployed in the field should possess sufficient environmental tolerance and data verification capabilities. Lightweight communication standards should be used for data transmission protocols to ensure timely and complete data uploads. Data preprocessing includes outlier detection, missing value imputation, and data standardization. Missing value imputation uses time series interpolation or spatial interpolation methods, while data standardization converts indicators with different dimensions into a unified scale. The calculation of dynamic weighting coefficients needs to comprehensively consider the spatiotemporal resolution and measurement accuracy of the data source. Data sources with high spatiotemporal resolution are usually assigned higher weights, but the impact of sensor accuracy and stability on weight allocation must also be considered.

[0071] The division of feature dimensions is not static and needs to be adaptively adjusted according to crop type, planting pattern, and regional characteristics. For example, there are significant differences in the environmental feature dimensions between greenhouse agriculture and open-field planting. Greenhouse agriculture needs to add the microclimate feature dimension, while open-field planting needs to consider regional meteorological factors more. The collection of historical pest and disease occurrence records must ensure data accuracy and representativeness. The records should include detailed information such as pest and disease type, occurrence time, severity, and affected area. These records are the basis for calculating the contribution of feature correlation. The size of the sliding time window needs to balance the model's sensitivity and stability. A window that is too small will make the model overly sensitive to random fluctuations, while a window that is too large will reduce the model's response speed to trend changes. The selection of nonlinear weighted aggregation algorithms should consider the algorithm's interpretability and computational efficiency. For applications with high real-time requirements, algorithms with lower computational complexity are preferable, while for applications with high accuracy requirements, more complex ensemble learning algorithms can be used. The generation of regional-level pest and disease risk maps requires the support of a geographic information system. The spatial resolution of the map is determined according to actual management needs, generally set at the field level or a finer scale. The calculation of probability indices and confidence scores should be transparent and traceable to facilitate user understanding of the reliability of the model output. Stratified calibration of growth stages and climate zones requires the establishment of a detailed crop phenological calendar and climate zoning database, and the impact of current year's climate anomalies on crop phenological periods must also be considered during calibration.

[0072] Example 2: See Figure 3Extracting abnormal fluctuation characteristics is not simply a matter of judging whether data points exceed fixed thresholds; rather, it requires establishing a dynamic expected range based on historical data from the same period and crop growth models. For meteorological elements such as hourly temperature data, the system calculates the degree of deviation from the average of the same period over the past five years. This deviation is quantified using statistical methods, considering not only the absolute difference but also the duration of the deviation. Simultaneously, the system calculates the frequency of data mutations, focusing on the rate of change of indicator values ​​within a short period, such as a sharp increase in precipitation or a sudden increase in wind speed within three hours. Processing information on field biological activity is more complex. For example, the number of pests recorded by insect monitoring lamps needs to be distinguished between normal population fluctuations and genuine explosive growth. This requires time series analysis to identify abnormal growth patterns that exceed seasonal patterns. Based on the calculated deviation and mutation frequency, the system constructs a comprehensive risk sensitivity factor. This factor is a composite indicator; it is not a simple sum of various abnormal characteristics but rather assigns different weights to different environmental factors based on the occurrence patterns of different pests and diseases. For example, for certain moisture-loving diseases, a sustained high deviation in relative humidity is assigned a very high risk sensitivity weight, while for some migratory pests, sudden changes in wind speed under specific wind directions may become key indicator factors. The construction process incorporates logical judgment rules: when a specific environmental indicator at a monitoring point simultaneously meets the criteria of high deviation and high frequency of sudden changes, the data stream from that point will be marked as a priority data stream. Priority processing means that these data enjoy higher priority in subsequent calculation queues, and computing resources will be preferentially allocated to them, aiming to shorten the response time from anomaly detection to risk warning and gain the initiative in prevention and control decisions.

[0073] For data streams marked for priority processing, efficient noise filtering is essential to improve data quality and prevent false signals from interfering with risk assessment. Adaptive filtering algorithms dynamically adjust filtering parameters based on the data's inherent characteristics, rather than using a fixed filtering strength. The algorithm first identifies periodic interference patterns in the data stream. These interferences may originate from sensor operating cycles, regular changes due to day-night cycles, or systemic errors in the equipment itself. The identification process employs frequency domain analysis, converting the time-series data to the frequency domain and observing its power spectrum; significant peaks correspond to periodic interference. Based on the identified interference pattern strength, the system generates a dynamic filtering threshold range, with stronger interference bands corresponding to stricter filtering. Unlike traditional fixed-parameter filters, the core characteristic of adaptive filtering algorithms lies in their ability to dynamically adjust their filtering behavior and parameter settings based on the real-time characteristics of the input data stream. It does not pre-define a universal, unchanging filtering strength but uses the data itself as the basis for adjusting the filtering strategy. After the algorithm starts, it first performs preliminary feature analysis on the input priority data stream, a process that occurs simultaneously in the time and frequency domains. In the time domain, the algorithm detects continuous trends and abrupt changes in data points. In the frequency domain, it converts time-series data into frequency-domain signals using methods such as Fast Fourier Transform, thus clearly identifying periodic fluctuation patterns hidden within the data. These periodic patterns may originate from regular changes in the external environment, such as the diurnal temperature range caused by the Earth's rotation, or from the operational characteristics of the monitoring equipment itself, such as the minute reading oscillations caused by the sensor's timed self-calibration process. After identifying the intensity of the periodic interference pattern, the algorithm generates a dynamic and variable filtering threshold range. The intensity of the interference pattern is quantified by calculating the amplitude or power of the frequency component. For a periodic interference with a strong amplitude, such as the regular peak in temperature sensor readings at noon each day due to the strongest solar radiation, the algorithm determines it as a strong inherent pattern rather than abnormal noise, thus setting a relatively lenient filtering threshold to avoid misjudging such strong, predictable normal fluctuations as noise and over-smoothing, thereby preserving the true dynamics of the data. Conversely, for a fluctuation with a weak amplitude and variable frequency, the algorithm may use a stricter filtering threshold to effectively suppress it. This dynamic threshold range setting ensures the filter's ability to distinguish and process noise with different characteristics.

[0074] The next step is to perform segmented smoothing of the priority data stream using dynamic filtering coefficients. The determination of the filtering coefficients is closely related to the data acquisition frequency and sensor accuracy. For data sources with high acquisition frequency and high accuracy, smaller filtering coefficients can be used to retain more detailed fluctuations; for data with low frequency or limited accuracy, larger coefficients are used for stronger smoothing to suppress random errors. Smoothing is performed on a data segment basis, with each segment containing observations within a specific time window. After smoothing, a stability index needs to be calculated for each data segment. This index measures the degree of fluctuation within the segment and the smoothness of the smoothed curve. By setting a stability index threshold, the system automatically removes data segments below the threshold, which usually indicate unreliable data quality or the presence of strong noise that cannot be effectively filtered out. The high-confidence data segments that are ultimately retained represent cleaned and verified reliable environmental information. Reorganizing the selected high-confidence data segments according to the spatiotemporal dimension is a necessary step in generating the cleaned environmental feature vector. Spatiotemporal reassembly is not a simple stitching together; rather, it aligns and integrates data from different sensors that belong to the same geographic unit and time window, based on geographic coordinates and timestamps. For example, soil temperature data from different points within the same field, after considering spatial heterogeneity, are combined into a single feature value that represents the overall condition of the field. The reassembled data forms purified environmental feature vectors, each corresponding to a geographic grid unit and a specific time point, with each dimension representing a purified environmental variable. These purified environmental feature vectors are then systematically input into a pre-trained risk prediction network. This network is typically a deep learning model whose structure effectively captures the complex nonlinear mapping relationship between environmental features and the risk of pest and disease occurrence. The network receives feature vectors at the input layer, performs feature transformation and combination through hidden layers, and finally generates a numerical value representing the probability of pest and disease occurrence at the output layer. This output value is assigned corresponding spatial location and time stamps, and through techniques such as spatial interpolation, the risk values ​​of all grid points are visualized, ultimately forming a pest and disease risk probability distribution map with spatiotemporal markers.

[0075] Example 3: Analyzing the technical specifications for green pest control measures is the initial step in building a knowledge base. This process requires systematically sorting and converting technical documents from agricultural research institutions, plant protection departments, and production practices into machine-readable formats. Technical specification documents typically exist in unstructured text form, containing descriptive provisions for various technologies such as biological control, ecological regulation, and physical control. The analysis work first utilizes natural language processing technology to identify key constraints on technology implementation within the text. These constraints include environmental parameter thresholds, such as the average temperature range and relative humidity limits required when applying Trichogramma wasps to control stem borers, as well as constraints related to crop growth stages, such as deploying sex pheromone traps at the beginning of adult emergence. The analysis process requires precisely extracting the specific numerical ranges and logical relationships of the parameters, such as the composite condition "effective when the average temperature is above 20 degrees Celsius for five consecutive days," and converting it into a standard logical expression of "temperature > 20°C & duration >= 5 days."

[0076] Constructing an evaluation matrix for the effectiveness of control measures is the core of the knowledge base for quantitative decision-making. This matrix is ​​a multi-dimensional evaluation table, with each row representing a specific control measure and each column representing an evaluation dimension. Key dimensions include control efficiency, typically expressed using quantitative indicators such as insect population reduction rate or disease index reduction rate; environmental compatibility, assessing the impact of the measure on non-target organisms, soil health, and water bodies, which can be categorized as high, medium, or low; and resource consumption level indicators, comprehensively considering the human, material, financial, and time costs required to implement the measure. Each dimension requires a unified scoring standard. For example, control efficiency can be divided into multiple score ranges based on expected results, while environmental compatibility is rated based on its potential risk to the ecosystem. The data for the effectiveness evaluation matrix comes from field trial reports, literature meta-analysis, and expert experience, ensuring the scientific rigor and practicality of the evaluation results.

[0077] The initial selection set of control measures based on matching crop type and real-time growth status with constraints is a dynamic screening process. The system first reads the currently monitored farmland data, including crop variety names and their specific phenological stages. Then, it matches and verifies the constraints attached to each control measure in the knowledge base with the current farmland status. For example, for rice paddies in the flowering stage, the system automatically filters out all measures applicable to rice and permissible during the flowering stage, while excluding measures explicitly marked as usable only in the seedling or maturity stages, and measures completely unsuitable for rice. The matching process strictly follows constraint logic; a measure is only included in the initial selection set if the farmland environmental parameters fall entirely within the environmental threshold range required by the measure, and the crop growth stage meets its stage constraints. The initial selection set of control measures is sorted using a multi-dimensional utility function to identify the candidate solution with the best overall performance from multiple alternatives. This process first requires obtaining the current farmland environmental status data and calculating the matching degree between each measure in the initial selection set and the current environmental status. Matching degree calculation considers not only whether the conditions are met, but also the degree of compliance. For example, if the current temperature is exactly at the midpoint of the optimal temperature range for a certain measure, its matching degree score will be higher than those that meet the conditions but are at the edge of the range. Evaluating the ratio of the implementation cost to the expected prevention and control efficiency to generate a cost-benefit coefficient is key to measuring economic feasibility. Costs include direct material costs, labor costs, and machinery wear and tear, while the expected prevention and control efficiency comes from the effectiveness evaluation matrix. A low-cost, high-efficiency measure will achieve a very high cost-benefit coefficient.

[0078] Calculating the overall priority score by integrating matching degree and cost-effectiveness coefficient is a step in achieving the final ranking. A weighted comprehensive model is used here to integrate multiple indicators. The expression for this model can be described as:

[0079]

[0080] Where: symbol Representing the The overall priority score for each prevention and control measure is the direct basis for decision-making. (Symbol) Indicates the first The degree of matching between the measures and the current state of the farmland environment is represented by a value between 0 and 1, with a higher degree of matching indicating a closer match to 1. (Symbol) Indicates the first The cost-benefit coefficient of this measure is obtained by normalizing the ratio of expected prevention efficiency to implementation cost; a higher value indicates better economic benefits. (Symbol) and These are respectively assigned to the matching degree and cost-benefit ratio The weighting factors, and satisfying The specific values ​​of the weighting factors reflect the decision-makers' preferences between "technological applicability" and "economic efficiency," and can be dynamically adjusted according to actual policy guidance or farmers' specific needs. For example, in scenarios that pursue maximum ecological benefits, a higher weight may be assigned to the matching degree; while in situations with tight budgets, the cost-benefit coefficient may be given more importance.

[0081] After calculating the overall priority score for each measure, the system sorts them in descending order of score. The system selects measures with overall priority scores higher than a preset threshold and combines them into a candidate prevention and control scheme set. This preset threshold serves as a quality control measure, ensuring that all schemes entering the final decision-making process have a basically feasible overall performance. The threshold can be flexibly set according to the number and quality of candidate schemes, thereby outputting a moderately sized and high-quality set of candidate schemes, providing excellent input options for subsequent multi-objective optimization decisions.

[0082] Example 4: The core objectives of a multi-objective decision-making model are typically set as maximizing control effectiveness, minimizing environmental impact, and optimizing economic cost. These three objectives are pursued in parallel, without a single, overwhelming optimization objective. Maximizing control effectiveness focuses on the degree to which measures suppress target pests and diseases, aiming to control the population size or severity of pests and diseases below the economic threshold. Minimizing environmental impact emphasizes assessing the potential negative impacts of control measures on the farmland ecosystem and surrounding environment, including disturbances to non-target organisms, soil health, and water quality. Optimizing economic cost considers input and output from an input-output perspective, striving to achieve acceptable control effectiveness with the lowest possible economic cost. These three objectives together constitute the basic framework of the decision-making process.

[0083] The calculated pest and disease occurrence probability index, along with the previously generated set of candidate control measures, is input into a multi-objective decision-making model, which then begins searching the solution space. For example, suppose a rice-growing area is in the late tillering stage. The pest and disease risk dynamic assessment model calculates an outbreak probability index of 0.75 for rice planthoppers, indicating a high-risk level. Simultaneously, the green control measures knowledge base matches four candidate control measures based on the current conditions: Option A (releasing ladybugs), Option B (applying the plant-derived insecticide matrine), Option C (installing intelligent spectral insect-attracting lamps), and Option D (implementing scientific paddy field water management). The multi-objective decision-making model, for example using a multi-objective evolutionary algorithm, will use these options as a basis to generate a set of numerous non-dominated solutions—the Pareto optimal solution set—through a simulated "evolutionary" process. In this set, any improvement in any solution, such as further improving control effectiveness, will inevitably lead to the deterioration of at least one other objective (such as cost or environmental impact). Therefore, these solutions represent the best trade-offs under the current conditions.

[0084] Table 1 shows a simplified Pareto optimal solution set, which includes several representative policy options generated by the model and their predicted performance on each objective. It should be noted that the actual generated solution set may be more complex, containing more dimensions of information and more solutions.

[0085] Table 1: Pareto optimal solution set for high risk of rice planthopper

[0086]

[0087] Selecting the optimal control strategy from the Pareto optimal solution set based on decision preference weights is key to transforming general solutions into specific action plans. Decision preference weights are not fixed; they are dynamically allocated based on macro-level policy requirements, such as the local promotion of green control technologies and specific indicators for pesticide reduction actions, as well as micro-level farmers' actual needs, such as scale of operation, financial capacity, and the perceived quality of agricultural products. Analyzing current agricultural policy orientations and farmers' operational goals is the foundation for generating specific weights. For example, if current policies strongly favor the reduction of chemical pesticides, and farmers are committed to producing high-end organic rice, then the policy compliance coefficient will significantly favor measures with low environmental impact, while the economic preference coefficient may have a higher tolerance for long-term measures with higher initial investment. The system converts these qualitative orientations into quantitative coefficients, such as assigning a weight of 0.5 to environmental impact goals, 0.3 to economic costs, and 0.2 to control effectiveness. Calculating the strategy bias index for each solution in the Pareto optimal solution set is to quantify the extent to which each solution meets the various objectives. The strategy bias index is calculated by analyzing the position of the solution's values ​​in each objective dimension relative to the ideal point. For example, solution 1 performs well in terms of prevention and control effectiveness and environmental impact, but is not the lowest in cost; its strategy bias index shows that it focuses more on ecological benefits. Solution 2, on the other hand, has outstanding prevention and control effectiveness and low cost, but has a greater environmental impact; its index shows that it leans towards traditional, efficient prevention and control thinking. Integrating the policy compliance coefficient and the economic preference coefficient forms a multi-dimensional decision preference weight vector, which indicates the relative importance of each objective under the current decision-making environment.

[0088] The weighted projection algorithm is used to calculate the fit between each solution and the decision preference weight vector. Essentially, this algorithm measures how close each solution is to the ideal solution in the direction defined by the weight vector. The solution with the highest fit means that its overall performance across multiple objectives best aligns with the current decision preference. For example, under a decision preference emphasizing green development (high environmental weight), solutions 1 or 3 in Table 1 might receive a higher fit score due to their lower environmental impact index; while under a preference that prioritizes cost control (high economic weight), solution 4 might win due to its extremely low cost. The selected optimal control strategy is the best choice that balances the demands of multiple parties under the current conditions. The optimal control strategies are arranged into a control strategy sequence according to implementation time and regional priority. This step transforms the static plan into a dynamic action plan. For example, the selected optimal strategy might be "Solution 1: Prioritize the release of ladybugs, supplemented by spot treatment with a small amount of matrine." The system will plan a specific implementation schedule based on the occurrence patterns of rice planthoppers, the colonization time of ladybugs, and the efficacy characteristics of matrine: First, ladybugs will be released immediately upon detection of the infestation center. Then, 3-5 days later, based on the pest monitoring results, matrine will be used for targeted treatment of areas still exceeding the threshold. Simultaneously, if there is a significant risk gradient within the farmland, such as areas closer to the source of the infestation having a higher risk, the strategy sequence will indicate priority for implementing measures in those areas.

[0089] Example 5: Tracking the effectiveness of control strategies relies on a continuously operating farmland response data monitoring network. This network consists of various sensor nodes deployed in the field, regularly patrolling drones, and manual inspection records. The core farmland response data monitored includes the rise and fall of pests and diseases themselves, such as changes in the number of target pests automatically counted by intelligent pest monitoring lamps, the expansion trend of crop lesions obtained through hyperspectral camera inversion, and changes in crop physiological states, such as relative chlorophyll content, canopy temperature, and plant height. This data is transmitted to a central processing system at a set frequency, such as hourly or daily, forming a continuous data stream reflecting the actual situation in the field. The system synchronously compares the real-time acquired response data with the expected target values ​​set during the multi-objective optimization decision-making stage. The expected target values ​​are quantitative indicators set based on model predictions when formulating control strategies, such as expecting to reduce insect population density by 50% on the seventh day after implementation, or controlling the disease index below a certain threshold.

[0090] Calculating the deviation between the implementation effect and the expected target is a crucial diagnostic step. This calculation needs to consider both absolute and relative changes. Absolute deviation is the arithmetic difference between the real-time observed value and the expected target value, directly reflecting the scale of the deviation. The relative deviation ratio is the percentage obtained by dividing the absolute deviation by the expected target value, used to assess the severity of the deviation. For example, if the expected reduction is to 10 insects per 100 plants, but the actual monitoring value is 15, the absolute deviation is 5, and the relative deviation ratio is 50%. Setting a tolerance range for deviation is a prerequisite for determining whether intervention is necessary. This range is not fixed but dynamically adjusted based on the crop's sensitivity and the current level of the pest infestation. For crops in their critical flowering and pollination stages, and highly sensitive to pests and diseases, such as strawberries, the tolerance range is set very narrowly, as even a small adverse deviation may trigger an alarm. For more tolerant crops, or when pest and disease levels are low, the tolerance range can be appropriately widened to avoid overreaction. If either the calculated absolute deviation or the ratio of relative deviation exceeds the currently set tolerance range, the system will automatically trigger a strategy correction command. Analyzing the direction of deviation is equally important, as it indicates the direction of correction. If the actual pest and disease numbers consistently exceed the expected target, it indicates that the current control strategy is insufficient, potentially underestimating the outbreak or that the measures are inadequate. If the actual values ​​are significantly lower than expected, it may mean that the strategy is too conservative, resulting in unnecessary resource investment. Depending on the type and magnitude of the deviation, the system will select different correction paths. For small deviations where the system determines it is mainly due to model parameter drift, a parameter recalibration process will be initiated. For large deviations that may indicate a fundamental change in environmental conditions or that the original model structure is no longer applicable, a partial model reconstruction process will be considered.

[0091] Parameter recalibration primarily adjusts sensitive parameters in the pest and disease occurrence probability index calculation model. For example, if the predicted disease spread rate based on initial data is found to be significantly lower than the actual observed spread rate, the system may increase the weighting coefficient of the humidity factor in the model or correct the function parameters relating disease infection probability to temperature. These adjustments to sensitive parameters are based on a comparative analysis of the latest monitoring data and historical data, using optimization algorithms to automatically find new parameter combinations that allow the model output to most closely approximate the actual observations. This process is a continuous, adaptive learning process, enabling the risk assessment model to gradually approach the realities of the field. When parameter recalibration still fails to effectively reduce bias, or when bias characteristics indicate that the existing model relationships may be invalid, a deeper level of model reconstruction is initiated. This may involve changing the structure of the risk prediction network, such as adding new hidden layer nodes to capture more complex data features, or introducing new environmental factors as model inputs. Reconstruction is a more cautious process, typically retaining the old model as a reference and validating the performance of the new model on a small scale before a full update. Whether it's parameter recalibration or model reconstruction, the ultimate goal is to make risk assessment more accurate. Once the model is updated, the system will re-run the multi-objective optimization decision-making scheme. This process will use the updated, more realistic pest and disease occurrence probability index and the latest candidate control scheme set to regenerate a Pareto optimal solution set, and select a new optimal control strategy based on decision preference weights. This new strategy will be broken down into specific operational instructions, forming an updated control strategy sequence, which will be immediately issued to the execution units. For example, the original plan was to spray a biological pesticide all over the area, but the revised strategy might change to spraying in key areas and increasing the release of natural enemies.

[0092] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0093] Although embodiments of the invention 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 to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A big data-driven green pest control decision-making method, characterized in that, include: Collect multi-source heterogeneous agricultural environmental data, which includes at least meteorological elements, crop physiological indicators and field biological activity information, and generate dynamic weighting coefficients based on the spatiotemporal resolution and measurement dimension of the data source. A dynamic assessment model for pest and disease risk is constructed, and multi-source heterogeneous agricultural environmental data is integrated based on the dynamic weight coefficients to calculate the probability index of pest and disease occurrence in different regions at different growth stages. Establish a knowledge base for green prevention and control measures, which includes the applicable conditions and intensity parameters of biological control, ecological regulation and physical control technologies, and match a set of candidate control schemes according to crop type and growth cycle. A multi-objective optimization decision scheme is generated. Based on the pest and disease occurrence probability index and the candidate control scheme set, the optimal control strategy sequence is generated by balancing the control effect, environmental impact and economic cost. Implement a dynamic adjustment mechanism for prevention and control strategies, update the probability index of pest and disease occurrence based on real-time farmland feedback data, and trigger iterative correction of the optimal prevention and control strategy sequence; The constructed dynamic assessment model for pest and disease risk, based on the dynamic weighting coefficients and fusion of multi-source heterogeneous agricultural environmental data, calculates the probability index of pest and disease occurrence in different regions at different growth stages, including: The agricultural environmental data is divided into characteristic dimensions, including meteorological factors, soil factors and biological factors. The correlation contribution of each characteristic dimension is calculated based on historical records of pest and disease occurrence. Based on the correlation contribution, feature fusion coefficients are generated. The temporal volatility and spatial heterogeneity of multi-source heterogeneous agricultural environmental data are analyzed through a sliding time window, and the allocation ratio of the feature fusion coefficients is dynamically adjusted. A nonlinear weighted aggregation algorithm is used to integrate multi-source heterogeneous agricultural environmental data to generate a regional-level pest and disease risk map, which includes a probability index and a confidence assessment value. Based on crop growth stages and climate zone characteristics, the regional pest and disease risk map is stratified and calibrated to output a standardized pest and disease occurrence probability index. The method employs a nonlinear weighted aggregation algorithm to integrate multi-source heterogeneous agricultural environmental data and generate regional-level pest and disease risk maps, including: Extract anomalous fluctuation characteristics from multi-source heterogeneous agricultural environmental data, and calculate the deviation and mutation frequency between meteorological elements and biological activity information; Based on the deviation and mutation frequency, a risk sensitivity factor is constructed, and the monitoring data of high deviation areas and high mutation frequency areas are marked as priority data streams. An adaptive filtering algorithm is used to eliminate noise interference in the priority data stream and generate a purified environmental feature vector. The purified environmental feature vector is input into the risk prediction network, which outputs a probability distribution map of pest and disease risk with spatiotemporal labels.

2. The big data-driven green pest control decision-making method according to claim 1, characterized in that, The step of eliminating noise interference in the prioritized data stream using an adaptive filtering algorithm to generate a purified environmental feature vector includes: Identify and prioritize periodic interference patterns in the data stream, and generate a filtering threshold range based on the intensity of the interference patterns; A segmented smoothing process is performed on the priority data stream using dynamic filtering coefficients, which are related to the data acquisition frequency and sensor accuracy. Calculate the stability index of the smoothed data segments, remove data segments below the stability index threshold, and retain high-confidence data segments; The high-confidence data segments are reorganized according to the spatiotemporal dimensions to generate a purified environmental feature vector.

3. The big data-driven green pest control decision-making method according to claim 1, characterized in that, The establishment of a green control measures knowledge base includes the applicable conditions and effectiveness parameters of biological control, ecological regulation, and physical control technologies. A set of candidate control schemes is matched based on crop type and growth cycle, including: Analyze the technical specifications for green prevention and control measures, and extract the environmental parameter thresholds and crop growth stage constraints required for the implementation of the measures; Construct a prevention and control measure effectiveness evaluation matrix, which includes indicators of prevention and control efficiency, environmental compatibility, and resource consumption level. A preliminary set of control measures that meet the constraints based on matching crop type with real-time growth status; The initial set of prevention and control measures is sorted using a multi-dimensional utility function to generate a set of candidate prevention and control schemes.

4. The big data-driven green pest control decision-making method according to claim 3, characterized in that, The process of sorting the initial set of prevention and control measures using a multi-dimensional utility function to generate a candidate prevention and control scheme set includes: Obtain current farmland environmental status data and calculate the matching degree between each measure in the initial set of prevention and control measures and the environmental status; The ratio of the cost of implementing the assessment measures to the expected prevention and control efficiency is used to generate a cost-benefit coefficient. The overall priority score is calculated by combining the matching degree and cost-effectiveness coefficient, and the prevention and control measures are arranged in descending order of the score. Measures with a comprehensive priority score higher than a preset threshold are selected to form a set of candidate prevention and control measures.

5. The big data-driven green pest control decision-making method according to claim 1, characterized in that, The generation of multi-objective optimization decision schemes, based on the pest and disease occurrence probability index and the candidate control scheme set, generates an optimal control strategy sequence by balancing control effectiveness, environmental impact, and economic cost, including: A multi-objective decision-making model is established, with the parallel objectives of maximizing prevention and control effectiveness, minimizing environmental impact, and optimizing economic costs. Input the probability index of pest and disease occurrence and the set of candidate control schemes into the multi-objective decision-making model to generate the Pareto optimal solution set. The optimal prevention and control strategy is selected from the Pareto optimal solution set based on decision preference weights, and the decision preference weights are dynamically configured according to policy requirements and farmers' needs. The optimal prevention and control strategies are arranged into a prevention and control strategy sequence according to the order of implementation time and regional priority.

6. The big data-driven green pest control decision-making method according to claim 5, characterized in that, The selection of optimal prevention and control strategies from the Pareto optimal solution set based on decision preference weights includes: Analyze current agricultural policy orientations and farmers' business objectives to generate policy compliance coefficients and economic preference coefficients; Calculate the policy bias index of each solution in the Pareto optimal solution set, whereby the policy bias index reflects the degree to which the solution satisfies different objectives; The policy compliance coefficient and the economic preference coefficient are integrated into a decision preference weight vector; The solution with the highest degree of fit with the decision preference weight vector is calculated by weighted projection algorithm, and the solution with the highest degree of fit is selected as the optimal prevention and control strategy.

7. The big data-driven green pest control decision-making method according to claim 1, characterized in that, The aforementioned dynamic adjustment mechanism for implementing prevention and control strategies updates the probability index of pest and disease occurrence based on real-time farmland feedback data and triggers iterative correction of the optimal prevention and control strategy sequence, including: The farmland response data after the implementation of the monitoring and control strategy includes the rise and fall of pests and diseases and changes in crop physiological state. Calculate the deviation between the implementation effect and the expected goal. If the deviation exceeds the allowable range, initiate the strategy correction process. Adjust the sensitive parameters in the probability index of pest and disease occurrence based on the direction and magnitude of the deviation. Rerun the multi-objective optimization decision scheme to generate an updated sequence of prevention and control strategies.

8. The big data-driven green pest control decision-making method according to claim 7, characterized in that, The deviation between the calculated implementation effect and the expected target is determined. If the deviation exceeds the allowable range, a strategy correction process is initiated, including: A tolerance range for deviation is set, and the tolerance range is dynamically adjusted according to crop sensitivity and disaster level. Compare the real-time farmland response data with the expected target value, and calculate the ratio of absolute deviation to relative deviation. If either the absolute deviation ratio or the relative deviation ratio exceeds the allowable range, a strategy correction instruction will be triggered. Based on the type of deviation, select either parameter recalibration or model reconstruction to execute the strategy correction process.

Images