A wine grape powdery mildew early warning method and system based on the Internet of Things

Abstract

This invention discloses an IoT-based early warning method for powdery mildew in wine grapes. The method involves acquiring meteorological data for a future time period; inputting this future meteorological data into a trained mapping model from meteorological data to sensor data to obtain predicted sensor data. This sensor data includes the air temperature and humidity of the environment where the wine grapes are located, the duration of leaf moisture on the grape leaves, and the light intensity received by the grapes. Based on an index scoring function for each type of sensor data, an index score is calculated and weighted to obtain a comprehensive risk score. A dynamic early warning threshold is set, and the percentage of the comprehensive risk score relative to the dynamic early warning threshold is calculated. Different early warning signals are triggered based on different percentages. The invention also discloses a corresponding early warning system. This invention achieves accurate early warning of powdery mildew.

Description

Technical Field

[0001] This invention relates to the field of plant disease prediction technology, and in particular to an Internet of Things-based method and system for early warning of powdery mildew in wine grapes. Background Technology

[0002] Powdery mildew is a fungal disease caused by *Hylocereus undatus*, and it is one of the most common diseases in grape cultivation. Traditional monitoring of powdery mildew mainly relies on manual inspection and experience-based judgment, which has problems such as delayed response, limited coverage, and low identification accuracy.

[0003] Existing early warning models are mostly based on simplified meteorological parameters, ignoring the spatiotemporal dynamics of disease occurrence, and lack multi-factor coupling analysis, resulting in insufficient prediction accuracy. For example, some methods use only a single indicator such as temperature or humidity, which cannot effectively capture the spread patterns of powdery mildew in the microenvironment; while some machine learning models can process static data, they are difficult to adapt to real-time changes in the orchard environment. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an IoT-based early warning method for powdery mildew in wine grapes, solving the problems of weak generalization ability and coarse level classification in existing early warning models, which makes it difficult to achieve early warning. This invention also provides an IoT-based early warning system for powdery mildew in wine grapes.

[0005] The technical solution of the present invention is as follows:

[0006] An Internet of Things (IoT)-based method for early warning of powdery mildew in wine grapes, comprising:

[0007] Obtain meteorological data for a future time period;

[0008] Meteorological data for future time periods are input into a training model that maps meteorological data to sensor data to obtain predicted sensor data. The sensor data includes the air temperature and humidity of the environment in which the wine grapes are located, the duration of leaf wetness of the wine grapes, and the light intensity received by the wine grapes.

[0009] The index score corresponding to each sensor data is calculated based on the index scoring function of each sensor data, and then weighted and accumulated to obtain the comprehensive risk score;

[0010] Set a dynamic early warning threshold, calculate the percentage of the comprehensive risk score relative to the dynamic early warning threshold, and trigger different early warning signals based on different percentages.

[0011] Furthermore, the sensor data includes the soil temperature and soil moisture of the environment in which the wine grapes are located.

[0012] Furthermore, the index scoring function includes an air temperature scoring function. Air humidity scoring function Leaf surface wetting duration scoring function Light intensity scoring function ,in, Based on 7 days of air temperature in °C: the cumulative time of air temperature in the range [15, 35] is less than 60 hours. =20; The cumulative time when the air temperature is in [15,20) or (30,35] is ≥60 hours. =50; the cumulative time for air temperature in [20,25) is in [60,120) hours. =70; the cumulative time for air temperature to be in the range of [25,30] is in the range of [60,120) hours. =90; The cumulative time the air temperature is in the range of [20,30] is ≥120 hours. =100, otherwise take =100. =10;

[0013] Based on 7 days of air humidity as a percentage: the cumulative time of air humidity in the range [45, 60) is less than 30 hours. =10; the cumulative time of air humidity in the range [45, 60) is ≥120 hours. =40; the cumulative time for air humidity in the range [60, 80) is in the range [60, 120) hours. =75; the cumulative time of air humidity in the range [60, 80) is ≥120 hours. =100; air humidity >80 or the cumulative time in [45,60) is in [30,120) hours or the cumulative time in [60,80) is <60 hours. =30; otherwise, take 30. =10;

[0014] Based on the duration of leaf surface moisture in hours over a 7-day period: Leaf surface moisture duration > 120, =10; the leaf surface remains moist for a duration of [60, 120]. =30; Leaf surface wetness duration <30 =60; Leaf surface wetness duration [30,60), =100;

[0015] Based on 7-day light intensity measured in Lux: The cumulative time of light intensity in the range [15000, 20000] > 60 hours. =30; the cumulative time of light intensity in the range [10000, 15000) is >60 hours. =60; Light intensity <Cumulative time of 10000> 60 hours =100; otherwise, take 100. =10.

[0016] Furthermore, the index scoring function includes a soil temperature scoring function. Soil moisture scoring function ,in Based on soil temperature in °C over 7 days: The cumulative time for soil temperature in [15,18) or [25,28) > 120 hours, or the cumulative time for soil temperature in [28,30) > [48,72] hours. =40; the cumulative time of soil temperature in [10,15) or [28,30) > 72 hours, =70; cumulative time of soil temperature <10 or >30 >48 hours, =100; otherwise, take 100. =10;

[0017] Soil moisture is categorized by percentage over 7 days: Soil moisture in the range of [50, 60) or [75, 85) for a cumulative period >120 hours, or soil moisture in the range of [40, 50) or [85, 90] for a cumulative period >48, 72 hours. =40; the cumulative time of soil moisture in [40,50) or [85,90] > 72 hours. =70; cumulative time of soil moisture <40% or >90% >48 hours. =100; otherwise, take 100. =10.

[0018] Furthermore, the dynamic early warning threshold is determined based on the product of the early warning threshold benchmark value and the dynamic coefficient. The dynamic coefficient is obtained by multiplying the ratio of the standard deviation of the predicted sensor data to the standard deviation of the historical prediction by the sensitivity coefficient and then adding 1. The sensitivity coefficient has different values ​​based on the prediction uncertainty.

[0019] Furthermore, the warning threshold benchmark value is set based on historical disease data statistics, including: extracting sensor data before the onset of each historical disease case and calculating the corresponding comprehensive risk score, arranging the comprehensive risk scores of all historical cases in ascending order, and obtaining the comprehensive risk score that was lower than 90% of the historical disease time before the occurrence as the warning threshold benchmark value.

[0020] Furthermore, the prediction uncertainty is determined based on the ratio of the difference between the conditional quantiles Q95 and Q05 of the sensor data predicted by the mapping model to Q50.

[0021] Furthermore, during the weighted accumulation, the weighting coefficients of the index scores corresponding to each sensor data are adjusted according to different seasons.

[0022] Furthermore, the mapping model is established by the gradient boosting decision tree algorithm, and the trained mapping model is incrementally trained according to time periods.

[0023] Another technical solution of the present invention is: an Internet of Things-based early warning system for powdery mildew in wine grapes, comprising:

[0024] The data acquisition module is used to acquire meteorological data for future time periods;

[0025] The data prediction module is used to input meteorological data for future time periods into the trained meteorological data to sensor data mapping model to obtain predicted sensor data. The sensor data includes the air temperature and humidity of the environment where the wine grapes are located, the duration of leaf wetness of the wine grapes, and the light intensity received by the wine grapes.

[0026] The risk score calculation module is used to calculate the index score corresponding to each sensor data based on the index scoring function of each sensor data, and perform weighted summation to obtain the comprehensive risk score;

[0027] The early warning module is used to set a dynamic early warning threshold, calculate the percentage of the comprehensive risk score relative to the dynamic early warning threshold, and trigger different early warning signals based on different percentages.

[0028] Compared with the prior art, the present invention has the following advantages:

[0029] This invention designs a multi-source data fusion prediction model that uses future meteorological data to predict specific sensor data of the environment in which wine grapes are located. Based on the influence of environmental parameters on powdery mildew of wine grapes, the data from each sensor are scored and weighted to obtain a comprehensive risk score that can be used to distinguish warning levels by setting thresholds, thereby achieving accurate early warning of powdery mildew. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating the Internet of Things-based early warning method for powdery mildew in wine grapes according to the present invention. Detailed Implementation

[0031] The present invention will be further described below with reference to embodiments, but these are not intended to limit the scope of the invention.

[0032] The IoT-based early warning system for powdery mildew in wine grapes in this embodiment includes: a data acquisition module, a data prediction module, a risk score calculation module, and an early warning module.

[0033] The data acquisition module is used to acquire meteorological data for future time periods.

[0034] The data prediction module is used to input meteorological data for future time periods into the trained meteorological data to sensor data mapping model to obtain predicted sensor data. The sensor data includes the air temperature and humidity of the environment where the wine grapes are located, the duration of leaf wetness of the wine grapes, and the light intensity received by the wine grapes.

[0035] The risk score calculation module is used to calculate the index score corresponding to each sensor data based on the index scoring function of each sensor data, and perform weighted summation to obtain the comprehensive risk score;

[0036] The early warning module is used to set dynamic early warning thresholds, calculate the percentage of the comprehensive risk score relative to the dynamic early warning threshold, and trigger different early warning signals based on different percentages.

[0037] The system is physically deployed using an edge-cloud collaborative architecture to ensure efficient operation: a lightweight model is deployed on the local gateway in the orchard to achieve real-time forecasting and low-latency response. The cloud platform is responsible for data storage, model optimization, and global monitoring, supporting unified management across multiple sites. Sensor accuracy is periodically calibrated, weather forecast interfaces are updated, and system performance is optimized through log analysis.

[0038] For the early warning method implemented by the above-mentioned IoT-based early warning system for powdery mildew in wine grapes, please refer to [link to relevant documentation]. Figure 1 As shown, it includes the following steps:

[0039] Data acquisition and processing includes obtaining meteorological data for future time periods, as well as historical meteorological data, historical environmental data of the wine grapes' environment obtained through an IoT sensor network, and historical disease data. The data obtained through the IoT sensor network is referred to as sensor data. The historical disease data includes records of powdery mildew occurrences in the wine grapes and manual inspection tags. Meteorological data is provided by the local weather station and specifically includes temperature, daily average temperature, maximum temperature, minimum temperature, relative humidity, rainfall, wind speed, sunlight intensity, and duration, with the future time period being the next 7 days. Sensor data includes the air temperature and humidity of the wine grapes' environment, the duration of leaf moisture, and the light intensity received by the wine grapes. In a preferred embodiment, the sensor data also includes the soil temperature and soil moisture of the wine grapes' environment.

[0040] The duration of leaf surface humidity is calculated using air temperature and humidity data. The dew point temperature is calculated from the air temperature and humidity. When the leaf temperature is lower than the dew point temperature, it indicates that the leaf surface is in a humid state. The formula for calculating the dew point temperature is:

[0041] ,

[0042] Where a is a constant of 243.12; b is a constant of 17.62; T d Dew point temperature (°C);

[0043] ,

[0044] RH is the relative humidity of the air (%), and T is the air temperature (°C).

[0045] Historical sensor data from the same period was integrated with meteorological data released by the local meteorological bureau at the corresponding time points. Data was aligned at an hourly granularity, and the sensor data collected every 10 minutes was aggregated by hourly average to generate an hourly average as a representative value. The meteorological bureau data itself is hourly granular and was directly matched with the sensor data by timestamp to ensure consistency. To ensure the generalization ability of the mapping model, historical sensor data and historical meteorological data were divided into training, validation, and test sets in chronological order. 70% was used for the training set, 15% for the validation set, and 15% for the test set. The training set was used for initial fitting of the mapping model, the validation set was used for hyperparameter optimization and early stopping to control the number of iterations, and the test set was used to independently evaluate the performance of the mapping model. Considering the characteristics of time series data, a time-series partitioning method was adopted to strictly ensure that the timestamps of the training data were earlier than those of the validation data.

[0046] Meteorological data for future time periods are input into a trained meteorological data-to-sensor data mapping model to obtain predicted sensor data. The mapping model uses the Gradient Boosting Decision Tree (GBDT) algorithm to predict environmental parameters of the environment in which the wine grapes are located. The specific implementation is as follows:

[0047] During model construction, the input feature X is a vector composed of time-aligned meteorological data, including temperature, daily average temperature, maximum temperature, minimum temperature, relative humidity, rainfall, wind speed, sunlight intensity, and duration. The output target Y is time-aligned sensor data, including air temperature, air humidity, soil temperature, soil moisture, leaf wetness duration, and light intensity.

[0048] By iteratively training multiple decision trees, each tree learns the residuals of its predecessor model, and the final prediction result is a weighted sum of all the trees:

[0049] ,

[0050] in, This represents the k-th decision tree. For decision tree sets, For the first time after time alignment and preprocessing A sample vector of meteorological data. To and Simultaneously aligned Predicted values ​​of sensor monitoring data.

[0051] The training process of the model includes:

[0052] 1. Initialize the model, the formula is as follows:

[0053] ,

[0054] Among them, L , For the first The true target value corresponding to each training sample is the sensor monitoring data. For all training samples, the target value is... The mean.

[0055] For the squared loss function, the initial model is the mean of the target values ​​of the training samples.

[0056] 2. Iterative optimization process, for each iteration m=1,2,…,M:

[0057] Calculate pseudo residuals: For the squared loss function, .

[0058] Fit a new decision tree using training data Training decision trees The leaf node region is obtained. , =1,2,…,J m .

[0059] Calculate leaf node weights For the squared loss function, .

[0060] Update model .in, For learning rate, It is an indicator function, when the sample Belongs to the region The value is 1 when the time is right, and 0 otherwise. For the m-th tree Leaf node regions.

[0061] A Bayesian optimization method is used to automatically search for the optimal hyperparameter combination. The optimization objective is the average performance of k-fold cross-validation. The number of trees M is 100-200, and the optimal value is determined by early stopping. The learning rate ν is 0.1-0.3 to balance training speed and generalization ability. The subsampling ratio is 0.8-1.0 to control the proportion of training samples used. The feature sampling ratio is 0.8 to increase the diversity of base learners. The optimization process uses Gaussian process regression to model the relationship between hyperparameters and model performance, and the search direction is guided by the acquisition function.

[0062] The model performance is evaluated on the test set. R² should be greater than 0.85, RMSE should be less than 5% of the predicted sensor index physical range, and MAE should be less than 3% of the predicted sensor index physical range. If the model's performance on the validation or test set does not meet the above standards, the optimization process needs to be initiated. If the prediction accuracy indicators such as R², RMSE, and MAE do not meet the standards, the training data should be checked and cleaned first, and outliers and missing values ​​should be handled. If the prediction accuracy is still insufficient after data optimization, the process returns to the hyperparameter tuning stage. Bayesian optimization or grid search is then performed again using the validation set to adjust parameters such as tree depth, learning rate, and number of subtrees in the GBDT model.

[0063] The trained prediction model is also subjected to incremental training periodically. As a preferred embodiment, this embodiment uses newly added data to incrementally train the model every quarter to adapt to the slow changes in the environment and disease patterns. At the same time, when the model's performance indicators are detected to be declining continuously and approaching the threshold, the model retraining and comprehensive evaluation process is immediately triggered.

[0064] Input the weather data for the next seven days into the trained prediction model The forward prediction is performed, and the prediction results are inversely normalized to obtain the actual physical quantities, generating a sensor data prediction sequence for the next seven days: In addition, the quantile regression GBDT (QR-GBDT) method is used to simultaneously predict the conditional quantiles Q05, Q50, and Q95 of the sensor indicators, providing a prediction interval estimate. The relative uncertainty index is then calculated based on the prediction interval width.

[0065] .

[0066] Based on the output of the aforementioned mapping model, a comprehensive risk score is obtained through different calculations of the indicator scores of various sensor data, and a dynamic design of the warning threshold is implemented to achieve refined warning levels and trigger differentiated responses. This includes two steps: first, calculating the indicator score corresponding to each sensor data based on its indicator scoring function, and then weighting and accumulating the scores to obtain a comprehensive risk score; second, setting a dynamic warning threshold, calculating the percentage of the comprehensive risk score relative to the dynamic warning threshold, and triggering different warning signals based on different percentages.

[0067] The formula for calculating the comprehensive risk score is:

[0068] ,

[0069] in, For the i-th predicted sensor data, This is the scoring function for the indicator, calculated based on the cumulative effective time. The weighting coefficients are set as follows: air temperature 20%, air humidity 20%, soil moisture 10%, soil temperature 15%, leaf surface wet duration 20%, and light intensity 15%. The weighting coefficients are dynamically adjusted based on environmental characteristics. The adjustment mechanism includes:

[0070] Different dynamic adjustment factors (each value) are applied according to the growing season: Spring (March-May): Air temperature base weight × 1.2, Air humidity base weight × 1.3, Soil moisture base weight × 0.9, Soil temperature base weight × 1.1, Leaf surface wet duration base weight × 1.4, Light intensity base weight × 0.8; Summer (June-August): Air temperature base weight × 1.4, Air humidity base weight × 1.1, Soil moisture base weight × 0.7, Soil temperature base weight × 1.3, Leaf surface wet duration base weight × 1.2, Light intensity base weight × 1.0; Autumn (September-November): Air temperature base weight × 1.1, Air humidity base weight × 1.2, Soil moisture base weight × 1.1, Soil temperature base weight × 1.0, Leaf surface wet duration base weight × 1.3, Light intensity base weight × 0.9; Warnings are stopped after leaf fall until bud break in the following year, and only environmental indicators are monitored.

[0071] The dynamically adjusted weights need to be normalized to ensure that the sum of all weight coefficients is 100%. The normalization formula is as follows:

[0072] ,

[0073] in The base weights are A and A is the dynamic adjustment factor. These are the normalized actual weights.

[0074] The scoring criteria are as follows, based on the cumulative effective time of the predicted sensor data over seven days:

[0075] Regarding air temperature (°C), temperatures below 15°C or above 35°C significantly inhibit pathogen activity. 20–25°C is the optimal period for spore germination, while 25–30°C is the optimal period for mycelial growth and disease outbreaks.

[0076] Air temperature scoring function The settings are based on 7 days of air temperature in °C: if the cumulative time of air temperature in the range of [15, 35] is less than 60 hours, it is considered a suboptimal temperature, and the disease develops slowly. =20; When the cumulative time for air temperature to be in the range of [15,20) or (30,35] is ≥60 hours, the suboptimal temperature persists and begins to pose a risk. =50; when the cumulative time of air temperature in the range of [20, 25) is in the range of [60, 120) hours, spore activity is enhanced, resulting in efficient germination. =70; the cumulative time of air temperature in the range of [25,30] is in the range of [60,120) hours, indicating a rapid mycelial growth period and a higher risk. =90; if the cumulative time of air temperature in the range of [20,30] is ≥120 hours, the mycelium may have already completed infection, posing an extremely high risk. =100, otherwise take =100. =10.

[0077] Regarding air humidity (%), humidity is crucial for spore germination. Humidity below 45% is too low, making spore germination difficult; 60%–80% is the optimal range; humidity above 80% may inhibit conidia formation but favors other diseases, thus reducing its contribution to powdery mildew risk.

[0078] Air humidity scoring function The setting is based on 7 days of air humidity as a percentage: when the cumulative time of air humidity in the range of [45, 60) is less than 30 hours, spore activity is inhibited. =10; the cumulative time of air humidity in the range [45, 60) is ≥120 hours, which is considered sub-suitable humidity. =40; the cumulative time of air humidity in the range of [60, 80) is in the range of [60, 120) hours, which is the optimal humidity period for spore germination. =75; the cumulative time of air humidity in the range [60, 80) is ≥120 hours, and the optimal humidity lasts throughout the entire infection cycle. =100; air humidity >80 or the cumulative time in [45,60) is in [30,120) hours or the cumulative time in [60,80) is <60 hours, high humidity may inhibit conidia, but the plant is in a susceptible environment. =30; otherwise, take 30. =10.

[0079] Regarding the duration of leaf surface moisture (h), the presence of a water film or high humidity on the leaf surface is a necessary condition for spore germination, but powdery mildew is not prone to outbreaks in humid environments. There is an optimal risk window for the daily duration of moisture.

[0080] Leaf surface moisture duration scoring function The setting is: >120, =10; [60,120], =30; <30, =60; [30,60), =100.

[0081] Regarding light intensity (Lux), low light and diffused light conditions are conducive to the occurrence of powdery mildew, while strong light conditions are unfavorable to its occurrence.

[0082] Light intensity scoring function The setting is based on 7 days of light intensity measured in Lux: the cumulative time of light intensity in the range of [15000, 20000] is >60 hours. =30; the cumulative time of light intensity in the range [10000, 15000) is >60 hours. =60; Light intensity <Cumulative time of 10000> 60 hours =100; otherwise, take 100. =10.

[0083] Soil temperature (°C) directly affects root vitality and nutrient absorption efficiency, which in turn affects the growth rate of new shoots, leaf maturity, and the synthesis of disease-resistant compounds, indirectly determining the susceptibility of aboveground tissues to pathogens.

[0084] Soil temperature scoring function The soil temperature is defined as 7 days in °C: the cumulative time of soil temperature in [15,18) or [25,28) is >120 hours, or the cumulative time of soil temperature in [28,30) is >48,72 hours. =40; the cumulative time of soil temperature in [10,15) or [28,30) > 72 hours, =70; cumulative time of soil temperature <10 or >30 >48 hours, =100; otherwise, take 100. =10.

[0085] Soil moisture (%) significantly affects the leaf surface microenvironment and the overall physiological resistance of the plant by causing water stress in the plant, altering the turgor pressure of leaf cells, stomatal opening and closing, and the concentration of osmotic regulators.

[0086] Soil Moisture Scoring Function The soil moisture content is defined as a percentage over 7 days: the cumulative time for soil moisture in the range of [50, 60) or [75, 85) is >120 hours, or the cumulative time for soil moisture in the range of [40, 50) or [85, 90] is >48, 72 hours. =40; the cumulative time of soil moisture in [40,50) or [85,90] > 72 hours. =70; cumulative time of soil moisture <40% or >90% >48 hours. =100; otherwise, take 100. =10.

[0087] The dynamic early warning threshold is based on a three-level dynamic adjustment mechanism that adjusts the threshold according to the level of uncertainty. The calculation formula is:

[0088] ,

[0089] in: To predict the standard deviation, the dispersion of multiple predicted values ​​from the same sensor data output is calculated using a mapping model. In this embodiment, quantile regression is used to obtain the predicted values ​​for Q05, Q50, and Q95. Therefore, The standard deviation of these quantile predictions can be estimated using the sample standard deviation, calculated as follows: ,in, This represents the predicted value of the k-th quantile in the model output. The average of these predicted values ​​is given by N, where N is the number of quantiles used in the calculation.

[0090] This is the historical prediction standard deviation, calculated based on historical predictions from sensor data. The formula is: Where M represents the historical prediction range. It is the residual of the j-th historical prediction. It is the average of these residuals.

[0091] The sensitivity coefficient is determined by the relative uncertainty index (Uncertainty) obtained during forecasting. When Uncertainty < 15%, it represents low uncertainty, and the sensitivity coefficient is [value missing]. =0.2; when 15% ≤ Uncertainty <35%, it is considered medium uncertainty, and the sensitivity coefficient is [value missing]. =0.3; when Uncertainty ≥ 35%, it is considered high uncertainty, and the sensitivity coefficient is [value missing]. =0.4.

[0092] The basic early warning threshold is set based on historical disease data statistics. The calculation method is as follows: For each historical disease case j, environmental data for the 7 days prior to the onset of the disease is extracted, and the comprehensive risk score S before the onset of the disease is calculated. The comprehensive risk scores S of all historical cases are used to form a sample set, which is then sorted in ascending order and determined using the 90th percentile method. , This means that 90% of historical disease events occurred before the risk score was below this threshold.

[0093] The warning level classification is based on the comprehensive risk score S and the dynamic threshold. Comparison: When When, no warning is triggered; when When this occurs, a Level 1 warning is triggered, prompting precautions to prevent disease outbreaks; when When a Level 2 warning is triggered, indicating a moderate risk, a patrol notification is sent; when When a Level 3 warning is triggered, indicating a high risk, prepare for targeted spraying; when When a Level 4 warning is triggered, targeted medication is immediately administered.

[0094] The predictions presented in the above embodiments, employing forward-looking forecasts based on meteorological data and sensor indicators, combined with dynamic early warning grading, provide reliable technical support for early warning of grape powdery mildew. The early warning accuracy rate is calculated within the corresponding time period of the test set. Specifically, it is the proportion of predicted disease occurrence events that match actual disease occurrence records. If a prediction triggers an early warning at a future time period, and the disease actually occurs within a 7-day time window following that period, it is counted as a correct early warning. That is, early warning accuracy rate = (number of correctly predicted disease events / total number of actual disease events) × 100%. Testing shows that the early warning accuracy rate can reach over 90%.

Claims

1. A method for early warning of powdery mildew in wine grapes based on the Internet of Things, characterized in that, include: Obtain meteorological data for a future time period; Meteorological data for future time periods are input into a training model that maps meteorological data to sensor data to obtain predicted sensor data. The sensor data includes the air temperature and humidity of the environment in which the wine grapes are located, the duration of leaf wetness of the wine grapes, and the light intensity received by the wine grapes. The index score corresponding to each sensor data is calculated based on the index scoring function of each sensor data, and then weighted and accumulated to obtain the comprehensive risk score; Set a dynamic early warning threshold, calculate the percentage of the comprehensive risk score relative to the dynamic early warning threshold, and trigger different early warning signals based on different percentages.

2. The method for early warning of powdery mildew in wine grapes based on the Internet of Things according to claim 1, characterized in that, The sensor data includes the soil temperature and soil moisture of the environment in which the wine grapes are located.

3. The method for early warning of powdery mildew in wine grapes based on the Internet of Things according to claim 1, characterized in that, The index scoring function includes the air temperature scoring function. Air humidity scoring function Leaf surface wetting duration scoring function Light intensity scoring function ,in, Based on 7 days of air temperature in °C: the cumulative time of air temperature in the range [15, 35] is less than 60 hours. =20; The cumulative time when the air temperature is in [15,20) or (30,35] is ≥60 hours. =50; the cumulative time for air temperature in [20,25) is in [60,120) hours. =70; the cumulative time for air temperature to be in the range of [25,30] is in the range of [60,120) hours. =90; The cumulative time the air temperature is in the range of [20,30] is ≥120 hours. =100, otherwise take =100. =10; Based on 7 days of air humidity as a percentage: the cumulative time of air humidity in the range [45, 60) is less than 30 hours. =10; the cumulative time of air humidity in the range [45, 60) is ≥120 hours. =40; the cumulative time for air humidity in the range [60, 80) is in the range [60, 120) hours. =75; the cumulative time of air humidity in the range [60, 80) is ≥120 hours. =100; air humidity >80 or the cumulative time in [45,60) is in [30,120) hours or the cumulative time in [60,80) is <60 hours. =30; otherwise, take 30. =10; Based on the duration of leaf surface moisture in hours over a 7-day period: Leaf surface moisture duration >120, =10; the leaf surface remains moist for a duration of [60, 120]. =30; Leaf surface wetness duration <30 =60; Leaf surface wetness duration [30,60), =100; Based on 7-day light intensity measured in Lux: The cumulative time of light intensity in the range [15000, 20000] > 60 hours. =30; the cumulative time of light intensity in the range [10000, 15000) is >60 hours. =60; Light intensity <Cumulative time of 10000> 60 hours =100; otherwise, take 100. =10.

4. The method for early warning of powdery mildew in wine grapes based on the Internet of Things according to claim 2, characterized in that, The index scoring function includes the soil temperature scoring function. Soil moisture scoring function ,in Based on soil temperature in °C over 7 days: A cumulative time of soil temperature in the range [15, 18) or [25, 28) > 120 hours, or a cumulative time of soil temperature in the range [28, 30) > [48, 72] hours. =40; the cumulative time of soil temperature in [10,15) or [28,30) > 72 hours, =70; cumulative time of soil temperature <10 or >30 >48 hours, =100; otherwise, take 100. =10; Soil moisture is categorized by percentage over 7 days: Soil moisture in the range of [50, 60) or [75, 85) for a cumulative period >120 hours, or soil moisture in the range of [40, 50) or [85, 90] for a cumulative period >48, 72 hours. =40; the cumulative time of soil moisture in [40,50) or [85,90] > 72 hours. =70; cumulative time of soil moisture <40% or >90% >48 hours. =100; otherwise, take 100. =10.

5. The method for early warning of powdery mildew in wine grapes based on the Internet of Things according to claim 1, characterized in that, The dynamic early warning threshold is determined based on the product of the early warning threshold benchmark value and the dynamic coefficient. The dynamic coefficient is obtained by multiplying the standard deviation of the predicted sensor data by the standard deviation of the historical prediction by the sensitivity coefficient and then adding 1. The sensitivity coefficient has different values ​​based on the prediction uncertainty.

6. The method for early warning of powdery mildew in wine grapes based on the Internet of Things according to claim 5, characterized in that, The warning threshold benchmark is set based on historical disease data statistics, including: extracting sensor data before the onset of each historical disease case and calculating the corresponding comprehensive risk score, arranging the comprehensive risk scores of all historical cases in ascending order, and obtaining the comprehensive risk score that was lower than 90% of the historical disease time before the occurrence as the warning threshold benchmark.

7. The method for early warning of powdery mildew in wine grapes based on the Internet of Things according to claim 5, characterized in that, The prediction uncertainty is determined by the ratio of the difference between the conditional quantiles Q95 and Q05 of the sensor data predicted by the mapping model to Q50.

8. The method for early warning of powdery mildew in wine grapes based on the Internet of Things according to claim 1, characterized in that, During the weighted accumulation, the weighting coefficients for the indicator scores corresponding to each sensor data are adjusted according to different seasons.

9. The method for early warning of powdery mildew in wine grapes based on the Internet of Things according to claim 1, characterized in that, The mapping model is established by the gradient boosting decision tree algorithm, and the trained mapping model is incrementally trained according to time periods.

10. An Internet of Things-based early warning system for powdery mildew in wine grapes, characterized in that, include: The data acquisition module is used to acquire meteorological data for future time periods; The data prediction module is used to input meteorological data for future time periods into the trained meteorological data to sensor data mapping model to obtain predicted sensor data. The sensor data includes the air temperature and humidity of the environment where the wine grapes are located, the duration of leaf wetness of the wine grapes, and the light intensity received by the wine grapes. The risk score calculation module is used to calculate the index score corresponding to each sensor data based on the index scoring function of each sensor data, and perform weighted summation to obtain the comprehensive risk score; The early warning module is used to set a dynamic early warning threshold, calculate the percentage of the comprehensive risk score relative to the dynamic early warning threshold, and trigger different early warning signals based on different percentages.

Images