Method for predicting tobacco field yield based on multi-source data analysis

Abstract

The invention discloses a method for realizing tobacco field yield prediction based on multi-source data analysis, and specifically relates to the technical field of tobacco yield prediction. Firstly, a tobacco planting area is divided into monitoring areas, and multi-source data such as meteorology, soil, remote sensing images and actual yield are acquired in real time. The collected data are pre-processed, and spatiotemporal consistency features and information integrity features are extracted. A comprehensive matching index is calculated and weighted to generate. On this basis, a yield prediction model is established, and the predicted value is compared and analyzed with the actual yield. If a large deviation is found, the model parameters are dynamically adjusted through fuzzy logic based on the comprehensive matching index and the actual yield data, and the data processing and fusion algorithms are optimized, thereby significantly improving the prediction accuracy and reducing the field management risk caused by data heterogeneity.

Description

Technical Field

[0001] The invention relates to the technical field of tobacco yield prediction, and in particular to a method for realizing tobacco field yield prediction based on multi-source data analysis. Background Art

[0002] Tobacco field yield prediction based on multi-source data analysis refers to the use of data from multiple sources to scientifically predict the actual yield of tobacco fields through data analysis and modeling techniques. This method combines multi-dimensional information sources, such as meteorological data, soil characteristics, agronomic measures, pest and disease monitoring, and remote sensing images, and integrates and analyzes these data to reveal the key factors affecting tobacco yield and establish a yield prediction model. This prediction can help tobacco growers understand the potential yield of the field in advance and optimize management decisions.

[0003] For example, in a tobacco-growing area, farmers want to know the tobacco yield of the season. The research team collected meteorological data (such as rainfall, temperature, and light), soil data (such as pH value, organic matter content), and tobacco yield data in previous years in the area. In addition, drones were used to obtain field remote sensing images to analyze crop growth conditions and vegetation index (NDVI). By inputting these data into the machine learning model, the research team successfully predicted the tobacco yield of the season. The forecast results show that due to insufficient rainfall in the season, tobacco yields are expected to be low in some areas. Based on this analysis, farmers can take measures in advance, such as increasing irrigation frequency or adjusting fertilization strategies, to make up for potential yield losses.

[0004] The prior art has the following deficiencies:

[0005] There may be inconsistencies in the acquisition time, spatial resolution, and format of multi-source data. For example, remote sensing images may be provided in a form of monthly updates and low resolution, which makes it difficult to capture vegetation details within a small area of ​​the field, while meteorological data are updated daily but have wide-area coverage, and soil data rely on point sampling and are sparsely distributed. This heterogeneity leads to data mismatches in time and space, and direct integration may introduce biases, resulting in the prediction model being unable to accurately reflect actual field conditions. In addition, due to the mismatch of data in time and space, the model may make incorrect assessments of tobacco conditions at critical growth stages. For example, low-resolution remote sensing data cannot capture small areas of drought in the field, and point soil data can hardly reveal local nutrient deficiencies. This can cause the model to mistakenly believe that the field is growing well and ignore the yield risks in problem areas. If growers do not take timely remedial measures (such as local irrigation or fertilization) based on incorrect predictions, it may result in a reduction in yield in the entire field or even a complete loss of local tobacco plants. Summary of the invention

[0006] The purpose of the present invention is to provide a method for predicting tobacco field yield based on multi-source data analysis to overcome the shortcomings of the background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a method for predicting tobacco field yield based on multi-source data analysis, comprising the following steps:

[0008] S1: Divide the tobacco planting area into several monitoring areas, and obtain multi-source data in different time periods in each monitoring area in real time, the multi-source data including meteorological data, soil data and remote sensing image data, as well as actual yield data in each monitoring area:

[0009] S2: After preprocessing the collected multi-source data, extract the spatiotemporal consistency features and information integrity features in the multi-source data, determine the weight assignment of the temporal and spatial matching degree of the multi-source data in each time period, and calculate the weighted average sum to obtain the comprehensive matching index of the multi-source data;

[0010] S3: Establish a yield prediction model and compare and analyze the actual yield with the prediction results. If the yield prediction results have a large deviation, dynamically adjust the parameters of the yield prediction model through fuzzy logic based on the calculated comprehensive matching index of multi-source data and the actual yield data in each monitoring area, and optimize the data processing and fusion algorithm.

[0011] Preferably, in S2, the extracted time consistency feature is analyzed to generate a time coverage anomaly index, and the method for obtaining the time coverage anomaly index is:

[0012] Set the input time coverage data as a time series ;in: is the coverage rate of the Nth day in the time series, ranging from [0, 1], where N is the total number of days; define the window size w, which determines the time range used to calculate the coverage feature each time, set the window step size s, which represents the step size of the window movement, and calculate the window mean for each window [t, t+w−1]: ; is the coverage rate of the i-th day in the time series, is the average coverage in the tth window, and the window standard deviation is calculated: ; is the standard deviation of coverage in the t-th window, for each time point , calculate the coverage anomaly index, the expression is: ; Where HZK is the time coverage anomaly index.

[0013] Preferably, in S2, the integrity of the extracted spatial sampling points is analyzed to generate a sampling point deviation index, and the sampling point deviation index is obtained by:

[0014] Set the geographic coordinates of the sampling points to and sample values , n is the total number of sampling points, the range of the target interpolation area is [Xmin, Xmax] × [Ymin, Ymax], the target area is divided into uniform grids, and the center of each grid is the point to be interpolated;

[0015] Calculate any two points and The Euclidean distance , the expression is: ; Calculate the corresponding semivariance value , the expression is: ; For point The attribute value at represents the sampling point The observed value of Indicate point The attribute value at represents the sampling point Observed value; According to the relationship between the calculated semivariance value and the distance h, the semivariance function model is selected: ; In the formula, is the base, indicating the measurement error of the sampling point, C is the step value, indicating the maximum variation value, and a is the range, indicating the maximum influence distance of spatial correlation;

[0016] For the points to be interpolated , set the weights of n sampling points to , the Kriging equation is: ; In the formula, is the semivariance between sampling points i and j, is the semi-variance value between the sampling point i and the point to be interpolated, μ is the Lagrange multiplier, solve the equations and get the weight ;

[0017] According to weight , calculate the predicted value of the point to be interpolated , the expression is: ; Calculate interpolation error , the calculation formula is: ; In the formula, is the sill value of the semivariogram, is the semi-variance value between the sampling point j and the interpolation point ∗, which measures the degree of variation between the two points. The calculation results are filled into the target area grid, the error distribution map is generated, and the sampling point deviation index is calculated. The expression is: ; In the formula, It is a small positive number to prevent the denominator from being zero, and GHZ is the sampling point deviation exponent.

[0018] Preferably, in S2, the weighted assignment of the degree of temporal and spatial matching of multi-source data in each time period is determined, specifically: the time coverage anomaly index and the sampling point deviation index are converted into a comprehensive feature vector, and the comprehensive feature vector is used as the input of the machine learning model. The machine learning model uses each group of comprehensive feature vectors to predict the weighted assignment label of the degree of temporal and spatial matching of multi-source data in each time period as the prediction target, and takes minimizing the sum of the prediction errors of the weighted assignment labels of the degree of temporal and spatial matching of multi-source data in all time periods as the training target. The machine learning model is trained until the sum of the prediction errors converges and the model training is stopped. The weighted assignment of the degree of temporal and spatial matching of multi-source data in each time period is determined according to the model output results, wherein the machine learning model is a polynomial regression model, and the weighted average summation calculation of the weighted assignments of the degree of temporal and spatial matching of multi-source data in each time period is performed to obtain the comprehensive matching index of the multi-source data.

[0019] Preferably, in S3, the actual output is compared and analyzed with the prediction result, specifically: collecting multi-source data as input features, selecting a prediction model according to the data scale and feature dimension, and dividing the data set into a training set, a validation set, and a test set; fitting model parameters using the training set, adjusting hyperparameters using the validation set, and evaluating model performance using the test set;

[0020] Using the trained model, input the validation set or new multi-source data into the model to generate the predicted tobacco yield of each monitoring area; collect the actual yield data of the corresponding area from the plot monitoring records, and calculate the yield deviation of each area, that is, the difference between the predicted tobacco yield and the actual tobacco yield;

[0021] According to the actual output, the area is divided into: Low-yield area: , where P25 is the lower quartile of actual yield; medium-yield area: actual yield is between P25 and P75; high-yield area: , where P75 is the upper quartile;

[0022] If the yield deviation ΔY is less than the preset yield deviation threshold of the low-yield area, it means that the yield prediction result has a large deviation and the model fails to fully capture the impact of key characteristics;

[0023] If the yield deviation value ΔY is greater than the preset yield deviation threshold of the high-yield area, it means that the yield prediction result has a large deviation, the model overfits certain features, and there is overfitting.

[0024] Preferably, in S3, according to the calculated comprehensive matching index of multi-source data and the actual production data in each monitoring area, the parameters of the production prediction model are dynamically adjusted through fuzzy logic, specifically:

[0025] The comprehensive matching index LZ of multi-source data and the actual production data Y are used as the input items of fuzzy logic, and the adjustment results of the prediction model parameters are used as the output items of fuzzy logic;

[0026] The comprehensive matching index LZ, actual production data Y and the parameter adjustment results of the model are fuzzified;

[0027] According to the combination relationship between LZ and Y, fuzzy rules are formulated;

[0028] Convert LZ and Y into fuzzy sets, and use membership functions to define the fuzziness of different states. The membership function defines: Low matching degree: ; Medium match: defined using triangular or trapezoidal membership functions; High match: increasing membership functions, applicable to LZ ≥ 0.7;

[0029] According to the fuzzy rule table, the centroid method is used to infer the output;

[0030] Calculate the matching degree between the input fuzzy set and the rule conditions, select the rules that meet the conditions and calculate the activation degree of the output membership function;

[0031] Convert the fuzzy output set into crisp values ​​and use the centroid method to calculate the specific value of the parameter adjustment result , is the clarity value adjusted by the model, expressed as: ; x is all possible values ​​of the output variable, indicating the range of the parameter adjustment amount, μ(x) is the membership function, indicating the membership of the output variable x in the fuzzy set;

[0032] According to the output of the fuzzy logic system, the optimization strategy of the model is determined: O<0.3: no adjustment is required; 0.3≤O<0.6: slight adjustment; O≥0.6: significant adjustment.

[0033] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0034] 1. The present invention improves the accuracy and reliability of tobacco field yield prediction through comprehensive analysis and dynamic optimization of multi-source data. By dividing the monitoring area and collecting meteorological, soil, remote sensing images and actual yield data in real time, the problem of inconsistent time and spatial resolution and format of multi-source data is solved; the comprehensive matching index is calculated using the spatiotemporal consistency characteristics and information integrity characteristics, and the yield prediction model parameters are dynamically adjusted in combination with fuzzy logic, which significantly reduces the impact of data heterogeneity on model accuracy, and optimizes data processing and fusion algorithms to effectively capture the dynamic changes of key growth stages.

[0035] 2. The present invention can conduct in-depth analysis on the characteristic impact of low-yield and high-yield areas according to yield deviation, and timely adjust the model to avoid underestimation or overfitting. Through the fuzzy logic system combined with the matching index and actual yield data, a clear model optimization strategy is output, which realizes the intelligent adjustment of the yield prediction model, significantly improves the adaptability and robustness of the model, reduces the risk of field management, and ultimately improves the efficiency and economic benefits of tobacco planting. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0037] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] For examples, see Figure 1 As shown, the method for realizing tobacco field yield prediction based on multi-source data analysis described in this embodiment includes the following steps:

[0040] S1: Divide the tobacco planting area into several monitoring areas, and obtain multi-source data in different time periods in each monitoring area in real time, the multi-source data including meteorological data, soil data and remote sensing image data, as well as actual yield data in each monitoring area:

[0041] S2: After preprocessing the collected multi-source data, extract the spatiotemporal consistency features and information integrity features in the multi-source data, determine the weight assignment of the temporal and spatial matching degree of the multi-source data in each time period, and calculate the weighted average sum to obtain the comprehensive matching index of the multi-source data;

[0042] S3: Establish a yield prediction model and compare and analyze the actual yield with the prediction results. If the yield prediction results have a large deviation, dynamically adjust the parameters of the yield prediction model through fuzzy logic based on the calculated comprehensive matching index of multi-source data and the actual yield data in each monitoring area, and optimize the data processing and fusion algorithm.

[0043] Divide the tobacco growing area into several monitoring areas:

[0044] The tobacco-growing areas are divided into several monitoring units according to the geographical scope, topographical characteristics and planting density of the growing areas.

[0045] The division of each monitoring area can be reasonably defined based on GPS coordinates, partition area or natural boundaries (such as irrigation systems, ridges).

[0046] Use geographic information system (GIS) technology or high-precision mapping by drones to generate a digital boundary map of the monitoring area, providing a basis for subsequent data collection and analysis.

[0047] Real-time acquisition of meteorological data in different time periods in each monitoring area:

[0048] Deploy automatic weather stations in the monitoring area or connect to third-party meteorological services to collect meteorological data such as rainfall, temperature, humidity, light intensity, wind speed and evaporation in different time periods.

[0049] The frequency of meteorological data collection can be set to minutes or hours to reflect the dynamic impact of real-time weather changes on tobacco growth.

[0050] Through wireless sensor networks (WSN) or Internet of Things (IoT) technology, meteorological data is transmitted to the data center in real time to ensure timeliness and integrity.

[0051] Real-time acquisition of soil data in different time periods in each monitoring area:

[0052] A soil sensor network is deployed in each monitoring area to monitor the dynamic changes of soil pH, organic matter content, nutrient concentration (such as nitrogen, phosphorus, potassium), water and salinity in real time.

[0053] The sensor data is supplemented by ground sampling to generate a spatial distribution map of soil characteristics to make up for the shortcomings of point sensor deployment.

[0054] The sensor data is uploaded via wireless transmission technology and combined with timestamps to record soil conditions in different time periods, providing accurate input for the prediction model.

[0055] Real-time acquisition of remote sensing image data in different time periods in each monitoring area:

[0056] Use multispectral and hyperspectral sensors carried by drones or satellites to regularly obtain remote sensing image data in the monitoring area, such as vegetation index (NDVI, EVI) and canopy coverage.

[0057] The frequency of remote sensing image acquisition can be adjusted according to the tobacco growth stage. For example, it can be increased to once a week during the growth period to ensure that key growth information is obtained.

[0058] Remote sensing image analysis can be used to generate plot-level crop health maps that reflect differences in crop growth and potential problem areas (such as drought or pests and diseases).

[0059] Obtain the actual production data of each monitoring area in real time:

[0060] After tobacco is harvested, actual yield data (such as dry tobacco leaf weight, quality grade, etc.) of each monitoring area are counted through plot-level records.

[0061] Use precise weighing equipment or automatic recording systems to reduce manual statistical errors and ensure the accuracy of the correspondence between production data and regions.

[0062] Correlation analysis is performed between yield data and multi-source data of growth stages to provide basic data support for model correction and optimization of management measures.

[0063] Integrate multi-source data and build a time series database:

[0064] According to the monitoring area and time period, meteorological, soil, remote sensing and yield data are integrated into spatiotemporal sequences and stored in a unified database.

[0065] The database supports regional queries and multi-time period comparisons, providing multi-dimensional data support for prediction and decision-making.

[0066] The spatiotemporal consistency feature is used to measure the degree of alignment of multi-source data in time and space to ensure the accuracy and reliability of data fusion.

[0067] The time consistency feature is used to measure the synchronization of different data sources in terms of collection time and the matching degree of update frequency, including: Calculating the time deviation between data sources: For each data point, calculate the difference between its timestamp and the timestamp of the benchmark data. Statistical distribution of time deviation (such as mean, standard deviation) as time consistency feature. Calculating time coverage: The ratio of the number of time points of actual data collection in a certain period to the total number of theoretical time points, which is used to evaluate the integrity of time sampling.

[0068] Spatial consistency features are used to measure the matching degree of multi-source data in spatial resolution and coverage, including: Spatial overlap rate: Calculate the degree of spatial coverage overlap between each data source and the target area: Spatial resolution difference: Compare the resolutions of different data sources (such as the pixel size of remote sensing images and the spacing between soil sampling points), and record the statistical characteristics of resolution differences. Point density distribution: Analyze the distribution density of soil sensors or sampling points, calculate the number of effective points per unit area, and quantify the coverage uniformity of spatial data.

[0069] Information integrity features are used to assess the data's temporal and spatial coverage, as well as the integrity and validity of collected data.

[0070] The time integrity feature is used to measure the continuity and missingness of multi-source data in the time dimension, including: Time interval integrity: calculate the interval distribution in the time series, and count the number and proportion of interval anomalies (such as too long sampling intervals or mutations). Effective time coverage: Count the ratio of the actual number of data collection points to the theoretical number of data collection points in the target time period. Missing segment length distribution: Analyze the length of the missing segment of the data (the time of continuous missing), and record the maximum, minimum and distribution characteristics of the missing segment.

[0071] Spatial integrity features are used to measure the coverage and effectiveness of multi-source data in the spatial dimension, including: Regional coverage integrity: Calculate the proportion of valid data points in the target area to the total area; Spatial sampling point integrity: Analyze the distribution of point data sources (such as soil sensors) and count the area and proportion of uncovered areas. Remote sensing data image integrity: Calculate the proportion of valid pixels to total pixels through cloud interference or image missing area detection.

[0072] Data validity features are used to measure the proportion of outliers or invalid data points in multi-source data, including: Outlier detection: use statistical methods (such as box plot method) or machine learning models (such as isolation forest) to mark outliers and calculate the proportion of outliers. Data repetition rate: calculate the proportion of data points recorded repeatedly in time or space to evaluate the independence of the data.

[0073] After analyzing the extracted time consistency features, the time coverage anomaly index is generated. The method for obtaining the time coverage anomaly index is as follows:

[0074] Set the input time coverage data to a time series: ;in: is the coverage rate of the Nth day in the time series, ranging from [0, 1], and N is the total number of days.

[0075] Define the window size w: w determines the time range used for each calculation of the coverage feature, usually 5 to 15 days (for example, w=7 means a window of 7 days). Window step s: s represents the step size of the window movement, usually 1, that is, moving day by day.

[0076] For each window [t, t+w−1], calculate the window mean: ; is the coverage rate of the i-th day in the time series, is the average coverage in the tth window, and the window standard deviation is calculated: ; is the standard deviation of the coverage rate in the tth window, reflecting volatility.

[0077] For each time point , calculate the coverage anomaly index, the expression is: ; Where HZK is the time coverage anomaly index.

[0078] A large anomaly index indicates that the coverage rate at a certain point in time deviates greatly from the average value of its time window. There may be problems such as inconsistent data collection frequency, missing data from a certain source, and uneven time distribution. For example, low-frequency remote sensing data and high-frequency meteorological data cannot be synchronized, resulting in a sudden increase or decrease in coverage rate. Poor matching will cause data fusion bias, making it difficult for the model to accurately capture the dynamic changes in key growth stages, thereby reducing the reliability of the prediction.

[0079] A smaller anomaly index indicates that the temporal coverage fluctuates less in different time periods, the frequency of multi-source data collection is relatively stable and the temporal distribution is uniform, and the coverage is consistent with expectations. In this case, data fusion can more accurately reflect the dynamic changes in the field environment and provide reliable input for subsequent analysis and prediction. In addition, a high degree of matching helps to identify small regional changes, improve the ability to identify problem areas and improve the accuracy of decision-making.

[0080] After analyzing the integrity of the extracted spatial sampling points, the sampling point deviation index is generated. The sampling point deviation index is obtained as follows:

[0081] Set the geographic coordinates of the sampling points to and sample values , n is the total number of sampling points. The target interpolation area ranges from [Xmin, Xmax] × [Ymin, Ymax], and the target area is divided into uniform grids, with each grid center being the point to be interpolated.

[0082] Calculate any two points and The Euclidean distance , the expression is: ; Calculate the corresponding semivariance value , the expression is: ; For point The attribute value at represents the sampling point Observed values ​​(such as soil nutrient concentration, vegetation index NDVI, etc.), Indicate point The attribute value at represents the sampling point Observed value of

[0083] Based on the relationship between the calculated semivariogram and the distance h, select the semivariogram model (such as spherical model, exponential model, Gaussian model). Example spherical model: ; In the formula, is the Nugget, which indicates the measurement error of the sampling point; C is the Sill, which indicates the maximum variation value; a is the Range, which indicates the maximum influence distance of spatial correlation.

[0084] For the points to be interpolated , set the weights of n sampling points to , the Kriging equation is: ; In the formula, is the semivariance between sampling points i and j, is the semivariance between sampling point i and the point to be interpolated, μ is the Lagrange multiplier, which is used to ensure that the weight sum is 1. Solve the equations to get the weight .

[0085] According to weight , calculate the predicted value of the point to be interpolated , the expression is: ; Calculate interpolation error , represents the uncertainty of interpolation, and the calculation formula is: ; In the formula, is the sill value (i.e. Nugget) of the semivariogram, is the semi-variance value between the sampling point j and the interpolation point ∗, which measures the degree of variation between the two points. The calculation results are filled into the target area grid, the error distribution map is generated, and the sampling point deviation index is calculated. The expression is: ; In the formula, It is a small positive number to prevent the denominator from being zero, and GHZ is the sampling point deviation exponent.

[0086] A large sampling point deviation index indicates that there is a significant deviation in the spatial distribution of sampling points in the target area, resulting in a high interpolation error. This may be because the sampling points are too concentrated in certain areas and fail to evenly cover the entire target area, or some areas lack the necessary sampling point support. In addition, inconsistent data in the time dimension (such as changes in the location of sampling points in different time periods) may also lead to a decrease in the degree of matching. As a result, the fusion quality of multi-source data is reduced, affecting the accurate assessment of dynamic changes in local areas.

[0087] A smaller sampling point deviation index means that the sampling points are evenly distributed in the target area, which can better reflect the overall characteristics of space and time, and the interpolation error is low. This shows that the multi-source data has a high degree of matching in spatial coverage and time synchronization, especially the high degree of coordination between point data (such as soil sampling points) and regional data (such as remote sensing images). A good degree of matching helps to improve the accuracy of multi-source data fusion, thereby providing a reliable basis for subsequent analysis and prediction.

[0088] The time coverage anomaly index and the sampling point deviation index are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as the input of the machine learning model. The machine learning model uses each group of comprehensive feature vectors to predict the weighted labels of the time and space matching degree of multi-source data in each time period as the prediction target, and takes minimizing the sum of the prediction errors of the weighted labels of the time and space matching degree of multi-source data in all time periods as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The weighted assignment of the time and space matching degree of multi-source data in each time period is determined according to the model output results. Among them, the machine learning model is a polynomial regression model, and the weighted average summation of the weighted assignments of the time and space matching degree of multi-source data in each time period is calculated to obtain the comprehensive matching index of the multi-source data.

[0089] The method for obtaining the weighted values ​​of the temporal and spatial matching degree of multi-source data in each time period is as follows: the corresponding function expression is obtained from the comprehensive feature vector training data of the trained machine learning model: ; In the formula, is the output function of the model, HZK is the time coverage anomaly index, GHZ is the sampling point deviation index, Assign weights to the temporal and spatial matching degrees of multi-source data in each time period.

[0090] S3: Establish a yield prediction model and compare and analyze the actual yield with the prediction results. If the yield prediction results have a large deviation, dynamically adjust the parameters of the yield prediction model through fuzzy logic based on the calculated comprehensive matching index of multi-source data and the actual yield data in each monitoring area, and optimize the data processing and fusion algorithm.

[0091] Collect multi-source data as input features, including: meteorological data (such as rainfall, temperature, light intensity). Soil data (such as pH, organic matter content, nutrient concentration). Remote sensing data (such as NDVI, EVI, vegetation coverage). Historical yield data (such as dry tobacco yield per unit area). Normalize or standardize feature data to eliminate dimensional differences. Use interpolation algorithms or machine learning methods to fill in missing values. Ensure that data matches consistently in time and space.

[0092] According to the data size and feature dimension, choose a suitable prediction model, such as regression model (linear regression, random forest regression, XGBoost). Deep learning model (LSTM, Transformer, for time series data).

[0093] Divide the dataset: Divide the dataset into a training set (70%), a validation set (15%), and a test set (15%). Feature selection: Use correlation analysis or feature importance indicators (such as Shap value) to screen key input variables. Model training: Use the training set to fit model parameters, and use the validation set to adjust hyperparameters (such as learning rate, tree depth). Evaluate model performance through the test set, including mean square error (MSE) and mean absolute error (MAE).

[0094] Using the trained model, input the validation set or new multi-source data into the model to generate the predicted tobacco yield for each monitoring area; collect the actual yield data for the corresponding area from the plot monitoring records. Calculate the yield deviation for each area, that is, the difference between the predicted tobacco yield and the actual tobacco yield;

[0095] According to the actual output, the area is divided into: Low-yield area: , where P25 is the lower quartile of actual yield; medium-yield area: actual yield is between P25 and P75; high-yield area: , where P75 is the upper quartile.

[0096] Analysis of low-yield areas: If the yield deviation ΔY is less than the preset yield deviation threshold of the low-yield area, it means that the yield prediction result has a large deviation, and the model may not fully capture the impact of key features (such as insufficient soil fertility, insufficient rainfall, etc.). Check the value range of the input features to ensure that the features of the low-yield area cover similar situations in the training data.

[0097] High-yield area analysis: If the yield deviation value ΔY is greater than the preset yield deviation threshold of the high-yield area, it means that the yield prediction result has a large deviation, which may be because the model overfits some features, resulting in an over-prediction. Check whether the feature value of the high-yield area falls outside the range of the training data. If so, it may be necessary to expand the training data or use regularization methods to prevent overfitting.

[0098] For areas or stages with large deviations, supplement the corresponding multi-source data (such as increasing the sampling frequency or introducing new features). Introduce new features (such as pest and disease data or irrigation records) to improve the model's understanding of influencing factors. Use ensemble learning methods (such as boosted trees and bagging methods) to improve the robustness of the model. Retrain with optimized data and models, and verify the improvement effect.

[0099] According to the calculated comprehensive matching index of multi-source data and the actual production data in each monitoring area, the parameters of the production prediction model are dynamically adjusted through fuzzy logic to optimize the data processing and fusion algorithm. Specifically:

[0100] The comprehensive matching index LZ of multi-source data and the actual production data Y are used as the input items of fuzzy logic, and the adjustment results of the prediction model parameters are used as the output items of fuzzy logic;

[0101] The comprehensive matching index LZ indicates the matching degree of multi-source data in time and space, and the range is usually [0,1]. Fuzzification range: Low matching (Low): LZ < 0.4. Medium matching (Medium): 0.4 ≤ LZ < 0.7. High matching (High): LZ ≥ 0.7.

[0102] The actual yield data Y represents the actual yield data (yield per unit area) in each monitoring area. Fuzzy range: Low yield (Low): The actual yield is lower than 25% of the regional average. Medium yield (Medium): The actual yield is close to the regional average (±25%). High yield (High): The actual yield is higher than 25% of the regional average.

[0103] Parameter adjustment results of the prediction model: indicates the degree to which the model needs to be optimized or the specific optimization strategy, ranging from [0,1]. Fuzzy range: No adjustment required (No Adjustment): The model performance is stable and the matching degree is high. Minor adjustment (MinorAdjustment): The data fusion algorithm or feature weight needs to be slightly optimized. Major adjustment (Major Adjustment): The model parameters need to be significantly modified or new features need to be introduced.

[0104] According to the combination of LZ and Y, fuzzy rules are formulated. For example: If LZ is low and Y is low, the model may fail to capture the characteristics of low-yield areas, and the output is significantly adjusted. If LZ is medium and Y is high, the data match is relatively good and only slight adjustments are needed. If LZ is high and Y is high, the model performs well and the output does not need to be adjusted. For example, LZ: low, Y: low → Model adjustment: significant adjustment. LZ: medium, Y: medium → Model adjustment: slight adjustment. LZ: high, Y: high → Model adjustment: no adjustment is needed. LZ: low, Y: high → Model adjustment: significant adjustment (may be data fusion failure or feature missing).

[0105] Convert LZ and Y into fuzzy sets, and use membership functions to define the fuzziness of different states. The membership function defines: Low matching degree: ; Medium match: defined using triangular or trapezoidal membership functions. High match: increasing membership functions, applicable to LZ ≥ 0.7.

[0106] According to the fuzzy rule table, use the maximum-minimum synthesis method or the centroid method to infer the output. Calculate the matching degree between the input fuzzy set and the rule conditions. Select the rules that meet the conditions and calculate the activation degree of the output membership function.

[0107] Convert a set of fuzzy outputs into crisp values, e.g. using the centroid method to calculate the specific value of the parameter adjustment result , is the clarity value adjusted by the model (range [0, 1]), expressed as: ; x is a possible value of the output variable, indicating the value within the possible range of the fuzzy logic system output. In this application, the value of x indicates the possible parameter adjustment amount, which is usually within a certain range (such as [0,1]). μ(x) is the membership function, which indicates the membership of the output variable x in the fuzzy set. The value of μ(x) is between [0,1], indicating the degree to which x belongs to a fuzzy set (such as "no adjustment", "slight adjustment", "significant adjustment"). The shape of the membership function can be a triangle, trapezoid, Gaussian function, etc.

[0108] According to the output of the fuzzy logic system, the optimization strategy of the model is determined: O<0.3: No adjustment is required. 0.3≤O<0.6: Minor adjustment (such as optimizing the data fusion algorithm). O≥0.6: Significant adjustment (such as modifying the model structure, adding new features).

[0109] Optimize data processing and fusion algorithms: Improve interpolation algorithms or introduce new data sources to solve data matching problems when LZ is low. Optimize model parameters: Adjust feature weights or introduce regularization to control overfitting in areas where Y deviates.

[0110] In this embodiment, the tobacco planting area is first divided into several monitoring areas, and multi-source data in different time periods of each area are obtained in real time, including meteorological data, soil data, remote sensing image data and actual yield data. Then, the collected data is preprocessed, the spatiotemporal consistency characteristics and information integrity characteristics are extracted, the time and space matching weights of the multi-source data in each time period are calculated, and the comprehensive matching index of the multi-source data is generated by weighted average summation. On this basis, a yield prediction model is established, and the prediction results are compared and analyzed with the actual yield. If the prediction deviation is found to be large, the model parameters are dynamically adjusted based on the comprehensive matching index and the actual yield data using fuzzy logic, and the data processing and fusion algorithms are optimized to improve the prediction accuracy and robustness of the model.

[0111] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0112] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0113] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the objects associated with each other are in an "or" relationship, but it may also indicate an "and / or" relationship, which can be understood by referring to the context. A person of ordinary skill in the art can appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this article can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0114] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A method for predicting tobacco field yield based on multi-source data analysis, characterized in that: The following steps are involved: S1: Divide the tobacco planting area into several monitoring areas, and obtain multi-source data in different time periods in each monitoring area in real time. The multi-source data includes meteorological data, soil data and remote sensing image data, as well as actual yield data in each monitoring area: S2: After preprocessing the collected multi-source data, extract the spatiotemporal consistency features and information integrity features in the multi-source data, determine the weight assignment of the temporal and spatial matching degree of the multi-source data in each time period, and calculate the weighted average sum to obtain the comprehensive matching index of the multi-source data; After analyzing the extracted time consistency features, the time coverage anomaly index is generated. The method for obtaining the time coverage anomaly index is as follows: set the input time coverage data as a time series ;in: is the coverage rate of the Nth day in the time series, ranging from [0, 1], where N is the total number of days; define the window size w, which determines the time range used to calculate the coverage feature each time, set the window step size s, which represents the step size of the window movement, and calculate the window mean for each window [t, t+w−1]: ; is the coverage rate of the i-th day in the time series, is the average coverage rate in the tth window, and the window standard deviation is calculated: ; is the standard deviation of coverage in the t-th window, for each time point , calculate the coverage anomaly index, the expression is: ; Where HZK is the time coverage anomaly index; After analyzing the integrity of the extracted spatial sampling points, the sampling point deviation index is generated. The sampling point deviation index is obtained by setting the geographic coordinates of the sampling point as and sample values , n is the total number of sampling points, the range of the target interpolation area is [Xmin, Xmax] × [Ymin, Ymax], the target area is divided into uniform grids, and the center of each grid is the point to be interpolated; Calculate any two points and The Euclidean distance , the expression is: ; Calculate the corresponding semivariance value , the expression is: ; For point The attribute value at represents the sampling point The observed value of Indicate point The attribute value at represents the sampling point Observed value of According to the relationship between the calculated semivariogram value and the distance h, the semivariogram function model is selected: ; In the formula, is the base, indicating the measurement error of the sampling point, C is the step value, indicating the maximum variation value, and a is the range, indicating the maximum influence distance of spatial correlation; For the points to be interpolated , set the weights of n sampling points to , the Kriging equation is: ; In the formula, is the semivariance between sampling points i and j, is the semi-variance value between the sampling point i and the point to be interpolated, μ is the Lagrange multiplier, solve the equations and get the weight ; According to weight , calculate the predicted value of the point to be interpolated , the expression is: ; Calculate interpolation error , the calculation formula is: ; In the formula, is the sill value of the semivariogram, is the semi-variance value between the sampling point j and the interpolation point ∗, which measures the degree of variation between the two points. The calculation results are filled into the target area grid, the error distribution map is generated, and the sampling point deviation index is calculated. The expression is: ; In the formula, is a small positive number to prevent the denominator from being zero, and GHZ is the sampling point deviation index; The time coverage anomaly index and the sampling point deviation index are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as the input of the machine learning model. The machine learning model predicts the weighted labels of the time and space matching degree of the multi-source data in each time period with each group of comprehensive feature vectors as the prediction target, and minimizes the sum of the prediction errors of the weighted labels of the time and space matching degree of the multi-source data in all time periods as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The weighted values ​​of the time and space matching degree of the multi-source data in each time period are determined according to the output results of the model, wherein the machine learning model is a polynomial regression model, and the weighted average summation calculation of the weighted values ​​of the time and space matching degree of the multi-source data in each time period is performed to obtain the comprehensive matching index of the multi-source data; S3: Establish a yield prediction model and compare and analyze the actual yield with the prediction results. If the yield prediction results have a large deviation, dynamically adjust the parameters of the yield prediction model through fuzzy logic based on the calculated comprehensive matching index of multi-source data and the actual yield data in each monitoring area, and optimize the data processing and fusion algorithm.

2. The method for predicting tobacco field yield based on multi-source data analysis according to claim 1, characterized in that: In S3, the actual production is compared and analyzed with the predicted results, specifically: collect multi-source data as input features, select the prediction model according to the data scale and feature dimension, and divide the data set into training set, validation set and test set; use the training set to fit the model parameters, adjust the hyperparameters through the validation set, and evaluate the model performance through the test set; Using the trained model, input the validation set or new multi-source data into the model to generate the predicted tobacco yield of each monitoring area; collect the actual yield data of the corresponding area from the plot monitoring records, and calculate the yield deviation of each area, that is, the difference between the predicted tobacco yield and the actual tobacco yield; According to the actual output, the area is divided into: Low-yield area: , where P25 is the lower quartile of actual yield; medium-yield area: actual yield is between P25 and P75; high-yield area: , where P75 is the upper quartile; If the yield deviation ΔY is less than the preset yield deviation threshold of the low-yield area, it means that the yield prediction result has a large deviation and the model fails to fully capture the impact of key characteristics; If the yield deviation value ΔY is greater than the preset yield deviation threshold of the high-yield area, it means that the yield prediction result has a large deviation, the model overfits certain features, and there is overfitting.

3. The method for predicting tobacco field yield based on multi-source data analysis according to claim 2, characterized in that: In S3, according to the calculated comprehensive matching index of multi-source data and the actual production data in each monitoring area, the parameters of the production prediction model are dynamically adjusted through fuzzy logic, specifically: The comprehensive matching index LZ of multi-source data and the actual production data Y are used as the input items of fuzzy logic, and the adjustment results of the prediction model parameters are used as the output items of fuzzy logic; The comprehensive matching index LZ, actual production data Y and the parameter adjustment results of the model are fuzzified; According to the combination relationship between LZ and Y, fuzzy rules are formulated; Convert LZ and Y into fuzzy sets, and use membership functions to define the fuzziness of different states. The membership function defines: Low matching degree: ; Medium match: defined using triangular or trapezoidal membership functions; High matching degree: the membership function increases, applicable to LZ ≥ 0.7; According to the fuzzy rule table, the centroid method is used to infer the output; Calculate the matching degree between the input fuzzy set and the rule conditions, select the rules that meet the conditions and calculate the activation degree of the output membership function; Convert the fuzzy output set into crisp values ​​and use the centroid method to calculate the specific value of the parameter adjustment result , is the clarity value adjusted by the model, expressed as: ; x is all possible values ​​of the output variable, indicating the range of the parameter adjustment amount, μ(x) is the membership function, indicating the membership of the output variable x in the fuzzy set; According to the output of the fuzzy logic system, the optimization strategy of the model is determined: O<0.3: no adjustment is required; 0.3≤O<0.6: slight adjustment; O≥0.6: significant adjustment.

Images