An improved traceability method for agricultural production processes based on artificial intelligence

Abstract

The present invention discloses an improved traceability method for agricultural production processes based on artificial intelligence. S1: Collect multi-dimensional agricultural production data; S2: Perform data cleaning, noise removal, outlier detection, and format conversion on the multi-dimensional agricultural production data; S3: Divide the multi-dimensional agricultural production data into short-term data, medium-term data, and long-term data according to the time attributes of the multi-dimensional agricultural production data; S4: Divide the multi-dimensional agricultural production data according to the spatial positions of plots, regions, and crop types; S5: Classify the multi-dimensional agricultural production data according to multiple production stages of crops from sowing, fertilizing, watering, weeding, and harvesting; S6: Feed back the information of abnormal agricultural production nodes to the management system; S7: Generate a traceability report; S8: Dynamically adjust the irrigation frequency, fertilization amount, or equipment maintenance plan of crops according to the feedback information. The present invention realizes the flexible management of agricultural production processes and improves the overall efficiency and quality of agricultural production.

Description

Technical Field

[0001] The present invention relates to the field of agricultural technologies, and in particular, to an improved method for tracing the agricultural production process based on artificial intelligence. Background Art

[0002] In the prior art, the data management and tracing methods for the intelligent agricultural production process mainly rely on traditional data collection and analysis tools. For example, the growth of crops, soil conditions, meteorological environment, and equipment operation status are traced through fixed rules or manual monitoring. The traditional methods usually collect data based on a distributed sensor network and then process it through manual work or simple algorithms. However, with the continuous improvement of the intelligence and automation levels of agricultural production, the data generated in the agricultural production process is becoming more and more complex, dynamic, and diverse. The agricultural data generated by different devices has a high degree of heterogeneity and uncertainty. Traditional processing methods often show low efficiency and are unable to cope with the dynamic changes of data in the face of such complex data scenarios, easily leading to problems such as inaccurate data classification and untimely identification of abnormal nodes.

[0003] Existing agricultural data classification methods usually rely on fixed algorithm models, such as rule-based classification or simple clustering algorithms. When dealing with heterogeneous data, simple clustering algorithms are difficult to automatically adapt to the changing characteristics of the data. In the face of emergencies (such as sudden climate changes, outbreaks of pests and diseases, and equipment failures) in the agricultural production process, it often requires manual intervention and it is difficult to adjust the model parameters in a timely manner to reflect real-time changes. The robustness of traditional algorithms is poor and they are easily limited by a single data source, unable to effectively integrate and process multi-dimensional data, resulting in insufficient accuracy of classification results and abnormal node identification, and may miss key data, causing potential production risks.

[0004] Another disadvantage of the prior art is that in the multi-stage process management of agricultural production, traditional methods are difficult to flexibly adapt to the requirements of different production stages. Whether it is from sowing, fertilizing to harvesting, the data in each link has significant differences. Traditional processing tools cannot be dynamically adjusted according to the specific characteristics of each stage and cannot accurately identify the key characteristics in the production process, resulting in low efficiency in the monitoring and management of the production process.

[0005] In summary, the prior art mainly has the following disadvantages: First, the efficiency of traditional agricultural data classification algorithms in processing multi-source and heterogeneous data is low and they cannot adapt to the dynamic changes of data; second, the prior art has limited capabilities in the identification and real-time feedback of abnormal nodes and cannot respond in a timely manner to emergencies occurring in the agricultural production process; third, existing production management tools cannot adapt to the multi-stage process of agricultural production and lack a flexible dynamic adjustment mechanism, affecting the management and monitoring efficiency of the production process. Summary of the Invention

[0006] An object of the present invention is to propose an improved traceability method for agricultural production processes based on artificial intelligence. The present invention realizes flexible management of agricultural production processes, improving the overall efficiency and quality of agricultural production.

[0007] An improved traceability method for agricultural production processes based on artificial intelligence according to an embodiment of the present invention includes the following steps:

[0008] S1. Multidimensional agricultural production data regarding crop growth, soil conditions, meteorological environment, and equipment operation status are collected in real time through devices deployed in the agricultural production environment;

[0009] S2. The multidimensional agricultural production data is subjected to data cleaning, noise removal, outlier detection, and format conversion;

[0010] S3. According to the time attributes of the multidimensional agricultural production data, the multidimensional agricultural production data is divided into short-term data, medium-term data, and long-term data, and an adaptive clustering model is used to analyze the data characteristics in different time periods respectively;

[0011] S4. According to the spatial distribution characteristics of the multidimensional agricultural production data, the multidimensional agricultural production data is divided according to the spatial positions of plots, regions, and crop types, and adaptive clustering analysis is performed on the data at each spatial position, and the parameters of the adaptive clustering model are dynamically adjusted so that the classification results of the multidimensional agricultural production data reflect the crop growth, environmental conditions, and production equipment status in different regions;

[0012] S5. According to the specific links of agricultural production, in accordance with multiple production stages of crops from sowing, fertilizing, watering, weeding, and harvesting, the multidimensional agricultural production data is classified, and dynamic clustering is performed on the data in each production stage to identify key features in each stage, and the parameters of the adaptive clustering model are adjusted in real time to cope with data changes in agricultural production links;

[0013] S6. Based on the multi-level clustering results in steps S3 to S5, abnormal nodes existing in the agricultural production process are automatically identified, including abnormal crop growth, abnormal soil conditions, sudden changes in the meteorological environment, and equipment failures, and an abnormal alarm is issued in real time, and the information of the abnormal agricultural production nodes is fed back to the management system;

[0014] S7. When an abnormality occurs in the agricultural production process, based on the multi-level clustering results of the multidimensional agricultural production data, traceability analysis is automatically performed. By tracing the data sources and historical changes of the abnormal nodes, the time, location, and cause of the abnormality are determined, and a traceability report is generated;

[0015] S8. The abnormal nodes and their causes in the traceability management are fed back to the agricultural production management system in real time, and the irrigation frequency, fertilization amount, or equipment maintenance plan of the crops is dynamically adjusted according to the feedback information.

[0016] Optionally, S1 includes:

[0017] S11. Real-time collect multi-dimensional crop growth data D c (t) through a sensor network arranged in the agricultural production environment, including leaf area, stem length, photosynthesis rate, and nutrient content, where D c is a multi-dimensional vector of crop growth parameters, and t is the acquisition time point;

[0018] S12. Real-time collect multi-dimensional soil condition data D s (t) using sensor devices deployed in the soil, including soil temperature, humidity, pH value, and nutrient content, where D s is a multi-dimensional vector of soil conditions;

[0019] S13. Real-time collect meteorological environment data D w (t) through a weather station, including air temperature, humidity, wind speed, light intensity, and rainfall, where D w is a multi-dimensional vector of the meteorological environment;

[0020] S14. Real-time collect equipment operation status data D e (t) related to agricultural production equipment through an equipment operation status monitoring system, including the working current, working voltage, temperature, and vibration conditions of the equipment, where D e is a multi-dimensional vector of the equipment operation status;

[0021] S15. Integrate the multi-dimensional agricultural production data collected in steps S11 to S14 into multi-dimensional agricultural production data D(t):

[0022] D(t) = {D c (t), D s (t), D w (t), D e (t)}.

[0023] Optionally, S2 includes:

[0024] S21. Clean the collected multi-dimensional agricultural production data D(t), remove invalid data and duplicate data, and obtain the cleaned multi-dimensional agricultural production data D clean (t);

[0025] S22. Remove noise from the cleaned multi-dimensional agricultural production data D clean (t), eliminate the noise generated by sensor devices and environmental interference, and obtain the noise-removed multi-dimensional agricultural production data D denoise (t);

[0026] S23. Perform outlier detection on the multi-dimensional agricultural production data D denoise (t) after noise removal, and identify the outlier data points of the multi-dimensional agricultural production data by using the distribution-based detection method. The outlier data points include the crop growth parameters, soil conditions, and meteorological environment data that exceed the reasonable range. The outlier detection result is represented by the function E(t):

[0027] E(t) = {e1(t), e2(t), …, e n (t)};

[0028] where, e n (t) is the outlier data point;

[0029] S24. Analyze the outlier data in E(t), automatically correct or delete the outlier data, and obtain the multi-dimensional agricultural production data D corrected (t) after outlier processing;

[0030] S25. Perform format conversion on the multi-dimensional agricultural production data D corrected (t) after outlier processing, and convert the multi-dimensional agricultural production data into a unified format according to the requirements of data analysis and clustering models, and obtain the converted multi-dimensional agricultural production data D formatted (t).

[0031] Optionally, the S3 includes:

[0032] S31. According to the time attribute of the multi-dimensional agricultural production data, divide the multi-dimensional agricultural production data D formatted (t) into short-term data D short (t), medium-term data D mid (t), and long-term data D long (t):

[0033] D short (t) represents the agricultural production data collected within a short time range and is used for real-time monitoring and feedback;

[0034] D mid (t) represents the agricultural production data collected within the medium-term time period and is used for trend identification;

[0035] D long (t) represents the agricultural production data collected within the long time range and is used for periodic pattern analysis;

[0036] S32. Perform classification analysis on the short-term data D short (t) by using an adaptive clustering model, and identify the change patterns of crop growth, soil conditions, meteorological environment, and equipment operation status in agricultural production in the short term:

[0037]

[0038] Among them, C short (t) is the short-term clustering result, representing the i-th short-term data point, μ j (t) is the clustering center, the mean vector of the j-th class, w ij (t) is the sample belonging to the membership function of the j-th class, representing the distance weight between the short-term data point and the clustering center;

[0039] S33. Perform trend recognition on the medium-term data D mid (t), use the adaptive clustering model to analyze the data change trend, and capture the data change trend over time by integrating the trend within the time interval [t0, t1], which is used for the analysis of crop growth change trend, soil nutrient fluctuation trend, and meteorological change trend:

[0040]

[0041] Among them, T mid (t) is the medium-term trend recognition result, representing the i-th medium-term data point, μ j (t) is the change trend of the clustering center, w ij (t) is the membership function;

[0042] S34. Perform periodic pattern analysis on the long-term data D long (t), use the adaptive clustering model to detect and analyze the periodic fluctuations of the long-term data in agricultural production, and capture the periodic fluctuations in the long-term data through the periodic term cos(2πf j (t - t0)):

[0043]

[0044] Among them, P long (t) is the result of the long-term periodic pattern analysis, is the i-th long-term data point, μ j (t) is the clustering center, f j represents the periodic frequency, w ij (t) is the membership function.

[0045] Optionally, the S4 includes:

[0046] S41. According to the spatial distribution characteristics of the multi-dimensional agricultural production data D formatted (t), divide the multi-dimensional agricultural production data spatially according to the plot, region, and crop type to obtain the agricultural production data D loc (t) at different spatial positions:

[0047] D loc1 D(t) represents the data set of plot 1.

[0048] D loc2 D(t) represents the data set of plot 2.

[0049] D locn D(t) represents the data set of plot n;

[0050] S42. Analyze the agricultural production data D(t) at each spatial location using an adaptive clustering model. The clustering process dynamically adjusts the clustering center μ(t) locn and the membership function w(t) j according to different spatial locations, so that the adaptive clustering model reflects the crop growth, environmental conditions, and production equipment status corresponding to the spatial location; ij

[0051] S43. According to the differences in spatial locations, dynamically adjust the agricultural production data D(t) at different spatial locations, so that the clustering results reflect the agricultural production characteristics of each region. The dynamic adjustment of the adaptive clustering model is achieved by minimizing the function L(t): loc loc ij where,

[0052]

[0053] where, represents the multi-dimensional agricultural production data point at the i-th spatial location, w(t) ij is the membership function of the multi-dimensional agricultural production data point belonging to the clustering center j, λ is the regularization parameter, represents the dynamic adjustment amount of the clustering center over time, which adapts to the changes in agricultural production data at different time periods, and captures the acceleration changes of production data at different spatial locations through the second derivative term;

[0054] S44. According to the results of the clustering analysis, reflect the crop growth status, soil conditions, meteorological environment, and production equipment status in different regions in real time, and generate the classification result C(t) of the spatial location: spatial (t):

[0055]

[0056] where, is the difference between the multi-dimensional agricultural production data point and the clustering center μ(t), reflecting the spatial difference of agricultural production data at time t, j (t), For describing dynamic adjustment in the time dimension, involving multi-dimensional agricultural production data points The time derivative reflects the growth change rate of crops. β is the adjustment coefficient, and the angular parameter θ jij (t) is used to capture the relative angular change between different cluster centers, reflects the influence of the second-order time derivative on the cluster centers, capturing the acceleration change of crop growth or equipment status. δ is the spatial decay coefficient, indicating the acceleration influence that decays as the spatial distance |x i -x j | increases.

[0057] Optionally, the S5 includes:

[0058] S51. According to the specific links of agricultural production, classify the multi-dimensional agricultural production data D formatted (t) according to the production stages of crops, including multiple production stages such as sowing, fertilizing, watering, weeding, and harvesting, to obtain the production stage data set D stage (t), where D seed (t) represents the data in the sowing stage, D fertil (t) represents the data in the fertilizing stage, D water (t) represents the data in the watering stage, D weed (t) represents the data in the weeding stage, D harvest (t) represents the data in the harvesting stage;

[0059] S52. Perform dynamic clustering analysis on each production stage data set D stage (t) using an adaptive clustering model, identify the key features in each production stage, and dynamically adjust the parameters of the adaptive clustering model according to the changes in the data during the agricultural production process;

[0060] S53. Identify the key features in the agricultural production process according to the clustering results of each production stage, including the soil conditions in the sowing stage, the nutrient requirements in the fertilizing stage, the humidity changes in the watering stage, the weed coverage rate in the weeding stage, and the crop maturity in the harvesting stage, and cope with the changes in the data in the production stage by adjusting the parameters μ j (t) and w ij (t);

[0061] S54. Real-time adjust the parameters of the adaptive clustering model for each production stage, and optimize the clustering process by minimizing the objective function L stage (t):

[0062]

[0063] Among them, λ is a regulation parameter, representing the dynamic change of the clustering center over time.

[0064] Optionally, the S6 includes:

[0065] S61. Automatically identify the abnormal nodes existing in the agricultural production process based on the multi-level clustering results of steps S3 to S5, including abnormal crop growth, abnormal soil conditions, sudden changes in meteorological environment, and equipment failure conditions. By comparing the clustering results of short-term data C short (t), medium-term trend recognition data T mid (t), and long-term periodic data P long (t), data C spatial (t) at different spatial positions, and data C stage (t) at each production stage, detect the deviation from the normal production mode. The result of abnormal node detection is represented by the function A(t):

[0066] A(t) = {a1(t), a2(t), …, a m (t)};

[0067] Among them, aa m (t) represents the m-th abnormal node, and the detected abnormal nodes include abnormalities in crop growth, soil conditions, meteorological environment, and equipment status;

[0068] S62. Classify each abnormal node a m (t), and determine the type of abnormality in combination with various clustering results, including abnormal crop growth, abnormal soil conditions, abnormal meteorological environment, and equipment failure conditions. According to different types of abnormalities, automatically generate a classification label L m (t):

[0069]

[0070] Among them, L m (t) is the classification label of the abnormal node, representing the variances of the clustering data at each stage respectively. The classification process depends on the differences from the normal clustering centers at each stage;

[0071] S63. According to the classification results of the abnormal nodes, combined with the clustering data in the short term, medium term, and long term, as well as the clustering results at different spatial positions and production stages, automatically calculate the degree of abnormality E m (t) and evaluate the deviation degree of the abnormality from the normal production state;

[0072] S64. The detected abnormal nodes a m (t), the type of abnormality L m (t), and the degree of abnormality E m(t) It is fed back to the agricultural production management system in real time. The system generates an alarm based on the feedback information, and combines the data clustering characteristics of different time periods, spatial locations, and production stages to prompt the manager to handle abnormal situations and provide corresponding handling suggestions according to the severity of the abnormality.

[0073] The beneficial effects of the present invention are as follows:

[0074] (1) The improved adaptive clustering algorithm adopted by the present invention can dynamically adjust the clustering model parameters according to the changes in time, space, and production stage of agricultural production data. By dividing multi-dimensional agricultural data into short-term, medium-term, and long-term data, and performing adaptive classification and trend analysis on the data in each stage, it can more accurately handle data changes at different time scales, overcome the limitation that traditional clustering algorithms are difficult to adapt to data variability, and can update the clustering model in real time in the scenario where meteorological environment, soil conditions, and crop growth data change rapidly during the agricultural production process, ensuring the accuracy and sensitivity of data classification, and effectively avoiding production risks caused by data lag or failure to identify abnormal points in time in traditional algorithms.

[0075] (2) The present invention automatically identifies abnormal nodes in agricultural production through a multi-level clustering model, including abnormalities in crop growth, soil conditions, meteorological environment, and equipment operation status. Different from the existing monitoring methods that rely on manual or fixed rules, the improved adaptive clustering algorithm can dynamically identify abnormalities by real-time analyzing the clustering results of short-term, medium-term, and long-term data in combination with data of spatial location and production stage. It can not only improve the recognition accuracy of abnormal nodes, but also classify, quantify, and provide real-time feedback on abnormal situations by calculating the deviation degree between abnormal and normal states, generating automated alarm information, reducing manual intervention, and improving the efficiency of dealing with emergencies during the agricultural production process.

[0076] (3) The present invention proposes to classify data according to different links (sowing, fertilizing, watering, weeding, and harvesting) of agricultural production and perform dynamic clustering. The data characteristics in each production stage are different, and it is difficult for traditional methods to handle the differences between stages. By adjusting the clustering model parameters in real time in different production stages to make it adapt to the data characteristics of each stage, flexible management of the agricultural production process is realized. The dynamic adjustment mechanism can ensure accurate data classification in each production stage, making production management more targeted and time-effective, and improving the overall efficiency and quality of agricultural production. Description of the Drawings

[0077] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0078] Figure 1Flowchart of an improved agricultural production process traceability method based on artificial intelligence proposed by the present invention;

[0079] Figure 2 Flowchart of dynamic clustering analysis of data at each stage of agricultural production (sowing, fertilizing, watering, weeding, harvesting) in an improved agricultural production process traceability method based on artificial intelligence proposed by the present invention. Detailed implementation manners

[0080] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0081] Reference Figure 1-2 , an improved agricultural production process traceability method based on artificial intelligence, includes the following steps:

[0082] S1. Real-time collect multi-dimensional agricultural production data on crop growth, soil conditions, meteorological environment and equipment operation status through devices deployed in the agricultural production environment;

[0083] S2. Perform data cleaning, noise removal, outlier detection and format conversion on the multi-dimensional agricultural production data;

[0084] S3. According to the time attributes of the multi-dimensional agricultural production data, divide the multi-dimensional agricultural production data into short-term data, medium-term data and long-term data, and adopt an adaptive clustering model to analyze the data characteristics in different time periods respectively;

[0085] S4. According to the spatial distribution characteristics of the multi-dimensional agricultural production data, divide the multi-dimensional agricultural production data according to the spatial positions of plots, regions and crop types, and perform adaptive clustering analysis on the data at each spatial position, dynamically adjusting the parameters of the adaptive clustering model so that the classification results of the multi-dimensional agricultural production data reflect the crop growth, environmental conditions and production equipment status in different regions;

[0086] S5. According to the specific links of agricultural production, classify the multi-dimensional agricultural production data according to multiple production stages of crops from sowing, fertilizing, watering, weeding to harvesting, and perform dynamic clustering on the data at each production stage, identify the key features in each stage, and adjust the parameters of the adaptive clustering model in real time to cope with the data changes in the agricultural production process;

[0087] S6. Automatically identify abnormal nodes existing in the agricultural production process based on the multi-level clustering results of steps S3 to S5, including abnormal crop growth, abnormal soil conditions, sudden changes in the meteorological environment and equipment failures, and issue abnormal alarms in real time, and feedback the information of the abnormal agricultural production nodes to the management system;

[0088] S7. When an anomaly occurs during the agricultural production process, based on the multi-level clustering results of multi-dimensional agricultural production data, perform traceability analysis automatically. By tracing the data sources of the anomaly nodes and their historical changes, determine the time, location, and cause of the anomaly, and generate a traceability report.

[0089] S8. Real-time feedback the anomaly nodes and their causes in the traceability management to the agricultural production management system, and dynamically adjust the irrigation frequency, fertilization amount, or equipment maintenance plan of the crops according to the feedback information.

[0090] In this embodiment, S1 includes:

[0091] S11. Real-time collect the multi-dimensional data D c (t) of crop growth through the sensor network deployed in the agricultural production environment, including leaf area, stem length, photosynthesis rate, and nutrient content. D c is a multi-dimensional vector of crop growth parameters, and t is the collection time point.

[0092] S12. Real-time collect the multi-dimensional data D s (t) of soil conditions through the sensor devices deployed in the soil, including soil temperature, humidity, pH value, and nutrient content. D s is a multi-dimensional vector of soil conditions.

[0093] S13. Real-time collect the meteorological environment data D w (t) through the weather station, including air temperature, humidity, wind speed, light intensity, and rainfall. D w is a multi-dimensional vector of the meteorological environment.

[0094] S14. Real-time collect the equipment operation status data D e (t) related to agricultural production equipment through the equipment operation status monitoring system, including the working current, working voltage, temperature, and vibration conditions of the equipment. D e is a multi-dimensional vector of the equipment operation status.

[0095] S15. Integrate the multi-dimensional agricultural production data collected in steps S11 to S14 into the multi-dimensional agricultural production data D(t):

[0096] D(t) = {D c (t), D s (t), D w (t), D e}

[0097] In this embodiment, S2 includes:

[0098] S21. Clean the collected multi-dimensional agricultural production data D(t), remove invalid and duplicate data, and obtain the cleaned multi-dimensional agricultural production data D(t); clean (t);

[0099] S22. Remove noise from the cleaned multi-dimensional agricultural production data D(t), eliminate the noise generated from sensor devices and environmental interference, and obtain the multi-dimensional agricultural production data D(t) after noise removal; clean (t); denoise (t);

[0100] S23. Detect outliers from the multi-dimensional agricultural production data D(t) after noise removal, adopt a distribution-based detection method to identify the outlier data points of the multi-dimensional agricultural production data. The outlier data points include crop growth parameters, soil conditions, and meteorological environment data that exceed the reasonable range. The outlier detection result is represented by the function E(t): denoise (t);

[0101] E(t) = {e1(t), e2(t), …, e(t)}; n (t);

[0102] where e(t) is the outlier data point; n (t);

[0103] S24. Analyze the outlier data in E(t), automatically correct or delete the outlier data, and obtain the multi-dimensional agricultural production data D(t) after outlier processing; corrected (t);

[0104] S25. Convert the format of the multi-dimensional agricultural production data D(t) after outlier processing, convert the multi-dimensional agricultural production data into a unified format according to the requirements of data analysis and clustering models, and obtain the converted multi-dimensional agricultural production data D(t). corrected (t); formatted (t).

[0105] In this embodiment, S3 includes:

[0106] S31. According to the time attribute of the multi-dimensional agricultural production data, divide the multi-dimensional agricultural production data D(t) into short-term data D(t), medium-term data D(t), and long-term data D(t): formatted (t); short (t); mid (t); long (t):

[0107] D(t) represents the agricultural production data collected within a short time range and is used for real-time monitoring and feedback; short (t);

[0108] D(t) represents the agricultural production data collected within the medium-term time period and is used for trend identification; mid (t);

[0109] D long D(t) represents the agricultural production data collected over a long - term time range for periodic pattern analysis;

[0110] S32. Classify and analyze the short - term data D short (t) using an adaptive clustering model to identify the change patterns of crop growth, soil conditions, meteorological environment, and equipment operation status in short - term agricultural production:

[0111]

[0112] Among them, C short (t) is the short - term clustering result, represents the i - th short - term data point, μ j (t) is the clustering center, the mean vector of the j - th class, w ij (t) is the sample belonging to the membership function of the j - th class, indicating the distance weight between the short - term data point and the clustering center;

[0113] S33. Identify the trend of the medium - term data D mid (t), analyze the data change trend using an adaptive clustering model, and capture the data change trend over time by integrating the trend within the time interval [t0, t1] for the analysis of crop growth change trend, soil nutrient fluctuation trend, and meteorological change trend:

[0114]

[0115] Among them, T mid (t) is the medium - term trend identification result, represents the i - th medium - term data point, μ j (t) is the change trend of the clustering center, w ij (t) is the membership function;

[0116] S34. Conduct periodic pattern analysis on the long - term data D long (t), detect and analyze the periodic fluctuations in the long - term data in agricultural production using an adaptive clustering model, and capture the periodic fluctuations in the long - term data through the periodic term cos(2πf j (t - t0)):

[0117]

[0118] Among them, P long (t) is the result of long - term periodic pattern analysis, is the i - th long - term data point, μ j (t) is the clustering center, fj represents the periodic frequency, ω ij (t) is the membership function.

[0119] In this embodiment, S4 includes:

[0120] S41. According to the spatial distribution characteristics of the multi-dimensional agricultural production data D formatted (t), the multi-dimensional agricultural production data is spatially divided according to plots, regions, and crop types to obtain the agricultural production data D loc (t) at different spatial positions:

[0121] D loc1 (t) represents the data set of plot 1,

[0122] D loc2 (t) represents the data set of plot 2,

[0123] D locn (t) represents the data set of plot n;

[0124] S42. Analyze the agricultural production data D locn (t) at each spatial position using an adaptive clustering model. The clustering process dynamically adjusts the clustering center μ j (t) and the membership function ω ij (t) according to different spatial positions, so that the adaptive clustering model reflects the crop growth, environmental conditions, and production equipment status at the corresponding spatial position;

[0125] S43. According to the differences in spatial positions, dynamically adjust the parameters of the agricultural production data D loc (t) at different spatial positions, so that the clustering results reflect the agricultural production characteristics of each region. The dynamic adjustment of the adaptive clustering model is achieved by minimizing the function L loc (t):

[0126]

[0127] Among them, represents the multi-dimensional agricultural production data point at the i-th spatial position, ω ij (t) is the membership function of the multi-dimensional agricultural production data point belonging to the clustering center j, λ is the regularization parameter, represents the dynamic adjustment amount of the clustering center changing with time. Through this term, it adapts to the changes in agricultural production data at different time periods, and captures the acceleration changes of production data at different spatial positions through the second derivative term;

[0128] S44. According to the results of cluster analysis, the growth status of crops, soil conditions, meteorological environment and production equipment status in different regions are reflected in real time, and the classification result C of spatial positions is generated. spatial (t):

[0129]

[0130] Among them, is the multi-dimensional agricultural production data point and the difference between the cluster center μ j (t), which reflects the spatial difference of agricultural production data at time t, is used to describe the dynamic adjustment in the time dimension and involves the time derivative of the multi-dimensional agricultural production data point reflects the growth change rate of crops. β is the adjustment coefficient, and the angular parameter θ ij (t) is used to capture the relative angular change between different cluster centers, reflects the influence of the second-order time derivative on the cluster center and captures the acceleration change of crop growth or equipment status. δ is the spatial attenuation coefficient, represents the acceleration influence that decays as the spatial distance |x i -x j | increases.

[0131] In this embodiment, S5 includes:

[0132] S51. According to the specific links of agricultural production, the multi-dimensional agricultural production data D formatted (t) is classified according to the production stages of crops, including multiple production stages such as sowing, fertilizing, watering, weeding and harvesting, and the production stage data set D stage (t) is obtained. Among them, D seed (t) represents the data in the sowing stage, D fertil (t) represents the data in the fertilizing stage, D water (t) represents the data in the watering stage, D weed (t) represents the data in the weeding stage, D harvest (t) represents the data in the harvesting stage;

[0133] S52. The adaptive clustering model is used to perform dynamic cluster analysis on each production stage data set D stage (t) to identify the key features in each production stage, and the parameters of the adaptive clustering model are dynamically adjusted according to the changes in the data during the agricultural production process;

[0134] S53. Identify the key features in the agricultural production process based on the clustering results of each production stage, including soil conditions in the sowing stage, nutrient requirements in the fertilization stage, humidity changes in the watering stage, weed coverage in the weeding stage, and crop maturity in the harvesting stage. By adjusting the parameters μ j (t) and w ij (t) to cope with the changes in data during the production stage;

[0135] S54. Adjust the parameters of the adaptive clustering model for each production stage in real time, and optimize the clustering process by minimizing the objective function L stage (t):

[0136]

[0137] where λ is the adjustment parameter, represents the dynamic change of the cluster center over time.

[0138] In this embodiment, S6 includes:

[0139] S61. Automatically identify the abnormal nodes existing in the agricultural production process based on the multi-level clustering results of steps S3 to S5, including abnormal crop growth, abnormal soil conditions, sudden changes in meteorological environment, and equipment failures. By comparing the clustering results of short-term data C

[0140] (t), medium-term trend identification data T short (t), and long-term periodic data P mid (t), data C long at different spatial positions, and data C spatial (t) at each production stage, detect the deviation from the normal production mode. The result of abnormal node detection is represented by the function A(t): stage (t):

[0141] A(t) = {a1(t), a2(t), …, a m (t)};

[0142] where a m (t) represents the mth abnormal node, and the detected abnormal nodes include abnormalities in crop growth, soil conditions, meteorological environment, and equipment status;

[0143] S62. Classify each abnormal node a m (t), and determine the type of abnormality in combination with various clustering results, including abnormal crop growth, abnormal soil conditions, abnormal meteorological environment, and equipment failures. According to different types of abnormalities, automatically generate a classification label L m (t):

[0144]

[0145] Among them, L m (t) is the classification label of the abnormal node, respectively representing the variances of the clustering data at each stage. The classification process depends on the differences from the normal clustering centers at each stage;

[0146] S63. According to the classification results of the abnormal nodes, combined with the clustering data in the short term, medium term, and long term, as well as the clustering results at different spatial positions and production stages, automatically calculate the degree of abnormality E m (t) and evaluate the deviation degree of the abnormality from the normal production state;

[0147] S64. Transmit the detected abnormal node a m (t), the abnormal type L m (t), and the degree of abnormality E m (t) to the agricultural production management system in real time. The system generates an alarm based on the feedback information, and combines the data clustering characteristics at different time periods, spatial positions, and production stages to prompt the manager to handle the abnormal situation and provide corresponding handling suggestions according to the severity of the abnormality.

[0148] Example 1:

[0149] In this example, based on the improved adaptive clustering algorithm, data collection, abnormal node identification, and traceability management are carried out for the intelligent agricultural production process of a certain agricultural base. The base is located in City A, covers an area of about 1200 mu, and the planted crops are soybeans and corn. By deploying agricultural Internet of Things devices, including soil sensors, weather stations, and unmanned aerial vehicle systems, the base can monitor the growth of crops, soil conditions, and meteorological environment in real time, and track the operating status of agricultural equipment. The entire production process is compared and tested by the method of the present invention and the traditional method to verify the advantages of the present invention in the intelligent agricultural production process.

[0150] At a certain stage, the farm management system receives a piece of data from the soil sensor, indicating that the soil humidity in a certain area continues to decline. In the next 10 hours, the soil humidity data fails to return to the normal range. The system discovers through the clustering model that the soil humidity in this area deviates significantly from other areas, and the abnormal score reaches 0.92, exceeding the system - set threshold of 0.85. Further analysis shows that the abnormality of the humidity data has not spread to adjacent areas, and it is speculated that it may be due to a failure of the irrigation equipment. The system generates an abnormal report, recording the location information of the humidity - abnormal area (plot ID: #045), the abnormal period from 15:30 on May 15, 2024 to 01:30 on May 16, 2024, the abnormal score of 0.92, and the predicted abnormal reason: the irrigation equipment may be faulty. The report is sent to the agricultural management center in real time through the Internet of Things platform.

[0151] Technicians at the agricultural management center went to the site for inspection within 15 minutes after receiving the report and found that there was indeed a malfunction in an irrigation device. The device had not been working properly for the past 12 hours, resulting in the crop roots not being properly irrigated for a long time. After repairing the device, the soil moisture data gradually returned to normal. In contrast, in the adjacent area using the traditional monitoring system, due to the failure to detect the device malfunction in time, the crops were damaged, and the soil moisture in the area did not recover, affecting the growth of the crops, and the expected yield was reduced by about 8%.

[0152] Comparison of device failure detection:

[0153] Implementation of the method of the present invention: At 15:30 on May 15, the system detected abnormal soil moisture in plot ID#045, with an abnormal score of 0.92. The system automatically generated a report and sent it to the management center. The technician repaired the malfunction at 16:00, and the soil moisture data returned to the normal range.

[0154] Using the traditional method: The device failure in the same area was discovered during the manual inspection the next day, and the repair time was 11:00 on May 16, with a repair delay of 15 hours. The crops were severely affected by water shortage, and the expected regional yield was reduced by 8%.

[0155] Comparison of crop growth anomaly identification:

[0156] Implementation of the method of the present invention: On June 3, 2024, the crop height monitoring data showed that the corn in plot ID#078 was growing slowly, and the plant height was only 80% of the average height. The system found through cluster analysis that the soil nutrient content in the area was lower than that in other areas, with an abnormal score of 0.87, and recommended increasing the fertilization amount. After receiving the notice, the management staff immediately adjusted the fertilization strategy for the area. Four days later, the growth rate of the crops returned to normal.

[0157] Using the traditional method: Since the traditional system failed to detect the growth anomaly in time, it was not until the regular inspection on June 12 that it was found that the growth of the regional crops was significantly behind. Irreversible growth losses had already occurred to the crops when remedial measures were taken, and the expected yield decreased by 5%.

[0158] Comparison of meteorological environment anomaly response:

[0159] Implementation of the method of the present invention: On July 10, 2024, the weather station recorded heavy rainfall, and the rainfall exceeded 100 mm in a short time. The system analyzed and predicted through the adaptive clustering model that some low-lying areas would face the risk of waterlogging, generated a warning and recommended that the manager activate the drainage system. The report included the specific locations of the low-lying areas (plot ID: #011 and #017), the estimated rainfall, and the possible waterlogging range. The management staff immediately took measures to prevent the crops from being soaked in water.

[0160] Using traditional methods: Due to the failure to detect the risk of waterlogging in a timely manner, some areas had excessive water accumulation after rain, causing the roots of crops to be flooded. The management personnel only discovered this during the inspection the next day. It took nearly 10 hours to remove the water, and the crops in the area were damaged, with an estimated 6% decrease in yield.

[0161] Table 1 Comparative data of intelligent agricultural production

[0162]

[0163] It can be clearly seen from Example 1 that the present invention significantly improves the efficiency and accuracy of data processing in the process of intelligent agricultural production. Especially in the aspects of abnormal node recognition, crop growth monitoring, and abnormal event response, it can handle emergencies more quickly than traditional methods, reducing crop losses caused by equipment failures or environmental changes. The above comparative data fully demonstrate that the adaptive clustering algorithm of the present invention has strong real-time processing capabilities, flexibility, and accuracy, effectively solving the limitations existing in the processing of agricultural production data by traditional methods.

[0164] The present invention adopts an improved adaptive clustering algorithm that can dynamically adjust the clustering model parameters according to the changes in time, space, and production stage of agricultural production data. By dividing multi-dimensional agricultural data into short-term, medium-term, and long-term data, and performing adaptive classification and trend analysis on the data at each stage, it can more accurately handle data changes at different time scales, overcoming the limitation that traditional clustering algorithms are difficult to adapt to the variability of data. During the agricultural production process, in scenarios where meteorological environments, soil conditions, and crop growth data change rapidly, the clustering model can be updated in real time to ensure the accuracy and sensitivity of data classification, effectively avoiding production risks caused by data lag or failure to timely identify abnormal points in traditional algorithms.

[0165] The present invention automatically identifies abnormal nodes in agricultural production through a multi-level clustering model, including abnormalities in crop growth, soil conditions, meteorological environments, and equipment operation status. Different from the existing monitoring methods that rely on manual or fixed rules, the improved adaptive clustering algorithm can dynamically identify abnormalities by combining the clustering results of short-term, medium-term, and long-term data in real-time analysis with data on spatial location and production stage. It can not only improve the recognition accuracy of abnormal nodes but also classify, quantify, and provide real-time feedback on abnormal situations by calculating the deviation degree between abnormal and normal states, generating automated alarm information, reducing manual intervention, and improving the efficiency of dealing with emergencies in the process of agricultural production.

[0166] The present invention proposes to classify data according to different links of agricultural production (sowing, fertilizing, watering, weeding and harvesting) and perform dynamic clustering. The data characteristics in each production stage are different, and it is difficult for traditional methods to cope with the differences between stages. By adjusting the clustering model parameters in real time in different production stages to make it adapt to the data characteristics of each stage, flexible management of the agricultural production process is achieved. The dynamic adjustment mechanism can ensure accurate data classification in each production stage, making production management more targeted and time-effective, and improving the overall efficiency and quality of agricultural production.

[0167] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. An improved traceability method for agricultural production process based on artificial intelligence, characterized in that, It includes the following steps: S1. Real-time collect multi-dimensional agricultural production data on crop growth, soil conditions, meteorological environment, and equipment operation status through devices deployed in the agricultural production environment; S2. Perform data cleaning, noise removal, outlier detection, and format conversion on the multi-dimensional agricultural production data; S3. According to the time attributes of the multi-dimensional agricultural production data, divide the multi-dimensional agricultural production data into short-term data, medium-term data, and long-term data, and use an adaptive clustering model to analyze the data characteristics of different time periods respectively; S4. According to the spatial distribution characteristics of the multi-dimensional agricultural production data, divide the multi-dimensional agricultural production data according to the spatial positions of plots, regions, and crop types, and conduct adaptive clustering analysis on the data of each spatial position, dynamically adjusting the parameters of the adaptive clustering model so that the classification results of the multi-dimensional agricultural production data reflect the crop growth, environmental conditions, and production equipment status of different regions; S5. According to the specific links of agricultural production, classify the multi-dimensional agricultural production data according to multiple production stages of crops from sowing, fertilizing, watering, weeding, and harvesting, and conduct dynamic clustering on the data of each production stage, identify the key features in each stage, and dynamically adjust the parameters of the adaptive clustering model to cope with the data changes in the agricultural production process; S6. Automatically identify abnormal nodes existing in the agricultural production process based on the multi-level clustering results of steps S3 to S5, including abnormal crop growth, abnormal soil conditions, sudden changes in the meteorological environment, and equipment failures, and issue abnormal alarms in real time, and feedback the information of abnormal agricultural production nodes to the management system; S7. When an abnormality occurs in the agricultural production process, based on the multi-level clustering results of the multi-dimensional agricultural production data, automatically conduct traceability analysis, determine the time, location, and cause of the abnormality by tracing the data source and its historical changes of the abnormal node, and generate a traceability report; S8. Real-time feedback the abnormal nodes and their causes in the traceability management to the agricultural production management system, and dynamically adjust the irrigation frequency, fertilization amount, or equipment maintenance plan of the crops according to the feedback information; Among them, S3 includes: S31. Divide the multi-dimensional agricultural production data D formatted (t) into short-term data D short (t), medium-term data D mid (t) and long-term data D long (t) according to the time attribute of the multi-dimensional agricultural production data: D short (t) represents agricultural production data collected within a short time range for real-time monitoring and feedback; D mid (t) represents the agricultural production data collected during the medium-term time period for trend identification; D long (t) represents agricultural production data collected over a long time period for periodic pattern analysis; S32. Classify and analyze the short-term data D short (t) using an adaptive clustering model to identify the change patterns of crop growth, soil conditions, meteorological environment, and equipment operation status in agricultural production in the short term: Among them, C short (t) is the short-term clustering result, represents the i-th short-term data point, μ j (t) is the clustering center, the mean vector of the j-th class, w ij (t) is the sample belonging to the membership function of the j-th class, representing the distance weight between the short-term data point and the clustering center; S33. Perform trend recognition on the intermediate data D mid (t), analyze the data change trend using an adaptive clustering model, and capture the data change trend over time by integrating the trend within the time interval [t0, t1], which is used for the analysis of the crop growth change trend, soil nutrient fluctuation trend, and meteorological change trend: Among them, T mid (t) is the recognition result of the medium-term trend, represents the i-th medium-term data point, μ j (t) is the changing trend of the cluster center, W ij (t) is the membership function; S34. Analyze the long-term data D long (t) periodically and regularly, and use the adaptive clustering model to detect and analyze the periodic fluctuations of the long-term data in agricultural production. Capture the periodic fluctuations in the long-term data through the periodic term cos(2πf j (t - t0)): Among them, P long (t) is the analysis result of long-term periodic law, is the i-th long-term data point, μ j (t) is the clustering center, f j represents the periodic frequency, w ij (t) is the membership function 2. An improved traceability method for agricultural production processes based on artificial intelligence according to claim 1, characterized in that, S1 includes: S11. Real-time collect multi-dimensional data D of crop growth through a sensor network arranged in an agricultural production environment c (t), including leaf area, stem length, photosynthesis rate, and nutrient content. D c is a multi-dimensional vector of crop growth parameters, and t is the acquisition time point; S12. Use the sensor devices deployed in the soil to collect the multi-dimensional data D of soil conditions in real time s (t), including soil temperature, humidity, pH value and nutrient content, where D s is a multi-dimensional vector of soil conditions; S13. Real-time collect meteorological environment data D w (t), including air temperature, humidity, wind speed, light intensity and rainfall, D w is a multi-dimensional vector of the meteorological environment; S14. Real-time collect the device operation status data D related to agricultural production equipment through the device operation status monitoring system e (t), including the working current, working voltage, temperature and vibration condition of the device, D e is the multi-dimensional vector of the device operation status; S15. Integrate the multi-dimensional agricultural production data collected in steps S11 to S14 into multi-dimensional agricultural production data D(t): D(t) = (D c (t), D s (t), D w (t), D e (t)}.

3. An improved traceability method for agricultural production processes based on artificial intelligence according to claim 1, characterized in that, S2 includes: S21. Clean the collected multi-dimensional agricultural production data D(t) to remove invalid data and duplicate data, and obtain the cleaned multi-dimensional agricultural production data D clean (t); S22. Remove noise from the cleaned multi-dimensional agricultural production data D clean (t) to eliminate the noise generated from sensor devices and environmental interference, and obtain the multi-dimensional agricultural production data D denoise (t) after noise removal; S23. Perform outlier detection on the multi-dimensional agricultural production data D denoise (t) after noise removal, and use a distribution-based detection method to identify the outlier data points in the multi-dimensional agricultural production data. The outlier data points include crop growth parameters, soil conditions, and meteorological environment data that exceed the reasonable range. The outlier detection result is represented by the function E(t): E(t) = {e1(t), e2(t),..., e n (t)}; Among them, e n (t) is an abnormal data point; S24. Analyze the abnormal data in E(t), automatically correct or delete the abnormal data, and obtain the multi-dimensional agricultural production data D corrected (t) after abnormal data processing; S25. Perform format conversion on the multi-dimensional agricultural production data D corrected (t) after exception handling, and convert the multi-dimensional agricultural production data into a unified format according to the requirements of data analysis and the clustering model, so as to obtain the converted multi-dimensional agricultural production data D formatted (t).

4. An improved traceability method for agricultural production processes based on artificial intelligence according to claim 1, characterized in that, S4 includes: S41. According to the spatial distribution characteristics of the multi-dimensional agricultural production data D formatted (t), the multi-dimensional agricultural production data is spatially divided according to plots, regions, and crop types to obtain the agricultural production data D loc (t) at different spatial positions: D loc1 (t) represents the data set of plot 1, D loc2 (t) represents the data set of plot 2, D locn (t) represents the data set of plot n; S42. Analyze the agricultural production data D locn (t) at each spatial location using an adaptive clustering model. During the clustering process, dynamically adjust the cluster centers μ j (t) and the membership function w ij (t) according to different spatial locations, so that the adaptive clustering model reflects the growth of crops, environmental conditions, and the status of production equipment at the corresponding spatial locations; S43. According to the differences in spatial positions, dynamic parameter adjustment is performed on the agricultural production data D loc (t) at different spatial positions, so that the clustering results reflect the agricultural production characteristics of each region. The dynamic adjustment of the adaptive clustering model is achieved by minimizing the function L loc (t): Among them, represents the multi-dimensional agricultural production data point at the i-th spatial position, and w ij (t) is the membership function of the multi-dimensional agricultural production data point belonging to the cluster center j, λ is the regularization parameter, represents the dynamic adjustment amount of the cluster center over time. By this term, it adapts to the changes of agricultural production data in different time periods, and captures the acceleration changes of production data at different spatial positions through the second derivative term; S44. According to the results of cluster analysis, the growth conditions of crops, soil conditions, meteorological environment, and production equipment status in different regions are reflected in real time, and the classification result C of spatial positions is generated spatial (t): Among them, is the difference between the multi-dimensional agricultural production data points and the clustering center μ j (t), reflecting the spatial difference of agricultural production data at time t, used to describe the dynamic adjustment in the time dimension, involving the time derivative of the multi-dimensional agricultural production data points reflecting the change speed of crop growth, β is the adjustment coefficient, and the angular parameter θ ij (t) is used to capture the relative angular change between different clustering centers, reflecting the influence of the second-order time derivative on the clustering center, capturing the acceleration change of crop growth or equipment state, and δ is the spatial attenuation coefficient, indicating the acceleration influence that decays as the spatial distance |x i -x j | increases.

5. An improved method for tracing the agricultural production process based on artificial intelligence according to claim 1, characterized in that, S5 includes: S51. Classify the multi-dimensional agricultural production data D formatted (t) according to the specific links of agricultural production into multiple production stages including sowing, fertilizing, watering, weeding, and harvesting to obtain the production stage data set D stage (t), where D seed (t) represents the data in the sowing stage, D fertil (t) represents the data in the fertilizing stage, D water (t) represents the data in the watering stage, D weed (t) represents the data in the weeding stage, D harvest (t) represents the data in the harvesting stage; S52. For each production stage dataset D stage (t), perform dynamic clustering analysis using an adaptive clustering model, identify the key features in each production stage, and dynamically adjust the parameters of the adaptive clustering model according to the changes in data during the agricultural production process; S53. Identify the key features in the agricultural production process based on the clustering results of each production stage, including the soil conditions in the sowing stage, the nutrient requirements in the fertilization stage, the humidity changes in the watering stage, the weed coverage rate in the weeding stage, and the crop maturity in the harvesting stage, and adjust the parameters μ j (t) and w ij (t) to cope with the changes in data during the production stage; S54. Adjust the parameters of the adaptive clustering model at each production stage in real time, and optimize the clustering process by minimizing the objective function L stage (t): where λ is a tuning parameter, representing the dynamic change of the cluster center over time.

6. An improved traceability method for agricultural production process based on artificial intelligence according to claim 1, characterized in that, S6 includes: S61. Automatically identify abnormal nodes existing in the agricultural production process based on the multi-level clustering results of steps S3 to S5, including abnormal crop growth, abnormal soil conditions, sudden changes in meteorological environment, and equipment failures. By analyzing the short-term data C short (t), the medium-term trend recognition data T mid (t), and the long-term periodic data P long (t), the data C spatial at different spatial positions, and the data C stage (t) at each production stage, compare the clustering results to detect deviations from the normal production mode. The results of abnormal node detection are represented by the function A(t): A(t) = (a1(t), a2(t),..., a m (t)}; Among them, a m (t) represents the m-th abnormal node, and the detected abnormal nodes include abnormalities in crop growth, soil conditions, meteorological environment, and equipment status; S62. For each abnormal node a m (t), classify it, and determine the type of abnormality by combining the clustering results of each category, including abnormal crop growth, abnormal soil conditions, abnormal meteorological environment, and equipment failure conditions. According to different types of abnormalities, automatically generate a classification label L m (t): Among them, L m (t) is the classification label of the abnormal node, which respectively represent the variances of the clustering data at each stage, and the classification process depends on the differences from the normal clustering centers at each stage; S63. According to the classification results of abnormal nodes, combined with the clustering data in the short term, medium term, and long term, as well as the clustering results at different spatial positions and production stages, automatically calculate the degree of abnormality E m (t) and evaluate the degree of deviation of the abnormality from the normal production state; S64. The detected abnormal node a m (t), the abnormal type L m (t) and the abnormal degree E m (t) are fed back to the agricultural production management system in real time. The system generates an alarm based on the feedback information, and combines the data clustering characteristics of different time periods, spatial positions and production stages to prompt the manager to handle the abnormal situation and provide corresponding handling suggestions according to the severity of the abnormality.

Images