AI-assisted prediction method and system for plant disease occurrence patterns

Abstract

The present invention discloses an AI-assisted prediction method and system for the occurrence law of plant diseases, which relates to the field of machine learning. The method includes: establishing a multi-source data set, performing data time series disease situation analysis; using any analysis result as a reference node, performing node similarity matching; using the similarity value of the reference node and the first-level adjacent node to establish an attenuation weight, performing node similarity matching, establishing second-level adjacent nodes, until constructing N-level adjacent nodes; using the N-level adjacent nodes as the starting point, performing backtracking aggregation, using the backtracking aggregation results to enhance the reference node data, using the data enhancement results as identification data, performing reinforcement learning of the deep prediction model; and predicting the occurrence law of plant diseases. The method solves the technical problem that the existing prediction of the occurrence law of plant diseases is limited by the data source and model capability, resulting in the prediction accuracy being difficult to meet the requirements of actual applications, and achieves the technical effect of improving the prediction accuracy of the occurrence law of diseases.

Description

Technical Field

[0001] The present application relates to the field of machine learning, and in particular to an AI-assisted method and system for predicting the occurrence patterns of plant diseases. Background Art

[0002] In agricultural production, the occurrence of plant diseases poses a serious threat to crop yield and quality. In order to effectively prevent and control plant diseases, it is necessary to explore the occurrence patterns of diseases and formulate scientific prevention and control strategies based on them. At present, many plant disease prediction methods mainly rely on a single data source. Although this method is simple and easy to implement, it is often difficult to fully reflect the complex mechanism of disease occurrence due to the limitations of data information. Some traditional statistical models and preliminary machine learning methods have limitations in processing complex and nonlinear data, and it is difficult to accurately capture the dynamic changes of disease occurrence.

[0003] In the current relevant technologies, the prediction of plant disease occurrence patterns is limited by data sources and model capabilities, resulting in technical problems that the prediction accuracy is difficult to meet the requirements of practical applications. Summary of the invention

[0004] This application provides an AI-assisted method and system for predicting the occurrence patterns of plant diseases. By integrating multi-source data sets (including meteorological data, physical network sensor data, remote sensing image data, etc.), and utilizing technical means such as data preprocessing, data time series disease situation analysis, step-by-step node similarity matching, and step-by-step node backtracking aggregation, a deep prediction model is constructed. Reinforcement learning is performed using the deep prediction model to achieve dynamic prediction and early warning of disease occurrence patterns, thereby achieving the technical effect of improving the prediction accuracy of disease occurrence patterns.

[0005] This application provides an AI-assisted plant disease occurrence law prediction method, including:

[0006] A multi-source data set is established, wherein the multi-source data set includes meteorological data, physical network sensor data, and remote sensing image data. After data preprocessing of the multi-source data set, a data time series disease situation analysis is performed, and the data time series disease situation analysis is realized through historical disease data and an expert database; any data time series disease situation analysis result is used as a reference node, and node similarity matching is performed on the multi-source data set after data preprocessing to establish first-level adjacent nodes; an attenuation weight is established using the similarity value of the reference node and the first-level adjacent node, and node similarity matching of the multi-source data set after data processing is performed based on the attenuation weight and the first-level adjacent node, and second-level adjacent nodes are established, and adjacent node traversal is performed until N-level adjacent nodes are constructed, and adjacent construction is terminated; with the N-level adjacent nodes as the starting point, backtracking aggregation is performed, and backtracking aggregation results are established, and the reference node data is enhanced using the backtracking aggregation results, and the data enhancement results are used as identification data to perform reinforcement learning of the deep prediction model; the enhanced deep prediction model is used to predict the occurrence law of plant diseases, and disease warnings are issued.

[0007] In a possible implementation, the following processing is performed:

[0008] A similarity screening threshold is established before the adjacent nodes are constructed, and the similarity screening threshold is used to constrain the generation of adjacent nodes.

[0009] In a possible implementation, the adjacent node traversal is performed until N levels of adjacent nodes are constructed, and the adjacent construction is completed, and the following processing is performed:

[0010] The initial weight value is configured according to the experience database; the initial weight value is attenuated and fitted using the similarity values ​​of the reference node and the first-level adjacent node, a step-by-step attenuation result is established, the N value is configured with the step-by-step attenuation result, and the traversal end point is established; when the second-level adjacent nodes are constructed, the corresponding step-by-step attenuation result is called with the second-level adjacent nodes, the node similarity matching of the multi-source data set after data processing is performed, and the third-level adjacent nodes are established; the matching traversal is continued until the N-level adjacent nodes are constructed or any adjacent nodes do not exist and meet the similarity screening threshold, and the construction is terminated.

[0011] In a possible implementation, the corresponding step-by-step attenuation result is called with the secondary adjacent nodes, and node similarity matching of the multi-source data set after data processing is performed, and the following processing is performed:

[0012] The reference node is subjected to principal component analysis to establish a principal component analysis result; the principal component analysis result is subjected to attenuation analysis using the secondary attenuation result to establish an additional feature; the secondary adjacent nodes are used as matching features, and node similarity matching of the multi-source data set after data processing is performed based on the matching features and the additional features.

[0013] In a possible implementation, the N-level adjacent nodes are used as the starting point, backtracking aggregation is performed, a backtracking aggregation result is established, and the following processing is performed:

[0014] Initialize the features of N-level adjacent nodes to establish the N-th level aggregate feature set; establish exponential decay, use the exponential decay to trace the initial aggregate feature set back to the N-1-level adjacent nodes, initialize the features of the N-1-level adjacent nodes after backtracking, and establish the N-1-th level aggregate feature set; use exponential decay to backtrack the N-1-level aggregate feature set to the N-2-level adjacent nodes; perform level-by-level repetition of backtracking to complete backtracking aggregation.

[0015] In a possible implementation, the following processing is performed:

[0016] A time series monitoring data set is established, wherein the time series monitoring data set is a crop data set within a target growth area; after processing the time series monitoring data set, the data set is input as input data into an enhanced deep prediction model to establish an output result; an early warning level trigger analysis is performed based on the output result, and a disease early warning is issued using the early warning level trigger analysis result.

[0017] In a possible implementation, the following processing is performed:

[0018] Configure a timing update database, which is used to record multi-source data of timing updates; use the timing update database to analyze the amount of new data and establish a first optimization trigger result; use the timing update database to perform consistency analysis of multi-source data sets and establish a second optimization trigger result; perform update trigger evaluation based on the first optimization trigger result and the second optimization trigger result, and use the update trigger evaluation result and the timing update database to optimize the deep prediction model.

[0019] In a possible implementation, the data preprocessing is performed on the multi-source data set, and the following processing is performed:

[0020] After the multi-source data set is time-series aligned, data missing value filling and noise removal are performed to establish a first preprocessing result; data standardization is performed on the first preprocessing result to establish a second preprocessing result, and the second preprocessing result is used as the result after data preprocessing.

[0021] This application also provides an AI-assisted plant disease occurrence law prediction system, including:

[0022] A data time series disease situation analysis module, which is used to establish a multi-source data set, which includes meteorological data, physical network sensor data, and remote sensing image data. After data preprocessing of the multi-source data set, data time series disease situation analysis is performed, and the data time series disease situation analysis is implemented through historical disease data and expert database; a node similarity matching module, which is used to use any data time series disease situation analysis result as a reference node, perform node similarity matching on the multi-source data set after data preprocessing, and establish a first-level adjacent node; an adjacent node traversal module, which is used to use the reference node and the first-level adjacent node A decay weight is established based on the similarity value, and node similarity matching of the multi-source data set after data processing is performed based on the decay weight and the first-level adjacent nodes, and a second-level adjacent node is established. The adjacent node traversal is performed until the N-level adjacent nodes are constructed and the adjacent construction is completed; a reinforcement learning module is used to perform backtracking aggregation with the N-level adjacent nodes as the starting point, establish backtracking aggregation results, use the backtracking aggregation results to enhance the benchmark node data, and use the data enhancement results as identification data to perform reinforcement learning of the deep prediction model; a plant disease occurrence law prediction module is used to predict the occurrence law of plant diseases with the enhanced deep prediction model and issue a disease warning.

[0023] The AI-assisted plant disease occurrence law prediction method and system proposed in this application first establishes a multi-source data set, which includes meteorological data, physical network sensor data, and remote sensing image data. After data preprocessing of the multi-source data set, data time series disease situation analysis is performed. The data time series disease situation analysis is realized through historical disease data and expert database. Then, any data time series disease situation analysis result is used as a reference node, and node similarity matching is performed on the multi-source data set after data preprocessing to establish first-level adjacent nodes. Then, the attenuation is established using the similarity values ​​of the reference node and the first-level adjacent nodes. Weight, based on the attenuation weight and the first-level adjacent nodes, the node similarity matching of the multi-source data set after data processing is performed, the second-level adjacent nodes are established, and the adjacent node traversal is performed until the N-level adjacent nodes are constructed. After the adjacent construction is completed, the N-level adjacent nodes are used as the starting point to perform backtracking aggregation, and the backtracking aggregation results are established. The backtracking aggregation results are used to enhance the benchmark node data, and the data enhancement results are used as identification data to perform reinforcement learning of the deep prediction model. Finally, the enhanced deep prediction model is used to predict the occurrence law of plant diseases and issue disease warnings, achieving the technical effect of improving the prediction accuracy of disease occurrence laws. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0025] Figure 1 A flowchart of the AI-assisted plant disease occurrence pattern prediction method provided in an embodiment of the present application.

[0026] Figure 2 A schematic diagram of the structure of the AI-assisted plant disease occurrence pattern prediction system provided in an embodiment of the present application.

[0027] Explanation of the accompanying drawings: data time series disease situation analysis module 10, node similarity matching module 20, adjacent node traversal module 30, reinforcement learning module 40, plant disease occurrence law prediction module 50. DETAILED DESCRIPTION

[0028] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0029] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0030] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0031] The present application embodiment provides an AI-assisted method for predicting the occurrence pattern of plant diseases, such as Figure 1 As shown, the method includes:

[0032] Step S100, establishing a multi-source data set, the multi-source data set includes meteorological data, physical network sensor data, and remote sensing image data, performing data preprocessing on the multi-source data set, and performing data time series disease situation analysis, which is realized through historical disease data and expert database. The data preprocessing of the multi-source data set, step S100 further includes: step S110, after performing time series alignment on the multi-source data set, performing data missing value filling and noise removal, and establishing a first preprocessing result; step S120, performing data standardization on the first preprocessing result, establishing a second preprocessing result, and using the second preprocessing result as the result after data preprocessing.

[0033] Specifically, data is collected from multiple sources, including meteorological data (such as temperature, humidity, rainfall, etc.), physical network sensor data (such as soil moisture, light intensity, CO2 concentration, etc.), and remote sensing image data (such as satellite images, farmland images taken by drones, etc.). The collected data is preprocessed. Since the multi-source data sets come from different data sources, their timestamps are not completely consistent. These data are aligned in chronological order through interpolation, time window aggregation, etc., so as to facilitate subsequent time series analysis. Missing values ​​in the data are processed by forward or backward filling, mean, median or mode filling, etc.; noise in the data is processed by smoothing filtering, detrending, outlier detection and removal, etc., to improve the integrity and accuracy of the data. After completing time series alignment, missing value filling and noise removal, the obtained data set is the first preprocessing result. Standardization methods such as Z-score standardization (converting data to a distribution with a mean of 0 and a standard deviation of 1) and Min-Max standardization (converting data to values ​​between 0 and 1) are used to convert the data to a unified scale, eliminate the dimensional differences between different features, and improve the stability and accuracy of the model. After completing data standardization, the obtained data set is the second preprocessing result. Using historical disease data and expert databases, the occurrence trends and patterns of diseases at different time points are analyzed, including identifying seasonal patterns of disease occurrence and the impact of climate conditions on disease occurrence.

[0034] Step S200, taking any data time series disease situation analysis result as a reference node, performing node similarity matching in the multi-source data set after data preprocessing, and establishing first-level adjacent nodes. Specifically, taking any result in the data time series disease situation analysis result as a reference node, that is, a starting point for establishing adjacent nodes, finding data points similar to the reference node in the preprocessed multi-source data set, and establishing first-level adjacent nodes, wherein the similarity is obtained by calculating the Euclidean distance of the node attribute values.

[0035] Step S300, using the similarity value of the reference node and the first-level adjacent node to establish the attenuation weight, based on the attenuation weight and the first-level adjacent node to perform node similarity matching of the multi-source data set after data processing, establish the second-level adjacent node, perform adjacent node traversal, until the N-level adjacent nodes are constructed, and the adjacent construction is completed. Specifically, according to the similarity value of the reference node and the first-level adjacent node, the attenuation weight is calculated to reflect the change of similarity with distance. Using the attenuation weight and the first-level adjacent node, node similarity matching is performed to establish the second-level adjacent node. Repeat the above process until the N-level adjacent nodes (the adjacent node hierarchy established by multiple similarity matching) are constructed.

[0036] In a possible implementation, step S300 further includes step S310 of establishing a similarity screening threshold before constructing adjacent nodes, wherein the similarity screening threshold is used to constrain the generation of adjacent nodes.

[0037] Specifically, a similarity screening threshold is set to determine which nodes are similar enough to the previous node and can be considered as adjacent nodes. The calculated similarity is screened using the set similarity screening threshold, and only nodes whose similarity exceeds the threshold are selected as adjacent nodes of the previous node. This implementation method ensures that only nodes that are similar enough to the previous node are selected as adjacent nodes by setting a similarity screening threshold, reducing the interference of noise data on the analysis results and improving the accuracy and reliability of the data.

[0038] In a possible implementation, the adjacent node traversal is performed until N levels of adjacent nodes are constructed, and the adjacent construction is completed. Step S300 further includes step S320, configuring the initial weight value according to the experience database. Specifically, a database containing historical data, expert knowledge and field experience is accessed, and this database contains information about plant diseases, environmental factors (such as meteorological conditions, soil moisture, etc.) and the relationship between them and the occurrence of diseases. Based on the information in the experience database, an initial weight value is configured for the similarity between the reference node and the first-level adjacent node, and this weight value reflects the importance of the similarity between the nodes to the subsequent analysis. Step S330, the initial weight value is attenuated and fitted using the similarity value of the reference node and the first-level adjacent node, and a step-by-step attenuation result is established. The N value is configured with the step-by-step attenuation result to establish the traversal end point. Specifically, the similarity value between the reference node and the first-level adjacent node is calculated, and the initial weight value is attenuated according to this similarity value. This attenuation is based on an assumption that as the distance between nodes increases (the increase of node level), the correlation or similarity between them will gradually weaken. For each level of adjacent nodes, a new weight value is calculated based on the similarity value decay result of the previous level. In this way, a weight sequence of step-by-step decay is formed. Based on the stability of the step-by-step decay result, an N value is determined as the end point of the adjacent node construction, which is used to terminate the traversal during the adjacent node construction process. Step S340, when the secondary adjacent nodes are constructed, the corresponding step-by-step decay result is called with the secondary adjacent nodes, the node similarity matching of the multi-source data set after data processing is performed, and the third-level adjacent nodes are established. The matching traversal is continued until the N-level adjacent nodes are constructed or any adjacent nodes do not exist to meet the similarity screening threshold, and the construction is terminated. Specifically, after the secondary adjacent nodes are constructed, the step-by-step decay result corresponding to the secondary adjacent nodes is called. Based on the step-by-step decay result of the call and the multi-source data set after data processing, the node similarity matching is performed to establish the third-level adjacent nodes. The matching traversal is continued to establish higher-level adjacent nodes in turn until the N-level adjacent nodes are reached or any adjacent nodes no longer meet the similarity screening threshold. Once the traversal end point is reached or the similarity screening threshold is no longer met, the construction process of the adjacent nodes is terminated. This implementation method configures the initial weight value based on the empirical database and uses similar values ​​for attenuation fitting, which more accurately reflects the actual relationship between nodes, reduces the impact of noise data on the analysis results, and improves the accuracy of the prediction.

[0039] In a possible implementation, the corresponding step-by-step attenuation result is called with the secondary adjacent nodes, and the node similarity matching of the multi-source data set after data processing is performed. Step S340 further includes step S341, performing principal component analysis on the reference node to establish the principal component analysis result. Specifically, the principal component analysis is performed on the reference node. The principal component analysis extracts the main features of the data (i.e., the principal components) by projecting the original data into a new coordinate system. These principal components are sorted by importance. The first principal component contains the most information in the data, the second principal component is second, and so on. The result of the principal component analysis is saved as the principal component analysis result of the reference node. This result is a matrix containing principal component scores and principal component loads, where the score represents the projection of the original data on the principal component, and the load represents the contribution of the original variable to the principal component. Step S342, using the secondary attenuation result, the principal component analysis result is subjected to attenuation analysis to establish additional features. Specifically, the attenuation weight associated with the secondary adjacent node is obtained. This result reflects the attenuation degree of the similarity between the reference node and the primary adjacent node. By multiplying the principal component score with the corresponding attenuation weight, the secondary attenuation result is applied to the principal component analysis result to reflect the trend that the importance of the principal component information gradually decreases as the node level increases. The results of the attenuation analysis are used as additional features, including the attenuated principal component score, attenuation coefficient, etc. Step S343, using the secondary adjacent nodes as matching features, and performing node similarity matching of the multi-source data set after data processing based on the matching features and the additional features. Specifically, the features related to the secondary adjacent nodes are extracted, and the extracted matching features are combined with the additional features established in step S342 to form a complete feature set. This feature set contains direct similarity information (embodied by matching features) and indirect similarity information (embodied by additional features) between nodes. Based on the combined feature set, node similarity matching is performed in the multi-source data set after data processing. This implementation method extracts the main features of the data by performing principal component analysis on the reference node, reduces the influence of noise and redundant information, and at the same time, uses the secondary attenuation results to perform attenuation analysis on the principal component analysis results, reflects the attenuation trend of similarity between nodes, and improves the accuracy of adjacent node construction.

[0040] Step S400, taking N-level adjacent nodes as the starting point, perform backtracking aggregation, establish backtracking aggregation results, use the backtracking aggregation results to enhance the benchmark node data, and use the data enhancement results as identification data to perform reinforcement learning of the deep prediction model. Specifically, starting from N-level adjacent nodes, trace back to the benchmark node to aggregate and integrate the data. Use the backtracking aggregation results to enhance the benchmark node data to improve the richness and accuracy of the data. The enhanced data is used as identification data and input into the deep prediction model for reinforcement learning.

[0041] In a possible implementation, the N-level adjacent nodes are used as the starting point to perform backtracking aggregation and establish the backtracking aggregation result. Step S400 further includes step S410, initializing the features of the N-level adjacent nodes and establishing the N-level aggregated feature set. Specifically, feature extraction and initialization are performed on the N-level adjacent nodes (i.e., the outermost or most distant adjacent nodes), that is, key information is extracted from the node data, such as temperature and humidity in meteorological data, soil moisture and light intensity in physical network sensor data, and vegetation index in remote sensing image data. The extracted features are integrated into a feature vector to form the N-level aggregated feature set. This feature set represents the overall feature state of the N-level adjacent nodes. Step S420, establish exponential decay, use the exponential decay to trace the initial aggregated feature set back to the N-1-level adjacent nodes, initialize the features of the N-1-level adjacent nodes after backtracking, and establish the N-1-level aggregated feature set. Specifically, the exponential decay function is used to simulate the attenuation effect when information propagates from the N-level adjacent nodes to the N-1-level adjacent nodes. The exponential decay function is used to represent the phenomenon of gradually decreasing with distance. By applying the exponential decay function, the feature values ​​in the N-level aggregate feature set are adjusted to reflect their degree of influence on the N-1 level adjacent nodes. Then, the features of the N-1 level adjacent nodes after backtracking are initialized to establish the N-1 level aggregate feature set. Step S430, use exponential decay to backtrack the N-1 level aggregate feature set to the N-2 level adjacent nodes. Specifically, this step is similar to step 420, but the goal is to backtrack the information in the N-1 level aggregate feature set to the N-2 level adjacent nodes. The exponential decay function is also used to adjust the feature values, and the features are re-initialized to establish the N-2 level aggregate feature set. Step S440, perform a level-by-level repetition of backtracking to complete the backtracking aggregation. Specifically, this step is a repetitive process of steps 420 and 430, but the level of each backtracking will be reduced by one level until the reference node is reached. Through level-by-level backtracking and feature initialization, a backtracking aggregation result containing information of adjacent nodes of all levels is finally established. This implementation method integrates information from adjacent nodes at different levels through backtracking aggregation to form a comprehensive feature representation, which helps the model to more accurately understand the complexity and diversity of plant disease occurrence patterns. The exponential decay function in the backtracking aggregation process simulates the attenuation effect of information when it propagates between nodes, enhances the expressive power of features, and helps the model better capture subtle changes in disease occurrence patterns, thereby improving the prediction accuracy of deep prediction models for plant disease occurrence patterns.

[0042] Step S500, use the enhanced deep prediction model to predict the occurrence law of plant diseases and report disease warnings. Specifically, the enhanced deep prediction model is applied to the new multi-source data set to predict the occurrence law of plant diseases. According to the prediction results, it is determined whether the disease is about to occur or has occurred, and corresponding warning information is issued so that timely measures can be taken to prevent and control it. The embodiment of the present application integrates multi-source data sets (including meteorological data, physical network sensor data, remote sensing image data, etc.), and uses data preprocessing, data time series disease situation analysis, step-by-step node similarity matching, step-by-step node backtracking aggregation and other technical means to construct a deep prediction model, and uses the deep prediction model for reinforcement learning to achieve dynamic prediction and early warning of the occurrence law of diseases, thereby achieving the technical effect of improving the prediction accuracy of the occurrence law of diseases.

[0043] In a possible implementation, step S500 further includes step S510, establishing a time series monitoring data set, wherein the time series monitoring data set is a crop data set in the target growth area. Specifically, a crop data set in the target growth area is collected, including crop growth environment information (such as meteorological data, soil data, etc.) and crop growth status information (such as leaf color, growth height, yield, etc.). These data sets are arranged in time series to form a time series monitoring data set. Step S520, after processing the time series monitoring data set, it is input as input data into the enhanced deep prediction model to establish an output result. Specifically, before the time series monitoring data set is input into the model, it is preprocessed, including time series alignment, data missing value filling, noise removal and data standardization. The preprocessed data is input as input data into the enhanced deep prediction model, and the model predicts the occurrence law of plant diseases based on these data. Step S530, based on the output result, an early warning level trigger analysis is performed, and a disease early warning is reported using the early warning level trigger analysis result. Specifically, the result output by the model is a plurality of numerical values ​​or probability values, indicating the possibility or degree of disease occurrence. The output results are compared with the preset thresholds to determine the warning level (such as low risk, medium risk, high risk, etc.). According to the results of the warning level trigger analysis, the corresponding disease warning is reported by sending text messages, emails, push notifications, etc. This implementation method realizes real-time monitoring of the growth status of crops in the target growth area and disease warning by establishing a time-series monitoring data set and inputting it into the enhanced deep prediction model. Through the warning level trigger analysis, the prediction results are converted into specific warning levels, providing users with intuitive and easy-to-understand warning information, which helps users better understand disease risks and take corresponding countermeasures.

[0044] In a possible implementation, step S500 further includes step S540, configuring a time series update database, which is used to record multi-source data of time series updates. Specifically, a database specifically used to store time series update data is designed and constructed, that is, a time series update database, which can efficiently handle the insertion, query and update operations of a large amount of time series data. New data from different data sources (such as weather stations, physical network sensors, remote sensing satellites, etc.) are cleaned, formatted and standardized to ensure the quality and consistency of the data. The processed new data are stored in the time series update database in the order of the time series. Step S550, using the time series update database to analyze the amount of new data and establish a first optimization trigger result. Specifically, the amount of new data in the time series update database is regularly counted and analyzed to evaluate the growth rate and scale of the data. According to the analysis results of the amount of new data, a threshold is set. When the amount of new data reaches or exceeds this threshold, it is considered that the data is growing rapidly, and the deep prediction model may need to be updated to adapt to the new data conditions, triggering the first optimization operation, that is, considering updating or retraining the model. Step S560, using the time series update database to perform consistency analysis of the multi-source data set, and establish a second optimization trigger result. Specifically, a consistency check is performed on the multi-source data set in the time series update database, including steps such as data comparison, outlier detection, and data integrity verification to ensure that the data provided by different data sources are consistent in content and format. If the data is found to be inconsistent or abnormal, it is believed that the data quality may be affected, thereby affecting the accuracy of the deep prediction model. At this time, the second optimization operation is triggered, that is, considering updating or adjusting the model to adapt to changes in data quality. Step S570, based on the first optimization trigger result and the second optimization trigger result, an update trigger evaluation is performed, and the deep prediction model is optimized using the update trigger evaluation result and the time series update database. Specifically, the first optimization trigger result and the second optimization trigger result are combined to perform an update trigger evaluation, including considering factors such as the growth rate of data, data quality, and model performance. If the evaluation result shows that the model needs to be updated, the latest data in the time series update database is used to train or update the model, including adjusting model parameters, introducing new features, and improving model structure. Steps to improve the prediction performance and accuracy of the model. This implementation method ensures that the deep prediction model can adapt to data changes and growth, and improves the reliability and stability of the model by regularly analyzing the amount of new data and the consistency of multi-source data sets, and triggering model optimization operations based on the analysis results.

[0045] In the above, refer to Figure 1 The AI-assisted plant disease occurrence law prediction method according to an embodiment of the present invention is described in detail. Figure 2An AI-assisted plant disease occurrence pattern prediction system according to an embodiment of the present invention is described.

[0046] The AI-assisted plant disease occurrence law prediction system according to the embodiment of the present invention is used to solve the technical problem that the existing plant disease occurrence law prediction is limited by the data source and model capability, resulting in the prediction accuracy being difficult to meet the requirements of practical applications, thereby achieving the technical effect of improving the prediction accuracy of the disease occurrence law. The AI-assisted plant disease occurrence law prediction system includes: a data time series disease situation analysis module 10, a node similarity matching module 20, an adjacent node traversal module 30, a reinforcement learning module 40, and a plant disease occurrence law prediction module 50.

[0047] The data time series disease situation analysis module 10 is used to establish a multi-source data set, which includes meteorological data, physical network sensor data, and remote sensing image data. After data preprocessing of the multi-source data set, data time series disease situation analysis is performed, and the data time series disease situation analysis is implemented through historical disease data and expert database; the node similarity matching module 20 is used to use any data time series disease situation analysis result as a reference node, perform node similarity matching on the multi-source data set after data preprocessing, and establish first-level adjacent nodes; the adjacent node traversal module 30 is used to use the similarity values ​​of the reference node and the first-level adjacent nodes to establish Attenuation weight, node similarity matching of multi-source data sets after data processing based on attenuation weight and first-level adjacent nodes, establish second-level adjacent nodes, perform adjacent node traversal, until N-level adjacent nodes are constructed, and adjacent construction is completed; reinforcement learning module 40 is used to use N-level adjacent nodes as the starting point, perform backtracking aggregation, establish backtracking aggregation results, use the backtracking aggregation results to enhance the benchmark node data, use the data enhancement results as identification data, and perform reinforcement learning of the deep prediction model; plant disease occurrence law prediction module 50 is used to predict the occurrence law of plant diseases with the enhanced deep prediction model and issue disease warnings.

[0048] The following will describe in detail the specific configuration of the neighbor node traversal module 30. As described above, the neighbor node traversal module 30 may further include: a similarity screening threshold establishing unit for establishing a similarity screening threshold before neighbor node construction, wherein the similarity screening threshold is used to constrain the generation of neighbor nodes.

[0049] Among them, the adjacent node traversal is performed until N levels of adjacent nodes are constructed and the adjacent construction is completed. The adjacent node traversal module 30 may further include: an initial weight value configuration unit is used to configure the initial weight value according to the experience database; an attenuation fitting unit is used to use the similarity value of the reference node and the first-level adjacent node to perform attenuation fitting on the initial weight value, establish a step-by-step attenuation result, configure the N value with the step-by-step attenuation result, and establish the traversal end point; a third-level adjacent node establishment unit is used to call the corresponding step-by-step attenuation result with the second-level adjacent node after the second-level adjacent node is constructed, perform node similarity matching of the multi-source data set after data processing, and establish a third-level adjacent node; a construction termination unit is used to continue to execute matching traversal until N levels of adjacent nodes are constructed or any adjacent node does not exist to meet the similarity screening threshold, and the construction is terminated.

[0050] Among them, the corresponding step-by-step attenuation results are called by the secondary adjacent nodes to perform node similarity matching of the multi-source data set after data processing. The third-level adjacent node establishment unit can further include: a principal component analysis subunit is used to perform principal component analysis on the reference node and establish a principal component analysis result; an additional feature establishment subunit is used to use the secondary attenuation result to perform attenuation analysis on the principal component analysis result and establish additional features; a similarity matching subunit is used to use the secondary adjacent nodes as matching features, and perform node similarity matching of the multi-source data set after data processing based on the matching features and the additional features.

[0051] The specific configuration of the reinforcement learning module 40 will be described in detail below. As described above, taking the N-level adjacent nodes as the starting point, performing backtracking aggregation, and establishing the backtracking aggregation result, the reinforcement learning module 40 may further include: a feature initialization unit for initializing the features of the N-level adjacent nodes and establishing the N-th level aggregate feature set; a node backtracking unit for establishing exponential decay, using the exponential decay to backtrack the initial aggregate feature set to the N-1-level adjacent nodes, performing feature initialization on the backtracked N-1-level adjacent nodes, establishing the N-1-th level aggregate feature set, using exponential decay to backtrack the N-1-th level aggregate feature set to the N-2-level adjacent nodes, and performing level-by-level repetition of backtracking to complete the backtracking aggregation.

[0052] The specific configuration of the plant disease occurrence law prediction module 50 will be described in detail below. As described above, the plant disease occurrence law prediction module 50 may further include: a time series monitoring data set establishment unit for establishing a time series monitoring data set, wherein the time series monitoring data set is a crop data set in the target growth area; a prediction unit for processing the time series monitoring data set and inputting it into the enhanced deep prediction model as input data to establish an output result; and a warning level trigger analysis unit for performing a warning level trigger analysis based on the output result, and reporting a disease warning using the warning level trigger analysis result.

[0053] Among them, the plant disease occurrence law prediction module 50 can further include: a timing update database configuration unit is used to configure a timing update database, and the timing update database is used to record multi-source data of timing updates; a new data volume analysis unit is used to use the timing update database to analyze the new data volume and establish a first optimization trigger result; a consistency analysis unit is used to use the timing update database to perform consistency analysis of multi-source data sets and establish a second optimization trigger result; a deep prediction model optimization unit is used to perform update trigger evaluation according to the first optimization trigger result and the second optimization trigger result, and use the update trigger evaluation result and the timing update database to optimize the deep prediction model.

[0054] The specific configuration of the data time series disease situation analysis module 10 will be described in detail below. As described above, the multi-source data set is subjected to data preprocessing, and the data time series disease situation analysis module 10 may further include: a first preprocessing unit is used to perform data missing value filling and noise removal after the multi-source data set is time-series aligned, and establish a first preprocessing result; a second preprocessing unit is used to perform data standardization on the first preprocessing result, establish a second preprocessing result, and use the second preprocessing result as the result after data preprocessing.

[0055] The AI-assisted plant disease occurrence pattern prediction system provided in the embodiment of the present invention can execute the AI-assisted plant disease occurrence pattern prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0056] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0057] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. AI-assisted plant disease occurrence law prediction method, characterized in that: The method comprises: Establish a multi-source data set, the multi-source data set includes meteorological data, physical network sensor data, and remote sensing image data, perform data time series disease situation analysis after data preprocessing on the multi-source data set, and the data time series disease situation analysis is realized through historical disease data and expert database; The results of disease situation analysis of any data time series are used as reference nodes, and node similarity matching is performed on multi-source data sets after data preprocessing to establish first-level adjacent nodes; The attenuation weight is established by using the similarity value of the reference node and the first-level adjacent node. The node similarity matching of the multi-source data set after data processing is performed based on the attenuation weight and the first-level adjacent node, and the second-level adjacent node is established. The adjacent node traversal is performed until N-level adjacent nodes are constructed, and the adjacent construction is completed. Taking N-level adjacent nodes as the starting point, perform backtracking aggregation, establish backtracking aggregation results, use the backtracking aggregation results to enhance the benchmark node data, use the data enhancement results as identification data, and perform reinforcement learning of the deep prediction model; The enhanced deep prediction model is used to predict the occurrence patterns of plant diseases and issue disease warnings.

2. The AI-assisted plant disease occurrence law prediction method according to claim 1, characterized in that: A similarity screening threshold is established before the adjacent nodes are constructed, and the similarity screening threshold is used to constrain the generation of adjacent nodes.

3. The AI-assisted plant disease occurrence law prediction method according to claim 2, characterized in that: The performing of adjacent node traversal until N levels of adjacent nodes are constructed and adjacent construction is completed also includes: Configure initial weight values ​​based on the empirical database; Perform attenuation fitting on the initial weight value using similarity values ​​of the reference node and the first-level adjacent nodes, establish a step-by-step attenuation result, configure the N value based on the step-by-step attenuation result, and establish a traversal end point; When the second-level adjacent nodes are constructed, the corresponding step-by-step attenuation results are called with the second-level adjacent nodes, and the node similarity matching of the multi-source data set after data processing is performed to establish the third-level adjacent nodes; Continue to perform matching traversal until N levels of adjacent nodes are constructed or no adjacent node meets the similarity screening threshold, and the construction is terminated.

4. The AI-assisted plant disease occurrence law prediction method according to claim 3, characterized in that: The method of calling the corresponding step-by-step attenuation result with the secondary adjacent nodes and performing node similarity matching of the multi-source data set after data processing also includes: Performing principal component analysis on the reference node to establish a principal component analysis result; Use the secondary decay results to perform decay analysis on the principal component analysis results and establish additional features; The second-level adjacent nodes are used as matching features, and node similarity matching is performed on multi-source data sets after data processing based on matching features and additional features.

5. The AI-assisted plant disease occurrence regularity prediction method according to claim 1, characterized in that: The method of taking the N-level adjacent nodes as the starting point, performing backtracking aggregation, and establishing the backtracking aggregation result also includes: Initialize the features of N-level adjacent nodes and establish the N-level aggregated feature set; Establishing exponential decay, using the exponential decay to trace back the initial aggregate feature set to the N-1 level adjacent nodes, initializing the features of the traced back N-1 level adjacent nodes, and establishing the N-1 level aggregate feature set; Use exponential decay to trace back the N-1 level aggregated feature set to the N-2 level adjacent nodes; Perform level-by-level repetition of backtracking to complete the backtracking aggregation.

6. The AI-assisted plant disease occurrence regularity prediction method according to claim 1, characterized in that: The method further comprises: Establishing a time series monitoring data set, wherein the time series monitoring data set is a crop data set in a target growth area; After processing the time series monitoring data set, the data is input into the enhanced depth prediction model as input data to establish an output result; Based on the output results, an early warning level trigger analysis is performed, and the early warning level trigger analysis results are used to issue a disease early warning.

7. The AI-assisted plant disease occurrence regularity prediction method according to claim 1, characterized in that: The method further comprises: Configuring a time series update database, wherein the time series update database is used to record multi-source data of time series update; Utilizing the time series update database to analyze the amount of newly added data and establish a first optimization trigger result; Using the time series update database to perform consistency analysis of multi-source data sets and establish a second optimization trigger result; An update trigger evaluation is performed based on the first optimization trigger result and the second optimization trigger result, and the deep prediction model is optimized using the update trigger evaluation result and the timing update database.

8. The AI-assisted plant disease occurrence regularity prediction method according to claim 1, characterized in that: The performing data preprocessing on the multi-source data set further includes: After performing time series alignment on the multi-source data sets, performing data missing value filling and noise removal to establish a first preprocessing result; The first preprocessing result is subjected to data standardization to establish a second preprocessing result, and the second preprocessing result is used as a result after data preprocessing.

9. AI-assisted plant disease occurrence pattern prediction system, characterized by: The system is used to implement the AI-assisted plant disease occurrence law prediction method according to any one of claims 1 to 8, and the system comprises: A data time series disease situation analysis module, which is used to establish a multi-source data set, which includes meteorological data, physical network sensor data, and remote sensing image data. After data preprocessing of the multi-source data set, data time series disease situation analysis is performed. The data time series disease situation analysis is implemented through historical disease data and an expert database; A node similarity matching module, which is used to use any data time series disease situation analysis result as a reference node, perform node similarity matching on a multi-source data set after data preprocessing, and establish a first-level adjacent node; An adjacent node traversal module, wherein the adjacent node traversal module is used to establish an attenuation weight using similarity values ​​of a reference node and a first-level adjacent node, perform node similarity matching of a multi-source data set after data processing based on the attenuation weight and the first-level adjacent node, establish a second-level adjacent node, and perform adjacent node traversal until N levels of adjacent nodes are constructed, and adjacent construction is completed; A reinforcement learning module, wherein the reinforcement learning module is used to perform backtracking aggregation with N-level adjacent nodes as the starting point, establish backtracking aggregation results, use the backtracking aggregation results to perform benchmark node data enhancement, use the data enhancement results as identification data, and perform reinforcement learning of the deep prediction model; A plant disease occurrence law prediction module is used to predict the plant disease occurrence law with an enhanced deep prediction model and issue a disease warning.

Images