Peach yield dynamic prediction method based on large model and space-time mapping knowledge domain

Abstract

The invention discloses a peach yield dynamic prediction method based on a large model and a space-time mapping knowledge domain, and relates to the technical field of intelligent agriculture. The method comprises the following steps: firstly, acquiring multi-source heterogeneous data of a peach orchard, screening key influence factors by using an Xgboost-PSO-BP neural network, calculating basic association weights of the key influence factors, and constructing a space-time knowledge graph with dynamic association strength; domain knowledge injection is carried out on the basic large model based on the space-time knowledge graph, and a peach professional domain large model with the agronomic knowledge reasoning ability is obtained; further, output predicted values and knowledge calibration suggestions are output through two-way interactive reasoning of the space-time knowledge graph and the peach professional field large model, the space-time knowledge graph is fed back and updated after confirmation of a verification module, and a closed loop is formed; meanwhile, the system monitors risk factors in real time, dynamically corrects the space-time knowledge graph when a threshold value is triggered, conducts reasoning again, and outputs a corrected prediction result. According to the method, the problems of lack of agricultural causality, poor model interpretability and insufficient dynamic response in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent agriculture, and in particular to a peach yield dynamic prediction method based on a large model and a spatiotemporal knowledge graph. BACKGROUND

[0002] As an important economic fruit tree variety in China, the yield stability of peaches directly affects the economic benefits of fruit farmers, the regulation of fresh supply chains, and the efficiency of agricultural insurance actuarial calculations. Therefore, accurate prediction of peach yield plays a key supporting role in the development of intelligent agriculture. The formation process of peach yield is influenced by multiple dynamic coupling factors, including weather, soil, phenological stages, farming operation records, and variety characteristics. Therefore, a multi-dimensional and dynamic prediction system needs to be established to cope with complex agricultural production scenarios.

[0003] Current peach yield prediction technologies and related research have laid a certain foundation, but there are still the following technical defects, which cannot meet the actual needs of the industry: There are still problems such as low intelligence level in peach orchard management, lack of multi-dimensional monitoring data, low management efficiency and low energy utilization efficiency, and lack of support from big data platforms. In recent years, general large language models have been used in the agricultural field in a trial application, but due to the lack of vertical knowledge injection in the peach industry, even if some research attempts to combine knowledge graphs, they only achieve one-way calling of static knowledge, and do not establish a closed loop of knowledge-guided reasoning to feedback and complete knowledge. It is also unable to dynamically correct the understanding bias of the model for the field knowledge.

[0004] In summary, current peach yield prediction technologies have problems such as missing agronomic causality, insufficient multi-source data fusion, model and knowledge fragmentation, insufficient dynamic response, and poor interpretability. Therefore, it is necessary to construct a peach yield dynamic prediction method based on a large model and a spatiotemporal knowledge graph to integrate spatiotemporal dynamic knowledge modeling, domain large model reasoning, and risk real-time response, in order to improve prediction accuracy and industry applicability. SUMMARY

[0005] The present application aims to solve the deficiencies of existing peach yield prediction technologies and provides a peach yield dynamic prediction method based on a large model and a spatiotemporal knowledge graph. Through multi-source data fusion, spatiotemporal knowledge modeling, domain large model reasoning, and risk dynamic correction, accurate prediction of peach yield and interpretable output are achieved, providing support for digital management and control of peach orchards.

[0006] The technical solution of the present application: a peach yield dynamic prediction method based on a large model and a spatiotemporal knowledge graph, comprising the following steps: Obtain multi-source heterogeneous data of a peach orchard, use a neural network model to filter out key influencing factors from the multi-source heterogeneous data and calculate their basic correlation weights, and build a spatiotemporal knowledge graph containing dynamic correlation strength based on the key influencing factors and basic correlation weights; A fine-tuned dataset was constructed using the spatiotemporal knowledge graph, and domain knowledge was injected into the LLaMA-2B basic model to obtain a large model for the peach major domain. Based on real-time data, spatiotemporal subgraphs are extracted from the spatiotemporal knowledge graph and input into the peach professional domain big model for bidirectional interactive reasoning to obtain yield predictions, evidence chains, and knowledge calibration suggestions. After the validity of the knowledge calibration suggestions is confirmed by the verification module, they are used to update the spatiotemporal knowledge graph to form a closed loop. The system monitors preset risk factors in real time. When the monitored value triggers a risk threshold, it immediately updates the corresponding dynamic correlation strength in the spatiotemporal knowledge graph and re-predicts to output the corrected prediction result.

[0007] As a further improvement to this application, the multi-source heterogeneous data includes static data of peach varieties, dynamic meteorological data, soil monitoring data, agricultural operation record data, and UAV image data. The multi-source heterogeneous data is subjected to phenological period adaptation preprocessing and standardization processing, and missing data is processed by the phenological period similarity completion method to unify the spatiotemporal granularity of the data.

[0008] As a further improvement to this application, the phenological period adaptation preprocessing of multi-source heterogeneous data includes: based on the flowering phenological stage of peach trees, focusing on retaining and enhancing the processing of rainfall and pollinating insect density data; based on the fruit enlargement phenological stage of peach trees, focusing on retaining and enhancing the processing of soil moisture content and fruit enlargement rate data.

[0009] As a further improvement of this application, the formula for calculating the dynamic correlation strength is: Dynamic correlation strength = Basic correlation weight × Phenological period sensitivity coefficient × Data timeliness factor; The basic association weights are generated by the neural network model, the phenological period sensitivity coefficient is predefined by the agricultural expert knowledge base, and the data timeliness factor is calculated to decay based on the data acquisition time.

[0010] As a further improvement to this application, the neural network model is Xgboost-PSO-BP, which is used to screen key influencing factors, calculate basic correlation weights, and output predicted yield values.

[0011] As a further improvement to this application, the injection of domain knowledge into the LLaMA-2B basic model specifically includes: Extract the triplet data from the spatiotemporal knowledge graph, and combine it with the publicly available agricultural knowledge graph dataset and the historical yield data of the peach orchard to construct a model training dataset; Using the LLaMA-2B model as the base model, the base model is fine-tuned through knowledge masking and structured evidence chain generation tasks to obtain the large model for the peach professional field.

[0012] As a further improvement to this application, the knowledge masking task specifically includes: constructing a professional text corpus containing agronomic rules for the peach industry; performing random masking processing on key entities in the professional text corpus; training the basic large model to predict the masked key entities, thereby completing the deep injection of professional knowledge in the peach field.

[0013] As a further improvement to this application, the step of extracting a spatiotemporal subgraph from the spatiotemporal knowledge graph based on real-time data and inputting it into the peach professional domain large model for bidirectional interactive reasoning specifically includes: Receive a peach yield prediction request input by a user, the prediction request including the spatial identifier of the target peach orchard and the predicted phenological period; Based on the spatial identifiers and predicted phenological periods, the corresponding spatiotemporal subgraphs are retrieved from the spatiotemporal knowledge graph; The spatiotemporal subgraph is transformed into structured prompt words embedded with peach agronomy rules and input into the large model of the peach professional field; The large-scale model for the peach specialty performs multi-hop reasoning along the entity relationship path of the spatiotemporal knowledge graph, and outputs preliminary peach yield predictions and knowledge calibration suggestions.

[0014] As a further improvement to this application, the verification module verifies the effectiveness of the knowledge calibration recommendation by comparing it with the latest small-batch real-time data or control group experimental data.

[0015] As a further improvement to this application, the real-time monitoring of preset risk factors and re-prediction includes: Real-time monitoring data of the target peach orchard during the predicted phenological period is collected, including soil moisture content, ambient temperature, and incidence of pests and diseases; The monitoring data is compared with the preset risk factor thresholds. If any risk factor threshold is triggered, the dynamic association strength of the corresponding entity in the spatiotemporal knowledge graph is automatically updated, and a re-prediction is triggered. The final output includes: the final peach yield forecast, the structured chain of evidence affecting the forecast, and agricultural operation suggestions generated based on the chain of evidence.

[0016] Compared with existing technologies, the advantages of this invention are: 0. By introducing a spatiotemporal knowledge graph, the numerical mapping relationships in the traditional black-box model are elevated to a causal relationship network with clear agronomical significance. The Xgboost-PSO-BP hybrid model is used to quantify factor weights and, combined with phenological sensitivity and data timeliness, construct dynamic association strengths. This allows the model to more precisely and scientifically characterize the complex mechanisms of peach tree growth, significantly improving the accuracy of yield prediction. Furthermore, through the collaborative work of a large-scale peach-specific model infused with domain knowledge and the spatiotemporal knowledge graph, the single numerical prediction output is expanded into a comprehensive output of predicted values ​​and structured evidence chains. Decision-makers can clearly trace the key factors influencing the prediction and their contribution paths, understanding the agronomical basis for the model's judgments, greatly enhancing the model's credibility and acceptability in agricultural production practice.

[0017] 1. This invention utilizes a closed-loop mechanism of reasoning, suggestion, verification, and updating. The knowledge calibration suggestions generated during the reasoning process of a large-scale model in a professional field, after rigorous verification by a validation module, can be used to update the spatiotemporal knowledge graph. This transforms the entire system from a static model into an intelligent system capable of continuously learning, correcting, and enriching its knowledge base from practical applications and data, effectively avoiding model aging and knowledge lag issues. Attached Figure Description

[0018] Figure 1 A flowchart illustrating a method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph, provided for an embodiment of this application; Figure 2 A schematic diagram of the garden site environment data acquisition system and transmission provided in the embodiments of this application; Figure 3 This is a schematic diagram illustrating the bidirectional interactive reasoning between the spatiotemporal knowledge graph and the large-scale model of the peach professional field provided in the embodiments of the present invention; Figure 4 This is a schematic diagram illustrating a method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph, provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0020] See attached document Figures 1-4 A method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph includes the following steps: Step S1: Obtain multi-source heterogeneous data from Taoyuan, use a neural network model to screen out key influencing factors from the multi-source heterogeneous data and calculate their basic association weights, and construct a spatiotemporal knowledge graph containing dynamic association strength based on the key influencing factors and basic association weights. Step S2: Construct a fine-tuning dataset using the spatiotemporal knowledge graph, inject domain knowledge into the LLaMA-2B basic model, and obtain a large model for the peach professional domain; Step S3: Based on real-time data, extract spatiotemporal subgraphs from the spatiotemporal knowledge graph and input them into the peach professional domain big model for bidirectional interactive reasoning to obtain yield predictions, evidence chains, and knowledge calibration suggestions. After the knowledge calibration suggestions are verified by the verification module, they are used to update the spatiotemporal knowledge graph to form a closed loop. Step S4: Monitor the preset risk factors in real time. When the monitored value triggers the risk threshold, update the corresponding dynamic correlation strength in the spatiotemporal knowledge graph immediately and re-predict to output the corrected prediction result.

[0021] It should be noted that in step S1, multi-source heterogeneous data of peach orchards are acquired and trained using the Xgboost-PSO-BP neural network model. The purpose is to screen out environmental factors that have a significant impact on peach yield and output yield prediction values, and to deeply explore the nonlinear coupling relationship between each factor and yield.

[0022] The multi-source data includes static data on peach varieties, dynamic meteorological data, soil monitoring data, agricultural operation record data, and UAV imagery data. Static data on peach varieties was obtained through a public agricultural data platform; dynamic meteorological data was obtained through a meteorological station service interface; soil monitoring data was obtained through agricultural IoT sensors; agricultural operation record data was entered through farmer interactive terminals; and UAV imagery data was obtained through periodic aerial photography of peach orchards. The multi-source data underwent phenological period adaptation preprocessing, specifically retaining rainfall and pollinating insect density data for the peach flowering period, and soil moisture content and fruit enlargement rate data for the peach fruit enlargement period. The preprocessed multi-source data was then standardized to unify the spatiotemporal granularity of the data. A phenological period similarity completion method was used to process missing data, resulting in standardized multi-source data with spatiotemporal labels.

[0023] Specifically, the process involves acquiring heterogeneous data from multiple sources, including peach orchard images, peach varieties, sensor monitoring data, agricultural operation records, and phenological information. First, data on the entire peach production process in a specific peach-producing area over the past three years is collected, including peach varieties, weather station data, IoT sensor data, drone images, agricultural operation records, and final yield data.

[0024] The collected data underwent simple data cleaning, including deleting or correcting errors, inconsistencies, or duplicates. The cleaned data was then normalized: by standardizing the entire production process data at the Taoyuan site, the data was transformed to a specific interval or standard value range. Specifically:

[0025] Let the optimal interval be [c, d]: ; in, The bottom line is intolerable. If the upper limit is unacceptable, then in, The transformation matrix of the standard sample vector. This represents the sample size for each parameter throughout the entire production process in Taoyuan. This represents the average of the total sample size for all parameters throughout the entire production process in Taoyuan. The standard deviation of the total sample size for each parameter in the entire production process in Taoyuan; The normalized data is managed and stored using Neo4j and HBase+MongoDB based on its source, usage scenario, and data type.

[0026] It should also be noted that the Xgboost algorithm was used to perform feature importance analysis on this historical dataset, identifying the top 10 key influencing factors. These factors were used as input, and yield as output, to construct and train an Xgboost-PSO-BP neural network model. After training, the model was analyzed using methods such as SHAP to quantify the correlation weight of each key factor with the basic yield at different phenological stages.

[0027] Data was processed using feature extraction and feature selection methods. The Xgboost algorithm, which combines the advantages of Lasso and Ridge regression, was used to build the model. The model aimed to address variable selection and multicollinearity, with the goal of identifying environmental factors that significantly impact peach yield within the orchard production environment.

[0028] The objective function for environmental factors is: ; in The number of samples; The number of environmental factors; For the first The regression coefficients corresponding to each environmental factor; For the first The corresponding sample of the th sample One influencing factor; As a penalty factor; This represents the magnitude of the regularization parameter, ranging from 0 to 1. When = 0, its cost function is equivalent to the cost function of ridge regression. When = 1, its cost function is equivalent to the cost function of Lasso regression.

[0029] It should also be noted that in step S1, the basic association weights of each influencing factor in different phenological periods are quantified and extracted from the trained model. Based on this, a spatiotemporal knowledge graph ST-KG is constructed. The relationship between entities in the graph contains a dynamic association strength determined by the basic association weights, phenological period sensitivity coefficients, and data timeliness factors. This stage solves the problem of difficulty in determining weights in the early stages of knowledge graph construction.

[0030] Specifically, the Xgboost-PSO-BP neural network algorithm is used for training and prediction to explore the coupling relationship of multiple environmental factors in peach orchards, fit the nonlinear relationship between multiple environmental factors and peach yield, and construct a peach yield prediction model.

[0031] Specifically, this includes: selecting a set of peach yield data and inputting it as sample data. and target sample data Provided to network algorithms.

[0032] Use input samples Connection weights and threshold Calculate the input values ​​of each unit in the hidden layer. Then use Through the transfer function Calculate the output values ​​of each unit in the hidden layer. : ; ; Among them, the output values ​​of the hidden layer are used weight and threshold Calculate the output value of each unit in the output layer. The actual output value of each unit in the output layer is calculated using the transfer function. The specific calculation is as follows: ; ; Secondly, target samples are used in network algorithms. Compared with the actual output value of the network Calculate the training error of each unit in the output layer. The specific calculation is as follows: ; Among them, using connection weights Training error of the output layer and the output value of the intermediate layer Calculate the training error of each unit in the hidden layer. The specific calculation is as follows: ; Subsequently, the particle swarm optimization algorithm was used to optimize the weights and thresholds: ; ; in, , It is the total number of particles in this group. For the velocity of the particle, A random number between (0, 1) The particle's current position. and As a learning factor, The optimal position for each particle, This is the best position within the entire group.

[0033] Then, the training error of the hidden layer units is used. The input of each unit in the input layer To correct the connection weights and threshold The specific calculation method is as follows: ; ; The next set of peach yield data samples is selected and provided to the network algorithm. The process is repeated from step two until the error is within an acceptable range.

[0034] The trained and optimized neural network model was used to predict new peach yields. Finally, entities and relationships are defined in the Neo4j graph database, and the calculated basic association weights are used as initial attributes to construct a spatiotemporal knowledge graph ST-KG.

[0035] Furthermore, it should be noted that in step S1, the formula for calculating the dynamic correlation strength is: Dynamic correlation strength = Basic correlation weight × Phenological period sensitivity coefficient × Data timeliness factor; wherein, the basic correlation weight is generated by the Xgboost-PSO-BP model, the phenological period sensitivity coefficient is predefined by the agricultural expert knowledge base, and the data timeliness factor is calculated to decay based on the data acquisition time.

[0036] In step S2, fact triples and relation paths are extracted from the initialized spatiotemporal knowledge graph ST-KG, and combined with agronomic rule text to construct a fine-tuning dataset. Domain knowledge is injected into the general-purpose large language model, including knowledge masking and structured evidence chain generation tasks, ensuring that when outputting prediction results, it simultaneously generates a logically sound reasoning process conforming to a preset format.

[0037] See Figure 2 It should be noted that, in this embodiment, based on the peach tree growth cycle, the needs for monitoring the orchard environment, soil fertility testing, and processing management, multimodal acquisition equipment is deployed, and various sensors and network cameras are installed, including temperature and humidity sensors, carbon dioxide sensors, light sensors, and sensors for nitrogen, phosphorus, potassium, and EC values. This allows for the real-time acquisition of various data indicators throughout the entire peach orchard production process, with real-time data transmission achieved via a 5G transmission module and edge computing nodes. Climate data and soil moisture data are also acquired, including data on orchard temperature and humidity, carbon dioxide, light intensity, relative humidity in the soil, nitrogen, phosphorus, potassium, and EC values. Finally, based on the peach orchard production management auxiliary system, data such as fruit yield are obtained.

[0038] From the constructed spatiotemporal knowledge graph ST-KG, tens of thousands of knowledge triples and relation paths are extracted, and combined with publicly available agricultural knowledge texts to construct a domain knowledge dataset. Specifically, the spatiotemporal knowledge graph is extracted.

[0039] Subsequently, LLaMA-2B was selected as the base model. A structured evidence chain in JSON format was designed, and domain knowledge was injected into LLaMA-2B using this dataset. When training it to answer yield prediction questions, it outputs according to this format, thus obtaining a large-scale peach domain model, Peach-LLM, which possesses professional knowledge and interpretability. This is the large-scale peach domain model in this invention.

[0040] In step S3, a spatiotemporal subgraph is extracted from the spatiotemporal knowledge graph based on real-time data and input into the peach professional domain big model for reasoning, specifically including: Receive a peach yield prediction request input by a user, the prediction request including the spatial identifier of the target peach orchard and the predicted phenological period; Based on the spatial identifiers and predicted phenological periods, the corresponding spatiotemporal subgraphs are retrieved from the spatiotemporal knowledge graph; The spatiotemporal subgraph is transformed into structured prompt words embedded with peach agronomy rules and input into the large model of the peach professional field; The large-scale model for the peach specialty performs multi-hop reasoning along the entity relationship path of the spatiotemporal knowledge graph, and outputs preliminary peach yield predictions and knowledge calibration suggestions.

[0041] Furthermore, in step S3, the verification module verifies the effectiveness of the knowledge calibration recommendation by comparing it with the latest small-batch real-time data or control group experimental data.

[0042] Specifically, the spatiotemporal knowledge graph ST-KG and the large-scale peach domain model Peach-LLM engage in bidirectional interactive reasoning: The system receives a peach yield prediction request from a user, which includes the spatial identifier of the target peach orchard and the predicted phenological period. Based on the spatial identifier and the predicted phenological period, the system retrieves the corresponding spatiotemporal subgraph from the ST-KG, which contains the third-order entities of the target peach orchard during the predicted phenological period and their association strengths. The spatiotemporal subgraph is then transformed into structured prompts embedded with agronomic rules for the peach industry and input into the large-scale peach domain model Peach-LLM. The large-scale peach domain model Peach-LLM performs multi-hop reasoning along the entity relationship paths of the ST-KG, providing knowledge completion suggestions for the ST-KG.

[0043] Meanwhile, the large-scale model Peach-LLM in the peach professional field may output a knowledge calibration suggestion such as: "Based on reasoning, the utilization rate of phosphate fertilizer by the new variety Honey Snow Peach seems to be higher than the basic weight of 0.7 set in the current spatiotemporal knowledge graph ST-KG."

[0044] Subsequently, the verification module receives this suggestion and retrieves soil phosphorus consumption data from experimental plots recently planted with "Honey Peach" for comparison. If the data shows that the phosphorus fertilizer utilization rate is indeed higher, the corresponding basic association weight in the spatiotemporal knowledge graph ST-KG is increased from 0.7 to 0.75. The knowledge completion suggestion includes missing entities or relations in ST-KG; the spatiotemporal knowledge graph ST-KG is updated based on the knowledge completion suggestion.

[0045] In step S3, real-time monitoring of preset risk factors and re-prediction include: Real-time monitoring data of the target peach orchard during the predicted phenological period is collected, including soil moisture content, ambient temperature, and incidence of pests and diseases; The monitoring data is compared with the preset risk factor thresholds. If any risk factor threshold is triggered, the dynamic association strength of the corresponding entity in the spatiotemporal knowledge graph is automatically updated, and a re-prediction is triggered. The final output includes: the final peach yield forecast, the structured chain of evidence affecting the forecast, and agricultural operation suggestions generated based on the chain of evidence.

[0046] Specifically, the system collects real-time monitoring data of the target peach orchard during the predicted phenological period. This monitoring data includes soil moisture content, ambient temperature, and the incidence of pests and diseases. The monitoring data is then compared with preset peach risk factor thresholds, which include soil moisture content below 50% during the fruit enlargement stage, daily rainfall above 30mm during the flowering stage, and the incidence of pests and diseases above 5%. If any peach risk factor threshold is triggered by the monitoring data, the system will immediately update the dynamic correlation strength between relevant entities in the spatiotemporal knowledge graph (ST-KG) and automatically trigger a new round of collaborative reasoning to quickly output the corrected prediction results and emergency agricultural advice, ensuring the timeliness of the prediction.

[0047] In summary, by acquiring and storing multi-source heterogeneous data, including data on the entire peach orchard production process, the orchard's meteorological environment, soil nutrients, and peach tree growth, and by constructing a peach yield prediction model based on the coupling of multiple environmental factors using an improved XGBOOST-PSO-BP neural network algorithm, we can predict peach yield. This provides strong support for the digital control and refined management of the entire peach orchard production process, and helps improve the quality and efficiency of the entire peach orchard industry chain.

[0048] In the description of this invention, it should be understood that the above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0049] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0050] In this application, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0051] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0052] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0053] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0054] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0055] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0056] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0057] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0058] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph, characterized in that, Includes the following steps: Obtain multi-source heterogeneous data from Taoyuan, use a neural network model to screen out key influencing factors from the multi-source heterogeneous data and calculate their basic correlation weights, and construct a spatiotemporal knowledge graph containing dynamic correlation strength based on the key influencing factors and basic correlation weights. A fine-tuned dataset was constructed using the spatiotemporal knowledge graph, and domain knowledge was injected into the LLaMA-2B basic model to obtain a large model for the peach major domain. Based on real-time data, spatiotemporal subgraphs are extracted from the spatiotemporal knowledge graph and input into the peach professional domain big model for bidirectional interactive reasoning to obtain yield predictions, evidence chains, and knowledge calibration suggestions. After the validity of the knowledge calibration suggestions is confirmed by the verification module, they are used to update the spatiotemporal knowledge graph to form a closed loop. The system monitors preset risk factors in real time. When the monitored value triggers a risk threshold, it immediately updates the corresponding dynamic correlation strength in the spatiotemporal knowledge graph and re-predicts to output the corrected prediction result.

2. The method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph according to claim 1, characterized in that, The multi-source heterogeneous data includes static data of peach varieties, dynamic meteorological data, soil monitoring data, agricultural operation record data, and UAV image data. The multi-source heterogeneous data is preprocessed and standardized according to phenological period, and missing data is processed by the phenological period similarity completion method to unify the spatiotemporal granularity of the data.

3. The method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph according to claim 2, characterized in that, The preprocessing of multi-source heterogeneous data for phenological adaptation includes: based on the phenological stage of peach tree flowering, focusing on retaining and enhancing the processing of rainfall and pollinating insect density data; based on the phenological stage of peach tree fruit enlargement, focusing on retaining and enhancing the processing of soil moisture content and fruit enlargement rate data.

4. The method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph according to claim 2, characterized in that, The formula for calculating the dynamic correlation strength is: Dynamic correlation strength = Basic correlation weight × Phenological period sensitivity coefficient × Data timeliness factor; The basic association weights are generated by the neural network model, the phenological period sensitivity coefficient is predefined by the agricultural expert knowledge base, and the data timeliness factor is calculated to decay based on the data acquisition time.

5. The method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph according to claim 1, characterized in that, The neural network model is Xgboost-PSO-BP. The Xgboost-PSO-BP neural network model is used to screen key influencing factors, calculate basic correlation weights, and output production prediction values.

6. The method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph according to claim 1, characterized in that, The domain knowledge injection into the LLaMA-2B base model specifically includes: Extract the triplet data from the spatiotemporal knowledge graph, and combine it with the publicly available agricultural knowledge graph dataset and the historical yield data of the peach orchard to construct a model training dataset; Using the LLaMA-2B model as the base model, the base model is fine-tuned through knowledge masking and structured evidence chain generation tasks to obtain the large model for the peach professional field.

7. The method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph according to claim 6, characterized in that, The knowledge masking task specifically includes: constructing a professional text corpus containing agronomic rules for the peach industry; performing random masking on key entities in the professional text corpus; and training the basic large model to predict the masked key entities, thereby completing the deep injection of professional knowledge in the peach field.

8. The method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph according to claim 1, characterized in that, The process of extracting spatiotemporal subgraphs from the spatiotemporal knowledge graph based on real-time data and inputting them into the peach-specific domain-wide model for bidirectional interactive reasoning specifically includes: Receive a peach yield prediction request input by a user, the prediction request including the spatial identifier of the target peach orchard and the predicted phenological period; Based on the spatial identifiers and predicted phenological periods, the corresponding spatiotemporal subgraphs are retrieved from the spatiotemporal knowledge graph; The spatiotemporal subgraph is transformed into structured prompt words embedded with peach agronomy rules and input into the large model of the peach professional field; The large-scale model for the peach specialty performs multi-hop reasoning along the entity relationship path of the spatiotemporal knowledge graph, and outputs preliminary peach yield predictions and knowledge calibration suggestions.

9. The method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph according to claim 1, characterized in that, The verification module verifies the effectiveness of the knowledge calibration recommendations by comparing them with the latest small-batch real-time data or control group experimental data.

10. The method for dynamic prediction of peach yield based on a large model and spatiotemporal knowledge graph according to claim 1, characterized in that, The real-time monitoring of preset risk factors and re-prediction includes: Real-time monitoring data of the target peach orchard during the predicted phenological period is collected, including soil moisture content, ambient temperature, and incidence of pests and diseases; The monitoring data is compared with the preset risk factor thresholds. If any risk factor threshold is triggered, the dynamic association strength of the corresponding entity in the spatiotemporal knowledge graph is automatically updated, and a re-prediction is triggered. The final output includes: the final peach yield forecast, the structured chain of evidence affecting the forecast, and agricultural operation suggestions generated based on the chain of evidence.

Images