A facility agriculture intelligent management system and method based on trusted data

Abstract

The application discloses a facility agriculture intelligent management system and method based on credible data, relates to the technical field of facility agriculture intelligent management, and comprises the following steps: collecting relevant information of target facility agriculture subjects, and configuring initial scheduling weights for data uploading of the target facility agriculture subjects; acquiring historical data of the subjects according to subject access time, and performing credible notarization on access process information; forming a standardized unified data view; preparing training data and jointly training a global model in cooperation with multiple subjects; determining individualized performance references of the subjects according to subject business attributes; after deploying the global model, combining real-time inference performance with the references, and performing first adaptive adjustment on configuration weights of the subjects; collecting global model operation feedback, comprehensively determining contribution degrees of inference performance and current configuration states of the subjects, and performing second dynamic adjustment on configuration of the subjects, so that the application realizes facility agriculture intelligent management.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology for facility agriculture, specifically an intelligent management system and method for facility agriculture based on trusted data. Background Technology

[0002] With the rapid development of new-generation information technologies such as the Internet of Things, big data, and artificial intelligence, facility agriculture is accelerating its transformation from a traditional experience-driven model to a data-driven intelligent management model. In the daily operation of facility agriculture, environmental monitoring, energy consumption management, and agricultural operations continuously generate a large amount of multi-source heterogeneous data. The efficient aggregation, reliable circulation, and intelligent application of this data have become important links supporting the management of facility agriculture.

[0003] However, traditional facility agriculture management models face the following prominent problems when dealing with the processing and collaborative application of agricultural data from multiple entities, sources, and scenarios: First, data collection standards are not unified, and multi-source heterogeneous data is difficult to share. Currently, the facility agriculture field lacks unified data standards and interface specifications, and sensor equipment produced by different manufacturers has significant differences in communication protocols, data formats, and units of measurement. Second, data circulation lacks a reliable mechanism. Facility agriculture data involves sensitive commercial data such as the operating status, planting varieties, and yield information of agricultural production entities, which faces security threats during cross-entity circulation and sharing. Most existing agricultural management systems lack sound identity authentication, data storage, and usage control mechanisms. In addition, there is a lack of replicable and scalable collaborative modeling and intelligent decision-making mechanisms. Existing facility agriculture management systems are mostly customized for single farms or single crops, and model training relies on centralized data aggregation. Due to concerns about data privacy and security, various entities are unwilling to share raw data, making cross-entity collaborative modeling difficult to carry out. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent management system and method for facility agriculture based on trusted data, so as to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent management of facility agriculture based on trusted data, the method comprising the following steps: Collect relevant information about the target facility agriculture entities and form a standardized configuration dataset, and configure the initial scheduling weight for data upload of each target facility agriculture entity based on the relevant information; Historical data is retrieved based on the access trigger timestamp, and after timestamp verification, a historical operation data sequence is formed. The access process information is then written into the blockchain for evidence storage. Historical operational data is structured and analyzed, and agricultural public data is integrated to construct a historical multi-source heterogeneous data fusion set; this set is then normalized and stored in chronological order after being combined with the real-time fusion set to form a standardized data view. Extract parameter sequences from the standardized view and sort and normalize them; initiate a joint modeling task, each subject trains locally and uploads gradient updates, and merges them to generate a global model; match the historical inference records of each subject according to crop variety and facility type, calculate the benchmark and fluctuation range; deploy the global model for local inference verification, compare the real-time accuracy with the benchmark, and dynamically configure the aggregation weights of each subject for the first time accordingly; Obtain the global model operation logs deployed locally by each subject, extract inference accuracy, inference execution rate and inference response latency, and obtain a comprehensive score according to the preset proportion. Combine the aggregate weight after the first configuration to determine the comprehensive contribution score. Use the average score of all subjects as the benchmark, compare the score of each subject with the average, and perform the second dynamic configuration accordingly.

[0006] Collect relevant information about the target facility agriculture entities and form a standardized configuration dataset. Then, configure the initial scheduling weights for data uploads from each target facility agriculture entity based on this information. Specific steps include: Identify the target facility agriculture entities, collect their unique identifiers, facility type parameters, planting scale parameters, and main crop varieties to form a standardized configuration dataset. The facility type parameters include multi-span greenhouses, solar greenhouses, and plastic greenhouses, and the crop varieties include strawberries, tomatoes, and cucumbers. The target facility agriculture entity is authenticated, and the standardized configuration dataset is stored after successful authentication. For each target facility agriculture entity that has passed identity authentication, select the corresponding preset facility complexity coefficient according to the facility type parameter of the target facility agriculture entity; The preset facility complexity coefficients include the facility complexity coefficients for multi-span greenhouses, solar greenhouses, and plastic greenhouses; the preset facility complexity coefficient for multi-span greenhouses is 1.0; the preset facility complexity coefficient for solar greenhouses is 0.7; and the preset facility complexity coefficient for plastic greenhouses is 0.5. The planting scale coefficient is determined based on the planting scale parameters of the target facility agriculture entity, whereby the planting scale coefficient is the planting area value of the target facility agriculture entity. Let S be the facility complexity coefficient of the i-th target facility agriculture entity. fac,i The planting scale coefficient is A i The product value P of the i-th target facility agriculture entity is calculated using the following formula. i The definition is as follows: P i =S fac,i ×Ai ; Let n be the total number of target facility agriculture entities that have passed identity authentication. The total product value P is obtained by calculating the sum of the product values ​​of all target facility agriculture entities using the following formula. total The definition is as follows: P total =Σ j=1 n P j ; The initial scheduling weight W for the i-th target facility agriculture entity is calculated using the following formula. i The definition is as follows: W i =P i / P total ; Based on the initial scheduling weight W of each target facility agriculture entity i Configure the data upload channel priority for each target facility agriculture entity and record the access trigger timestamp for each target facility agriculture entity. The initial scheduling weight is positively correlated with the data upload channel priority.

[0007] In one specific implementation, the preset baseline upload frequency is 12 times / hour, based on the initial scheduling weight W of each target facility agriculture entity. i Calculate the actual transmission frequency F i =F base ×W i , of which F base For example, the W frequency of target facility agriculture entity A. A =0.45, its actual transmission frequency is 12×0.45=5.4 times / hour.

[0008] Historical data is retrieved based on the access trigger timestamp, and after timestamp verification, a historical operation data sequence is formed. The access process information is then written into the blockchain for evidence storage. The specific steps include: Based on the access trigger timestamp and unique identifier of each target facility agriculture entity, a targeted data retrieval request is initiated through the data aggregation interface to retrieve the original collected data records of each target facility agriculture entity within a preset time window before the access trigger timestamp; The original collected data records are verified by data timestamp, and data records with timestamps earlier than the access trigger timestamp are filtered out to form a historical operation data sequence for each target facility agriculture entity; The access process information of the target facility agriculture entity is written into the blockchain for evidence storage and traceability, generating a unique evidence hash value.

[0009] The historical operational data is structured and analyzed, and then integrated with agricultural public data to construct a historical multi-source heterogeneous data fusion set. This fusion set is then normalized and stored chronologically to form a standardized data view. Specific steps include: The historical operational data sequence was structured and analyzed to extract environmental parameters, energy consumption parameters and crop yield records. The environmental parameters included greenhouse temperature, humidity, light intensity, CO2 concentration and soil EC value, and the energy consumption parameters included heating energy consumption, supplemental lighting energy consumption and water and fertilizer energy consumption. The agricultural public data of each target facility agriculture entity is obtained. Using the unique identifier of each target facility agriculture entity as the association index, the agricultural public data is associated with the extracted environmental parameters, energy consumption parameters, agricultural operation records and crop yield records to form a complete data record of each target facility agriculture entity. The agricultural public data includes agricultural subsidies, government services, farmland transfer, seed industry management and price monitoring. Based on the unique identifiers of each target facility agriculture entity, the data is classified and collected to form a historical multi-source heterogeneous data fusion set; The heterogeneous data in the historical multi-source heterogeneous data fusion set and the real-time multi-source heterogeneous data fusion set are normalized according to a unified data format standard. The unified data format standard includes data field naming conventions, unified rules for measurement units, and time encoding format. The normalized data is then stored in chronological order to form a standardized data view.

[0010] Extract parameter sequences from the standardized view and sort and normalize them; initiate a joint modeling task, with each subject training locally and uploading gradient updates, then merging to generate a global model; match historical inference records of each subject based on crop variety and facility type, calculate the baseline and fluctuation range; deploy the global model for local inference verification, compare the real-time accuracy with the baseline, and dynamically configure the aggregation weights of each subject for the first time accordingly. Specific steps include: Extract the environmental parameter sequence, energy consumption parameter sequence, and crop yield record of each target facility agriculture entity from the standardized data view. Using the unique identifier of the target facility agriculture entity as the search index, sort the environmental parameters, energy consumption parameters, and crop yield records of each target facility agriculture entity in the order of timestamps. Normalize each parameter after sorting so that the parameter values ​​of each target facility agriculture entity are in the range of 0 to 1. Based on the federated learning framework, a cross-subject joint modeling task is initiated, and collaborative training is carried out among no fewer than 5 target facility agriculture subjects. Each target facility agriculture subject independently trains the local parameters of the facility agriculture prediction model using its own standardized data, calculates the model gradient update based on the trained local parameters, and uploads the model gradient update to the central server. The central server performs a weighted fusion operation on the model gradient updates uploaded by each target facility agriculture subject based on the model training weight parameters to obtain a global facility agriculture prediction model. Obtain the crop variety identifier and facility type identifier of each target facility agriculture entity in the standardized data view. Match the historical inference records of each target facility agriculture entity using the local facility agriculture prediction model before access based on the crop variety identifier and facility type identifier. The historical inference records include the number of inference requests and the corresponding inference accuracy of each target facility agriculture entity in the historical time period before access. Remove outliers in the historical inference records, calculate the average inference accuracy of each crop variety and facility type in the remaining historical inference records as the standard inference performance benchmark for each target facility agriculture entity, and calculate the standard deviation of the inference accuracy to determine the fluctuation range. The global facility agriculture prediction model is deployed to the local inference environment of each target facility agriculture entity. The real-time inference accuracy of each target facility agriculture entity in local inference verification of the global facility agriculture prediction model is obtained. The real-time inference accuracy is compared with the standard inference performance benchmark of each target facility agriculture entity. Based on the comparison results, the model aggregation weight of each target facility agriculture entity is dynamically configured for the first time.

[0011] The model aggregation weights for each target facility agriculture entity are dynamically configured for the first time. The specific steps include: When the real-time inference accuracy of a target facility agriculture entity is higher than its standard inference performance benchmark and exceeds the fluctuation range, the aggregation weight of the target facility agriculture entity in the global model aggregation is increased. When the real-time inference accuracy of a target facility agriculture entity is lower than its standard inference performance benchmark and exceeds the fluctuation range, the aggregation weight of the target facility agriculture entity in the global model aggregation is reduced. When the real-time inference accuracy of a target facility agriculture entity is within its standard inference performance benchmark and fluctuation range, the current aggregation weight of the target facility agriculture entity remains unchanged, and the first dynamic configuration is completed.

[0012] Obtain the global model runtime logs deployed locally by each entity, extract inference accuracy, inference execution rate, and inference response latency, and synthesize them according to preset proportions to obtain a comprehensive score. Combine this with the aggregated weights after the first configuration to determine the comprehensive contribution score. Using the average score of all entities as a benchmark, compare each entity's score with the average, and perform a second dynamic configuration accordingly. The specific steps include: Obtain the operation logs of the global facility agriculture prediction model deployed locally by each target facility agriculture entity, and extract the inference accuracy, inference execution rate and inference response latency recorded in the operation logs; Based on the reasoning accuracy, reasoning execution rate, and reasoning response latency of each target facility agriculture entity, the reasoning accuracy, reasoning execution rate, and reasoning response latency are used as evaluation dimensions, and a weighted sum is performed according to a preset weight ratio to obtain the comprehensive score of each target facility agriculture entity; the comprehensive score is multiplied by the aggregate weight of each target facility agriculture entity after the first dynamic configuration to calculate the comprehensive contribution score of each target facility agriculture entity. Calculate the average of the comprehensive contribution scores of all target facility agriculture entities as the average comprehensive contribution score; compare the comprehensive contribution score of each target facility agriculture entity with the average comprehensive contribution score, and perform a second dynamic configuration of the aggregation weight of each target facility agriculture entity based on the comparison results.

[0013] The aggregation weights of each target facility agriculture entity are dynamically configured a second time, as follows: When the comprehensive contribution score of a target facility agriculture entity is higher than the average comprehensive contribution score, the aggregation weight of that target facility agriculture entity in the global model aggregation is increased. When the comprehensive contribution score of a target facility agriculture entity is lower than the average comprehensive contribution score, the aggregation weight of that target facility agriculture entity in the global model aggregation is reduced. When the comprehensive contribution score of a target facility agriculture entity is within the deviation range of the average comprehensive contribution score, the aggregation weight of the target facility agriculture entity in the global model aggregation remains unchanged, and the second dynamic configuration is completed.

[0014] A smart management system for facility agriculture based on trusted data, the system comprising: Trusted access module, data fusion module, federated modeling module, feedback configuration module; The trusted access module is used to collect relevant information of the target facility agriculture entities and configure the initial scheduling weight for data upload of each target facility agriculture entity; The data fusion module is used to obtain the subject's historical data based on the subject's access time and to reliably store access process information; forming a standardized unified data view. The federated modeling module is used to prepare training data and jointly train multiple entities to generate a global model; determine the personalized performance reference of each entity based on its business attributes; after deploying the global model, combine the real-time inference performance with the reference to make the first adaptive adjustment of the configuration weights of each entity; The feedback configuration module is used to collect global model operation feedback, determine the contribution by combining the reasoning performance of each subject with the current configuration status, and make a second dynamic adjustment to the configuration of each subject. The output of the trusted access module is connected to the input of the data fusion module; The output of the data fusion module is connected to the input of the federated modeling module; The output of the federated modeling module is connected to the input of the feedback configuration module.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. By collecting the facility type parameters and planting scale parameters of the target facility agriculture entities, and based on the proportion of the product of the facility complexity coefficient and the planting scale coefficient to the sum of the products of all entities, the initial scheduling weight of data upload for each entity is automatically configured, and the access trigger timestamp is recorded. Unlike the existing technology that uniformly configures the upload frequency for all entities or relies on manual settings, this invention realizes differentiated automatic allocation of data upload channel resources. 2. By performing structured analysis on historical operational data and integrating agricultural public data, a multi-source heterogeneous data fusion set is constructed, which is then normalized to form a standardized data view. At the same time, the access process information is written into the blockchain for evidence storage and traceability. Unlike the existing technology where multi-source heterogeneous agricultural data is difficult to integrate and lacks a reliable circulation mechanism, this invention achieves cross-domain integration and standardized unification of facility agriculture production data and public data. 3. Match the historical inference records of the local model used before accessing each subject with the crop variety and facility type, and calculate the personalized standard inference performance benchmark and fluctuation range; after deploying the global model, perform local inference verification, compare the real-time inference accuracy with the benchmark, and perform the first dynamic configuration of the subject aggregation weight; collect global model operation feedback, determine the contribution of each subject by combining the inference performance with the current configuration status, and perform the second dynamic adjustment with the overall average level as a reference. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for intelligent management of facility agriculture based on trusted data, as proposed in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] like Figure 1 As shown, the present invention provides a technical solution, a method for intelligent management of facility agriculture based on trusted data, the method comprising the following steps: Collect relevant information about the target facility agriculture entities and form a standardized configuration dataset, and configure the initial scheduling weight for data upload of each target facility agriculture entity based on the relevant information; Historical data is retrieved based on the access trigger timestamp, and after timestamp verification, a historical operation data sequence is formed. The access process information is then written into the blockchain for evidence storage. Historical operational data is structured and analyzed, and agricultural public data is integrated to construct a historical multi-source heterogeneous data fusion set; this set is then normalized and stored in chronological order after being combined with the real-time fusion set to form a standardized data view. Extract parameter sequences from the standardized view and sort and normalize them; initiate a joint modeling task, each subject trains locally and uploads gradient updates, and merges them to generate a global model; match the historical inference records of each subject according to crop variety and facility type, calculate the benchmark and fluctuation range; deploy the global model for local inference verification, compare the real-time accuracy with the benchmark, and dynamically configure the aggregation weights of each subject for the first time accordingly; Obtain the global model operation logs deployed locally by each subject, extract inference accuracy, inference execution rate and inference response latency, and obtain a comprehensive score according to the preset proportion. Combine the aggregate weight after the first configuration to determine the comprehensive contribution score. Use the average score of all subjects as the benchmark, compare the score of each subject with the average, and perform the second dynamic configuration accordingly.

[0019] Collect relevant information about the target facility agriculture entities and form a standardized configuration dataset. Then, configure the initial scheduling weights for data uploads from each target facility agriculture entity based on this information. Specific steps include: Identify the target facility agriculture entities, collect their unique identifiers, facility type parameters, planting scale parameters, and main crop varieties to form a standardized configuration dataset. The facility type parameters include multi-span greenhouses, solar greenhouses, and plastic greenhouses, and the crop varieties include strawberries, tomatoes, and cucumbers. The target facility agriculture entity is authenticated, and the standardized configuration dataset is stored after successful authentication. For each target facility agriculture entity that has passed identity authentication, select the corresponding preset facility complexity coefficient according to the facility type parameter of the target facility agriculture entity; The preset facility complexity coefficients include the facility complexity coefficients for multi-span greenhouses, solar greenhouses, and plastic greenhouses; the preset facility complexity coefficient for multi-span greenhouses is 1.0; the preset facility complexity coefficient for solar greenhouses is 0.7; and the preset facility complexity coefficient for plastic greenhouses is 0.5. The planting scale coefficient is determined based on the planting scale parameters of the target facility agriculture entity, whereby the planting scale coefficient is the planting area value of the target facility agriculture entity. Let S be the facility complexity coefficient of the i-th target facility agriculture entity. fac,i The planting scale coefficient is A i The product value P of the i-th target facility agriculture entity is calculated using the following formula. i The definition is as follows: P i =S fac,i ×A i ; Let n be the total number of target facility agriculture entities that have passed identity authentication. The total product value P is obtained by calculating the sum of the product values ​​of all target facility agriculture entities using the following formula. total The definition is as follows: P total =Σ j=1 n P j ; The initial scheduling weight W for the i-th target facility agriculture entity is calculated using the following formula. i The definition is as follows: W i =P i / P total ; Based on the initial scheduling weight W of each target facility agriculture entity i Configure the data upload channel priority for each target facility agriculture entity and record the access trigger timestamp for each target facility agriculture entity. The initial scheduling weight is positively correlated with the data upload channel priority.

[0020] In one specific implementation, the preset baseline upload frequency is 12 times / hour, based on the initial scheduling weight W of each target facility agriculture entity. i Calculate the actual transmission frequency F i =F base ×W i , of which F base For example, the W frequency of target facility agriculture entity A. A =0.45, its actual transmission frequency is 12×0.45=5.4 times / hour.

[0021] Historical data is retrieved based on the access trigger timestamp, and after timestamp verification, a historical operation data sequence is formed. The access process information is then written into the blockchain for evidence storage. The specific steps include: Based on the access trigger timestamp and unique identifier of each target facility agriculture entity, a targeted data retrieval request is initiated through the data aggregation interface to retrieve the original collected data records of each target facility agriculture entity within a preset time window before the access trigger timestamp; The original collected data records are verified by data timestamp, and data records with timestamps earlier than the access trigger timestamp are filtered out to form a historical operation data sequence for each target facility agriculture entity; The access process information of the target facility agriculture entity is written into the blockchain for evidence storage and traceability, generating a unique evidence hash value.

[0022] The historical operational data is structured and analyzed, and then integrated with agricultural public data to construct a historical multi-source heterogeneous data fusion set. This fusion set is then normalized and stored chronologically to form a standardized data view. Specific steps include: The historical operational data sequence was structured and analyzed to extract environmental parameters, energy consumption parameters and crop yield records. The environmental parameters included greenhouse temperature, humidity, light intensity, CO2 concentration and soil EC value, and the energy consumption parameters included heating energy consumption, supplemental lighting energy consumption and water and fertilizer energy consumption. The agricultural public data of each target facility agriculture entity is obtained. Using the unique identifier of each target facility agriculture entity as the association index, the agricultural public data is associated with the extracted environmental parameters, energy consumption parameters, agricultural operation records and crop yield records to form a complete data record of each target facility agriculture entity. The agricultural public data includes agricultural subsidies, government services, farmland transfer, seed industry management and price monitoring. Based on the unique identifiers of each target facility agriculture entity, the data is classified and collected to form a historical multi-source heterogeneous data fusion set; The heterogeneous data in the historical multi-source heterogeneous data fusion set and the real-time multi-source heterogeneous data fusion set are normalized according to a unified data format standard. The unified data format standard includes data field naming conventions, unified rules for measurement units, and time encoding format. The normalized data is then stored in chronological order to form a standardized data view.

[0023] Extract parameter sequences from the standardized view and sort and normalize them; initiate a joint modeling task, with each subject training locally and uploading gradient updates, then merging to generate a global model; match historical inference records of each subject based on crop variety and facility type, calculate the baseline and fluctuation range; deploy the global model for local inference verification, compare the real-time accuracy with the baseline, and dynamically configure the aggregation weights of each subject for the first time accordingly. Specific steps include: Extract the environmental parameter sequence, energy consumption parameter sequence, and crop yield record of each target facility agriculture entity from the standardized data view. Using the unique identifier of the target facility agriculture entity as the search index, sort the environmental parameters, energy consumption parameters, and crop yield records of each target facility agriculture entity in the order of timestamps. Normalize each parameter after sorting so that the parameter values ​​of each target facility agriculture entity are in the range of 0 to 1. Based on the federated learning framework, a cross-subject joint modeling task is initiated, and collaborative training is carried out among no fewer than 5 target facility agriculture subjects. Each target facility agriculture subject independently trains the local parameters of the facility agriculture prediction model using its own standardized data, calculates the model gradient update based on the trained local parameters, and uploads the model gradient update to the central server. The central server performs a weighted fusion operation on the model gradient updates uploaded by each target facility agriculture subject based on the model training weight parameters to obtain a global facility agriculture prediction model. Obtain the crop variety identifier and facility type identifier of each target facility agriculture entity in the standardized data view. Match the historical inference records of each target facility agriculture entity using the local facility agriculture prediction model before access based on the crop variety identifier and facility type identifier. The historical inference records include the number of inference requests and the corresponding inference accuracy of each target facility agriculture entity in the historical time period before access. Remove outliers in the historical inference records, calculate the average inference accuracy of each crop variety and facility type in the remaining historical inference records as the standard inference performance benchmark for each target facility agriculture entity, and calculate the standard deviation of the inference accuracy to determine the fluctuation range. The global facility agriculture prediction model is deployed to the local inference environment of each target facility agriculture entity. The real-time inference accuracy of each target facility agriculture entity in local inference verification of the global facility agriculture prediction model is obtained. The real-time inference accuracy is compared with the standard inference performance benchmark of each target facility agriculture entity. Based on the comparison results, the model aggregation weight of each target facility agriculture entity is dynamically configured for the first time.

[0024] The model aggregation weights for each target facility agriculture entity are dynamically configured for the first time. The specific steps include: When the real-time inference accuracy of a target facility agriculture entity is higher than its standard inference performance benchmark and exceeds the fluctuation range, the aggregation weight of the target facility agriculture entity in the global model aggregation is increased. When the real-time inference accuracy of a target facility agriculture entity is lower than its standard inference performance benchmark and exceeds the fluctuation range, the aggregation weight of the target facility agriculture entity in the global model aggregation is reduced. When the real-time inference accuracy of a target facility agriculture entity is within its standard inference performance benchmark and fluctuation range, the current aggregation weight of the target facility agriculture entity remains unchanged, and the first dynamic configuration is completed.

[0025] Obtain the global model runtime logs deployed locally by each entity, extract inference accuracy, inference execution rate, and inference response latency, and synthesize them according to preset proportions to obtain a comprehensive score. Combine this with the aggregated weights after the first configuration to determine the comprehensive contribution score. Using the average score of all entities as a benchmark, compare each entity's score with the average, and perform a second dynamic configuration accordingly. The specific steps include: Obtain the operation logs of the global facility agriculture prediction model deployed locally by each target facility agriculture entity, and extract the inference accuracy, inference execution rate and inference response latency recorded in the operation logs; Based on the reasoning accuracy, reasoning execution rate, and reasoning response latency of each target facility agriculture entity, the reasoning accuracy, reasoning execution rate, and reasoning response latency are used as evaluation dimensions, and a weighted sum is performed according to a preset weight ratio to obtain the comprehensive score of each target facility agriculture entity; the comprehensive score is multiplied by the aggregate weight of each target facility agriculture entity after the first dynamic configuration to calculate the comprehensive contribution score of each target facility agriculture entity. Calculate the average of the comprehensive contribution scores of all target facility agriculture entities as the average comprehensive contribution score; compare the comprehensive contribution score of each target facility agriculture entity with the average comprehensive contribution score, and perform a second dynamic configuration of the aggregation weight of each target facility agriculture entity based on the comparison results.

[0026] The aggregation weights of each target facility agriculture entity are dynamically configured a second time, as follows: When the comprehensive contribution score of a target facility agriculture entity is higher than the average comprehensive contribution score, the aggregation weight of that target facility agriculture entity in the global model aggregation is increased. When the comprehensive contribution score of a target facility agriculture entity is lower than the average comprehensive contribution score, the aggregation weight of that target facility agriculture entity in the global model aggregation is reduced. When the comprehensive contribution score of a target facility agriculture entity is within the deviation range of the average comprehensive contribution score, the aggregation weight of the target facility agriculture entity in the global model aggregation remains unchanged, and the second dynamic configuration is completed.

[0027] In Example 1: The target entity is a facility agriculture cooperative with a certain scale of multi-span greenhouses, whose main crop is tomatoes. First, the target entity is identified, and its unique identifier, facility type parameters, planting scale parameters, and main crop varieties are collected to form a standardized configuration dataset. The facility type parameters are preset with corresponding complexity coefficients. Multi-span greenhouses correspond to the highest complexity coefficient, solar greenhouses correspond to the medium complexity coefficient, and plastic greenhouses correspond to the lowest complexity coefficient. The planting scale coefficient is directly adopted from the planting area value of the entity. The product of the facility complexity coefficient and the planting scale coefficient of the entity is calculated, and then the proportion of this product to the sum of the products of all entities that have passed identity authentication is calculated to obtain the initial scheduling weight of the entity. According to the weight, the priority of the data upload channel is configured for the entity. The larger the weight, the higher the priority of the upload channel. At the same time, the access trigger timestamp of the entity is recorded as a time reference benchmark for subsequent retrieval of historical data. After the entity completes the access, it initiates a targeted data retrieval request through the data aggregation interface based on its access trigger timestamp and unique identifier. The request retrieves the original data records collected by the entity within a preset time window before the access time point. The retrieved original data records are then verified by data timestamp, and data records with timestamps earlier than the access trigger timestamp are filtered out to form the entity's historical operation data sequence. At the same time, the entity's access process information is written into the blockchain for evidence storage and traceability, generating a unique evidence hash value to ensure the immutability and traceability of the access process. The historical operational data sequence is structured and parsed to extract environmental parameters, energy consumption parameters, and crop yield records. Environmental parameters include greenhouse temperature, humidity, light intensity, carbon dioxide concentration, and soil conductivity. Energy consumption parameters include heating energy consumption, supplemental lighting energy consumption, and water and fertilizer energy consumption. Simultaneously, agricultural public data for the entity is acquired. Using the entity's unique identifier as the association index, the agricultural public data is linked with the extracted environmental parameters, energy consumption parameters, and crop yield records to form a complete data record for the entity. This data is then categorized and aggregated based on the unique identifier to construct a historical multi-source heterogeneous data fusion set. The heterogeneous data from this historical multi-source heterogeneous data fusion set and the real-time multi-source heterogeneous data fusion set are normalized according to a unified data format standard. This standard includes data field naming conventions, unified unit of measurement rules, and time encoding formats, ensuring that data from different devices and different sources can be fused under the same standard. The normalized data is stored sequentially to form a standardized data view. The environmental parameter sequence, energy consumption parameter sequence, and crop yield record of each target subject are extracted from the standardized data view. The subject's unique identifier is used as the search index. The parameters are sorted in timestamp order, and each sorted parameter is normalized to make its value fall within a standardized range. Based on the federated learning framework, a cross-subject joint modeling task is initiated, and collaborative training is carried out among no less than five target subjects. Each subject independently trains the local parameters of the facility agriculture prediction model using its own standardized data. The model gradient update is calculated based on the trained local parameters, and the gradient update is uploaded to the central server. The central server performs weighted fusion calculation on the gradient updates uploaded by each subject based on the model training weight parameters to obtain the global facility agriculture prediction model. Obtain the crop variety identifier and facility type identifier of each subject. Based on these identifiers, match the historical records of each subject's inference using the local model before access, including the number of inference requests and the corresponding inference accuracy. Remove outliers from the historical inference records, calculate the average inference accuracy of each crop variety and facility type in the remaining records, and use it as the standard inference performance benchmark for each subject. Calculate the standard deviation of the inference accuracy to determine the fluctuation range. The global model is deployed to the local inference environment of each subject. The real-time inference accuracy of each subject in local inference verification of the global model is obtained. The real-time accuracy is compared with the standard inference performance benchmark of each subject. When the real-time accuracy of a subject is higher than its benchmark and exceeds the fluctuation range, the aggregation weight of the subject in the global model aggregation is increased; when it is lower than the benchmark and exceeds the fluctuation range, the aggregation weight of the subject is decreased; when it is within the benchmark and the fluctuation range, the current aggregation weight of the subject remains unchanged, completing the first dynamic configuration. Obtain the global model operation logs deployed locally by each subject, extract inference accuracy, inference execution rate, and inference response latency. Use these three indicators as evaluation dimensions, and combine them according to the preset weight ratio to obtain the comprehensive score of each subject. Combine the comprehensive score with the aggregate weight after the first dynamic configuration to determine the comprehensive contribution score of each subject. Calculate the average of the comprehensive contribution scores of all subjects, compare each subject's score with this average, and increase the aggregate weight if it is higher than the average, decrease the aggregate weight if it is lower than the average, and keep the aggregate weight unchanged if it is within the mean deviation range, thus completing the second dynamic configuration.

[0028] A smart management system for facility agriculture based on trusted data, the system comprising: Trusted access module, data fusion module, federated modeling module, feedback configuration module; The trusted access module is used to collect relevant information of the target facility agriculture entities and configure the initial scheduling weight for data upload of each target facility agriculture entity; The data fusion module is used to obtain the subject's historical data based on the subject's access time and to reliably store access process information; forming a standardized unified data view. The federated modeling module is used to prepare training data and jointly train multiple entities to generate a global model; determine the personalized performance reference of each entity based on its business attributes; after deploying the global model, combine the real-time inference performance with the reference to make the first adaptive adjustment of the configuration weights of each entity; The feedback configuration module is used to collect global model operation feedback, determine the contribution by combining the reasoning performance of each subject with the current configuration status, and make a second dynamic adjustment to the configuration of each subject. The output of the trusted access module is connected to the input of the data fusion module; The output of the data fusion module is connected to the input of the federated modeling module; The output of the federated modeling module is connected to the input of the feedback configuration module.

[0029] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for intelligent management of facility agriculture based on trusted data, characterized in that: The method includes the following steps: Collect relevant information about the target facility agriculture entities and form a standardized configuration dataset, and configure the initial scheduling weight for data upload of each target facility agriculture entity based on the relevant information; Historical data is retrieved based on the access trigger timestamp, and after timestamp verification, a historical operation data sequence is formed. The access process information is then written into the blockchain for evidence storage. Historical operational data is structured and analyzed, and agricultural public data is integrated to construct a historical multi-source heterogeneous data fusion set; this set is then normalized and stored in chronological order after being combined with the real-time fusion set to form a standardized data view. Extract parameter sequences from the standardized view and sort and normalize them; initiate a joint modeling task, each subject trains locally and uploads gradient updates, and merges them to generate a global model; match the historical inference records of each subject according to crop variety and facility type, calculate the benchmark and fluctuation range; deploy the global model for local inference verification, compare the real-time accuracy with the benchmark, and dynamically configure the aggregation weights of each subject for the first time accordingly; Obtain the global model operation logs deployed locally by each subject, extract inference accuracy, inference execution rate and inference response latency, and obtain a comprehensive score according to the preset proportion. Combine the aggregate weight after the first configuration to determine the comprehensive contribution score. Use the average score of all subjects as the benchmark, compare the score of each subject with the average, and perform the second dynamic configuration accordingly.

2. The intelligent management method for facility agriculture based on trusted data according to claim 1, characterized in that: Collect relevant information about the target facility agriculture entities and form a standardized configuration dataset. Then, configure the initial scheduling weights for data uploads from each target facility agriculture entity based on this information. Specific steps include: Identify the target facility agriculture entities, collect their unique identifiers, facility type parameters, planting scale parameters, and main crop varieties to form a standardized configuration dataset; The target facility agriculture entity is authenticated, and the standardized configuration dataset is stored after successful authentication. For each target facility agriculture entity that has passed identity authentication, select the corresponding preset facility complexity coefficient according to the facility type parameter of the target facility agriculture entity; The planting scale coefficient is determined based on the planting scale parameters of the target facility agriculture entity, whereby the planting scale coefficient is the planting area value of the target facility agriculture entity. Let S be the facility complexity coefficient of the i-th target facility agriculture entity. fac,i The planting scale coefficient is A i The product value P of the i-th target facility agriculture entity is calculated using the following formula. i The definition is as follows: P i =S fac,i ×A i ; Let n be the total number of target facility agriculture entities that have passed identity authentication. The total product value P is obtained by calculating the sum of the product values ​​of all target facility agriculture entities using the following formula. total The definition is as follows: P total =Σ j=1 n P j ; The initial scheduling weight W for the i-th target facility agriculture entity is calculated using the following formula. i The definition is as follows: W i =P i / P total ; Based on the initial scheduling weight W of each target facility agriculture entity i Configure the data upload channel priority for each target facility agriculture entity and record the access trigger timestamp for each target facility agriculture entity. The initial scheduling weight is positively correlated with the data upload channel priority.

3. The intelligent management method for facility agriculture based on trusted data according to claim 2, characterized in that: Historical data is retrieved based on the access trigger timestamp, and after timestamp verification, a historical operation data sequence is formed. The access process information is then written into the blockchain for evidence storage. The specific steps include: Based on the access trigger timestamp and unique identifier of each target facility agriculture entity, a targeted data retrieval request is initiated through the data aggregation interface to retrieve the original collected data records of each target facility agriculture entity within a preset time window before the access trigger timestamp; The original collected data records are verified by data timestamp, and data records with timestamps earlier than the access trigger timestamp are filtered out to form a historical operation data sequence for each target facility agriculture entity; The access process information of the target facility agriculture entity is written into the blockchain for evidence storage and traceability, generating a unique evidence hash value.

4. The intelligent management method for facility agriculture based on trusted data according to claim 3, characterized in that: Structured analysis of historical operational data and integration with agricultural public data are performed to construct a fusion set of historical multi-source heterogeneous data; After normalizing it with the real-time fused set, it is stored in time sequence to form a standardized data view. The specific steps include: The historical operational data sequence was structured and parsed to extract environmental parameters, energy consumption parameters, and crop yield records from the data records; Obtain agricultural public data for each target facility agriculture entity, and use the unique identifier of each target facility agriculture entity as the association index to associate the agricultural public data with the extracted environmental parameters, energy consumption parameters, agricultural operation records and crop yield records to form a complete data record for each target facility agriculture entity. Based on the unique identifiers of each target facility agriculture entity, the data is classified and collected to form a historical multi-source heterogeneous data fusion set; The heterogeneous data in the historical multi-source heterogeneous data fusion set and the real-time multi-source heterogeneous data fusion set are normalized according to a unified data format standard. The unified data format standard includes data field naming conventions, unified rules for measurement units, and time encoding format. The normalized data is then stored in chronological order to form a standardized data view.

5. The intelligent management method for facility agriculture based on trusted data according to claim 4, characterized in that: Extract parameter sequences from the standardized view and sort and normalize them; initiate a joint modeling task, each subject trains locally and uploads gradient updates, and merges them to generate a global model; match the historical inference records of each subject according to crop variety and facility type, and calculate the baseline and fluctuation range; Deploy the global model for local inference validation, compare the real-time accuracy with the benchmark, and dynamically configure the aggregation weights of each subject for the first time. The specific steps include: Extract the environmental parameter sequence, energy consumption parameter sequence, and crop yield record of each target facility agriculture entity from the standardized data view. Using the unique identifier of the target facility agriculture entity as the search index, sort the environmental parameters, energy consumption parameters, and crop yield records of each target facility agriculture entity in the order of timestamps. Normalize each parameter after sorting so that the parameter values ​​of each target facility agriculture entity are in the range of 0 to 1. Based on the federated learning framework, a cross-subject joint modeling task is initiated, and collaborative training is carried out among no fewer than 5 target facility agriculture subjects. Each target facility agriculture subject independently trains the local parameters of the facility agriculture prediction model using its own standardized data, calculates the model gradient update based on the trained local parameters, and uploads the model gradient update to the central server. The central server performs a weighted fusion operation on the model gradient updates uploaded by each target facility agriculture subject based on the model training weight parameters to obtain a global facility agriculture prediction model. Obtain the crop variety identifier and facility type identifier of each target facility agriculture entity in the standardized data view. Match the historical inference records of each target facility agriculture entity using the local facility agriculture prediction model before access based on the crop variety identifier and facility type identifier. The historical inference records include the number of inference requests and the corresponding inference accuracy of each target facility agriculture entity in the historical time period before access. Remove outliers in the historical inference records, calculate the average inference accuracy of each crop variety and facility type in the remaining historical inference records as the standard inference performance benchmark for each target facility agriculture entity, and calculate the standard deviation of the inference accuracy to determine the fluctuation range. The global facility agriculture prediction model is deployed to the local inference environment of each target facility agriculture entity. The real-time inference accuracy of each target facility agriculture entity in local inference verification of the global facility agriculture prediction model is obtained. The real-time inference accuracy is compared with the standard inference performance benchmark of each target facility agriculture entity. Based on the comparison results, the model aggregation weight of each target facility agriculture entity is dynamically configured for the first time.

6. The intelligent management method for facility agriculture based on trusted data according to claim 5, characterized in that: The model aggregation weights for each target facility agriculture entity are dynamically configured for the first time. The specific steps include: When the real-time inference accuracy of a target facility agriculture entity is higher than its standard inference performance benchmark and exceeds the fluctuation range, the aggregation weight of the target facility agriculture entity in the global model aggregation is increased. When the real-time inference accuracy of a target facility agriculture entity is lower than its standard inference performance benchmark and exceeds the fluctuation range, the aggregation weight of the target facility agriculture entity in the global model aggregation is reduced. When the real-time inference accuracy of a target facility agriculture entity is within its standard inference performance benchmark and fluctuation range, the current aggregation weight of the target facility agriculture entity remains unchanged, and the first dynamic configuration is completed.

7. The intelligent management method for facility agriculture based on trusted data according to claim 5, characterized in that: Obtain the global model operation logs deployed locally by each subject, extract inference accuracy, inference execution rate and inference response latency, and obtain a comprehensive score according to the preset proportion. Combine the aggregate weight after the first configuration to determine the comprehensive contribution score. Using the average score of all subjects as a benchmark, the scores of each subject are compared with the average, and a second dynamic configuration is performed accordingly. The steps include: Obtain the operation logs of the global facility agriculture prediction model deployed locally by each target facility agriculture entity, and extract the inference accuracy, inference execution rate and inference response latency recorded in the operation logs; Based on the reasoning accuracy, reasoning execution rate, and reasoning response latency of each target facility agriculture entity, the reasoning accuracy, reasoning execution rate, and reasoning response latency are used as evaluation dimensions, and a weighted sum is performed according to a preset weight ratio to obtain the comprehensive score of each target facility agriculture entity; the comprehensive score is multiplied by the aggregate weight of each target facility agriculture entity after the first dynamic configuration to calculate the comprehensive contribution score of each target facility agriculture entity. Calculate the average of the comprehensive contribution scores of all target facility agriculture entities as the average comprehensive contribution score; compare the comprehensive contribution score of each target facility agriculture entity with the average comprehensive contribution score, and perform a second dynamic configuration of the aggregation weight of each target facility agriculture entity based on the comparison results.

8. The intelligent management method for facility agriculture based on trusted data according to claim 7, characterized in that: The aggregation weights of each target facility agriculture entity are dynamically configured a second time, and the specific process is as follows: When the comprehensive contribution score of a target facility agriculture entity is higher than the average comprehensive contribution score, the aggregation weight of that target facility agriculture entity in the global model aggregation is increased. When the comprehensive contribution score of a target facility agriculture entity is lower than the average comprehensive contribution score, the aggregation weight of that target facility agriculture entity in the global model aggregation is reduced. When the comprehensive contribution score of a target facility agriculture entity is within the deviation range of the average comprehensive contribution score, the aggregation weight of the target facility agriculture entity in the global model aggregation remains unchanged, and the second dynamic configuration is completed.

9. A smart management system for facility agriculture based on trusted data, applied to the smart management method for facility agriculture based on trusted data as described in any one of claims 1-8, characterized in that: The system includes: Trusted access module, data fusion module, federated modeling module, feedback configuration module; The trusted access module is used to collect relevant information of the target facility agriculture entities and configure the initial scheduling weight for data upload of each target facility agriculture entity. The data fusion module is used to obtain the subject's historical data based on the subject's access time and to reliably store access process information; forming a standardized unified data view. The federated modeling module is used to prepare training data and jointly train multiple entities to generate a global model; determine the personalized performance reference of each entity based on its business attributes; after deploying the global model, combine the real-time inference performance with the reference to make the first adaptive adjustment of the configuration weights of each entity; The feedback configuration module is used to collect global model operation feedback, determine the contribution by combining the reasoning performance of each subject with the current configuration status, and make a second dynamic adjustment to the configuration of each subject. The output of the trusted access module is connected to the input of the data fusion module; The output of the data fusion module is connected to the input of the federated modeling module; The output of the federated modeling module is connected to the input of the feedback configuration module.

Images