Honey production line digital twin management and control method and system

Abstract

The invention relates to the technical field of digital control systems, in particular to a digital twin management and control method and system for a honey production line. The method comprises the following steps: deploying an edge unit integrated with a multi-modal sensor at a beehive end to realize multi-source acquisition and end-side semantic preprocessing of ecological and physical information of a swarm, and generating a bee state abstract packet with confidence; a multi-scale bee colony state matrix is constructed, behavior pattern recognition and anomaly detection are carried out, and then dynamic digital twinborn mapping is established; performing multi-objective optimization on the honey collection rhythm, beehive management and environment regulation and control through twin simulation, generating a production strategy, and issuing and executing the production strategy; meanwhile, real-time monitoring and intelligent grading packaging are conducted on honey quality based on production data and twinborn prediction, and finally a production strategy and a twinborn model are continuously optimized through a closed-loop feedback mechanism. According to the invention, end-to-end intelligent, controllable and efficient management of the whole process from bee colony state sensing to honey production is realized.

Description

Technical Field

[0001] This invention relates to the field of digital control system technology, specifically to a digital twin control method and system for a honey production line. Background Technology

[0002] Traditional honey production and management have long relied on human experience to judge the state of bee colonies and the timing of honey harvesting. With the development of technologies such as the Internet of Things and edge computing, the agricultural sector is evolving towards refinement and data-driven approaches, and the industry's demand for achieving full-process perception and intelligent management of bee colony ecology is growing.

[0003] Chinese invention patent application CN118569570A discloses a machine learning-based digital AI beekeeping management system and method. The system aims to improve the precision management and production efficiency of beekeeping through intelligent means. The system includes a closed-loop management process, from hardware installation and configuration to data analysis and decision support. Initial setup involves deploying intelligent beehives integrated with weight sensors, anti-theft alarms, far-infrared photoelectric sensors, and meteorological monitoring equipment, ensuring seamless data integration with the cloud. The system collects data around the clock, covering beehive weight changes, bee movement, and meteorological parameters. Based on the massive amounts of collected data, machine learning algorithms are used for in-depth analysis to assess the specific impact of environmental factors on honey production. By constructing predictive models, the system can accurately estimate honey production and guide intelligent adjustment of the beehive environment to maintain optimal bee colony condition. The system can also issue timely health management warnings to assist beekeepers in taking early intervention measures.

[0004] Existing technologies have made some progress in environmental monitoring. How to deeply integrate multi-source sensor data, construct dynamic models of bee colonies, and link them with the production process to achieve closed-loop management from state perception to decision execution has become an important direction for the technological upgrading of the bee industry under the background of smart agriculture. At the same time, realizing real-time evaluation and feedback of honey quality during the production process is also a key link to improve product standardization and value. Therefore, exploring a comprehensive management and control method for honey production lines that integrates advanced sensing, digital twin, and intelligent decision-making technologies is of great significance. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing a digital twin control method and system for honey production lines.

[0006] The technical solution of this invention: a digital twin control method for a honey production line, comprising the following specific implementation steps: S1. By deploying an edge unit integrating multimodal sensors at the beehive end, multi-source acquisition of bee colony ecological and physical information and edge-side semantic preprocessing are realized to generate a bee state summary package with confidence. S2. Construct a multi-scale bee colony state matrix based on bee state summary packages, perform behavior pattern recognition and anomaly detection, and establish a dynamic digital twin mapping; S3. Through digital twin simulation, multi-objective optimization of honey harvesting rhythm, hive management and environmental control is carried out to generate production strategies and send them to the end for execution; S4. Based on production execution data and digital twin predictions, real-time quality monitoring and intelligent grading and packaging of honey are carried out, and production strategies and digital twin models are continuously optimized through a closed-loop feedback mechanism.

[0007] Preferably, multi-source acquisition and edge semantic preprocessing specifically include: Weighing sensors, microclimate sensors, acoustic sensors, vibration sensors, and near-infrared sensors are deployed in each beehive; A regular inspection layer and an event perception layer are set up for hierarchical sampling. During the active period of flower source, a short-term collaborative group is formed by low-energy broadcast handshake between the edge units of neighboring boxes to achieve group-level synchronous sampling. The signals from each modal sensor are filtered, principal component extracted, and cross-modal consistency verified. The original multimodal data is transformed into semantic feature vectors and a confidence label is attached to each feature. Sparse dictionary encoding and privacy-aware hashing are performed on semantic feature vectors to generate bee-state digest packets, and a high-priority flag is set for the digest packets set by the event window to achieve low-latency reporting.

[0008] Preferably, the edge-side semantic preprocessing also includes performing lightweight adaptive learning on the edge units; Lightweight adaptive learning includes saving high-quality samples locally and generating reporting requests through model residual triggers; Lightweight adaptive learning also includes short-term online fine-tuning of the lightweight model using limited cached samples; Lightweight adaptive learning also includes periodically swapping summary samples with neighboring edge units to perform consistency checks and correct sensor baseline drift.

[0009] Preferably, constructing a multi-scale bee colony state matrix includes: The bee state summary package is used to construct a multi-scale state matrix according to time series, event priority, individual bee box and group window; Symmetrical repetition increment, ingress and egress peak rate, and vibrational main peak energy characteristics are fused in a spatial scale using neighboring boxes to form a preliminary mapping of bee colony behavior, and confidence level is used to perform soft weight correction on low consistency observations. By combining microclimate and near-infrared spectral data, the environment and bee colony behavior are coupled and modeled to obtain the state vector of each hive under environmental driving.

[0010] Preferably, behavioral pattern recognition and anomaly detection include: Peak activity patterns, diurnal honey-collecting rhythms, weight gain fluctuation patterns, and acoustic vibration characteristic patterns of bee colonies were extracted from the multi-scale state matrix. End-side confidence is introduced as pattern weight, and a bee colony behavior pattern library is generated through clustering and short-time dynamic recognition algorithms. In the behavior pattern library, potential abnormal patterns such as honey collection anomalies, bee colony disturbances, or microclimate maladaptation are marked, and group synchronization events and abnormal windows are specially identified.

[0011] Preferably, establishing a dynamic digital twin mapping specifically includes: The identified behavioral patterns are mapped to a virtual beehive model, and digital twin state nodes are generated through a time-space grid and behavioral pattern labels. Each node contains beehive state, environmental conditions, behavioral patterns and event priority flags. A lightweight graph structure predictor is used to predict future bee colony behavior based on a multi-scale state matrix and historical pattern sequences. A simulation of beehive operation strategies was conducted based on digital twin mapping, and the simulation results were optimized with the objectives of honey collection efficiency, bee colony health, and energy consumption to generate feasible operational suggestions.

[0012] Preferably, generating a production strategy and sending it to the endpoint for execution includes: Based on the state nodes and behavior predictions of digital twins, the beehive dynamics are divided into high-activity, high-risk, and low-activity categories; Prioritize honey harvesting and export scheduling for highly active beehives, implement microclimate intervention and honey source replenishment measures for low-activity beehives, and increase early warning inspections and environmental control for high-risk beehives; Generate a priority-sorted list of beehive operation strategies and distribute it to the beehive executor.

[0013] Preferably, step S3 further includes production control through a closed-loop feedback mechanism: The execution of the strategy is monitored in the edge unit and the beehive actuator, and compared with the prediction of the digital twin. Immediately identify and trigger adjustment strategies for execution deviations or abnormal situations; All deviations and adjustments are recorded and sent back to the digital twin system for real-time correction of model parameters and behavioral pattern weights.

[0014] Preferably, real-time quality monitoring and intelligent grading and packaging of honey include: The honey moisture, sugar content, pH value, enzyme activity and color distribution of each beehive were collected, and end-side preprocessing and feature extraction were performed in combination with trace spectral information to generate structured feature vectors. The feature vectors are input into a multi-level quality analysis model for classification, including high-quality, standard, and low-grade categories, and potential abnormal batches are identified. Digital twin prediction and edge confidence information are introduced to dynamically correct the level determination; Packaging plans are automatically generated based on the classification results and batch attributes, including planning the packaging type, label information, and storage conditions.

[0015] The technical solution of this invention: A digital twin control system for a honey production line, used to execute the aforementioned digital twin control method for a honey production line, comprising: The bee colony status acquisition and digital twin mapping module is used to collect multi-source data through edge units deployed at the beehive end and perform edge-side semantic preprocessing to generate bee state summary packages with confidence, and then construct digital twin mapping; The hive behavior analysis and production strategy generation module is used to construct a multi-scale hive state matrix based on digital twin mapping, perform behavior pattern recognition and anomaly detection, generate a hive behavior pattern library, and output production control suggestions. The intelligent production control and closed-loop optimization module is used to realize hierarchical control and dynamic resource scheduling of beehive operation strategy, and to monitor the execution of strategy through end-side actuators and form closed-loop feedback optimization. The honey quality monitoring and intelligent grading and packaging module is used to perform real-time multimodal quality monitoring and feature extraction of honey, combine digital twin prediction for intelligent grading and anomaly identification, and generate packaging plans and batch management strategies.

[0016] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: This invention designs a digital twin control method and system for a honey production line. By introducing edge computing and semantic preprocessing into the multi-source data acquisition layer, it achieves a high-confidence structured representation of bee colony ecological information, improving data quality and reliability from the source, while significantly reducing system communication overhead and end-side energy consumption. A multi-scale digital twin model built based on high-quality data can accurately map and predict bee colony behavior and health status, enabling early identification and soft intervention of risks such as abnormal honey harvesting and bee colony disturbances, greatly improving the predictability and controllability of the production process. Furthermore, by deeply integrating digital twins with production strategy simulation... This invention integrates various technologies to form an intelligent control loop, encompassing state perception, strategy optimization, and closed-loop execution. It achieves precise adaptation of hive hierarchical management, dynamic resource scheduling, and production rhythm, thereby optimizing honey yield and production efficiency while ensuring bee colony welfare. By connecting quality monitoring, intelligent grading, and production control, a quality traceability and feedback optimization mechanism is established throughout the entire process, effectively improving the consistency and traceability of honey product quality. Overall, this invention constructs an adaptive and iterative intelligent production system, providing effective technical support for the refined, intelligent, and sustainable development of the honey industry. Attached Figure Description

[0017] Figure 1 This is a flowchart of a digital twin control method for a honey production line proposed in this invention; Figure 2 This is a system architecture diagram of a digital twin control system for a honey production line proposed in this invention. Detailed Implementation

[0018] Example 1, as Figure 1 As shown, the present invention proposes a digital twin control method for a honey production line, which includes the following specific implementation steps: S1. By deploying multimodal edge units at the beehive end, multi-source acquisition of beehive ecological and physical information, edge-side semantic preprocessing, compression encoding, event-priority reporting, and lightweight adaptive learning are achieved. This provides high-quality, confidence-based structured observations for subsequent bee colony modeling and supply twins, realizing a digital twin acquisition layer with data reliability, low energy consumption, and edge-side adaptability. The specific implementation process is as follows: S11. Weighing, microclimate, acoustic, vibration, and NIR sensors are deployed in each hive. Sampling is performed in a stratified manner, combining conventional and event-driven methods, with synchronous sampling of neighboring hives within a short time window. This captures ecological behavior while optimizing energy consumption, achieving event-aware multimodal data acquisition. Specifically: Each beehive is equipped with an integrated edge unit (BEU), and the hardware includes, but is not limited to: a high-resolution weighing module (continuous or periodic snapshots), microclimate sensors (temperature, relative humidity, CO2, light intensity), an array of miniature microphones (sampling vibration / wing sound spectrum), a near-infrared (NIR) microspectrometer, and a short-range acceleration / vibration sensor; the BEU also includes a low-power MCU and an optional NPU for local initial inference. Set up a collaborative sampling strategy: Different sensors are divided into a regular inspection layer (snapshot every 10 minutes) and an event perception layer (high-frequency sampling during anomalies or window periods) according to their functions and energy consumption. For example: weighing and temperature and humidity are the routine inspection layer, and the microphone array and NIR are the event sensing layer (short-term up-frequency sampling when there is a sudden increase in weight, peak honey collection day and night, or abnormal vibration spectrum). A group-level synchronization mechanism is introduced, in which several adjacent BEUs form a short-term cooperative group during the active period of flower source through low-energy broadcast handshake. Within this window, the sampling rate is synchronously increased and the timestamps are mutually calibrated to capture the spatial cooperative pattern of the bee colony. This cooperative pattern is only enabled in the short window to save energy. S12. Filter, extract principal components, perform cross-modal consistency verification, and semanticize the features of each modal signal. Transform the original multimodal data into semantic feature vectors and attach confidence labels to ensure that the uploaded features are reliable and contain uncertainty, providing robust input for subsequent modeling. Specifically: Different modes are preprocessed in layers: Weighing is performed by calculating the difference using a short-time sliding window and eliminating sudden measurement noise. Microclimate data were filtered using an adaptive Kalman filter to eliminate heat transfer interference. The acoustic / vibration signal is subjected to short-time Fourier transform and envelope analysis, and then the "behavioral frequency band" is extracted using the sparse basis representation of the spectral envelope. NIR is used for small-window principal component extraction and optical path drift correction (micro-temperature compensation is performed using temperature readings). Local spatiotemporal consistency verification is implemented by combining the phase / amplitude difference of the array acoustics with the weighing waveform to distinguish between natural vibrations within the honeycomb and external mechanical interference (such as wind noise). If the multimodal data is inconsistent, it should be labeled as a "low confidence observation" and soft-weighted. An interpretable feature package is introduced, and a set of semantic features (e.g., peak ingress and egress rates, mass growth rate slope, vibration main peak energy, NIR principal component distance) is output for each time window and a confidence label is attached (calculated by sensor consistency and historical model residuals). S13. Perform sparse dictionary encoding and privacy-aware hashing on the semantic feature vector to generate a beehive digest packet. Set a high-priority flag in the event window to achieve compressed transmission, privacy protection, and low-latency event-driven reporting. Specifically: Design a sparse coding and compressed packet structure adapted to beekeeping scenarios: At the BEU end, semantic feature vectors are dictionary-encoded (a small number of basis vectors represent typical behavioral patterns), and this encoding and confidence are packaged together into a unified message called "bee state summary packet". A lightweight privacy hashing mechanism is introduced to generate local hash signatures (non-reversible) for flower source fingerprints or sensitive features that may expose the location, and only the signature is uploaded instead of the original vector. The hashing adopts parameterizable local awareness hashing to preserve similarity information. For event windows (triggered by step S11), a high-priority flag is set in the header of the summary packet, and a brief "time-intensity" report is attached to support the real-time alarm channel; regular packets follow low-frequency uploading to optimize energy consumption; if a synchronization event occurs in the linked group, the summary packet will contain the group ID and synchronization identifier, prompting the center to perform group pattern analysis. S14. Perform residual-driven sample caching, short-term fine-tuning, and neighborhood consistency calibration in the edge units to achieve end-side model adaptation, long-term sensor reliability, and autonomous operation under communication interruptions or new scenarios. Simultaneously, provide high-value samples for the periodic retraining of the central model. Specifically: Implement two types of lightweight learning mechanisms at the BEU level: Model residual trigger: When the local inference residuals continue to deviate from the distillation model predictions issued by the center, the BEU will save high-quality samples locally and generate a reporting request to prompt the center model to be updated or fine-tuned during migration. Short-term online fine-tuning: Uses limited cached samples to make step-level fine-tuning of lightweight models to adapt to day-night or seasonal changes (only performed during periods of low energy consumption or when there is sufficient power). Regularly (e.g., during daily low-load periods), the BEU performs self-tests and time synchronization, exchanges a small number of summary samples with neighboring BEUs for consistency checks and corrects sensor baseline drift (e.g., zero-point weighing or NIR optical path drift) to reduce long-term system errors; the self-test results and necessary calibration parameters are uploaded to the center in encrypted summary format. To reduce communication burden and ensure system robustness, BEU supports a rollback strategy: when communication with the center is unavailable, BEU continues event determination and local action suggestions (such as short-term increase in sampling, triggering internal logs) using local rule sets (the latest version issued by the center), and performs batch reporting and model alignment after communication is restored.

[0019] S2. Based on the high-confidence bee state summary package provided in step S1, a multi-scale bee colony state matrix is ​​constructed to identify behavioral patterns and abnormal events, generate a digital twin mapping, and optimize production strategies through twin simulation. This achieves bee colony state visualization, short-term prediction, and end-to-end production decision-making closed loop, improving the intelligence and controllability of honey production management. The specific implementation process is as follows: S21. Construct a multi-scale state matrix from the bee state summary package according to time, event, and hive population. Integrate weighing, vibration, microclimate, and NIR features, and use confidence-weighted calculations to generate a dynamic mapping between single hives and the colony, reflecting the real-time state of the bee colony under environmental driving conditions. Specifically: The bee state summary package uploaded in step S1 is sorted according to time series and event priority, and a multi-scale state matrix is ​​constructed according to beehive, population window and region level; Features such as symmetric repetition increment, ingress and egress rate, and vibrational main peak energy are fused by neighboring boxes on a spatial scale to form a preliminary mapping of bee colony behavior, and the confidence level is used to perform soft weight correction on low consistency observations. By combining microclimate and NIR spectral data, the environment and bee colony behavior are coupled and modeled to obtain the state vector of each hive under environmental driving conditions; S22. Extract bee colony activity features from the multi-scale state matrix, generate a behavior pattern library through clustering and short-term dynamic recognition, and label abnormal patterns using confidence scores to achieve soft identification and high-sensitivity detection of honey harvesting anomalies, bee colony disturbances, and microclimate adaptation issues. Specifically: Behavioral pattern features, such as peak activity of bee colonies, diurnal honey collection rhythm, weight growth fluctuation pattern and acoustic vibration feature pattern, are extracted within a time-space window using a multi-scale state matrix. End-side confidence is introduced as a pattern weight, and a bee colony behavior pattern library is generated through clustering and short-term dynamic recognition algorithms, and potential abnormal patterns (such as abnormal honey collection, bee colony disturbance or microclimate maladaptation) are labeled. Group synchronization events and abnormal windows are specially marked so that subsequent digital twin and prediction modules can prioritize the analysis of high-risk time periods; S23. Map the identified behavioral patterns to a virtual beehive model to generate dynamic digital twin state nodes, embedding behavioral patterns, environmental conditions, and event priorities to achieve real-time visualization and short-term prediction. Mapping bias is corrected through a confidence-based closed-loop process. Specifically: The behavioral patterns identified in step S22 are mapped to the virtual beehive model. Digital twin state nodes are generated through time-space grid and behavioral pattern labels. Each node contains beehive state, environmental conditions, behavioral patterns and event priority flags. Establish short-term predictive capabilities for each node: Use a lightweight graph structure predictor to predict future bee colony behavior, such as weight increment, honey collection peak, and potential anomaly probability, based on multi-scale state matrices and historical pattern sequences. The dynamic twin and the actual bee colony status are aligned through a confidence-corrected closed loop. When the deviation between the actual observation and the prediction exceeds the threshold, a feedback signal is generated to provide a reference for edge data acquisition or production control. S24. Simulate honey harvesting, temperature regulation, and hive management strategies based on digital twins. Generate feasible operational suggestions through multi-objective optimization and feed back key strategy summaries to the end-user or management system to achieve closed-loop optimization of real-time production control and bee colony health. Specifically: Based on digital twin mapping, we simulate beehive operation strategies, such as nectar source scheduling, temperature regulation between beehives, or optimization of honey harvesting rhythm, and simulate the impact of potential anomalies on yield. Multi-objective optimization (honey harvesting efficiency, bee colony health, energy consumption) is performed on the simulation results to generate feasible operation suggestions, and key operation signals or strategy summaries are fed back to the edge unit or production management system. By combining the short-term prediction bias and confidence level from step S23, the operation strategy is dynamically adjusted to ensure that the digital twin is continuously corrected to match the actual bee colony state.

[0020] S3. Based on the digital twin mapping and bee colony behavior prediction generated in step S2, intelligent regulation and resource optimization are implemented in the honey production process to achieve hierarchical management of beehives, regulation of honey harvesting rhythm, optimization of temperature and humidity environment, and intervention in bee colony health. Through closed-loop execution monitoring and strategy iteration, an end-to-end adaptive, efficient, and controllable honey production management system is constructed. The specific implementation process is as follows: S31. Based on the digital twin state nodes and behavior predictions, beehives are divided into high-activity, high-risk, and low-activity categories. Differentiated honey harvesting cycles, environmental control, and bee colony intervention strategies are designed for different categories to achieve personalized beehive management and efficient resource utilization. Specifically: Using the digital twin mapping data from step S2, the status nodes, predicted behaviors, and historical performance of each beehive are comprehensively evaluated. Combined with environmental data, bee colony activity, and potential abnormal risks, beehives are dynamically classified into three categories: high-activity, high-risk, and low-activity. Differentiated operational strategies were designed for beehives of different grades: honey harvesting and export scheduling were prioritized for highly active beehives; microclimate intervention, nectar source supplementation, or colony incentive measures were adopted for low-activity beehives; and early warning checks and environmental control were added before operations were carried out on high-risk beehives to ensure colony safety. A list of hive operation strategies is generated, sorted by priority, and distributed to hive actuators through edge units to achieve personalized hive operations rather than uniform strategies; the strategy list can be updated in real time to ensure flexible response to changes in bee colony behavior and environmental fluctuations. Dynamically generate a list of beehive operation strategies and allocate terminal execution resources by prioritizing them to achieve differentiated operations at the beehive level rather than uniform processing, thereby improving resource utilization efficiency. S32. Map the hive operation strategy to resources such as honey collection time, temperature and humidity control, flower source allocation, and auxiliary intervention. Dynamically optimize resource allocation through a multi-dimensional scheduling algorithm, taking into account honey collection efficiency, bee colony health, and energy consumption control, to achieve real-time optimization of the production process. Specifically: Mapping hive operation strategies to operable resources, including honey harvesting time windows, hive temperature and humidity control equipment, flower source allocation, auxiliary feeding, and bee colony intervention measures; Based on the prediction of short-term behavior, event priority, and bee colony health status using digital twins, a dynamic resource scheduling scheme is designed: honey harvesting equipment is concentrated during peak honey harvesting periods, environmental monitoring frequency is increased in high-risk beehives, and local incentives are provided in low-activity beehives to improve overall production efficiency. Simultaneously considering microclimate and environmental data, and combining real-time temperature, humidity, ventilation, and light control with strategy execution, environmental optimization is achieved to ensure a balance between bee colony health and honey collection efficiency. The scheduling scheme is automatically optimized through edge units on the end side, and can adjust priorities under resource constraints to ensure that critical hives receive sufficient resources while reducing energy consumption; S33. Monitor strategy execution on the edge unit and beehive actuator, compare with digital twin predictions, provide immediate feedback on deviations or anomalies, trigger strategy adjustments and transmit the data back to the system for correction of the twin model and strategy library, forming real-time closed-loop control, specifically: The scheduling strategy is distributed to the edge unit and beehive actuator, and digital twin prediction is used to monitor honey harvesting progress, bee colony activity and environmental conditions in real time. The system can identify deviations or abnormal situations in real time, such as honey collection being lower than predicted, abnormal bee colony vibration, or temperature and humidity deviating from the safe zone. The system will automatically trigger adjustment strategies or increase local intervention. All deviations and adjustments are recorded and sent back to the digital twin system to correct model parameters, behavioral pattern weights, and hive status nodes in real time, ensuring that the digital twin mapping is consistent with the actual hive status. Through a closed-loop feedback mechanism, the execution results will influence the next round of resource scheduling and strategy generation, achieving end-to-end adaptation, dynamic control and risk prevention; S34. Multi-objective optimization is performed based on comprehensive hive feedback, twin prediction bias, bee colony health, and energy consumption data. This iteratively updates honey harvesting rhythms, hive control, and resource allocation strategies to provide adaptive optimization solutions for the next production cycle, achieving a continuous intelligent production closed loop. Specifically: By combining the performance feedback from each beehive, the prediction bias of the digital twin, the bee colony health indicators, energy consumption data and production output, a global optimization assessment of the entire honey production line is conducted. Using a multi-objective optimization algorithm, the honey harvesting rhythm, hive control scheme, and resource allocation strategy are adjusted to achieve the optimal balance between production efficiency, bee colony health, and energy consumption. Update the digital twin model and beehive strategy library to achieve strategy iteration and learning: the execution data of each production cycle is used to optimize the prediction model parameters and behavior pattern library, so that the system can maintain its adaptive ability in different seasons, flower source changes and bee colony status. The optimization results are sent to step S31 to form a closed loop, realizing end-to-end dynamic and intelligent production management.

[0021] S4. Based on the honey production data and digital twin mapping results after step S3, real-time quality monitoring, multimodal feature extraction, intelligent grading, and anomaly identification are performed on the honey. Combined with digital twin prediction and batch management, intelligent packaging, grading strategy optimization, and production closed-loop feedback are achieved, improving product quality controllability, packaging accuracy, and production efficiency. Simultaneously, data support is provided for the next round of production strategy iteration. The specific implementation process is as follows: S41. Collect honey moisture, sugar content, pH value, enzyme activity, and color distribution from each beehive. Combine this with trace spectral information for end-side preprocessing and feature extraction, generating structured feature vectors and mapping them to digital twin nodes. This enables dynamic visualization of honey quality, providing a data foundation for subsequent grading. Specifically: Using the production execution data and digital twin mapping predictions fed back from step S3, multimodal quality monitoring is performed on the honey collected from each beehive, including moisture content, sugar content, pH value, enzyme activity and color distribution. At the same time, trace spectroscopy (such as near-infrared or visible spectroscopy) is combined to detect characteristic indicators. The collected quality data undergoes preliminary preprocessing and feature extraction at the end-side, transforming complex physicochemical signals into structured feature vectors and labeling them with confidence levels and beehive origin information, providing reliable input for subsequent grading analysis; Data is stored in digital twin model nodes according to time series and honey harvesting batches to achieve dynamic mapping of honey quality, so that the quality status of each production batch can be visualized in real time and associated with bee colony behavior, honey harvesting rhythm and environmental conditions. S42. Utilize a multi-level analysis model to classify feature vectors into quality grades, including high-quality, standard, and low-quality, while identifying potentially abnormal batches. Combine digital twin prediction and dynamic confidence level correction to achieve data-driven intelligent grading and anomaly monitoring, and provide feedback for production optimization. Specifically: The feature vector generated in step S41 is input into the multi-level quality analysis model to classify honey, including but not limited to high-quality, standard, and low-grade categories, while identifying potentially abnormal batches, such as honey with excessive moisture, low sugar content, or abnormal color. By introducing digital twin prediction and edge confidence information, the grade determination is dynamically corrected, and different batches are assigned grade prediction probabilities, so that the grading results not only depend on static thresholds, but also combine production environment, bee colony behavior patterns and historical trends. Abnormal batches generate warning tags and trigger a backtracking mechanism to correlate and analyze the abnormal information with the honey harvesting strategy, hive health status and resource control records in step S3, providing feedback for production optimization. S43. Automatically generate packaging plans based on grading results and batch attributes, planning packaging types, labeling information, and storage conditions. Simultaneously, adjust packaging strategies using digital twin predictions to ensure consistency between the packaging process and the digital twin mapping, guaranteeing grading accuracy, batch traceability, and inventory optimization, thereby improving packaging efficiency and management precision. Specifically: Based on the grading results and batch attributes of step S42, a packaging plan is automatically generated, including packaging type, label information, batch sequence and storage conditions, while optimizing the packaging sequence to meet warehousing and logistics requirements. By using digital twin mapping and honey quality prediction, we can anticipate future grade trends and adjust packaging strategies accordingly. For example, high-end packaging can be planned in advance for premium honey, while lower-grade honey can be processed in batches to reduce waste and optimize inventory structure. Key parameters such as weight, density, and color are collected in real time during the packaging process, and closed-loop comparison and grading are used to ensure that the packaging process is consistent with the digital twin mapping, thereby improving packaging accuracy and batch traceability. S44. Based on comprehensive grading and judgment, abnormal batches, packaging, and batch execution data, a global quality assessment of the production process is conducted. Deviation information is fed back to the production control system in step S3 to adjust honey harvesting, beehive management, and resource scheduling strategies. Simultaneously, the digital twin model is updated to achieve iterative production strategy development and end-to-end closed-loop optimization, thereby improving honey quality and production efficiency. Specifically: By combining the classification judgment in step S42, abnormal batch information, and packaging and batch execution data in step S43, a global quality assessment of the production process is conducted, including honey harvesting strategy, hive health management, and resource utilization efficiency. Through closed-loop optimization, quality deviations and abnormal information are fed back to the production control system in step S3 to adjust the honey harvesting rhythm, hive environment control and resource allocation strategies to reduce the occurrence rate of low-grade honey and abnormal batches. The digital twin model is updated and iterated, incorporating the latest honey quality data, grading information, and feedback results into the model training to improve prediction accuracy and strategy adaptability. It outputs strategy optimization suggestions and operation instructions to provide data-driven intelligent control for the next production cycle, realizing end-to-end closed-loop quality optimization of honey production from honey harvesting to packaging.

[0022] Example 2, as Figure 2 As shown, the present invention proposes a digital twin control system for a honey production line, which is used to execute a digital twin control method for a honey production line proposed in Embodiment 1. The system includes: a bee colony status acquisition and digital twin mapping module, a bee colony behavior analysis and production strategy generation module, an intelligent production control and closed-loop optimization module, and a honey quality monitoring and intelligent grading and packaging module.

[0023] The bee colony status acquisition and digital twin mapping module is responsible for collecting multi-source data inside and outside the beehive, including bee colony activity vibration, weight change, temperature and humidity, flower source conditions and microclimate information. The module performs preliminary processing and summary extraction through the edge computing unit, and generates bee colony digital twin nodes from the processed high-confidence data to realize dynamic mapping of bee colony status and preliminary identification of behavior patterns, providing an accurate basis for subsequent analysis. The bee colony behavior analysis and production strategy generation module, based on digital twin mapping, constructs a multi-scale bee colony state matrix, performs behavior pattern recognition and anomaly detection, generates a beehive behavior pattern library, and outputs beehive operation strategies and production control suggestions based on bee colony health, activity and risk assessment. At the same time, it forms short-term forecasts and strategy priority ranking, providing decision-making basis for resource scheduling and production regulation. The intelligent production control and closed-loop optimization module enables hierarchical control of hive operation strategies, resource scheduling, and environmental optimization, including honey harvesting rhythm arrangement, temperature, humidity and light control, bee colony incentives, and risk intervention. Through end-side actuators and edge units, the module monitors the strategy execution in real time, feeding back deviations and anomalies to the digital twin system. Combined with global production indicators, the module performs closed-loop optimization, achieving strategy iteration and end-to-end adaptive control to ensure the dual optimization of production efficiency and bee colony health. The honey quality monitoring and intelligent grading and packaging module performs real-time multimodal quality monitoring and feature extraction on the collected honey, including moisture, sugar content, pH value, enzyme activity, and color distribution. It also combines digital twin prediction for grading and anomaly identification. Based on the grading results, it intelligently generates packaging plans and batch management strategies to achieve dynamic packaging, label planning, and inventory optimization. At the same time, it feeds back quality deviations to the intelligent production control and closed-loop optimization module to form a closed-loop optimization of production and quality, ensuring the quality and traceability of honey production.

[0024] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A digital twin control method for a honey production line, characterized in that, The specific implementation steps include the following: S1. By deploying an edge unit integrating multimodal sensors at the beehive end, multi-source acquisition of bee colony ecological and physical information and edge-side semantic preprocessing are realized to generate a bee state summary package with confidence. S2. Construct a multi-scale bee colony state matrix based on bee state summary packages, perform behavior pattern recognition and anomaly detection, and establish a dynamic digital twin mapping; S3. Through digital twin simulation, multi-objective optimization of honey harvesting rhythm, hive management and environmental control is carried out to generate production strategies and send them to the end for execution; S4. Based on production execution data and digital twin predictions, real-time quality monitoring and intelligent grading and packaging of honey are carried out, and production strategies and digital twin models are continuously optimized through a closed-loop feedback mechanism.

2. The digital twin control method for a honey production line according to claim 1, characterized in that, Multi-source acquisition and edge semantic preprocessing specifically include: Weighing sensors, microclimate sensors, acoustic sensors, vibration sensors, and near-infrared sensors are deployed in each beehive; A regular inspection layer and an event perception layer are set up for hierarchical sampling. During the active period of flower source, a short-term collaborative group is formed by low-energy broadcast handshake between the edge units of neighboring boxes to achieve group-level synchronous sampling. The signals from each modal sensor are filtered, principal component extracted, and cross-modal consistency verified. The original multimodal data is transformed into semantic feature vectors and a confidence label is attached to each feature. Sparse dictionary encoding and privacy-aware hashing are performed on semantic feature vectors to generate bee-state digest packets, and a high-priority flag is set for the digest packets set by the event window to achieve low-latency reporting.

3. The digital twin control method for a honey production line according to claim 2, characterized in that, Edge-side semantic preprocessing also includes performing lightweight adaptive learning on edge units; Lightweight adaptive learning includes saving high-quality samples locally and generating reporting requests through model residual triggers; Lightweight adaptive learning also includes short-term online fine-tuning of the lightweight model using limited cached samples; Lightweight adaptive learning also includes periodically swapping summary samples with neighboring edge units to perform consistency checks and correct sensor baseline drift.

4. The digital twin control method for a honey production line according to claim 3, characterized in that, Constructing a multi-scale bee colony state matrix includes: The bee state summary package is used to construct a multi-scale state matrix according to time series, event priority, individual bee box and group window; Symmetrical repetition increment, ingress / exit peak rate, and vibrational main peak energy characteristics are fused in a spatial scale using neighboring boxes to form a preliminary mapping of bee colony behavior, and confidence level is used to perform soft weight correction on low consistency observations. By combining microclimate and near-infrared spectral data, the environment and bee colony behavior are coupled and modeled to obtain the state vector of each hive under environmental driving.

5. The digital twin control method for a honey production line according to claim 4, characterized in that, Behavioral pattern recognition and anomaly detection include: Peak activity patterns, diurnal honey-collecting rhythms, weight gain fluctuation patterns, and acoustic vibration characteristic patterns of bee colonies were extracted from the multi-scale state matrix. End-side confidence is introduced as pattern weight, and a bee colony behavior pattern library is generated through clustering and short-time dynamic recognition algorithms. In the behavior pattern library, potential abnormal patterns such as honey collection anomalies, bee colony disturbances, or microclimate maladaptation are marked, and group synchronization events and abnormal windows are specially identified.

6. The digital twin control method for a honey production line according to claim 5, characterized in that, Establishing a dynamic digital twin mapping specifically includes: The identified behavioral patterns are mapped to a virtual beehive model, and digital twin state nodes are generated through a time-space grid and behavioral pattern labels. Each node contains beehive state, environmental conditions, behavioral patterns and event priority flags. A lightweight graph structure predictor is used to predict future bee colony behavior based on a multi-scale state matrix and historical pattern sequences. A simulation of beehive operation strategies was conducted based on digital twin mapping, and the simulation results were optimized with the objectives of honey collection efficiency, bee colony health, and energy consumption to generate feasible operational suggestions.

7. The digital twin control method for a honey production line according to claim 6, characterized in that, Generating production strategies and deploying them to the endpoints for execution includes: Based on the state nodes and behavior predictions of digital twins, the beehive dynamics are divided into high-activity, high-risk, and low-activity categories; Prioritize honey harvesting and export scheduling for highly active beehives, implement microclimate intervention and honey source replenishment measures for low-activity beehives, and increase early warning inspections and environmental control for high-risk beehives; Generate a priority-sorted list of beehive operation strategies and distribute it to the beehive executor.

8. The digital twin control method for a honey production line according to claim 7, characterized in that, Step S3 also includes production control through a closed-loop feedback mechanism: The execution of the strategy is monitored in the edge unit and the beehive actuator, and compared with the prediction of the digital twin. Immediately identify and trigger adjustment strategies for execution deviations or abnormal situations; All deviations and adjustments are recorded and sent back to the digital twin system for real-time correction of model parameters and behavioral pattern weights.

9. The digital twin control method for a honey production line according to claim 8, characterized in that, Real-time quality monitoring and intelligent grading and packaging of honey include: The honey moisture, sugar content, pH value, enzyme activity and color distribution of each beehive were collected, and end-side preprocessing and feature extraction were performed in combination with trace spectral information to generate structured feature vectors. The feature vectors are input into a multi-level quality analysis model for classification, including high-quality, standard, and low-grade categories, and potential abnormal batches are identified. Digital twin prediction and edge confidence information are introduced to dynamically correct the level determination; Packaging plans are automatically generated based on the classification results and batch attributes, including planning the packaging type, label information, and storage conditions.

10. A digital twin control system for a honey production line, used to execute the digital twin control method for a honey production line according to any one of claims 1 to 9, characterized in that, include: The bee colony status acquisition and digital twin mapping module is used to collect multi-source data through edge units deployed at the beehive end and perform edge-side semantic preprocessing to generate bee state summary packages with confidence, and then construct digital twin mapping; The hive behavior analysis and production strategy generation module is used to construct a multi-scale hive state matrix based on digital twin mapping, perform behavior pattern recognition and anomaly detection, generate a hive behavior pattern library, and output production control suggestions. The intelligent production control and closed-loop optimization module is used to realize hierarchical control and dynamic resource scheduling of beehive operation strategy, and to monitor the execution of strategy through end-side actuators and form closed-loop feedback optimization. The honey quality monitoring and intelligent grading and packaging module is used to perform real-time multimodal quality monitoring and feature extraction of honey, combine digital twin prediction for intelligent grading and anomaly identification, and generate packaging plans and batch management strategies.

Images