Whole-process digital twin management and control method and system for low-oxygen insecticidal treatment

Abstract

The application discloses a low-oxygen modified atmosphere insecticidal full-process digital twin management and control method and system, relates to the tobacco technical field, and provides a data base for full-process management and control by constructing a multi-source real-time time series data set, realizing unified collection of nitrogen production, pipe network, sealing and oxygen reduction link states; multi-source data fusion and noise filtering are realized through extended Kalman filtering, and the reliability of monitoring data is improved; a full-process digital twin is constructed, three-dimensional virtual mapping and visual management and control of storage space, modified atmosphere equipment and insect state are realized; multi-objective dynamic optimization of modified atmosphere parameters is realized through a deep deterministic policy gradient algorithm, and the blindness of empirical setting is eliminated; conversion and issuing of optimal strategies to equipment instructions are completed, precise execution of management and control strategies and linkage with a WMS system are realized; through execution deviation iteration and synchronization of the digital twin, a full-process closed-loop management and control and self-adaptive optimization mechanism is formed.

Description

Technical Field

[0001] This application relates to the field of tobacco technology, and in particular to a digital twin control method and system for the entire process of low-oxygen controlled insecticide. Background Technology

[0002] In the aging and storage of tobacco raw materials, pest control is a core aspect of ensuring raw material quality. Low-oxygen controlled nitrogen fumigation technology, due to its green and environmentally friendly nature and lack of chemical pollution, has become the mainstream technology in the tobacco industry to replace aluminum phosphide chemical fumigation. Large-scale tobacco logistics centers have large-scale tobacco leaf storage, multiple warehouse floors, and dense stacking, making the low-oxygen controlled nitrogen fumigation process complex. It involves multiple key aspects such as nitrogen generation equipment operation, pipeline transportation, tobacco stack sealing, and oxygen concentration control, requiring high operational precision, safety, and controllability.

[0003] Existing low-oxygen controlled insecticide methods have the following prominent drawbacks: Fragmented process control: Nitrogen production, pipeline network, sealing, oxygen reduction and other links are independent of each other, lack a unified control platform, and the status cannot be linked in real time. This can easily lead to problems such as uneven oxygen reduction, untimely nitrogen replenishment, and undetected sealing leaks, resulting in unstable pest control effect.

[0004] Lack of full-process visualization: The operational status and parameter changes of each link can only be viewed through decentralized instruments or manual records, which cannot be intuitively visualized and controlled, making it difficult to locate abnormal nodes and trace the process.

[0005] The monitoring is limited in scope and the early warning is delayed: most monitoring only focuses on oxygen concentration, ignoring key parameters such as insect infestation, temperature and humidity, and core temperature. There is no linkage early warning mechanism, which can easily lead to the spread of insect infestation, mold growth in tobacco stacks, and other hidden dangers. The early warning response is also delayed.

[0006] The control is highly unpredictable: there is no simulation and prediction capability, the operation parameters are set based on experience, the trial and error costs are high, the dynamic adjustment is impossible, and it is difficult to achieve precise control.

[0007] Disconnected from the warehouse management system: Pest control operations lack effective integration with the WMS system. Operation planning, process recording, and effect acceptance require manual intervention, resulting in low efficiency. Data cannot be interconnected, making it difficult to form a full-process traceability system. Summary of the Invention

[0008] The main purpose of this application is to provide a digital twin management method and system for the entire process of low-oxygen controlled insecticide, in order to solve the problems of fragmented process management, lack of full-process visualization, single monitoring dimension, delayed early warning, and strong blindness in existing low-oxygen controlled insecticide management methods.

[0009] To achieve the above objectives, this application provides the following technical solution: A digital twin management method for the entire process of low-oxygen controlled insecticide application, the method comprising: Step S1: Collect oxygen concentration, temperature, humidity, insect density, and controlled atmosphere equipment operating status in the storage space, and construct a multi-source real-time time series dataset. Step S2: Input the multi-source real-time time series dataset into the extended Kalman filter algorithm to perform multi-source data fusion and random noise filtering to generate calibrated multi-source time series data; Step S3: Obtain the digital twin basic model of the storage space, and input the calibrated multi-source time series data into the digital twin basic model to perform three-dimensional virtual mapping of the storage space, controlled atmosphere equipment, and insect infestation status, and obtain a full-process digital twin. Step S4: The full-process digital twin is connected to the deep deterministic strategy gradient algorithm to perform multi-objective optimization iteration of the air conditioning parameters to obtain the globally optimal air conditioning control strategy. Step S5: Convert the global optimal controlled atmosphere management strategy into standardized executable control commands for the controlled atmosphere equipment, send them to each controlled atmosphere execution device in the warehouse space through the industrial IoT gateway, and obtain the execution feedback data of each controlled atmosphere execution device. Step S6: Iterate the global optimal atmosphere control strategy based on the execution deviation between the execution feedback data and the global optimal atmosphere control strategy, and send all execution feedback data and the iterated control strategy back to the full-process digital twin to complete dynamic synchronization and full-process closed-loop control.

[0010] Beneficial effects of steps S1 to S6: This method forms a complete technical closed loop of digital twin control for the entire process of low-oxygen controlled pest control through steps S1 to S6, effectively overcoming the technical problems of fragmented process, lack of status linkage, lack of visualization, single monitoring dimension, delayed early warning, strong blind control, and disconnection from WMS system in the existing low-oxygen controlled pest control.

[0011] Step S1 involves constructing a multi-source real-time time-series dataset to achieve unified collection of status data for each stage of nitrogen production, pipeline network, sealing, and oxygen reduction, providing a data foundation for full-process control. Step S2 uses extended Kalman filtering to achieve multi-source data fusion and noise filtering, improving the reliability of monitoring data. Step S3 involves constructing a full-process digital twin to achieve three-dimensional virtual mapping and visualized control of storage space, controlled atmosphere equipment, and pest status. Step S4 uses a deep deterministic strategy gradient algorithm to achieve multi-objective dynamic optimization of controlled atmosphere parameters, eliminating the blindness of setting parameters based on experience. Step S5 completes the conversion and issuance of optimal strategies to equipment instructions, achieving precise execution of control strategies and linkage with the WMS system. Step S6 uses execution deviation iteration and synchronization with the digital twin to form a closed-loop control and adaptive optimization mechanism for the entire process.

[0012] As a further improvement to this application, step S1 involves collecting data on oxygen concentration, temperature, humidity, insect density, and controlled atmosphere equipment operation status within the storage space, and constructing a multi-source real-time time-series dataset, including: Step S1.1: Activate the multi-type data acquisition nodes deployed in the storage space for oxygen, temperature and humidity, insect infestation, and equipment status; Step S1.2: Extract the raw data signals from each data acquisition node through the real-time data acquisition channel at a fixed sampling frequency; Step S1.3: Perform analog-to-digital conversion and level standardization on all raw data signals to generate raw sensing data in a unified format; Step S1.4: Classify and collect all raw sensing data according to five dimensions: oxygen concentration, temperature, humidity, insect density, and equipment status. Step S1.5: Perform time-series alignment and data preprocessing on the raw sensor data after classification and aggregation to generate a multi-source real-time time-series dataset.

[0013] Beneficial effects of steps S1.1 to S1.5: This series of steps, through five stages—deploying multiple types of data acquisition nodes, real-time data retrieval, analog-to-digital conversion and standardization, dimensional classification and aggregation, and time-series alignment preprocessing—constructs a multi-source real-time time-series dataset covering oxygen concentration, temperature, humidity, insect density, and equipment status. This overcomes the shortcomings of existing methods that only monitor single dimensions and oxygen concentration, providing a unified data foundation for subsequent data fusion and digital twin mapping.

[0014] The process includes: step S1.1, starting and initializing multiple types of data acquisition nodes; step S1.2, real-time acquisition of raw data signals through a fixed sampling frequency; step S1.3, analog-to-digital conversion and level standardization to eliminate heterogeneous data format differences; step S1.4, classification and aggregation according to five dimensions to construct a structured data system; and step S1.5, time alignment and preprocessing to generate a multi-source real-time time series dataset that can be used for filtering and fusion.

[0015] As a further improvement to this application, step S2 involves inputting the multi-source real-time time-series dataset into an extended Kalman filter algorithm for multi-source data fusion and random noise filtering to generate calibrated multi-source time-series data, including: Step S2.1: Input the multi-source real-time time series dataset into the state prediction unit of the extended Kalman filter algorithm in the order of timestamps to calculate the prior state estimate of each dimension of the data; Step S2.2: Input the prior state estimates of each dimension of data into the covariance prediction unit of the extended Kalman filter algorithm to calculate the prior error covariance matrix; Step S2.3: Input the prior error covariance matrix into the gain calculation unit of the extended Kalman filter algorithm to calculate the optimal Kalman gain matrix; Step S2.4: Input the optimal Kalman gain matrix and the prior state estimates of each dimension of data into the state update unit to calculate the posterior state estimates; Step S2.5: The posterior state estimate is split and recombined according to the original data dimension to obtain calibrated multi-source time series data.

[0016] Beneficial effects of steps S2.1 to S2.5: This series of steps, through five stages—state prediction, covariance prediction, gain calculation, state update, and splitting and recombining—applies the extended Kalman filter algorithm to the fusion of multi-source time-series data. This effectively filters out random noise during the acquisition process, generates calibrated multi-source time-series data, significantly improves data reliability and subsequent digital twin mapping accuracy, and lays a data quality foundation for precise control throughout the entire process.

[0017] The process includes: step S2.1 calculating prior state estimates for each dimension of the data to achieve state prediction; step S2.2 calculating the prior error covariance matrix to quantify prediction uncertainty; step S2.3 calculating the optimal Kalman gain matrix to determine the correction weights of the observations on the state estimates; step S2.4 calculating the posterior state estimates to complete the state update; and step S2.5 splitting and reorganizing the original data according to its dimensions to obtain calibrated multi-source time series data.

[0018] As a further improvement to this application, step S3 involves obtaining a digital twin base model of the storage space and inputting the calibrated multi-source time-series data into the digital twin base model to perform a three-dimensional virtual mapping of the storage space, controlled atmosphere equipment, and pest situation, thereby obtaining a full-process digital twin, including: Step S3.1: Convert the calibrated multi-source time-series data to the unified spatial coordinate system of the digital twin basic model to obtain the coordinate-transformed spatial data; Step S3.2: Based on the digital twin basic model, establish the actual positional relationship between each sensor and the controlled atmosphere equipment in the warehouse space, and generate a spatial position mapping table; Step S3.3: Input the spatial data after coordinate transformation and the spatial location mapping table into the digital twin basic model, perform three-dimensional virtual mapping of the warehouse space, and generate the three-dimensional virtual mapping result of the warehouse space; Step S3.4: Input the controlled atmosphere equipment operating status data in the spatial data after coordinate transformation and the spatial position mapping table into the digital twin basic model, perform three-dimensional virtual mapping of the controlled atmosphere equipment status, and generate the three-dimensional virtual mapping result of the controlled atmosphere equipment status; Step S3.5: Input the insect infestation status data in the spatial data after coordinate transformation and the spatial location mapping table into the digital twin basic model, perform three-dimensional virtual mapping of insect infestation status, and generate three-dimensional virtual mapping result of insect infestation status; Step S3.6: Integrate the three-dimensional virtual mapping results of the storage space, the three-dimensional virtual mapping results of the controlled atmosphere equipment status, and the three-dimensional virtual mapping results of the insect infestation status to generate a full-process digital twin.

[0019] Beneficial effects of steps S3.1 to S3.6: This series of steps, through six stages—coordinate transformation, spatial location mapping, 3D mapping of storage space, mapping of controlled atmosphere equipment status, mapping of pest status, and integration of the entire process digital twin—maps calibrated multi-source time-series data to a basic digital twin model, constructing a full-process digital twin covering storage space, controlled atmosphere equipment, and pest status. This overcomes the shortcomings of existing methods, such as lack of full-process visualization, difficulty in locating abnormal nodes, and inconvenience in process traceability, achieving real-time synchronization and intuitive visual control between physical and virtual spaces.

[0020] The process includes: step S3.1 completing the coordinate transformation from multi-source data to a unified spatial coordinate system; step S3.2 establishing the actual positional relationship between the sensor and the controlled atmosphere equipment, and generating a spatial position mapping table; step S3.3 performing three-dimensional virtual mapping of the storage space; step S3.4 performing three-dimensional virtual mapping of the controlled atmosphere equipment status; step S3.5 performing three-dimensional virtual mapping of the insect infestation status; and step S3.6 integrating the three types of mapping results to generate a full-process digital twin.

[0021] As a further improvement to this application, step S4 involves connecting the full-process digital twin to a deep deterministic strategy gradient algorithm for multi-objective optimization iteration of the controlled atmosphere parameters to obtain the globally optimal controlled atmosphere management strategy, including: Step S4.1: Extract real-time status parameters of each monitoring dimension from the full-process digital twin and integrate them to obtain a set of real-time status parameters of the digital twin; Step S4.2: Input the real-time state parameter set of the digital twin into the state space definition unit of the deep deterministic policy gradient algorithm, and generate the optimization problem definition by configuring the state space dimension and the climate regulation parameter optimization objective; Step S4.3: Initialize the policy network parameters and value network parameters of the deep deterministic policy gradient algorithm based on the definition of the optimization problem, and complete the algorithm initialization; Step S4.4: Input the real-time state parameter set of the digital twin into the initialized deep deterministic strategy gradient algorithm to calculate the action output value of the weather regulation parameter; Step S4.5: Input the action output value and the real-time state parameter set of the digital twin into the value network to calculate the cumulative reward estimate of the current state action pair; Step S4.6: Based on the cumulative return estimate, perform policy gradient update and value network parameter update to achieve multi-objective optimization iteration; Step S4.7: Determine whether the multi-objective optimization iteration meets the convergence condition. If it does, output the globally optimal climate control strategy. If it does not, return to step S4.4 to continue iterating until the convergence condition is met.

[0022] Beneficial effects of steps S4.1 to S4.7: This series of steps, through seven stages—state parameter extraction, optimization problem definition, algorithm initialization, action output calculation, cumulative reward estimation, policy gradient update, and convergence judgment—applies a deep deterministic policy gradient algorithm to the multi-objective optimization of climate control parameters. It overcomes the shortcomings of existing methods, such as lack of simulation prediction capabilities, reliance on experience to set operational parameters, high trial-and-error costs, and inability to dynamically adjust, and achieves dynamic solution from the real-time state of the digital twin to the globally optimal climate control strategy.

[0023] The process includes the following steps: Step S4.1 Extracting real-time state parameters for each monitoring dimension from the full-process digital twin; Step S4.2 Generating an optimization problem definition using state space definition units; Step S4.3 Initializing the deep deterministic policy gradient algorithm; Step S4.4 Calculating the action output values ​​of the climate control parameters; Step S4.5 Calculating the cumulative reward estimate of the current state action pair; Step S4.6 Performing policy gradient updates and value network parameter updates to achieve multi-objective optimization iteration; and Step S4.7 Determining whether the convergence condition is met and outputting the globally optimal climate control strategy.

[0024] As a further improvement to this application, step S5 involves converting the globally optimal controlled atmosphere management strategy into standardized executable control commands for the controlled atmosphere equipment, distributing these commands to each controlled atmosphere execution device in the warehouse space via an industrial IoT gateway, and obtaining execution feedback data from each controlled atmosphere execution device, including: Step S5.1: Parse the instruction format of the global optimal atmosphere control strategy, extract the control parameters of each atmosphere control device, and obtain the instruction parsing result of the atmosphere control strategy. Step S5.2: Convert the parsing result of the controlled atmosphere management strategy instruction into the equipment control instruction corresponding to each controlled atmosphere execution device; Step S5.3: Send the control commands of each device to each controlled atmosphere execution device through the industrial IoT gateway; Step S5.4: In response to the response data returned by each controlled atmosphere actuator after receiving the control command from each device and performing the corresponding operation, execution feedback data is obtained.

[0025] Beneficial effects of steps S5.1 to S5.4: This series of steps, through four stages—instruction format parsing, equipment control instruction conversion, industrial IoT gateway transmission, and execution feedback data acquisition—transforms the globally optimal controlled atmosphere management strategy into standardized executable control instructions and distributes them to each controlled atmosphere execution device. This overcomes the fragmented shortcomings of existing methods, such as independent stages, lack of a unified management platform, and inability to link status in real time, thus achieving precise execution of the management strategy and a closed loop of equipment response.

[0026] Specifically, step S5.1 parses the instruction format of the global optimal controlled atmosphere management strategy and extracts the control parameters of each controlled atmosphere device; step S5.2 converts the parsing results into device control instructions corresponding to each controlled atmosphere execution device; step S5.3 sends the device control instructions through the industrial IoT gateway; and step S5.4 obtains the execution feedback data of each controlled atmosphere execution device.

[0027] As a further improvement to this application, step S6 involves iterating the globally optimal atmosphere control strategy based on the execution deviation between the execution feedback data and the globally optimal atmosphere control strategy, and then transmitting all execution feedback data and the iterated control strategy back to the full-process digital twin to complete dynamic synchronization and full-process closed-loop control, including: Step S6.1: Obtain the setting control parameters of each atmosphere control device in the global optimal atmosphere control strategy, and integrate them to obtain the strategy setting parameter set; Step S6.2: Compare each execution feedback data with the set of strategy settings item by item, calculate the execution deviation between the actual execution value and the set parameter value of each controlled atmosphere device, and obtain the execution deviation dataset; Step S6.3: Calculate the correction amount of each control parameter of the globally optimal air conditioning control strategy based on the execution deviation dataset; Step S6.4: Update the strategy setting parameter set based on the correction amount of each control parameter to obtain the iterative control strategy parameter set; Step S6.5: Send the execution feedback data and the set of control strategy parameters after iteration back to the full-process digital twin, triggering the synchronous update of the model parameters of the full-process digital twin; Step S6.6: Determine whether the preset termination conditions have been met for the closed-loop control of the entire process. If they have been met, output the final control result. If not, return to step S5 to continue execution and complete the closed-loop control of the entire process.

[0028] Beneficial effects of steps S6.1 to S6.6: This series of steps involves six stages: obtaining the strategy setting parameter set, calculating the execution deviation, calculating the correction amount, updating the strategy parameters, synchronizing the digital twin, and determining the closed-loop termination. By iterating the deviation between the execution feedback data and the globally optimal air conditioning control strategy, the control strategy is adaptively optimized. All execution feedback data and the iterated control strategy are then fed back to the full-process digital twin. This overcomes the shortcomings of existing methods, such as delayed early warning response, disconnection from the WMS system, and inability to interconnect data, thus forming a full-process closed-loop control and traceability system.

[0029] The process includes the following steps: Step S6.1: Obtain the set control parameters of each controlled atmosphere device in the global optimal controlled atmosphere management strategy; Step S6.2: Calculate the execution deviation between the actual execution value and the set parameter value of each controlled atmosphere device; Step S6.3: Calculate the correction amount of each control parameter based on the execution deviation; Step S6.4: Update the strategy set parameter set to obtain the iterative control strategy parameter set; Step S6.5: Send the execution feedback data and the iterative control strategy parameter set back to the full-process digital twin to trigger the synchronous update of model parameters; Step S6.6: Determine whether the full-process closed-loop management has reached the preset termination condition and complete the full-process closed-loop management.

[0030] To achieve the above objectives, this application also provides the following technical solutions: A low-oxygen controlled insecticide end-to-end digital twin management system, wherein the end-to-end digital twin management system is applied to the end-to-end digital twin management method described above, and the end-to-end digital twin management system includes: The multi-source real-time time-series data acquisition module is used to collect oxygen concentration, temperature, humidity, insect density, and controlled atmosphere equipment operating status in the storage space, and construct a multi-source real-time time-series dataset. The multi-source real-time time series data calibration module is used to input the multi-source real-time time series dataset into the extended Kalman filter algorithm to perform multi-source data fusion and random noise filtering, and generate calibrated multi-source time series data; The full-process digital twin construction module is used to obtain the digital twin basic model of the storage space and input the calibrated multi-source time series data into the digital twin basic model to perform three-dimensional virtual mapping of the storage space, controlled atmosphere equipment, and insect infestation status, thereby obtaining the full-process digital twin. The optimal atmosphere control strategy iteration module is used to connect the full-process digital twin to the deep deterministic strategy gradient algorithm to perform multi-objective optimization iteration of atmosphere control parameters, so as to obtain the global optimal atmosphere control strategy. The optimal controlled atmosphere control strategy execution module is used to convert the global optimal controlled atmosphere control strategy into standardized executable control commands for the controlled atmosphere equipment, and send them to each controlled atmosphere execution device in the warehouse space through the industrial Internet of Things gateway, and obtain the execution feedback data of each controlled atmosphere execution device. The optimal atmosphere control strategy correction module is used to iterate the global optimal atmosphere control strategy based on the execution deviation between the execution feedback data and the global optimal atmosphere control strategy, and to send all execution feedback data and the iterated control strategy back to the full-process digital twin to complete dynamic synchronization and full-process closed-loop control.

[0031] To achieve the above objectives, this application also provides the following technical solutions: An electronic device includes a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the full-process digital twin control method for low-oxygen controlled insecticide as described above.

[0032] To achieve the above objectives, this application also provides the following technical solutions: A computer-readable storage medium storing program instructions, which, when executed by a processor, enable the implementation of the aforementioned end-to-end digital twin control method for low-oxygen controlled pest control. Attached Figure Description

[0033] Figure 1 This is a schematic flowchart illustrating the steps of an embodiment of the low-oxygen controlled insecticide digital twin management method of this application. Figure 2 This is a schematic diagram of the functional modules of a full-process digital twin control system for low-oxygen controlled insecticide according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an embodiment of the electronic device of this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the storage medium of this application. Detailed Implementation

[0034] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0035] The terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (e.g., as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive having. For example, a process, method, system, product, or device having a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0036] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be present in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0037] It should be noted that, due to the limited types and number of symbols or letters that can represent specific meanings, for embodiments with many formulas or codes, there may be situations where symbols or letters cannot meet the usage requirements. Therefore, the interpretation of formula symbols in the steps or sub-steps of the embodiments is only valid for the current step or sub-step.

[0038] If the same symbol has different interpretations in different steps or sub-steps, the interpretation in the current step or sub-step shall prevail; if the same symbol appears in different steps or sub-steps, but no interpretation is given in subsequent steps or sub-steps after its first appearance, the interpretation in the first step or sub-step shall be used.

[0039] For example, "i=1,2,...,n" is a writing convention and a well-known meaning. If "i=1,2,...,n" has already appeared once in an embodiment and needs to be used again in formulas with different scenarios and meanings in the future, then the meaning of "i=1,2,...,n" should be interpreted according to the corresponding formulas. If "i=1,2,...,n" is not used for the first time and is changed to a symbol or letter with a non-well-known meaning, such as "p=1,2,...,q", to avoid repetition, it is more likely to cause confusion, ambiguity, and unclear problems.

[0040] like Figure 1 As shown, this embodiment provides an example of a full-process digital twin control method for low-oxygen modified insecticide. In this embodiment, the full-process digital twin control method includes the following steps: Step S1: Collect oxygen concentration, temperature, humidity, insect density, and controlled atmosphere equipment operating status in the storage space, and construct a multi-source real-time time series dataset.

[0041] Furthermore, step S1 specifically includes the following steps: Step S1.1: Activate the multi-type data acquisition nodes deployed in the storage space for oxygen, temperature and humidity, insect infestation, and equipment status.

[0042] Preferably, the core of this step is to start and initialize five types of data acquisition nodes within the warehouse space, and to enter the real-time acquisition working state through various types of sensors at a preset sampling frequency, providing the original data source for subsequent data fusion.

[0043] Among them, the oxygen concentration sensor adopts an electrochemical oxygen sensor with a measurement range of 0% to 25%, an accuracy of ±0.1%, a response time of less than 30 seconds, and is deployed in the core area inside the smoke stack sealing cover and key nodes of the pipeline network, with a sampling frequency of once per second.

[0044] For the temperature sensor, a platinum resistance temperature sensor is used, with a measurement range of -40 degrees Celsius to 85 degrees Celsius and an accuracy of ±0.3 degrees Celsius. It is deployed inside the smoke stack, in the sealed cover space, and on the outer wall of the pipeline network, with a sampling frequency of once per second.

[0045] Among them, the humidity sensor is a capacitive humidity sensor with a measurement range of 0%RH to 100%RH and an accuracy of ±2%RH. It is deployed inside the sealed enclosure and at the environmental monitoring point in the storage space, with a sampling frequency of once per second.

[0046] Among them, the insect density detection device adopts a combination of infrared trapping counter and image recognition. It counts crawling pests by blocking infrared beams, and collects images by industrial cameras and identifies moth pests through pre-trained convolutional neural networks. The counting accuracy is not less than 95%. It is deployed in the four corners of the storage space and around the tobacco stacks, with a sampling frequency of once per minute.

[0047] The equipment status acquisition module is connected to the nitrogen generator, valve controller, variable frequency fan, and sealing tester via a bus to collect real-time status parameters such as equipment operating current, voltage, speed, valve opening, and sealing pressure difference, with a sampling frequency of 10 times per second.

[0048] Preferably, the total number of data acquisition nodes of various types is not less than 50, including not less than 15 oxygen concentration sensors, not less than 15 temperature sensors, not less than 10 humidity sensors, not less than 5 insect density detection devices, and not less than 5 equipment status acquisition modules.

[0049] For example, in the low-oxygen controlled insect control operation area of ​​a tobacco logistics center, 18 oxygen concentration sensors, 20 temperature sensors, 12 humidity sensors, 6 insect density detection devices, and 5 equipment status acquisition modules are deployed, totaling 61 data acquisition nodes. Each node starts initialization at YYYY-MM-DD-HH:MM:SS and returns an online status indicator after initialization is completed.

[0050] Step S1.2: The raw data signals of each data acquisition node are extracted through the real-time data acquisition channel at a fixed sampling frequency.

[0051] Preferably, the core of this step is to complete the real-time acquisition of target parameters by various types of data acquisition nodes at a preset sampling frequency, and to aggregate the output signals of each node through the edge gateway to form a continuous raw data signal stream.

[0052] For the oxygen concentration sensor, it outputs an analog current signal of 4mA to 20mA, which is converted into a digital signal by a 16-bit analog-to-digital converter with a resolution of 65,536 levels and a range mapped to oxygen concentration values ​​of 0% to 25%.

[0053] For the temperature sensor, the output PT100 resistance value is converted into a 0V to 5V voltage signal by a bridge circuit and an instrumentation amplifier, then converted into a digital signal by a 14-bit analog-to-digital converter, and finally converted into a temperature value by looking up the PT100 calibration table.

[0054] For the humidity sensor, the output voltage signal is 0V to 10V, which is converted into a digital signal by a 12-bit analog-to-digital converter and linearly mapped to a humidity value of 0%RH to 100%RH.

[0055] Among them, the insect density detection device outputs digital counting pulses and directly accumulates the count.

[0056] Specifically, the device status acquisition module outputs ModbusRTU protocol data frames, and the values ​​of each register are extracted through protocol parsing.

[0057] Preferably, the sampling frequency is set according to Shannon's sampling theorem, and the minimum sampling frequency is not less than twice the highest frequency component of the measured signal; the edge gateway uploads data to the data center via the 4G wireless communication protocol.

[0058] For example, the oxygen concentration sensor collects data once per second, generating 3,600 data points per hour for a single node, and a total of 64,800 data points per hour for 18 oxygen concentration sensors; the equipment status acquisition module collects data 10 times per second, and a total of approximately 180,000 data points per hour for 5 modules.

[0059] Step S1.3: Perform analog-to-digital conversion and level standardization on all raw data signals to generate raw sensing data in a unified format.

[0060] Preferably, the core of this step is to complete the analog-to-digital conversion and normalization of the output signals of heterogeneous sensors, eliminate the differences in output formats of different sensors, and provide a unified data format for subsequent dimensional classification and aggregation.

[0061] Specifically, for analog-to-digital conversion, the oxygen concentration sensor outputs a 4mA to 20mA analog current signal, which is converted into a digital signal by a 16-bit analog-to-digital converter; the temperature sensor outputs a PT100 resistance value, which is converted into a voltage signal by a bridge circuit and an instrumentation amplifier, and then converted into a digital signal by a 14-bit analog-to-digital converter; the humidity sensor outputs a voltage signal, which is converted into a digital signal by a 12-bit analog-to-digital converter; and the insect density detection device and equipment status acquisition module output digital signals, which do not require analog-to-digital conversion.

[0062] For level standardization, oxygen concentration, temperature, humidity, and insect density values ​​are normalized by dividing them by their upper limits to obtain dimensionless values ​​in the range of 0 to 1; equipment status parameters are normalized to per-unit values ​​in the range of 0 to 1 according to the equipment's rated parameters; all normalized values ​​retain 6 significant decimal places.

[0063] Preferably, the standardized data structure includes four fields: sensor number, timestamp, normalized value, and data quality identifier. The data quality identifier is divided into three categories: valid, missing, and abnormal.

[0064] For example, the oxygen concentration sensor O2-001 collects a value of 12.5% ​​at time YYYY-MM-DD-HH:MM:SS-001, with an upper limit of 25% of the range, a normalized value of 0.5, and the data quality is marked as valid; the nitrogen generator exhaust pressure is 0.4MPa, the rated pressure is 0.8MPa, and the per-unit value is 0.5.

[0065] Step S1.4: Classify and collect all raw sensing data according to five dimensions: oxygen concentration, temperature, humidity, insect density, and equipment status.

[0066] Preferably, the core of this step involves standardizing the data and dividing it according to the monitoring parameter type to construct a five-dimensional classification dataset, providing a structured data system for subsequent time series alignment.

[0067] For the oxygen concentration dimension, standardized data from all oxygen concentration sensors are collected. The data structure includes sensor number, timestamp, and normalized oxygen concentration value.

[0068] For the temperature dimension, standardized data from all temperature sensors are collected, and the data structure includes sensor number, timestamp, and normalized temperature value.

[0069] For the humidity dimension, standardized data from all humidity sensors are collected, and the data structure includes sensor number, timestamp, and normalized humidity value.

[0070] For the insect infestation density dimension, standardized data from all insect infestation density detection devices are collected. The data structure includes monitoring point number, timestamp, and normalized insect infestation density value.

[0071] For the device status dimension, standardized data from all device status acquisition modules are collected. The data structure includes device number, timestamp, and normalized device status parameter vector.

[0072] Preferably, each dimension dataset is stored in ascending order of timestamps to support subsequent time-series alignment processing.

[0073] For example, the oxygen concentration dimension includes data from 18 sensors, totaling 64,800 records per hour; the temperature dimension includes data from 20 sensors, totaling 72,000 records per hour; the humidity dimension includes data from 12 sensors, totaling 43,200 records per hour; the insect density dimension includes data from 6 monitoring points, totaling 360 records per hour; and the equipment status dimension includes data from 5 devices, totaling approximately 180,000 records per hour.

[0074] Step S1.5: Perform time-series alignment and data preprocessing on the raw sensor data after classification and aggregation to generate a multi-source real-time time-series dataset.

[0075] Preferably, the core of this step is to unify the timestamps of the five dimensions of data and handle missing anomalies. Using the device status dimension with the highest sampling frequency as the baseline time axis, the other dimensions of data are interpolated and aligned to generate a multi-source real-time time series dataset that can be used for extended Kalman filter fusion.

[0076] For time-series alignment, linear interpolation alignment is performed on the four dimensions of oxygen concentration, temperature, humidity, and insect density, with a time interval of 0.1 seconds, based on the time axis of the device status dimension.

[0077] For missing data points, a forward filling method is used to fill them in, that is, the current missing value is filled with the valid value from the previous time step.

[0078] For outlier data points, the 3σ criterion is used for identification. If the data value exceeds the range of the mean plus or minus 3 times the standard deviation, it is marked as an outlier and replaced with forward filling.

[0079] After alignment, the timestamps of data in all dimensions are unified, with time precision at the millisecond level. Data with valid data quality indicators directly enters the subsequent processing flow, while missing and abnormal data are repaired by interpolation or padding before entering the subsequent processing flow.

[0080] Preferably, after time alignment, the time range of the dataset is a complete pest control operation cycle that is traced back from the current moment, and the duration of the operation cycle is set according to the actual operation plan.

[0081] For example, starting at YYYY-MM-DD-HH:MM:SS-000 and ending at YYYY-MM-DD-HH:MM:SS+3600s, with a time interval of 0.1 seconds, there are a total of 36001 time points. After time-series alignment, the five-dimensional dataset generates a multi-source real-time time-series dataset with a total of 36001 records multiplied by 61 acquisition nodes, totaling 2,196,061 records.

[0082] Beneficial effects of steps S1.1 to S1.5: This series of steps, through five stages—deploying multiple types of data acquisition nodes, real-time data retrieval, analog-to-digital conversion and standardization, dimensional classification and aggregation, and time-series alignment preprocessing—constructs a multi-source real-time time-series dataset covering oxygen concentration, temperature, humidity, insect density, and equipment status. This overcomes the shortcomings of existing methods that only monitor single dimensions and oxygen concentration, providing a unified data foundation for subsequent data fusion and digital twin mapping.

[0083] The process includes: step S1.1, starting and initializing multiple types of data acquisition nodes; step S1.2, real-time acquisition of raw data signals through a fixed sampling frequency; step S1.3, analog-to-digital conversion and level standardization to eliminate heterogeneous data format differences; step S1.4, classification and aggregation according to five dimensions to construct a structured data system; and step S1.5, time alignment and preprocessing to generate a multi-source real-time time series dataset that can be used for filtering and fusion.

[0084] Step S2: Input the multi-source real-time time series dataset into the extended Kalman filter algorithm to perform multi-source data fusion and random noise filtering, and generate calibrated multi-source time series data.

[0085] Step S2: Input the multi-source real-time time series dataset into the extended Kalman filter algorithm to perform multi-source data fusion and random noise filtering to generate calibrated multi-source time series data.

[0086] Furthermore, step S2 specifically includes the following steps: Step S2.1: Input the multi-source real-time time series dataset into the state prediction unit of the extended Kalman filter algorithm in the order of timestamps to calculate the prior state estimates of each dimension of the data.

[0087] Preferably, the core of this step is to complete the first iteration of the extended Kalman filter algorithm, and to make prior estimates of the data of each dimension at the current time through the state transition equation, so as to provide prior states for subsequent covariance prediction and gain calculation.

[0088] The state transition equation for the extended Kalman filter algorithm is as follows: .

[0089] in, Let be the prior state estimate at time k. This is the posterior state estimate at time k-1. This is the control input vector at time k-1. It is a nonlinear state transition function.

[0090] Among them, the nonlinear state transition function is established based on the physical characteristics of the hypoxia-induced insecticidal process, and the state vector is... The control input vector contains normalized values ​​of five parameters: oxygen concentration, temperature, humidity, insect density, and equipment status. It includes three control quantities: nitrogen generator exhaust flow rate, valve opening degree, and variable frequency fan frequency.

[0091] Specifically, the Jacobian matrix of the state vector is obtained by calculating the partial derivatives of the nonlinear state transition function with respect to each component of the state vector. : .

[0092] State transition matrix The calculation is performed dynamically at each time step, reflecting the linearized state transition relationship at the current operating point.

[0093] Preferably, the state vector dimension is the total number of acquisition nodes corresponding to the five types of parameters, i.e., 61 dimensions; the control input vector dimension is 3; the attenuation coefficient of oxygen concentration with nitrogen production flow rate in the nonlinear state transition function is taken as 0.015, and the heat transfer coefficient of temperature with fan frequency is taken as 0.008.

[0094] For example, at time k, the posterior state estimate from the previous time step. The normalized value of oxygen concentration is 0.020 (corresponding to an actual concentration of 0.5%), and the control input is... The normalized value of the nitrogen production flow rate is 0.750 (corresponding to 90 Nm³). 3 The prior state estimate is obtained by calculating the state transition function ( / h). The normalized value of oxygen concentration is 0.015 (corresponding to an actual concentration of 0.375%).

[0095] Step S2.2: Input the prior state estimates of each dimension of the data into the covariance prediction unit of the extended Kalman filter algorithm to calculate the prior error covariance matrix.

[0096] Preferably, the core of this step is to calculate the prior error covariance matrix, quantify the uncertainty of the prior state estimation, and provide a covariance basis for subsequent Kalman gain calculation.

[0097] The formula for calculating the prior error covariance matrix is ​​as follows: .

[0098] in, Let be the prior error covariance matrix at time k. Let be the posterior error covariance matrix at time k-1. Let be the process noise covariance matrix.

[0099] Among them, for the process noise covariance matrix It adopts a diagonal matrix form, with diagonal elements set according to the measurement uncertainties of various sensors, and the process noise variance of the oxygen concentration dimension is taken as... Temperature dimension Humidity dimension Insect density dimension Device status dimension .

[0100] Among them, for the initial posterior error covariance matrix It adopts a diagonal matrix form, with all diagonal elements being... This indicates that the initial state estimation has a large degree of uncertainty.

[0101] Preferably, the prior error covariance matrix The dimension is 61×61; the process noise covariance matrix The dimensions are 61×61.

[0102] For example, at time k, the state transition matrix The posterior error covariance matrix at the previous time step is a 61×61 matrix. The maximum value of the diagonal elements is 0.005, and the process noise covariance matrix... The maximum value of the diagonal elements is 0.0005, from which the prior error covariance matrix is ​​calculated. The maximum value of the diagonal element is 0.006.

[0103] Step S2.3: Input the prior error covariance matrix into the gain calculation unit of the extended Kalman filter algorithm to calculate the optimal Kalman gain matrix.

[0104] Preferably, the core of this step is to calculate the optimal Kalman gain matrix and determine the correction weight of the observations to the state estimate. The larger the gain, the greater the contribution of the observations to the state update.

[0105] The formula for calculating the optimal Kalman gain matrix is ​​as follows: .

[0106] in, Let K be the optimal Kalman gain matrix at time k. For the observation matrix, To observe the noise covariance matrix.

[0107] For the observation matrix, since the observation dimension of the multi-source real-time time series dataset is consistent with the state vector dimension, the observation matrix is ​​taken as a 61×61 identity matrix.

[0108] Among them, for the observation noise covariance matrix It adopts a diagonal matrix form, with diagonal elements set according to the measurement noise of various sensors. The observation noise variance of the oxygen concentration sensor is taken as... Temperature sensor Humidity sensor Insect density detection device The device status acquisition module takes .

[0109] Preferably, the optimal Kalman gain matrix The dimension is 61×61; the observation noise covariance matrix The dimensions are 61×61.

[0110] For example, at time k, the prior error covariance matrix The diagonal elements are 0.006, and the observation noise covariance matrix is... With diagonal elements of 0.00004 (oxygen concentration dimension), the optimal Kalman gain matrix is ​​calculated. The corresponding dimension has approximately 0.993 elements, indicating that the weight of the observation on the state update is close to 1.

[0111] Step S2.4: Input the optimal Kalman gain matrix and the prior state estimates of each dimension of data into the state update unit to calculate the posterior state estimates.

[0112] Preferably, the core of this step is the calculation of the posterior state estimate. The posterior state estimate is corrected by weighting the deviation between the actual observed value and the posterior state estimate using Kalman gain, resulting in a more accurate posterior state estimate.

[0113] The formula for calculating the posterior state estimate is as follows: .

[0114] in, Let be the posterior state estimate at time k. Let k be the actual observation vector at time k. For nonlinear observation functions, To observe the residuals.

[0115] For nonlinear observation functions, the state vector is mapped to the observation space. Since the state vector and the observation vector have the same dimension and corresponding physical meaning, the nonlinear observation function takes an identity mapping, i.e. .

[0116] Specifically, for the update of the posterior error covariance matrix: .

[0117] Where I is a 61×61 identity matrix, Let be the posterior error covariance matrix at time k, used for covariance prediction at the next time step.

[0118] Preferably, the observed residual If the absolute value of each dimension component exceeds 3 times the standard deviation of the observation noise, the observation value of that dimension is reduced in weight by multiplying the corresponding row of the gain matrix by 0.1.

[0119] For example, at time k, the normalized value of oxygen concentration in the prior state estimate is 0.015, the normalized value of oxygen concentration in the actual observation is 0.018, the dimension corresponding to the optimal Kalman gain is 0.993, the observation residual is 0.003, and the normalized value of oxygen concentration in the posterior state estimate is 0.015 + 0.993 × 0.003 = 0.018, corresponding to an actual concentration of 0.45%.

[0120] Step S2.5: The posterior state estimate is split and recombined according to the original data dimension to obtain calibrated multi-source time series data.

[0121] Preferably, the core of this step is to restore the posterior state estimate from the fused vector to the original five-dimensional data format. The 61-dimensional posterior state estimate vector is split and recombined according to five dimensions: oxygen concentration, temperature, humidity, insect density, and equipment status, to generate calibrated multi-source time series data with the same structure as the input data.

[0122] Specifically, for splitting and recombining, according to the original acquisition node number and dimension correspondence, each component in the posterior state estimation vector is restored to the corresponding sensor number and timestamp to generate calibrated time series data for each dimension.

[0123] For the calibrated data format, the data structure for each dimension includes sensor number, timestamp, normalized value after calibration, and data quality identifier after calibration, all of which are valid.

[0124] For calibration effectiveness evaluation, the mean square error of the data before and after calibration is calculated, and the mean square error of the oxygen concentration dimension does not exceed [a certain value]. Temperature dimension does not exceed Humidity dimension not exceeding Insect density dimension does not exceed The device status dimension does not exceed .

[0125] Preferably, the time range of the calibrated multi-source time series data is consistent with that of the input multi-source real-time time series dataset, with a time interval of 0.1 seconds.

[0126] For example, the normalized value of the oxygen concentration sensor O2-001 at time YYYY-MM-DD-HH:MM:SS-001 after calibration is 0.018, corresponding to an actual concentration of 0.45%. Compared with the original acquired value of 0.020 (0.5%), the deviation after noise filtering is 0.05%.

[0127] Beneficial effects of steps S2.1 to S2.5: This series of steps, through five stages—state prediction, covariance prediction, gain calculation, state update, and splitting and recombining—applies the extended Kalman filter algorithm to the fusion of multi-source time-series data. This effectively filters out random noise during the acquisition process, generates calibrated multi-source time-series data, significantly improves data reliability and subsequent digital twin mapping accuracy, and lays a data quality foundation for precise control throughout the entire process.

[0128] The process includes: step S2.1 calculating prior state estimates for each dimension of the data to achieve state prediction; step S2.2 calculating the prior error covariance matrix to quantify prediction uncertainty; step S2.3 calculating the optimal Kalman gain matrix to determine the correction weights of the observations on the state estimates; step S2.4 calculating the posterior state estimates to complete the state update; and step S2.5 splitting and reorganizing the original data according to its dimensions to obtain calibrated multi-source time series data.

[0129] Step S3: Obtain the digital twin basic model of the storage space, and input the calibrated multi-source time series data into the digital twin basic model to perform three-dimensional virtual mapping of the storage space, controlled atmosphere equipment, and insect infestation status, thereby obtaining a full-process digital twin.

[0130] Furthermore, step S3 specifically includes the following steps: Step S3.1: Convert the calibrated multi-source time-series data to the unified spatial coordinate system of the digital twin basic model to obtain the coordinate-transformed spatial data.

[0131] Preferably, the core of this step is to complete the coordinate transformation of the multi-source time-series data after calibration from the sensor's local coordinate system to the unified spatial coordinate system of the digital twin basic model, so as to provide data input with a consistent spatial reference for subsequent 3D virtual mapping.

[0132] For the unified spatial coordinate system, a right-handed rectangular coordinate system is established with the geometric center of the storage space as the origin, the X-axis along the length of the warehouse, the Y-axis along the width, and the Z-axis along the height. The coordinate unit is meters.

[0133] For coordinate transformation, each sensor has its three-dimensional coordinates in a unified spatial coordinate system calibrated during installation. After calibration, the sensor numbers in the multi-source time series data are matched with the coordinate calibration table. The normalized values ​​are then appended with the corresponding three-dimensional spatial coordinates to generate spatial data after coordinate transformation that includes spatial location information.

[0134] For the coordinate calibration table, a total station is used to measure the three-dimensional coordinates of each sensor installation location with a measurement accuracy of ±5mm. The calibration results are stored as a mapping table from sensor number to three-dimensional coordinates.

[0135] Preferably, the X-axis of the unified spatial coordinate system covers the entire length of the warehouse, the Y-axis covers the entire width of the warehouse, and the Z-axis covers the entire height of the warehouse; the data structure of the spatial data after coordinate transformation includes sensor number, timestamp, three-dimensional spatial coordinates, normalized values ​​after calibration, and data quality identifier.

[0136] For example, Warehouse No. 1 of a tobacco logistics center is 120 meters long, 40 meters wide, and 8 meters high. The origin of the unified spatial coordinate system is located at the geometric center of the warehouse. The X-axis range is -60 meters to 60 meters, the Y-axis range is -20 meters to 20 meters, and the Z-axis range is 0 meters to 8 meters. The calibration coordinates of the oxygen concentration sensor O2-001 are (10.5, 5.2, 3.0). After calibration, the normalized value is 0.018. After coordinate transformation, the spatial data record is (O2-001,YYYY-MM-DD-HH:MM:SS-001,10.5,5.2,3.0,0.018,valid).

[0137] Step S3.2: Based on the digital twin basic model, establish the actual positional relationship between each sensor and the controlled atmosphere equipment in the warehouse space, and generate a spatial position mapping table.

[0138] Preferably, the core of this step is to establish the actual positional relationship between the sensor and the controlled atmosphere equipment within the storage space. Through dual mapping of spatial proximity and functional association, it provides a spatial correspondence basis for the subsequent three-dimensional virtual mapping of the equipment and the sensor.

[0139] For spatial proximity relationships, the Euclidean distance from each sensor to each controlled atmosphere device is calculated. If the distance is less than the preset proximity threshold, a spatial proximity mapping between the sensor and the controlled atmosphere device is established. The proximity threshold is set according to the influence range of the device. The proximity threshold for the nitrogen generator is 5 meters, the proximity threshold for the variable frequency fan is 3 meters, and the proximity threshold for the valve controller is 2 meters.

[0140] Specifically, for functional associations, a mapping is established based on the functional correspondence between sensor types and controlled atmosphere equipment types. Oxygen concentration sensors are associated with nitrogen generators and valve controllers, temperature sensors are associated with variable frequency fans and sealing detectors, humidity sensors are associated with variable frequency fans, insect density detection devices are associated with sealing detectors, and equipment status acquisition modules are associated with the corresponding controlled atmosphere equipment.

[0141] The spatial location mapping table has a data structure containing four fields: sensor number, controlled atmosphere device number, spatial distance, and functional association type. It supports retrieving all sensors associated with a controlled atmosphere device number.

[0142] Preferably, the spatial location mapping table is incrementally updated when sensors or controlled atmosphere devices are added or removed, with an update frequency of once a day.

[0143] For example, the oxygen concentration sensor O2-001 is located at coordinates (10.5, 5.2, 3.0), and the nitrogen generator N2-001 is located at coordinates (12.0, 6.0, 1.5). The Euclidean distance is 1.97 meters, which is less than the proximity threshold of 5 meters. A spatial proximity mapping is established. The oxygen concentration sensor is functionally associated with the nitrogen generator, and a functional association mapping is established. The spatial location mapping table is recorded as (O2-001, N2-001, 1.97, oxygen concentration - nitrogen generator).

[0144] Step S3.3: Input the spatial data after coordinate transformation and the spatial location mapping table into the digital twin basic model, perform three-dimensional virtual mapping of the warehouse space, and generate the three-dimensional virtual mapping result of the warehouse space.

[0145] Preferably, the core of this step is to complete the three-dimensional virtual mapping of the storage space, presenting the warehouse building structure, smoke stack layout, and sealing cover shape in the form of a three-dimensional model, and mapping environmental parameters such as oxygen concentration, temperature, and humidity to the three-dimensional space in the form of color gradients or isosurfaces, thereby realizing the visualization of the environmental status of the storage space.

[0146] The digital twin basic model is constructed using a 3D geometric modeling engine. The warehouse building structure is represented by a cuboid geometry, the smoke stacks are represented by a cylindrical or cuboid geometry, and the sealing cover is represented by a semi-transparent closed curved surface. The model surface accuracy is no less than 100 triangular surfaces per square meter.

[0147] For the mapping of environmental parameters, oxygen concentration uses a four-color gradient mapping of blue-green-yellow-red, where blue indicates an oxygen concentration below 0.5% (insecticide control meets the standard), green indicates 0.5% to 2%, yellow indicates 2% to 5%, and red indicates above 5% (not meeting the standard); temperature uses a cool-warm color gradient mapping, where temperatures below 20 degrees Celsius are blue, 20 to 30 degrees Celsius are green, and above 30 degrees Celsius are red; humidity uses a light blue-dark blue gradient mapping.

[0148] For isosurface generation, three-dimensional spatial interpolation is performed on oxygen concentration, temperature, and humidity respectively. The inverse distance weighted interpolation method is used, with an interpolation radius of 5 meters and a weighting index of 2 to generate three-dimensional isosurfaces for each parameter.

[0149] Preferably, the update frequency of the three-dimensional virtual mapping result of the warehouse space is consistent with the time interval of the calibrated multi-source time series data, that is, it is updated once every 0.1 seconds.

[0150] For example, after the calibration values ​​of the 18 oxygen concentration sensors in Warehouse No. 1 are interpolated in three-dimensional space, an oxygen concentration isosurface is generated. The isosurface inside the sealed cover is blue (concentration below 0.5%), and the isosurface at the edge of the sealed cover is yellow (concentration 2% to 5%), indicating that there is a leak in the sealed cover.

[0151] Step S3.4: Input the controlled atmosphere equipment operating status data in the spatial data after coordinate transformation and the spatial location mapping table into the digital twin basic model, perform three-dimensional virtual mapping of the controlled atmosphere equipment status, and generate the three-dimensional virtual mapping result of the controlled atmosphere equipment status.

[0152] Preferably, the core of this step is to complete the three-dimensional virtual mapping of the operating status of the controlled atmosphere equipment, mapping the operating status parameters of the nitrogen generator, valve controller, variable frequency fan, and seal tester to the corresponding three-dimensional equipment models in the digital twin basic model with dynamic labels or animation effects, so as to realize the visual monitoring of equipment status.

[0153] For the 3D model of the controlled atmosphere equipment, the nitrogen generator is represented by a combined geometry containing the inlet pipe, the exhaust pipe, and the compressor; the valve controller is represented by a combined geometry containing the valve body, the valve stem, and the actuator; the variable frequency fan is represented by a combined geometry containing the fan casing, the impeller, and the motor; and the sealing tester is represented by a geometry containing the probe and the display screen.

[0154] Specifically, for equipment status mapping, the exhaust pressure of the nitrogen generator is displayed as a digital label on the exhaust pipe position, the exhaust flow rate is represented by an arrow animation, and the motor speed is represented by the impeller rotation speed; the valve opening of the valve controller is represented by a valve stem displacement animation, and the opening percentage is displayed as a digital label; the operating frequency of the variable frequency fan is displayed as a digital label, and the duct pressure is represented by a color gradient; the sealing cover pressure difference of the seal detector is displayed as a digital label, and the leakage rate is indicated by a flashing red indicator to show that leakage exceeds the standard.

[0155] Among them, for abnormal status indicators, when the equipment status parameters exceed the normal operating range, the outer contour of the equipment's 3D model is highlighted in red, and an abnormal parameter floating prompt box pops up.

[0156] Preferably, the update frequency of the three-dimensional virtual mapping result of the controlled atmosphere equipment status is consistent with the sampling frequency of the equipment status data, that is, it is updated once every 0.1 seconds.

[0157] For example, the exhaust pressure of nitrogen generator N2-001 is 0.4MPa, which is displayed as a digital label at the location of the exhaust pipe; the exhaust flow rate is 120Nm³ / h, which is represented by an arrow animation; the motor speed is 1480rpm, which is represented by the impeller rotation speed; and the outer contour of the equipment's 3D model is green (in normal operating condition).

[0158] Step S3.5: Input the insect infestation status data in the spatial data after coordinate transformation and the spatial location mapping table into the digital twin basic model, perform three-dimensional virtual mapping of insect infestation status, and generate three-dimensional virtual mapping result of insect infestation status.

[0159] Preferably, the core of this step is to complete the three-dimensional virtual mapping of the insect infestation status, mapping the monitoring data of the insect infestation density detection device to the corresponding position in the three-dimensional model of the storage space in the form of a heat map or scatter plot, so as to realize the visual monitoring of the insect infestation distribution.

[0160] For the insect infestation heat map, the normalized values ​​of the calibrated insect density detection device are interpolated in three dimensions using the inverse distance weighted interpolation method. The interpolation radius is 8 meters and the weighting index is 2 to generate a three-dimensional heat map of insect density. The density is represented by a green-yellow-red gradient from low to high.

[0161] Among them, for the scattered markers, when the count of the insect density detection device exceeds the preset warning threshold, the insect warning is indicated by red scattered markers at the corresponding three-dimensional position. The warning thresholds are set according to the insect species. The warning is triggered when the count of tobacco beetles exceeds 10 per hour at each monitoring point, and the warning is triggered when the count of tobacco mealybugs exceeds 5 per hour.

[0162] For the insect infestation trend indicator, the density change rate of the most recent 30 minutes is calculated for the insect infestation density time series data of each monitoring point. If the change rate is positive and exceeds the preset growth rate threshold, the insect infestation spread trend is indicated by a yellow upward arrow. The growth rate threshold is 50% per hour.

[0163] Preferably, the update frequency of the three-dimensional virtual mapping result of the insect infestation status is consistent with the sampling frequency of the insect infestation density detection device, that is, it is updated once per minute.

[0164] For example, the Pest-003 insect density detection device is located at coordinates (15.0, 8.0, 1.5). The count of tobacco beetles in the last hour is 15, which exceeds the warning threshold of 10. The insect infestation warning is indicated by red scatter dots at the corresponding three-dimensional position. The density change rate in the last 30 minutes is 80%, which exceeds the growth rate threshold of 50%. The insect infestation spread trend is indicated by yellow rising arrows.

[0165] Step S3.6: Integrate the three-dimensional virtual mapping results of the storage space, the three-dimensional virtual mapping results of the controlled atmosphere equipment status, and the three-dimensional virtual mapping results of the insect infestation status to generate a full-process digital twin.

[0166] Preferably, the core of this step is to integrate the three types of three-dimensional virtual mapping results, and to uniformly overlay the warehouse space environment mapping, controlled atmosphere equipment status mapping, and pest status mapping onto the digital twin basic model to generate a full-process digital twin covering both physical and virtual spaces, thereby achieving full-process visualized control.

[0167] The integration method uses a digital twin basic model as the base, overlaying the three-dimensional virtual mapping result of the storage space as an environmental layer onto the building structure geometry, overlaying the three-dimensional virtual mapping result of the controlled atmosphere equipment status as an equipment layer onto the equipment geometry, and overlaying the three-dimensional virtual mapping result of the insect infestation status as an insect infestation layer onto the interior of the storage space. The three layers are displayed by overlaying according to the transparency weight.

[0168] The data structure of the full-process digital twin includes four data modules: three-dimensional geometric model data, spatial distribution data of environmental parameters, real-time data of equipment status, and real-time data of insect distribution. These modules are linked together through a spatial coordinate system.

[0169] Among them, the interactive functions of the full-process digital twin support rotation, scaling, and translation of three-dimensional view operations, support filtering and displaying content by region, by device, and by dimension, and support clicking on the device model to view detailed operating parameters and historical curves.

[0170] Preferably, the rendering frame rate of the digital twin throughout the entire process is no less than 30 frames per second, and it supports access from multiple terminals (PC browser, mobile APP, large screen display system).

[0171] For example, the entire process of Warehouse No. 1 is displayed in a PC browser. The 3D model shows that the sealing covers of the six smoke stacks in the warehouse are represented by semi-transparent closed curved surfaces, the distribution of 18 oxygen concentration sensors is marked by color gradient dots, the nitrogen generator N2-001 is marked as running in green, and the pest monitoring point Pest-003 is marked as a warning with red scattered dots. The whole process of low-oxygen pest control is presented in real-time visualization.

[0172] Beneficial effects of steps S3.1 to S3.6: This series of steps, through six stages—coordinate transformation, spatial location mapping, 3D virtual mapping of storage space, 3D virtual mapping of controlled atmosphere equipment status, 3D virtual mapping of pest status, and integration of the entire process digital twin—maps calibrated multi-source time-series data to a basic digital twin model, constructing a full-process digital twin covering storage space, controlled atmosphere equipment, and pest status. This overcomes the shortcomings of existing methods, such as lack of full-process visualization, difficulty in locating abnormal nodes, and inconvenience in process traceability, achieving real-time synchronization and intuitive visual control between physical and virtual spaces.

[0173] The process includes: step S3.1 completing the coordinate transformation from multi-source data to a unified spatial coordinate system; step S3.2 establishing the actual positional relationship between the sensor and the controlled atmosphere equipment, and generating a spatial position mapping table; step S3.3 performing three-dimensional virtual mapping of the storage space; step S3.4 performing three-dimensional virtual mapping of the controlled atmosphere equipment status; step S3.5 performing three-dimensional virtual mapping of the insect infestation status; and step S3.6 integrating the three types of mapping results to generate a full-process digital twin.

[0174] Step S4: The entire process digital twin is connected to the deep deterministic strategy gradient algorithm to perform multi-objective optimization iteration of the gas control parameters, so as to obtain the globally optimal gas control strategy.

[0175] Furthermore, step S4 specifically includes the following steps: Step S4.1: Extract the real-time status parameters of each monitoring dimension from the full-process digital twin and integrate them to obtain the real-time status parameter set of the digital twin.

[0176] Preferably, the core of this step is to extract and integrate the real-time state parameters in the entire process digital twin, extracting the real-time values ​​of each dimension in the spatial distribution data of environmental parameters, real-time data of equipment status, and real-time data of insect distribution as state vectors, providing state space input for the deep deterministic strategy gradient algorithm.

[0177] Specifically, for the extraction of state parameters, the current time values ​​of the calibration values ​​of each oxygen concentration sensor, each temperature sensor, and each humidity sensor are extracted from the three-dimensional virtual mapping results of the storage space; the current time values ​​of the exhaust pressure, exhaust flow, motor speed, valve opening, frequency of each variable frequency fan, pressure difference of each sealing cover, and leakage rate of each nitrogen generator are extracted from the three-dimensional virtual mapping results of the controlled atmosphere equipment; and the current time values ​​of the counts of each insect density detection device are extracted from the three-dimensional virtual mapping results of the insect infestation status.

[0178] Specifically, for state vector construction, all extracted real-time state parameters are arranged in a fixed order to form a state vector. The state vector has a dimension of 61 (consistent with the state vector dimension of the extended Kalman filter), and each component corresponds to the current calibration value or current value of the device parameter of a data acquisition node.

[0179] For state normalization, each component of the state vector has been normalized to the 0 to 1 range in the previous step and can be directly used as input for the deep deterministic policy gradient algorithm without additional normalization processing.

[0180] Preferably, the extraction frequency of the real-time status parameter set of the digital twin is consistent with the update frequency of the entire digital twin, that is, it is extracted once every 0.1 seconds.

[0181] For example, at time YYYY-MM-DD-HH:MM:SS, a 61-dimensional state vector is extracted from the full-process digital twin, where the normalized value of the oxygen concentration sensor O2-001 is 0.018 (actual concentration 0.45%), the per-unit value of the exhaust pressure of the nitrogen generator N2-001 is 0.500 (actual 0.4MPa), and the normalized value of the insect density detection device Pest-003 is 0.300 (15 insects per hour).

[0182] Step S4.2: Input the real-time state parameter set of the digital twin into the state space definition unit of the deep deterministic policy gradient algorithm, and generate the optimization problem definition by configuring the state space dimension and the climate regulation parameter optimization objective.

[0183] Preferably, the core of this step is to define the optimization problem of the deep deterministic policy gradient algorithm. By configuring the state space, action space and reward function, the low-oxygen regulation and pest control problem is modeled as a Markov decision process in a continuous action space.

[0184] Among them, for the definition of state space, state space The state space is a 61-dimensional continuous space, with each dimension corresponding to a normalized parameter value of a data acquisition node. The state space range is... .

[0185] Among them, the definition of action space, action space The space is a 3-dimensional continuous space, corresponding to the normalized values ​​of three control variables: nitrogen generator exhaust flow rate, valve opening, and variable frequency fan frequency. The action space range is... .

[0186] For the reward function design, a multi-objective weighted reward function is adopted: .

[0187] in, , , , , These are the weighting coefficients; As an oxygen concentration reward, a value of 1 is applied when all oxygen concentration sensor values ​​are below 0.02 (corresponding to 0.5% of the actual concentration); otherwise, a negative deviation penalty is applied. For temperature-based rewards, a value of 1 is assigned when the temperature is within the normalized range of 20 to 25 degrees Celsius, and a value of 0 is assigned otherwise. For humidity bonus, a value of 1 is assigned when the humidity is within the normalized range of 50%RH to 65%RH, and a value of 0 is assigned otherwise. As a reward for insect infestation, a value of 1 is assigned when the normalized value of insect infestation density is below 0.05, and a value of 0 is assigned otherwise. For energy consumption rewards, the smaller the normalized value of nitrogen production flow rate and fan frequency, the higher the reward. ,in This is the normalized value for nitrogen production flow rate. This is the normalized value of the fan frequency.

[0188] For example, in the current state vector, the maximum normalized value of oxygen concentration is 0.018 (below 0.02 is considered acceptable), the normalized value of temperature is within the normal range, and the normalized value of insect density is 0.300 (above 0.05 is not considered acceptable). The calculated values ​​are... , , , , Weighted rewards .

[0189] Step S4.3: Initialize the policy network parameters and value network parameters of the deep deterministic policy gradient algorithm based on the definition of the optimization problem, and complete the algorithm initialization.

[0190] Preferably, the core of this step is to initialize the parameters of the policy network and value network of the deep deterministic policy gradient algorithm, providing initial network parameters for subsequent action output and policy iteration.

[0191] The policy network (Actor network) employs a 4-layer fully connected neural network structure: an input layer of dimension 61, two hidden layers each of dimension 256, and an output layer of dimension 3. The hidden layer activation function is ReLU, and the output layer activation function is Sigmoid (constraining the action output to the [0,1] interval). The policy network parameters are as follows: Uniform initialization using Xavier is employed.

[0192] The Critic network employs a 4-layer fully connected neural network structure. The input layer has a dimension of 64 (a concatenation of a 61-dimensional state vector and a 3-dimensional action vector), both hidden layers have a dimension of 256, and the output layer has a dimension of 1. The hidden layer activation function is ReLU, and the output layer has no activation function (outputting the Q-value). The parameters of the Critic network are as follows: Uniform initialization using Xavier is employed.

[0193] Among them, for the target network, the policy network target network parameters Initialize to The copy, value network target network parameters Initialize to The copy; the experience replay buffer capacity is 100,000 records, and the batch size is 64.

[0194] Preferably, the learning rate policy network takes Value network acquisition The target network soft update coefficient is set to 0.005.

[0195] For example, the policy network takes a 61-dimensional state vector as input, passes through two 256-dimensional hidden layers and a ReLU activation function, and outputs a 3-dimensional action vector constrained to [0,1] by a Sigmoid function, which corresponds to the normalized values ​​of nitrogen production flow rate, valve opening, and fan frequency, respectively.

[0196] Step S4.4: Input the real-time state parameter set of the digital twin into the initialized deep deterministic strategy gradient algorithm to calculate the action output value of the climate control parameters.

[0197] Preferably, the core of this step is to complete the forward computation of the policy network, input the current state vector into the policy network, output the normalized action values ​​of the three climate control parameters, and add exploration noise to ensure the policy exploration.

[0198] Specifically, for the forward computation of the policy network: .

[0199] in, For the deterministic actions output by the policy network, With a mean of 0 and a standard deviation of Gaussian exploration noise, This is the actual action output after adding noise.

[0200] Specifically, for action clipping, each component of the action vector after adding noise is clipped to the [0,1] interval to ensure that the action value is within the valid range.

[0201] Regarding the physical meaning of the action, The three components are: the normalized value of the nitrogen generator exhaust flow rate, mapped to 0 to 160 Nm³ / h; the normalized value of the valve opening, mapped to 0% to 100%; and the normalized value of the variable frequency fan frequency, mapped to 0 Hz to 50 Hz.

[0202] Preferably, the forward computation time of the policy network does not exceed 10 milliseconds per iteration, which meets the requirements of real-time control.

[0203] For example, after the current state vector is input into the policy network, the deterministic action output is (0.75, 0.80, 0.70), and after adding Gaussian exploration noise, it becomes (0.73, 0.82, 0.68). The pruned action output value... These correspond to a nitrogen production flow rate of 117 Nm³ / h, a valve opening of 82%, and a fan frequency of 34 Hz, respectively.

[0204] Step S4.5: Input the action output value and the real-time state parameter set of the digital twin into the value network to calculate the cumulative reward estimate of the current state action pair.

[0205] Preferably, the core of this step is to complete the forward computation of the value network, concatenate the current state vector and action vector and input them into the value network, and output the Q-value estimate of the current state-action pair, that is, the expected value of the cumulative reward, which is used to evaluate the effectiveness of the current strategy.

[0206] Specifically, for forward computation of the value network: .

[0207] in, It is a concatenation of a 61-dimensional state vector and a 3-dimensional action vector, with a total dimension of 64. For the nonlinear mapping function of the value network, This is the cumulative reward estimate for the current state action pair.

[0208] Specifically, the calculation of the target Q value is as follows: .

[0209] in, Let Q' be the discount factor and Q' be the target value network. For the target policy network, The target Q value is used to calculate the loss function of the value network.

[0210] For the loss function, mean squared error loss is used: .

[0211] Where N=64 is the batch size.

[0212] Preferably, each record in the experience playback buffer contains Quadruple; the target network parameters in the target Q value are updated via soft update.

[0213] For example, after the current state action is input into the value network, the Q-value estimate is 0.705, the target Q-value is 0.730, and the mean squared error loss is... .

[0214] Step S4.6: Based on the cumulative return estimate, perform policy gradient update and value network parameter update to achieve multi-objective optimization iteration.

[0215] Preferably, the core of this step is to update the parameters of the policy network and the value network, maximize the Q value through policy gradient ascent, and update the value network parameters by minimizing the TD error, thereby achieving multi-objective optimization iteration.

[0216] For updating the policy network parameters, policy gradient ascent is used: .

[0217] in, The gradient of the Q-value with respect to the action. The gradient of the policy network with respect to the parameters is updated end-to-end using the chain rule, and the learning rate of the policy network is... .

[0218] Specifically, for updating the value network parameters, the Adam optimizer is used to minimize the mean squared error loss, and the value network learning rate is... .

[0219] Specifically, for target network soft updates: .

[0220] .

[0221] Among them, soft update coefficient The target network parameters track the main network parameters with a very small step size, ensuring training stability.

[0222] Preferably, in each iteration, 64 records are randomly sampled from the experience replay buffer for parameter update; when the experience replay buffer is full, the old data is replaced by a first-in-first-out strategy.

[0223] For example, after the policy network parameters are updated once by gradient ascent, the policy network’s deterministic action for the current state output is adjusted from (0.75, 0.80, 0.70) to (0.76, 0.81, 0.69), the nitrogen production flow rate and valve opening are slightly increased, the fan frequency is slightly decreased, and the Q value estimate is increased from 0.705 to 0.712.

[0224] Step S4.7: Determine whether the multi-objective optimization iteration meets the convergence condition. If it does, output the globally optimal climate control strategy. If it does not, return to step S4.4 to continue iterating until the convergence condition is met.

[0225] Preferably, the core of this step is to complete the convergence judgment of multi-objective optimization iteration. By monitoring the rate of change of Q value and the rate of change of reward value, it is determined whether convergence has been achieved. After convergence, the globally optimal climate control strategy corresponding to the current strategy network parameters is output.

[0226] The convergence condition is determined using a dual-condition approach: Condition 1 is that the rate of change of the average reward value over 100 consecutive iterations is less than 10%. Condition two is that the mean rate of change of Q value is less than 1 / 2 in 100 consecutive iterations. Convergence is determined when both conditions are met simultaneously.

[0227] Among them, the maximum number of iterations is set to 5000 rounds. If the convergence condition is not met after 5000 rounds, the strategy with the highest current reward value is output as the global optimal climate control strategy.

[0228] The output format for the globally optimal atmospheric regulation control strategy includes the strategy network parameters. The policy network consists of three parts: input-output mapping relationship of the policy network, physical meaning mapping of action space, and policy network parameters. The input-output mapping relationship is the forward computation process from a 61-dimensional state vector to a 3-dimensional action vector, representing all weights and bias parameters of the policy network. The physical meaning of the action space is mapped as a linear mapping relationship from normalized action values ​​to actual atmospheric control parameter values.

[0229] Preferably, the output action value of the globally optimal atmospheric regulation control strategy is a deterministic action (without added exploration noise), that is... .

[0230] For example, after 3200 iterations, the average change rate of the reward value in the deep deterministic policy gradient algorithm over 100 consecutive iterations is... (less than) The mean change rate of Q value is (less than) If both conditions are met simultaneously, convergence is determined, and the globally optimal air conditioning control strategy is output. After the current state vector is input into the strategy network, the deterministic action output is (0.78, 0.85, 0.65), corresponding to a nitrogen production flow rate of 125 Nm³. 3 / h, valve opening 85%, fan frequency 32.5Hz.

[0231] Beneficial effects of steps S4.1 to S4.7: This series of steps, through seven stages—state parameter extraction, optimization problem definition, algorithm initialization, action output calculation, cumulative reward estimation, policy gradient update, and convergence judgment—applies a deep deterministic policy gradient algorithm to the multi-objective optimization of climate control parameters. It overcomes the shortcomings of existing methods, such as lack of simulation prediction capabilities, reliance on experience to set operational parameters, high trial-and-error costs, and inability to dynamically adjust, and achieves dynamic solution from the real-time state of the digital twin to the globally optimal climate control strategy.

[0232] The process includes the following steps: Step S4.1 Extracting real-time state parameters for each monitoring dimension from the full-process digital twin; Step S4.2 Generating an optimization problem definition using state space definition units; Step S4.3 Initializing the deep deterministic policy gradient algorithm; Step S4.4 Calculating the action output values ​​of the climate control parameters; Step S4.5 Calculating the cumulative reward estimate of the current state action pair; Step S4.6 Performing policy gradient updates and value network parameter updates to achieve multi-objective optimization iteration; and Step S4.7 Determining whether the convergence condition is met and outputting the globally optimal climate control strategy.

[0233] Step S5: Convert the global optimal controlled atmosphere management strategy into standardized executable control commands for the controlled atmosphere equipment, send them to each controlled atmosphere execution device in the warehouse space through the industrial IoT gateway, and obtain the execution feedback data of each controlled atmosphere execution device.

[0234] Furthermore, step S5 specifically includes the following steps: Step S5.1: Parse the instruction format of the global optimal atmosphere control strategy, extract the control parameters of each atmosphere control device, and obtain the instruction parsing result of the atmosphere control strategy.

[0235] Preferably, the core of this step is to analyze and allocate the action output value of the global optimal controlled atmosphere management strategy to the control parameters of each controlled atmosphere device, and decompose the 3D action vector into the specific control parameters corresponding to each controlled atmosphere device.

[0236] Among them, for action vector parsing, 3D action vector These correspond to: The first component is the normalized value of the nitrogen generator unit's exhaust flow rate, mapped to the actual flow rate. Nm 3 / h; The second component is the normalized value of the valve opening, mapped to the actual opening. The third component is the frequency normalization value of the variable frequency fan, mapped to the actual frequency. Hz.

[0237] Specifically, for control parameter allocation, the parsed control parameters are assigned to each gas regulation actuator according to the device number: nitrogen production flow rate. Distributed to nitrogen generator unit N2-001, valve opening degree Distributed to valve controllers Valve-001 to Valve-006, fan frequency Distributed to variable frequency fans Fan-001 and Fan-002.

[0238] The data structure for parsing the modified atmosphere control strategy instructions includes four fields: device number, control parameter name, target setpoint, and target setpoint unit.

[0239] Preferably, valve opening commands are uniformly issued to all valve controllers, and each valve performs the same opening adjustment; frequency commands for variable frequency fans are uniformly issued to all variable frequency fans.

[0240] For example, the global optimal atmosphere control strategy's action output value is (0.78, 0.85, 0.65), which, after analysis, corresponds to a nitrogen production flow rate of 125 Nm³ / h, a valve opening of 85%, and a fan frequency of 32.5 Hz. The command analysis result includes a target exhaust flow rate of 125 Nm³ / h for the nitrogen generator unit N2-001. 3 / h, the target opening degree of valve controllers Valve-001 to Valve-006 is 85%, and the target frequency of variable frequency fans Fan-001 and Fan-002 is 32.5Hz.

[0241] Step S5.2: Convert the parsing result of the controlled atmosphere management strategy instruction into the equipment control instruction corresponding to each controlled atmosphere execution device.

[0242] Preferably, the core of this step is to convert the parsed control parameters into the communication protocol format of each controlled atmosphere execution device, converting the unified format control parameters into protocol instructions that the device can recognize.

[0243] Specifically, the control instructions for the nitrogen generator unit are converted into ModbusTCP protocol write register instructions, with the target register address being the exhaust flow setting register, the data type being a 32-bit floating-point number, and the byte order being big-endian.

[0244] Among them, the valve controller control instructions are also converted into ModbusTCP protocol write register instructions. The target register address is the valve opening setting register, the data type is a 16-bit unsigned integer, and the range 0 to 1000 corresponds to 0% to 100% opening.

[0245] Among them, the variable frequency fan control instructions are also converted into Modbus TCP protocol write register instructions, the target register address is the frequency setting register, the data type is a 16-bit unsigned integer, and the range 0 to 500 corresponds to 0Hz to 50Hz.

[0246] For instruction verification, each control instruction includes a CRC16 checksum, and the check polynomial is: If the receiving end fails the verification, it discards the instruction and requests a retransmission.

[0247] Preferably, each device control instruction includes six fields: device address code, function code, register start address, number of data words, number of data bytes, and CRC checksum; the instruction sending timeout is 500 milliseconds, and the maximum number of retransmissions is 3.

[0248] For example, the target opening degree of Valve-001 valve controller is 85%. The converted Modbus TCP instruction is: device address 01, function code 06, register address 0064H, data value 0348H (decimal 850, corresponding to the range of 0 to 1000 for 85% opening degree), CRC check code XXXX.

[0249] Step S5.3: Send the control commands of each device to each controlled atmosphere execution device through the industrial IoT gateway.

[0250] Preferably, the core of this step is to reliably transmit equipment control commands to each controlled atmosphere execution device via the industrial IoT gateway, and to issue commands through the protocol conversion and routing functions of the industrial IoT gateway.

[0251] For the industrial IoT gateway, an industrial-grade gateway that supports ModbusTCP to ModbusRTU protocol conversion is adopted. The downlink interface is RS485 bus, the uplink interface is Ethernet, and the communication rate is configurable from 9600bps to 115200bps.

[0252] Specifically, the instruction sending strategy involves sending instructions sequentially according to the device address code, with an interval of no less than 50 milliseconds between adjacent instructions to avoid bus conflicts; all instructions are sent within a single round of transmission, and the single round of transmission does not exceed 2 seconds.

[0253] Specifically, for the confirmation mechanism, after each instruction is sent, the system waits for a confirmation frame from the device. The confirmation frame contains the device address code and function code. If no confirmation frame is received within 500 milliseconds, the instruction is retransmitted, up to a maximum of 3 times. If no confirmation frame is received after 3 retransmissions, the device is marked as having a communication error and the error is reported.

[0254] Preferably, the industrial IoT gateway is connected to the data center via Ethernet with a bandwidth of not less than 100Mbps; the gateway is connected to the controlled atmosphere actuator via RS485 bus with a bus length of not more than 1200 meters.

[0255] For example, 7 equipment control commands (1 nitrogen generator + 6 valve controllers + 2 variable frequency fans, a total of 9 commands) are sent sequentially according to the equipment address codes, with a total time of approximately 450 milliseconds. All equipment returns an acknowledgment frame within 500 milliseconds, indicating that the commands were successfully sent.

[0256] Step S5.4: In response to the response data returned by each controlled atmosphere actuator after receiving the control command from each device and performing the corresponding operation, execution feedback data is obtained.

[0257] Preferably, the core of this step is to collect the actual operating status data of each controlled atmosphere execution device after executing the control command, and to transmit the device response data back through the industrial IoT gateway to form an execution feedback closed loop.

[0258] Among them, for the response data of the execution equipment, the nitrogen generator unit returns the actual exhaust pressure, actual exhaust flow, actual motor speed, and actual motor current; the valve controller returns the actual valve opening and actual valve position feedback; the variable frequency fan returns the actual operating frequency, actual operating current, and actual duct pressure; and the sealing detector returns the actual sealing cover pressure difference and actual leakage rate.

[0259] Specifically, for the execution feedback data format, each execution feedback data entry includes five fields: device number, feedback parameter name, actual execution value, actual execution value unit, and feedback timestamp.

[0260] Regarding the timing of feedback data collection, after the device control command is issued, the device is allowed to stabilize for 5 seconds. After 5 seconds, the actual operating parameters of each device are collected as execution feedback data.

[0261] Preferably, the sampling frequency of the execution feedback data is consistent with the sampling frequency of the device status acquisition module, that is, 10 times per second; the execution feedback data acquisition cycle after a single round of instruction issuance is 5 seconds plus 0.1 seconds of acquisition time.

[0262] For example, the nitrogen generator unit N2-001 receives an exhaust flow rate of 125 Nm³. 3 Five seconds after the / h command, the actual exhaust flow rate of 122 Nm was returned. 3 / h, actual exhaust pressure 0.42MPa, actual motor speed 1485rpm, actual motor current 15.2A, execution feedback data recorded as (N2-001, exhaust flow, 122, Nm 3 / h,YYYY-MM-DD-HH:MM:SS-005).

[0263] Beneficial effects of steps S5.1 to S5.4: This series of steps, through four stages—instruction format parsing, equipment control instruction conversion, industrial IoT gateway transmission, and execution feedback data acquisition—transforms the globally optimal controlled atmosphere management strategy into standardized executable control instructions and distributes them to each controlled atmosphere execution device. This overcomes the fragmented shortcomings of existing methods, such as independent stages, lack of a unified management platform, and inability to link status in real time, thus achieving precise execution of the management strategy and a closed loop of equipment response.

[0264] Specifically, step S5.1 parses the instruction format of the global optimal controlled atmosphere management strategy and extracts the control parameters of each controlled atmosphere device; step S5.2 converts the parsing results into device control instructions corresponding to each controlled atmosphere execution device; step S5.3 sends the device control instructions through the industrial IoT gateway; and step S5.4 obtains the execution feedback data of each controlled atmosphere execution device.

[0265] Step S6: Iterate the global optimal atmosphere control strategy by analyzing the execution deviation between the execution feedback data and the global optimal atmosphere control strategy, and then transmit all execution feedback data and the iterated control strategy back to the full-process digital twin to complete dynamic synchronization and full-process closed-loop control.

[0266] Furthermore, step S6 specifically includes the following steps: Step S6.1: Obtain the setting control parameters of each atmosphere control device in the global optimal atmosphere control strategy, and integrate them to obtain the strategy setting parameter set.

[0267] Preferably, the core of this step is to extract and integrate the set control parameters of each air conditioning device in the global optimal air conditioning control strategy, so as to provide a set value benchmark for subsequent deviation calculation.

[0268] Specifically, for the strategy setting parameter set, the action output value of the globally optimal air conditioning control strategy is used. Extract the target set value of each controlled atmosphere device. The data structure of the strategy setting parameter set includes four fields: device number, control parameter name, target set value, and target set value unit.

[0269] Preferably, the set of strategy setting parameters is consistent with the parsing result of the air conditioning control strategy instruction in step S5.1, and can be directly reused.

[0270] For example, the output value of the globally optimal atmospheric regulation control strategy is (0.78, 0.85, 0.65), and the strategy setting parameter set includes a nitrogen production flow rate of 125 Nm³. 3 / h, valve opening 85%, fan frequency 32.5Hz.

[0271] Step S6.2: Compare each execution feedback data with the set of strategy settings item by item, calculate the execution deviation between the actual execution value and the set parameter value of each controlled atmosphere device, and obtain the execution deviation dataset.

[0272] Preferably, the core of this step is to compare the actual execution values ​​of each controlled atmosphere execution device with the strategy setting values ​​and calculate the deviation, quantify the execution accuracy of the control strategy, and identify devices and parameters with large deviations.

[0273] Among them, the formula for calculating the execution deviation is: .

[0274] in, The percentage of execution deviation for the i-th control parameter. This is the actual executed value. Set a value for the target.

[0275] The execution deviation dataset contains six fields: device number, control parameter name, target set value, actual execution value, execution deviation percentage, and deviation level.

[0276] Among them, the deviation level classification is as follows: a deviation percentage of no more than 2% is considered normal deviation, 2% to 5% is considered slight deviation, 5% to 10% is considered moderate deviation, and more than 10% is considered serious deviation.

[0277] Preferably, the execution deviation dataset is arranged in descending order of deviation level, with severe deviations listed first for priority processing.

[0278] For example, the target setpoint for the exhaust flow rate of the nitrogen generator unit N2-001 is 125 Nm. 3 / h, actual executed value 122Nm 3 / h, execution deviation percentage is The deviation level is slight deviation; the target setting value for the valve controller Valve-001 opening is 85%, the actual execution value is 84%, the execution deviation percentage is 1.2%, and the deviation level is normal deviation.

[0279] Step S6.3: Calculate the correction amount of each control parameter of the globally optimal air conditioning control strategy based on the execution deviation dataset.

[0280] Preferably, the core of this step is to calculate the correction amount of each control parameter. Based on the execution deviation, a proportional-integral correction strategy is adopted to calculate the correction amount of each parameter for the parameter update of the next round of control strategy.

[0281] Specifically, a proportional correction strategy is adopted for the formula used to calculate the correction amount: .

[0282] in, Let be the correction amount for the i-th control parameter. This is a proportional correction factor, with a value of 0.5, meaning the correction amount is 50% of the deviation value, to avoid over-correction leading to oscillation.

[0283] Regarding the correction limit, the correction amount in a single instance shall not exceed 10% of the target set value, i.e. Any corrections exceeding the limit will be truncated to 10%.

[0284] Specifically, for handling severe deviations, when the execution deviation exceeds 10%, the correction factor is adjusted. Increase it to 0.8 to accelerate convergence to the target value.

[0285] Preferably, the correction calculation is performed independently for each control parameter of each controlled atmosphere device; the corrected parameter values ​​are used to update the strategy setting parameter set.

[0286] For example, the exhaust flow deviation of nitrogen generator unit N2-001 is 3 Nm. 3 / h, correction factor 0.5, correction amount is 0.5×3=1.5Nm 3 / h; The corrected target setting is 125 + 1.5 = 126.5 Nm 3 / h; Valve-001 valve controller opening deviation is 1%, correction amount is 0.5×1%=0.5%, after correction the target setting value is 85%+0.5%=85.5%.

[0287] Step S6.4: Update the strategy setting parameter set based on the correction amount of each control parameter to obtain the iterative control strategy parameter set.

[0288] Preferably, the core of this step is to update the strategy setting parameter set, add the correction amount to the original strategy setting parameter set, and generate the iterative control strategy parameter set for the next round of control command issuance.

[0289] For the parameter update formula: .

[0290] in, Let be the set value of the i-th control parameter after iteration. Set the value for the original strategy. This is the correction amount calculated in step S6.3.

[0291] Among these, regarding parameter constraints, the setpoints of each control parameter after iteration must meet the physical constraints: nitrogen production flow rate range 0 to 160 Nm³. 3 / h, valve opening range 0% to 100%, fan frequency range 0Hz to 50Hz; parameter values ​​exceeding the constraint range are clipped to boundary values.

[0292] The data structure for the post-iteration control strategy parameter set includes five fields: device number, control parameter name, post-iteration set value, post-iteration set value unit, and iteration round.

[0293] Preferably, the iteration number is incremented by 1 after each iteration update; the control strategy parameter set after each iteration will be used as the input for the globally optimal air conditioning control strategy in the next step S5.

[0294] For example, the original set value for nitrogen production flow rate is 125 Nm. 3 / h, correction amount 1.5Nm 3 / h, the set value after iteration is 126.5Nm 3 / h; Valve opening original setting value 85%, correction amount 0.5%, after iteration setting value 85.5%; Fan frequency original setting value 32.5Hz, correction amount 0Hz (no correction if deviation is within the normal range), after iteration setting value 32.5Hz; The iteration round is the 3rd round.

[0295] Step S6.5: The execution feedback data and the set of control strategy parameters after iteration are sent back to the full-process digital twin, triggering the synchronous update of the model parameters of the full-process digital twin.

[0296] Preferably, the core of this step is to transmit the execution feedback data and the set of control strategy parameters after iteration back to the full-process digital twin, triggering the synchronous update of environmental parameters, equipment status, and pest distribution in the digital twin, so as to ensure that the virtual space and the physical space are consistent.

[0297] Specifically, for the feedback data transmission, the actual execution values ​​of each controlled atmosphere execution device are updated to the corresponding 3D model parameters of the device in the full-process digital twin according to the device number, and the 3D virtual mapping results of the device status are updated.

[0298] Specifically, for the feedback of the control strategy parameter set after iteration, the set values ​​of each control parameter after iteration are updated to the control parameter tags of the corresponding devices in the full-process digital twin, and the target operating status of the devices in the basic digital twin model is updated synchronously.

[0299] Among them, for the synchronous update of environmental parameters, based on the new control parameters corresponding to the parameter set of the iterative control strategy, the physical simulation module in the digital twin basic model is used to predict the changing trend of environmental parameters in the storage space, and update the isosurfaces of oxygen concentration, temperature, and humidity in the three-dimensional virtual mapping results of the storage space.

[0300] Regarding the synchronization update frequency, a full-process digital twin synchronization update is triggered after each step S6 is completed, and the synchronization update calculation time does not exceed 2 seconds.

[0301] Preferably, after the digital twin is updated synchronously throughout the entire process, the real-time status parameter set of the digital twin is updated accordingly for input in the next step S4.

[0302] For example, the N2-001 nitrogen generator unit was upgraded to a flow rate of 126.5 Nm³. 3 The exhaust flow rate is transmitted back to the full-process digital twin at / h, and the numerical label of exhaust flow rate in the equipment's 3D model is changed from 125Nm. 3 / h updated to 126.5Nm 3 / h; The physical simulation module predicts that the oxygen concentration will decrease by 0.05% in the next 10 minutes, and the color of the corresponding area in the three-dimensional virtual mapping result of the storage space transitions from light blue to dark blue.

[0303] Step S6.6: Determine whether the preset termination conditions have been met for the closed-loop control of the entire process. If they have been met, output the final control result. If not, return to step S5 to continue execution and complete the closed-loop control of the entire process.

[0304] Preferably, the core of this step is to determine whether to terminate the closed-loop management process by monitoring the achievement of pest control targets and the maximum number of closed-loop cycles.

[0305] The termination condition is determined by a combination of two conditions: Condition 1 is that the actual oxygen concentration value of all oxygen concentration sensors is lower than 0.5% (insect control meets the standard), and the count of the insect density detection device is lower than the warning threshold; Condition 2 is that the closed-loop control rounds reach the preset maximum rounds, which are set according to the insect control operation cycle, and are 360 ​​rounds (corresponding to 1 hour, 10 seconds per round).

[0306] The final control results output includes three parts: a snapshot of the final state of the digital twin of the entire process, the final operating parameters of each controlled atmosphere equipment, and an evaluation report on the effectiveness of pest control operations. The evaluation report on the effectiveness of pest control operations includes four indicators: operation start and end time, oxygen concentration compliance rate, pest density change rate, and energy consumption statistics.

[0307] For cases where the termination condition is not met, the process returns to step S5, using the iterated control strategy parameter set as the input for a new globally optimal atmospheric regulation control strategy, and continues to execute the instruction issuance and execution feedback closed loop.

[0308] Preferably, the oxygen concentration compliance rate requirement in the termination condition is that all 18 oxygen concentration sensors are below 0.5%, and partial compliance is not allowed to terminate the process; the insect density warning threshold is consistent with the setting in step S3.5.

[0309] For example, after the 320th round of closed-loop management, the actual values ​​of all 18 oxygen concentration sensors were below 0.5%, and the counts of all 6 insect density detection devices were below the warning threshold, thus meeting termination condition one and outputting the final management result. The insect control operation effect evaluation report showed that the operation start and end time was from YYYY-MM-DD-HH:MM:SS-000 to YYYY-MM-DD-HH:MM:SS+3200s, the oxygen concentration compliance rate was 100%, the insect density reduction rate was 75%, and the total energy consumption of the nitrogen generator unit was 18.5kWh.

[0310] Beneficial effects of steps S6.1 to S6.6: This series of steps involves six stages: obtaining the strategy setting parameter set, calculating the execution deviation, calculating the correction amount, updating the strategy parameters, synchronizing the digital twin, and determining the closed-loop termination. By iterating the deviation between the execution feedback data and the globally optimal air conditioning control strategy, the control strategy is adaptively optimized. All execution feedback data and the iterated control strategy are then fed back to the full-process digital twin. This overcomes the shortcomings of existing methods, such as delayed early warning response, disconnection from the WMS system, and inability to interconnect data, thus forming a full-process closed-loop control and traceability system.

[0311] The process includes the following steps: Step S6.1: Obtain the set control parameters of each controlled atmosphere device in the global optimal controlled atmosphere management strategy; Step S6.2: Calculate the execution deviation between the actual execution value and the set parameter value of each controlled atmosphere device; Step S6.3: Calculate the correction amount of each control parameter based on the execution deviation; Step S6.4: Update the strategy set parameter set to obtain the iterative control strategy parameter set; Step S6.5: Send the execution feedback data and the iterative control strategy parameter set back to the full-process digital twin to trigger the synchronous update of model parameters; Step S6.6: Determine whether the full-process closed-loop management has reached the preset termination condition and complete the full-process closed-loop management.

[0312] In summary, the beneficial effects of steps S1 to S6 in this embodiment are as follows: This method forms a complete technical closed loop of digital twin control for the entire process of low-oxygen controlled pest control through steps S1 to S6, effectively overcoming the technical problems of fragmented process, lack of status linkage, lack of visualization, single monitoring dimension, delayed early warning, strong blind control, and disconnection from WMS system in the existing low-oxygen controlled pest control.

[0313] Step S1 involves constructing a multi-source real-time time-series dataset to achieve unified collection of status data for each stage of nitrogen production, pipeline network, sealing, and oxygen reduction, providing a data foundation for full-process control. Step S2 uses extended Kalman filtering to achieve multi-source data fusion and noise filtering, improving the reliability of monitoring data. Step S3 involves constructing a full-process digital twin to achieve three-dimensional virtual mapping and visualized control of storage space, controlled atmosphere equipment, and pest status. Step S4 uses a deep deterministic strategy gradient algorithm to achieve multi-objective dynamic optimization of controlled atmosphere parameters, eliminating the blindness of setting parameters based on experience. Step S5 completes the conversion and issuance of optimal strategies to equipment instructions, achieving precise execution of control strategies and linkage with the WMS system. Step S6 uses execution deviation iteration and synchronization with the digital twin to form a closed-loop control and adaptive optimization mechanism for the entire process.

[0314] like Figure 2 As shown, this embodiment provides an example of a full-process digital twin control system for low-oxygen controlled insecticide. In this embodiment, the full-process digital twin control system is applied to the full-process digital twin control method described above.

[0315] Specifically, the end-to-end digital twin control system includes a multi-source real-time time-series data acquisition module 1, a multi-source real-time time-series data calibration module 2, an end-to-end digital twin construction module 3, an optimal controlled atmosphere control strategy iteration module 4, an optimal controlled atmosphere control strategy execution module 5, and an optimal controlled atmosphere control strategy correction module 6, which are connected electrically or through communication in sequence.

[0316] The system comprises the following modules: Multi-source real-time time-series data acquisition module 1 collects oxygen concentration, temperature, humidity, insect density, and controlled atmosphere equipment operating status within the storage space, constructing a multi-source real-time time-series dataset; Multi-source real-time time-series data calibration module 2 inputs the multi-source real-time time-series dataset into an extended Kalman filter algorithm for multi-source data fusion and random noise filtering, generating calibrated multi-source time-series data; Full-process digital twin construction module 3 acquires a digital twin base model of the storage space and inputs the calibrated multi-source time-series data into the digital twin base model to perform a three-dimensional virtual mapping of the storage space, controlled atmosphere equipment, and insect status, obtaining a full-process digital twin; Optimal controlled atmosphere control strategy iteration module 4 is used to... The end-to-end digital twin incorporates a deep deterministic strategy gradient algorithm to perform multi-objective optimization iterations of controlled atmosphere parameters, resulting in a globally optimal controlled atmosphere control strategy. The optimal controlled atmosphere control strategy execution module 5 converts the globally optimal controlled atmosphere control strategy into standardized executable control commands for the controlled atmosphere equipment, which are then distributed to each controlled atmosphere execution device in the warehouse space via an industrial IoT gateway, and the execution feedback data of each controlled atmosphere execution device is obtained. The optimal controlled atmosphere control strategy correction module 6 iterates the globally optimal controlled atmosphere control strategy based on the execution deviation between the execution feedback data and the globally optimal controlled atmosphere control strategy, and transmits all execution feedback data and the iterated control strategy back to the end-to-end digital twin, completing dynamic synchronization and end-to-end closed-loop control.

[0317] like Figure 3 As shown, the electronic device 7 includes a processor 71 and a memory 72 coupled to the processor 71.

[0318] The memory 72 stores program instructions for implementing the full-process digital twin control method for low-oxygen controlled insecticide in any of the above embodiments.

[0319] The processor 71 is used to execute program instructions stored in the memory 72 to perform full-process digital twin control of low-oxygen controlled insecticide.

[0320] The processor 71 can also be referred to as a CPU (Central Processing Unit). The processor 71 may be an integrated circuit chip with signal processing capabilities. The processor 71 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor.

[0321] Furthermore, Figure 4 This is a schematic diagram of the structure of a storage medium according to an embodiment of this application. See also: Figure 4 The storage medium 8 in this embodiment stores program instructions 81 capable of implementing all the above methods. These program instructions 81 can be stored in the storage medium as a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods in each embodiment of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0322] In the several embodiments provided in this application, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the end-to-end digital twin control of a unit is only one type of logical function end-to-end digital twin control. In actual implementation, there may be other end-to-end digital twin control methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces. The indirect coupling or communication connection between systems or units may be electrical, mechanical, signal, or other forms.

[0323] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A digital twin management method for the entire process of low-oxygen controlled insecticide application, characterized in that, The full-process digital twin control method includes: Step S1: Collect oxygen concentration, temperature, humidity, insect density, and controlled atmosphere equipment operating status in the storage space, and construct a multi-source real-time time series dataset. Step S2: Input the multi-source real-time time series dataset into the extended Kalman filter algorithm to perform multi-source data fusion and random noise filtering to generate calibrated multi-source time series data; Step S3: Obtain the digital twin basic model of the storage space, and input the calibrated multi-source time series data into the digital twin basic model to perform three-dimensional virtual mapping of the storage space, controlled atmosphere equipment, and insect infestation status, and obtain a full-process digital twin. Step S4: The full-process digital twin is connected to the deep deterministic strategy gradient algorithm to perform multi-objective optimization iteration of the air conditioning parameters to obtain the globally optimal air conditioning control strategy. Step S5: Convert the global optimal controlled atmosphere management strategy into standardized executable control commands for the controlled atmosphere equipment, send them to each controlled atmosphere execution device in the warehouse space through the industrial IoT gateway, and obtain the execution feedback data of each controlled atmosphere execution device. Step S6: Iterate the global optimal atmosphere control strategy based on the execution deviation between the execution feedback data and the global optimal atmosphere control strategy, and send all execution feedback data and the iterated control strategy back to the full-process digital twin to complete dynamic synchronization and full-process closed-loop control.

2. The end-to-end digital twin control method according to claim 1, characterized in that, Step S1: Collect data on oxygen concentration, temperature, humidity, insect density, and controlled atmosphere equipment operation status within the storage space, and construct a multi-source real-time time-series dataset, including: Step S1.1: Activate the multi-type data acquisition nodes deployed in the storage space for oxygen, temperature and humidity, insect infestation, and equipment status; Step S1.2: Extract the raw data signals from each data acquisition node through the real-time data acquisition channel at a fixed sampling frequency; Step S1.3: Perform analog-to-digital conversion and level standardization on all raw data signals to generate raw sensing data in a unified format; Step S1.4: Classify and collect all raw sensing data according to five dimensions: oxygen concentration, temperature, humidity, insect density, and equipment status. Step S1.5: Perform time-series alignment and data preprocessing on the raw sensor data after classification and aggregation to generate a multi-source real-time time-series dataset.

3. The end-to-end digital twin control method according to claim 1, characterized in that, Step S2 involves inputting the multi-source real-time time-series dataset into an extended Kalman filter algorithm for multi-source data fusion and random noise filtering to generate calibrated multi-source time-series data, including: Step S2.1: Input the multi-source real-time time series dataset into the state prediction unit of the extended Kalman filter algorithm in the order of timestamps to calculate the prior state estimate of each dimension of the data; Step S2.2: Input the prior state estimates of each dimension of data into the covariance prediction unit of the extended Kalman filter algorithm to calculate the prior error covariance matrix; Step S2.3: Input the prior error covariance matrix into the gain calculation unit of the extended Kalman filter algorithm to calculate the optimal Kalman gain matrix; Step S2.4: Input the optimal Kalman gain matrix and the prior state estimates of each dimension of data into the state update unit to calculate the posterior state estimates; Step S2.5: The posterior state estimate is split and recombined according to the original data dimension to obtain calibrated multi-source time series data.

4. The end-to-end digital twin control method according to claim 1, characterized in that, Step S3: Obtain the digital twin base model of the storage space, and input the calibrated multi-source time-series data into the digital twin base model to perform a three-dimensional virtual mapping of the storage space, controlled atmosphere equipment, and pest situation, thereby obtaining a full-process digital twin, including: Step S3.1: Convert the calibrated multi-source time-series data to the unified spatial coordinate system of the digital twin basic model to obtain the coordinate-transformed spatial data; Step S3.2: Based on the digital twin basic model, establish the actual positional relationship between each sensor and the controlled atmosphere equipment in the warehouse space, and generate a spatial position mapping table; Step S3.3: Input the spatial data after coordinate transformation and the spatial location mapping table into the digital twin basic model, perform three-dimensional virtual mapping of the warehouse space, and generate the three-dimensional virtual mapping result of the warehouse space; Step S3.4: Input the controlled atmosphere equipment operating status data in the spatial data after coordinate transformation and the spatial position mapping table into the digital twin basic model, perform three-dimensional virtual mapping of the controlled atmosphere equipment status, and generate the three-dimensional virtual mapping result of the controlled atmosphere equipment status; Step S3.5: Input the insect infestation status data in the spatial data after coordinate transformation and the spatial location mapping table into the digital twin basic model, perform three-dimensional virtual mapping of insect infestation status, and generate three-dimensional virtual mapping result of insect infestation status; Step S3.6: Integrate the three-dimensional virtual mapping results of the storage space, the three-dimensional virtual mapping results of the controlled atmosphere equipment status, and the three-dimensional virtual mapping results of the insect infestation status to generate a full-process digital twin.

5. The end-to-end digital twin control method according to claim 1, characterized in that, Step S4 involves integrating the full-process digital twin with a deep deterministic strategy gradient algorithm to perform multi-objective optimization iterations of the controlled atmosphere parameters, thereby obtaining the globally optimal controlled atmosphere management strategy, including: Step S4.1: Extract real-time status parameters of each monitoring dimension from the full-process digital twin and integrate them to obtain a set of real-time status parameters of the digital twin; Step S4.2: Input the real-time state parameter set of the digital twin into the state space definition unit of the deep deterministic policy gradient algorithm, and generate the optimization problem definition by configuring the state space dimension and the climate regulation parameter optimization objective; Step S4.3: Initialize the policy network parameters and value network parameters of the deep deterministic policy gradient algorithm based on the definition of the optimization problem, and complete the algorithm initialization; Step S4.4: Input the real-time state parameter set of the digital twin into the initialized deep deterministic strategy gradient algorithm to calculate the action output value of the weather regulation parameter; Step S4.5: Input the action output value and the real-time state parameter set of the digital twin into the value network to calculate the cumulative reward estimate of the current state action pair; Step S4.6: Based on the cumulative return estimate, perform policy gradient update and value network parameter update to achieve multi-objective optimization iteration; Step S4.7: Determine whether the multi-objective optimization iteration meets the convergence condition. If it does, output the globally optimal climate control strategy. If it does not, return to step S4.4 to continue iterating until the convergence condition is met.

6. The end-to-end digital twin control method according to claim 1, characterized in that, Step S5: The globally optimal controlled atmosphere management strategy is converted into standardized executable control commands for the controlled atmosphere equipment, and then sent to each controlled atmosphere execution device in the warehouse space through the industrial IoT gateway. Execution feedback data from each controlled atmosphere execution device is also obtained, including: Step S5.1: Parse the instruction format of the global optimal atmosphere control strategy, extract the control parameters of each atmosphere control device, and obtain the instruction parsing result of the atmosphere control strategy. Step S5.2: Convert the parsing result of the controlled atmosphere management strategy instruction into the equipment control instruction corresponding to each controlled atmosphere execution device; Step S5.3: Send the control commands of each device to each controlled atmosphere execution device through the industrial IoT gateway; Step S5.4: In response to the response data returned by each controlled atmosphere actuator after receiving the control command from each device and performing the corresponding operation, execution feedback data is obtained.

7. The end-to-end digital twin control method according to claim 1, characterized in that, Step S6: Iterate the global optimal atmosphere control strategy based on the execution deviation between the execution feedback data and the global optimal atmosphere control strategy, and send all execution feedback data and the iterated control strategy back to the full-process digital twin to complete dynamic synchronization and full-process closed-loop control, including: Step S6.1: Obtain the setting control parameters of each atmosphere control device in the global optimal atmosphere control strategy, and integrate them to obtain the strategy setting parameter set; Step S6.2: Compare each execution feedback data with the set of strategy settings item by item, calculate the execution deviation between the actual execution value and the set parameter value of each controlled atmosphere device, and obtain the execution deviation dataset; Step S6.3: Calculate the correction amount of each control parameter of the globally optimal air conditioning control strategy based on the execution deviation dataset; Step S6.4: Update the strategy setting parameter set based on the correction amount of each control parameter to obtain the iterative control strategy parameter set; Step S6.5: Send the execution feedback data and the set of control strategy parameters after iteration back to the full-process digital twin, triggering the synchronous update of the model parameters of the full-process digital twin; Step S6.6: Determine whether the preset termination conditions have been met for the closed-loop control of the entire process. If they have been met, output the final control result. If not, return to step S5 to continue execution and complete the closed-loop control of the entire process.

8. A full-process digital twin control system for low-oxygen controlled insecticide, wherein the full-process digital twin control system is applied to the full-process digital twin control method as described in any one of claims 1 to 7, characterized in that, The end-to-end digital twin management and control system includes: The multi-source real-time time-series data acquisition module is used to collect oxygen concentration, temperature, humidity, insect density, and controlled atmosphere equipment operating status in the storage space, and construct a multi-source real-time time-series dataset. The multi-source real-time time series data calibration module is used to input the multi-source real-time time series dataset into the extended Kalman filter algorithm to perform multi-source data fusion and random noise filtering, and generate calibrated multi-source time series data; The full-process digital twin construction module is used to obtain the digital twin basic model of the storage space and input the calibrated multi-source time series data into the digital twin basic model to perform three-dimensional virtual mapping of the storage space, controlled atmosphere equipment, and insect infestation status, thereby obtaining the full-process digital twin. The optimal atmosphere control strategy iteration module is used to connect the full-process digital twin to the deep deterministic strategy gradient algorithm to perform multi-objective optimization iteration of atmosphere control parameters, so as to obtain the global optimal atmosphere control strategy. The optimal controlled atmosphere control strategy execution module is used to convert the global optimal controlled atmosphere control strategy into standardized executable control commands for the controlled atmosphere equipment, and send them to each controlled atmosphere execution device in the warehouse space through the industrial Internet of Things gateway, and obtain the execution feedback data of each controlled atmosphere execution device. The optimal atmosphere control strategy correction module is used to iterate the global optimal atmosphere control strategy based on the execution deviation between the execution feedback data and the global optimal atmosphere control strategy, and to send all execution feedback data and the iterated control strategy back to the full-process digital twin to complete dynamic synchronization and full-process closed-loop control.

9. An electronic device, characterized in that, The method includes a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the end-to-end digital twin control method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions that, when executed by a processor, enable the implementation of the end-to-end digital twin control method as described in any one of claims 1 to 7.

Images