Intelligent preservation method and system for food based on digital twinning

Abstract

This invention discloses an intelligent food preservation method and system based on digital twins. The invention relates to the field of digital twin technology. It identifies corresponding dynamic preservation events based on the recognition of preservation data combinations of the target food; determines multiple preservation items for the target food based on semantic parsing of the dynamic preservation events; and dynamically extrapolates these items in the digital twin space by combining the morphological and surface features of the target food, improving the accuracy of optimization content for the target food during dynamic preservation. It also marks the optimization content for secondary foods during preservation and performs spatiotemporal alignment with the optimization content of the target food to output a corresponding preservation twin. This allows for the acquisition of faulty components in the food preservation chamber, the determination of corresponding preservation impact content based on the preservation twin, and the fusion of preservation compensation content and the current preservation status of the target food to determine backup work content for the food preservation chamber, thereby improving the accuracy of the dynamic work content of the food preservation chamber.

Description

Technical Field

[0001] This invention relates to the technical field of digital twins, and more particularly to a smart food preservation method and system based on digital twins. Background Technology

[0002] With the continuous development of cold chain logistics and warehousing technology, food preservation rooms have been widely used in the post-harvest storage of high-value-added foods such as fruits, vegetables, and meats. These food preservation rooms include controlled atmosphere storage and cold storage. In order to maintain the quality of food, existing technologies are gradually developing from single temperature control to multi-dimensional environmental regulation of temperature, humidity, gas composition, and other aspects.

[0003] Existing food preservation room monitoring systems typically use sensors arranged at single points or sparsely in space to acquire temperature, humidity, and gas data. This macroscopic data acquisition cannot accurately characterize the true physical field of the microenvironment in which the target food is located, such as local wind speed gradients and gas field distribution. At the same time, when performing image recognition, existing technologies often only perform coarse identification of surface defects and fail to perform high-precision cross-modal fusion of multidimensional environmental data and food images to determine the combination of preservation data. This results in the system being unable to keenly capture "dynamic preservation events" triggered by complex environmental changes, affecting the accuracy of the content to be optimized in the dynamic preservation process of the target food, and leading to low accuracy of the dynamic working content of the food preservation room. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a smart food preservation method and system based on digital twins.

[0005] This invention provides an intelligent food preservation method based on digital twins, comprising: The preservation areas of the food preservation room are marked, multi-dimensional preservation data of the preservation areas are obtained, and the preservation data combination of the target food is determined by combining the images of the preservation areas. Based on the identification of the preservation data combination of the target food, the corresponding dynamic preservation event is determined. Based on the semantic analysis of the dynamic preservation event, multiple preservation items of the target food are determined, and dynamic inference is carried out in the digital twin space in combination with the morphological and surface features of the target food, so as to determine the content to be optimized in the dynamic preservation process of the target food. Obtain secondary food items from the same preservation area, mark the optimization content of secondary food items during the preservation process, and perform spatiotemporal alignment with the optimization content of target food items. Introduce working data from the food preservation room for multi-factor fusion to output the corresponding preservation twin and the corresponding preservation compensation content. The system identifies faulty components in the food preservation compartment, determines the corresponding preservation impact based on the preservation twin, further integrates preservation compensation content and the current preservation status of the target food to determine backup work content for the food preservation compartment, and triggers dynamic work content for the food preservation compartment based on changes in the morphology of the target food.

[0006] This invention provides a digital twin-based intelligent food preservation system, which is applied to the aforementioned digital twin-based intelligent food preservation method. The digital twin-based intelligent food preservation system includes: The data processing module is used to mark the preservation area of ​​the food preservation room, acquire multi-dimensional preservation data of the preservation area, and determine the preservation data combination of the target food by combining the image of the preservation area. Based on the identification of the preservation data combination of the target food, the corresponding dynamic preservation event is determined. The digital twin module is used to determine multiple preservation items of the target food based on the semantic parsing of the dynamic preservation event, and to perform dynamic inference in the digital twin space in combination with the morphological and surface features of the target food, thereby determining the content to be optimized in the dynamic preservation process of the target food. The preservation compensation content module is used to acquire secondary food items in the same preservation area, mark the content to be optimized for secondary food items during the preservation process, and perform spatiotemporal alignment with the content to be optimized for target food items. It also incorporates working data from the food preservation room for multi-factor fusion to output the corresponding preservation twin and the corresponding preservation compensation content. The dynamic working module is used to acquire faulty components of the food preservation compartment, determine the corresponding preservation impact content by combining the preservation twin, further integrate the preservation compensation content and the current preservation status of the target food to determine the backup working content of the food preservation compartment, and trigger the dynamic working content of the food preservation compartment by combining the morphological changes of the target food.

[0007] Compared with the prior art, the beneficial effects of the present invention are: (1) Mark the preservation area of ​​the food preservation room, obtain multi-dimensional preservation data of the preservation area, and determine the preservation data combination of the target food by combining the image of the preservation area. Based on the recognition of the preservation data combination of the target food, determine the corresponding dynamic preservation event. Based on the semantic parsing of the dynamic preservation event, determine multiple preservation items of the target food, and combine the morphological features and surface features of the target food to perform dynamic deduction in the digital twin space, thereby determining the content to be optimized of the target food in the dynamic preservation process. Dynamic preservation events are introduced to further control multiple preservation items and improve the accuracy of the content to be optimized of the target food in the dynamic preservation process.

[0008] (2) Obtain secondary food items in the same preservation area and mark the content to be optimized in the preservation process of secondary food items. Combine the content to be optimized of target food items for spatiotemporal alignment. Introduce the working data combination of food preservation room for multi-factor fusion to output the corresponding preservation twin and output the corresponding preservation compensation content. Obtain the faulty parts of food preservation room and determine the corresponding preservation impact content in combination with the preservation twin. Further integrate the preservation compensation content and the current preservation status of target food items to determine the backup working content of food preservation room. Combine the morphological change content of target food items to trigger the dynamic working content of food preservation room. Further control the preservation twin, fully consider the preservation impact content, preservation compensation content and the current preservation status of target food items, and improve the accuracy of the dynamic working content of food preservation room. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the intelligent food preservation method based on digital twins in an embodiment of the present invention. Figure 2 This is a flowchart illustrating step S11 of the intelligent food preservation method based on digital twins in an embodiment of the present invention. Figure 3 This is a flowchart illustrating step S12 in the intelligent food preservation method based on digital twins according to an embodiment of the present invention. Figure 4 This is a flowchart illustrating step S13 in the intelligent food preservation method based on digital twins in an embodiment of the present invention. Figure 5 This is a flowchart illustrating step S14 of the intelligent food preservation method based on digital twins in an embodiment of the present invention. Figure 6 This is a schematic diagram of the structural composition of an intelligent food preservation system based on digital twins according to an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0011] Please see Figures 1 to 6 A smart food preservation method based on digital twins, applied to digital twin scenarios; the smart food preservation method based on digital twins includes: Step S11: Mark the preservation area of ​​the food preservation room, obtain multi-dimensional preservation data of the preservation area, and determine the preservation data combination of the target food by combining the image of the preservation area. Based on the identification of the preservation data combination of the target food, determine the corresponding dynamic preservation event. Step S12: Based on the semantic parsing of the dynamic preservation event, determine multiple preservation items for the target food, and combine the morphological and surface features of the target food to perform dynamic deduction in the digital twin space, thereby determining the content to be optimized for the target food in the dynamic preservation process. Step S13: Obtain secondary food items in the same preservation area, mark the content to be optimized in the preservation process of secondary food items, and perform spatiotemporal alignment with the content to be optimized of target food items. Introduce working data of food preservation room for multi-factor fusion to output the corresponding preservation twin and output the corresponding preservation compensation content. Step S14: Obtain the faulty component of the food preservation compartment, determine the corresponding preservation impact content in combination with the preservation twin, further integrate the preservation compensation content and the current preservation status of the target food to determine the backup work content of the food preservation compartment, and trigger the dynamic work content of the food preservation compartment in combination with the morphological changes of the target food.

[0012] refer to Figure 2 In step S11, the specific steps are as follows: S111: In the food preservation room, the three-dimensional distribution map of the food preservation room is spatially gridded, and the preservation area is topologically marked. Simultaneously, multi-dimensional preservation data of the preservation area is acquired using a sensor array. The multi-dimensional preservation data covers temperature and humidity gradients, gas field distribution, and wind field distribution. The multi-dimensional preservation data of the preservation area and the real-time image of the preservation area are fused across modes to determine the preservation data combination of the target food under a specific environment. S112: Dynamically identify the combination of preservation data, mark mutation features during the identification process, simultaneously obtain the food type of the target food, and further combine the corresponding mutation features to perform reverse tracing, thereby gradually determining multiple dynamic preservation contents at different levels during the tracing process, and determining the corresponding dynamic preservation event based on the fusion of multiple dynamic preservation contents.

[0013] In the embodiments of this application, in the food preservation room, the three-dimensional distribution map of the food preservation room is spatially gridded, and the preservation area is topologically marked. Simultaneously, multi-dimensional preservation data of the preservation area is acquired using a sensor array. The multi-dimensional preservation data covers temperature and humidity gradients, gas field distribution, and wind field distribution. The multi-dimensional preservation data of the preservation area and the real-time image of the preservation area are fused across modes to determine the preservation data combination of the target food under a specific environment. This approach takes into account both the multi-dimensional preservation data and the real-time image of the preservation area, ensuring the accuracy of the preservation data combination of the target food under a specific environment.

[0014] At this point, the system extracts the 3D structural CAD model or three-dimensional distribution map of the food preservation room, and uses voxelization technology to perform 3D spatial meshing of the entire physical space, discretizing it into a set of micro-cube meshes with specific volume resolution. In this process, the system not only assigns a unique spatial coordinate identifier to each mesh, but also introduces graph theory to perform topological labeling on the physical adjacency and connectivity between meshes. This topological labeling process will specifically identify and define the boundary attributes of the mesh, such as accurately distinguishing the cold air jet inlet mesh, the return air inlet mesh, and the blind zone mesh formed by the shelf obstruction, thereby constructing a structured spatial skeleton with a priori constraints of fluid dynamics and thermodynamics in the digital twin space.

[0015] The system drives a high-density distributed sensor array to acquire data synchronously, breaking through the limitations of traditional single-point measurement. Instead, it reconstructs discrete sensor readings into a continuous three-dimensional physical field based on spatial interpolation. Specifically, the system calculates and outputs temperature and humidity gradient fields in real time to characterize spatial thermodynamic stratification, reconstructs gas field distribution to map the concentration thermograms of gases with different specific gravities, and generates wind field distribution to present the velocity vector, turbulence intensity, and eddy distribution location of cold airflow, ultimately forming a high-fidelity multi-dimensional preservation environment matrix.

[0016] Meanwhile, multispectral or high-definition machine vision equipment deployed in the preservation area simultaneously captures real-time images of the current area; the system uses deep learning object detection and semantic segmentation to accurately extract the pixel-level mask of the target food from the complex background and extract its visual features such as appearance texture, color distribution and shape contour; using the camera's intrinsic and extrinsic parameter calibration matrix and depth information, the system precisely projects and maps the target food in the two-dimensional image onto the three-dimensional spatial grid established in the first sub-step, realizing the determination of the absolute coordinates and spatial occupancy declaration of the target food in physical space and twin space.

[0017] By aligning and fusing the multidimensional environmental physical field matrix with the visual features of the target food at the feature level, a cross-modal attention mechanism network is typically used. This allows the visual feature vector and the environmental tensor to interact in the latent space, thereby enabling the environmental data to "focus on the local area of ​​the target food". The fused high-dimensional tensor is then reduced in dimension and decoded to remove background environmental noise that is irrelevant to the target food. Finally, a composite data structure that deeply binds the "appearance state of the target food" with the "micrometer-level local environmental field in which it is located" is accurately output, which is determined as the preservation data combination of the target food under a specific environment.

[0018] Specifically, the food preservation room is a large commercial walk-in high-density dynamic controlled atmosphere preservation warehouse, and the target food is high-value premium blueberries; the system grids the three-dimensional shelf distribution map of the preservation warehouse, dividing the entire warehouse into micro-cube grids with a side length of 10 centimeters; in the topology marking stage, the system clearly marks the grid node (3-F-01) located at the front of the third shelf, which is directly below the central air supply duct of the warehouse and has no physical obstruction, belonging to the "strong convection exposed topology node".

[0019] The sensor array, combined with the simulation model, reconstructed multidimensional data of the area in real time: the temperature and humidity gradient showed that there was a vertical temperature difference of 0.5℃ in the area; the wind field distribution showed that the wind speed vector at grid (3-F-01) was as high as 2.5m / s, which belongs to the strong wind jet zone; the gas field distribution showed that due to the strong wind, the carbon dioxide concentration produced by the blueberries' respiration at this grid was only 800ppm, which is lower than the threshold of optimal modified atmosphere preservation. The vision system captures real-time images at the (3-F-01) grid, accurately outlines the blueberry packaging box through semantic segmentation, and extracts image features: the integrity of the blueberry skin bloom is partially missing, presenting dark reflective spots, and the depth measurement confirms that the blueberry box is located at the geometric center of the (3-F-01) grid.

[0020] The system performs attentional fusion calculations on the environmental physical field tensor of "strong wind speed vector + low CO2 concentration" and the visual feature vector of "local loss of blueberry bloom". The system instantly filters out irrelevant environmental data from distant shelves and directly causally associates "strong wind of 2.5 m / s" with the visual phenomenon of "bloom loss". The system outputs a specific combination of preservation data for this blueberry: {spatial coordinates: 3-F-01; food condition: wind erosion damage to the skin bloom; local environment: abnormally high wind speed of 2.5 m / s, low CO2 concentration of 800 ppm, presenting a typical "forced convection wind erosion microenvironment"}. This high-precision data combination lays an absolute data foundation for subsequent determination of "wind erosion and water loss preservation event".

[0021] Furthermore, the preservation data combination is dynamically identified, and mutation features are marked during the identification process. The food type of the target food is obtained simultaneously, and the corresponding mutation features are combined for reverse tracing. In this way, multiple dynamic preservation contents at different levels are gradually determined during the tracing process. The corresponding dynamic preservation event is determined based on the fusion of multiple dynamic preservation contents. This approach takes into account the overall consideration of the fusion of multiple dynamic preservation contents and ensures the accuracy of the corresponding dynamic preservation event.

[0022] At this point, after receiving the input combination of preservation data, the system immediately places it into a sliding time window and uses a long short-term memory network or a time-series anomaly detection model for highly sensitive dynamic identification. This is not a simple threshold comparison, but a deep analysis of the time-series evolution trajectory and covariance relationship between multiple variables within the data combination. When the system detects that the distribution of one or more dimensions in the data stream deviates from the historical stable baseline and the rate of change exceeds the set dynamic envelope, the system immediately triggers a feature capture mechanism to accurately mark the inflection point data of abnormal fluctuations and define it as a "mutation feature" that includes the time of occurrence, the magnitude of change, and the associated dimensions, which serves as the anchor point for subsequent tracing.

[0023] While capturing mutation characteristics, the system simultaneously retrieves the identity identifier of the target food through a digital twin base, and then accurately retrieves and obtains the "food type" attribute of the target food from the food physical property knowledge graph. The core of this step is to introduce the physicochemical sensitivity prior knowledge unique to this type, such as the freezing point temperature, jump type, water loss rate coefficient, or sensitivity threshold to specific gases of specific fruits and vegetables. The injection of this prior knowledge is equivalent to installing a "physical mechanism filter" for the subsequent traceability process, ensuring that all logical deductions are strictly limited to the biological and physical property boundaries of this type of food, thereby filtering out pseudo-anomalies that have no physical meaning.

[0024] The system activates the causal graph reverse reasoning engine, taking the marked mutation features as the result and combining them with the injected prior knowledge of food types as constraints. It then performs a layer-by-layer reverse peeling and tracing along the preset "appearance-microenvironment-device execution" causal link. The system analyzes the biological significance of the mutation features at the appearance level, and then, based on the sensitive characteristics of the food type, it reversely deduces the microenvironment parameter distortions that cause the appearance, such as abnormal temperature and humidity gradients or excessive gas field concentrations. Finally, it further penetrates to the underlying physical execution logic that causes microenvironment distortions, such as wind field disturbance sources or valve state deviations. In this reverse optimization process, the system gradually determines and outputs multiple "dynamic preservation contents" at different causal levels from the surface to the core.

[0025] The system inputs multiple dynamic preservation contents at different levels obtained through reverse tracing into the multi-source information fusion decision module. This module uses techniques such as Dirichlet allocation or Bayesian networks to evaluate the causal support strength and temporal consistency between contents at each level, eliminates weakly correlated isolated contents, and aggregates and reconstructs dynamic preservation contents with strong causal logical chains. This fusion is not a simple splicing, but rather the extraction of a global semantic structure that can fully summarize the "trigger source-evolution path-current performance". Finally, this structure is upgraded and identified as a "dynamic preservation event" with clear directionality and intervention guidance significance, completing the leap from discrete data to high-dimensional decision events.

[0026] Specifically, the system combined the blueberry data and placed it into a time window of the past 30 minutes for dynamic identification. The model found that in the last 5 minutes, the visual feature vector representing the "area of ​​missing fruit powder reflectivity" experienced an abnormal fluctuation with a sharp increase in slope. The system immediately marked this sharply declining visual feature, along with the "wind speed vector surging to 2.5 m / s" at the same moment, as a joint mutation feature.

[0027] The system uses RFID and vision to simultaneously identify the target as "high-value premium blueberries" and immediately retrieves prior knowledge about the species from the knowledge graph: the surface of blueberry fruit is covered with a layer of natural wax, whose biological function is to resist water loss and microbial infection; and this structure is extremely fragile and has a very high physical sensitivity to directional mechanical wind speeds greater than 1.5 m / s, making it extremely prone to irreversible wind erosion damage.

[0028] The system initiates reverse tracing, combining prior knowledge to analyze the mutation characteristics: the sudden reduction in bloom area is due to wind-induced mechanical abrasion exceeding the physical tolerance threshold, determining the first layer of dynamic preservation content: local physical wind erosion and water loss; then, it deduces the microenvironment: why did the local wind speed suddenly change to 2.5 m / s? The system checks the gas field and wind field topology and finds that the microenvironment parameters of the guide vane above the grid have shifted, determining the second layer of dynamic preservation content: micro-domain airflow organization distortion; finally, it continues to penetrate down to the equipment layer: tracing reveals that the return air valve in a certain area of ​​the Type A controlled atmosphere storage opened too wide during automatic adjustment, causing the cold air to form a short-circuited jet in the grid where the blueberries are located, determining the third layer of dynamic preservation content: the return air valve overload caused short-circuited airflow.

[0029] The system integrates the dynamic preservation content at the three levels mentioned above and makes decisions. Through Bayesian causal evaluation, it confirms that "overload of return air valve" strongly supports "airflow distortion", which in turn directly leads to "physical wind erosion". The system encapsulates and aggregates these three factors and finally outputs a high-dimensional dynamic preservation event: "Irreversible wind erosion loss event of blueberry powder driven by short-circuit airflow of return air valve".

[0030] refer to Figure 3 In step S12, the specific steps are as follows: S121: Perform semantic analysis on the dynamic preservation event and natural language processing during the analysis process to determine multiple preservation items of the target food, which include moisture loss, texture depression and color fading. S122: Obtain the morphological and surface characteristics of the target food, and import them into a digital twin space in combination with multiple preservation items of the target food. Perform corresponding dynamic simulations in the digital twin space to simulate the evolution of the internal physicochemical indicators of the target food under alternating environment. By comparing the ideal preservation curve in the digital twin space, use causal inference logic to determine the core disturbance factors that lead to quality degradation, thereby determining the content to be optimized in the dynamic preservation process of the target food.

[0031] In the embodiments of this application, the dynamic preservation event is semantically parsed and natural language processing is performed during the parsing process to determine multiple preservation items of the target food. The multiple preservation items cover moisture loss, texture depression and color fading, and moisture loss, texture depression and color fading are introduced.

[0032] At this point, after receiving the dynamic preservation events from the underlying fusion output, the system imports them into a semantic parsing engine based on a Large Language Model (LLM) or a domain-specific fine-tuning engine. Instead of performing simple keyword matching, it uses a self-attention mechanism to deeply mine the implicit dependencies and contextual logic between various semantic entities in the event text. Through lexical analysis and syntactic structure tree construction, the system accurately extracts the "trigger source entities: such as equipment malfunction, environmental distortion", "action path entities: such as fluid dynamics transfer, thermodynamic exchange", and "receptor result entities: such as tissue damage, physicochemical deterioration" from the event. This transforms the unstructured natural language description into a highly structured graph-based semantic representation, completing a machine-readable translation of the complex preservation mechanism.

[0033] Building upon structured semantic parsing, the system further utilizes core natural language processing technologies such as Named Entity Recognition (NER) and Dependency Parsing to perform deep drilling and degradation mapping on "recipient outcome entities." The system uses a professional ontology library in the food preservation field as a constraint boundary, precisely decomposing and mapping macroscopic event outcomes into specific biological tissue degradation dimensions at the microscopic level. During this process, the natural language processing model identifies and associates expressions such as "water loss / wind erosion / transpiration" with the "water loss" dimension, "wilt / softening / tissue collapse" with the "texture depression" dimension, and "browning / fading / pigment degradation" with the "color decay" dimension, thereby achieving precise anchoring from macroscopic event semantics to microscopic degradation indicators.

[0034] The system formally reorganizes and engineers the extracted entities of multiple degradation dimensions, identifying them as multiple preservation items corresponding to the target food. Each preservation item is not merely a label, but is endowed with a standardized quantitative representation framework, encompassing the current state threshold, safety boundary conditions, and monitoring feature operators of that degradation dimension. Through this step, the originally abstract single dynamic preservation event is completely decoupled and visualized as a set of multiple parallel and orthogonal preservation items, including "moisture loss," "texture depression," and "color decay." This lays a precise foundation of entity objects for subsequent independent parametric inference for each specific degradation dimension in the digital twin space.

[0035] Specifically, the system inputs the long text event into the semantic parsing engine; using the self-attention mechanism, the system accurately captures the deep causal dependencies between "short-circuit wind: trigger source entity", "driving force: action path entity, representing strong momentum transfer and surface shear force" and "irreversible wind erosion loss of fruit powder: receptor result entity, representing the physical peeling of the epidermal microstructure", and transforms it into a machine-readable structured triplet map, completing the deep semantic decoding of the complex physical process of "strong airflow shearing leading to epidermal damage".

[0036] The system calls upon the postharvest physiology ontology of fruits and vegetables and combines it with dependency parsing of natural language processing to drill down and break down the core receptor result of "irreversible wind erosion loss of fruit bloom". The system identifies that "fruit bloom" is a key barrier in blueberry biology that controls water transmembrane diffusion and resists photo-oxidation. Therefore, the NLP model intelligently maps and associates the macroscopic physical damage of "wind erosion loss" with three underlying biological degradation dimensions: accelerated transpiration of internal free water caused by fruit bloom peeling: mapped to the water loss dimension; decreased cell turgor pressure and tissue softening caused by loss of epidermal support: mapped to the texture depression dimension; and anthocyanin exposure to strong winds after the epidermal barrier disappears: mapped to the color fading dimension.

[0037] The system encapsulates the three degradation dimensions mentioned above in an engineered manner, and deterministically outputs three specific preservation items for the blueberry: the quantitative characterization of the moisture loss preservation item is: monitoring the slope of moisture content decrease and controlling the weight loss rate threshold based on near-infrared spectroscopy; the quantitative characterization of the texture depression preservation item is: calculating the blueberry sphericity fitting degree and monitoring the micro-deformation coefficient based on depth vision; and the quantitative characterization of the color decay preservation item is: extracting the anthocyanin reflectance ratio of the peel and evaluating the browning index based on multispectral imaging. Thus, for the blueberry wind erosion event, the system has successfully decomposed it into three orthogonal preservation items with independent monitoring entry points, ready to be put into subsequent twin-space dynamic simulation.

[0038] Furthermore, the morphological and surface characteristics of the target food are acquired, and multiple preservation items of the target food are imported into a digital twin space. Dynamic simulations are then performed within the digital twin space to model the evolution of the target food's internal physicochemical indicators under alternating environments. By comparing the ideal preservation curve in the digital twin space, causal inference logic is used to identify the core disturbance factors leading to quality degradation, thereby determining the optimization aspects of the target food during dynamic preservation. This approach incorporates the overall consideration of comparing the ideal preservation curve in the digital twin space. Simultaneously, dynamic preservation events are introduced to further control multiple preservation items, improving the accuracy of the optimization aspects of the target food during dynamic preservation.

[0039] At this point, the system performs high-precision three-dimensional phenotypic reconstruction of the target food through a multimodal visual perception array. In terms of morphological feature extraction, it uses structured light or time-of-flight depth camera to capture the spatial point cloud data of the target, and reconstructs its geometric topology through surface fitting, accurately quantifying its spatial geometric parameters such as volume, surface area, curvature distribution, and packing density. In terms of surface feature extraction, it relies on high-resolution multispectral imaging technology to capture the micro-texture distribution, gloss uniformity, and spectral reflectance characteristics of specific chemical components on the target's surface. These two types of features are deeply fused to construct a high-fidelity digital twin geometric precursor with physical scale and physicochemical property characterization.

[0040] The system synchronously imports the high-fidelity geometric precursor and multiple preservation items decoupled from S121 into a digital twin space. Within this space, the system uses real-time collected alternating environmental data as boundary conditions, combined with embedded physical information neural networks or micro-macro multiphysics coupling models, to perform cross-scale dynamic simulations of the target food. This simulation process uses the aforementioned morphological and surface features as spatial resistance parameters for heat and mass transfer and gas diffusion. For each preservation item's response characteristics under alternating environmental excitation, the system simulates the nonlinear evolution trajectory of the target food's internal physicochemical indicators over time with high spatiotemporal resolution.

[0041] After obtaining the simulated evolution trajectory of physicochemical indicators, the system performs multi-dimensional synchronous alignment and residual calculation with the "ideal preservation curve" pre-placed in the digital twin base. The "ideal preservation curve" is the benchmark model of quality degradation under the optimal static controlled atmosphere environment. For significant deviations, the system abandons traditional correlation analysis and instead introduces causal inference logic such as counterfactual reasoning and do-calculus. The system constructs multiple parallel intervention experiments in the virtual twin space, isolating or eliminating specific environmental input variables one by one, such as assuming the elimination of sudden wind speed changes or assuming the cutting off of local light, and observing the degree of blocking or mitigation of the quality deviation trajectory. By calculating the causal effect value of each environmental variable on quality degradation, the system accurately eliminates co-occurring factors of pseudo-correlation, penetrates the data appearance, and identifies the core disturbance factors that cause the curves of each preservation item to deviate from the ideal state.

[0042] Based on the core disturbance factors identified in the previous step, and combined with the current controllable actuator state of the system, the system performs strategic reverse solving and content refinement. The system transforms the abstract "causal effect value" into concrete engineering adjustment parameters, clearly defining which microenvironmental physical field distributions need to be reshaped and which equipment execution logics need to be modified during the dynamic preservation process. As a result, the system completely eliminates the need for lag compensation and instead outputs "content to be optimized" with forward-looking and fundamental characteristics. This content precisely points to the microenvironmental reconstruction path and equipment control parameter adjustment range required to eliminate the core disturbance factors.

[0043] Specifically, the system uses a 3D depth camera to scan the premium blueberries inside the Clamshell packaging box. Through point cloud processing, it extracts the average curvature of a single blueberry and the microscopic geometric features of the stem indentation. These features determine wind resistance and water vapor retention capacity. At the same time, it uses a multispectral camera to extract the distribution density of the bloom on the blueberry skin and the spectral reflectance of anthocyanins as surface features, constructing a high-fidelity three-dimensional digital twin of this batch of blueberries.

[0044] The system imports the blueberry twin and three preservation projects into the twin space, inputting the current 2.5m / s alternating wind speed field and low humidity environment caused by the "short circuit of the return air valve" as boundary conditions; combined with the morphological characteristics of blueberries, the peeling of fruit powder causes a drastic change in the thermal and moisture conductivity of the epidermis. The system uses the mass transfer equation to deduce that: under strong wind shear, internal water vapor penetrates the damaged epidermis at an abnormal rate and is lost: the water loss project evolves, causing the turgor pressure inside the cells to drop below the critical value for maintaining rigidity within 15 minutes: the texture depression project evolves, and at the same time, the micro-cracks caused by epidermal water loss accelerate oxygen invasion, triggering an abnormal surge in the activity of local polyphenol oxidase: the color fading project evolves.

[0045] The system compared the derived "water activity drop curve" with the "ideal low water loss curve" and found a huge negative deviation. The system then activated causal inference logic and generated counterfactual hypotheses in the twin space: assuming that the current temperature remains unchanged, but the local wind speed is forcibly reduced to 0.5 m / s, and assuming that local light is cut off. Through rigorous causal effect calculation, the system eliminated the spurious correlation of temperature fluctuations and determined that the core disturbance factor causing the simultaneous deterioration of the three major indicators of blueberry quality was not the global temperature anomaly, but "high-frequency turbulent shear stress exceeding the physiological tolerance limit of the epidermis within a specific spatial grid".

[0046] Based on this causal tracing result, the system accurately outputs the "content to be optimized" for the dynamic preservation process of blueberries: "The current optimization focus needs to be shifted from global temperature and humidity control to local flow field reconstruction. The turbulent shear stress at the (3-F-01) grid must be eliminated. The opening of the guide vane needs to be optimized or a microporous windbreak barrier needs to be introduced to force the local stagnation wind speed to be suppressed to below the non-destructive threshold of 1.0 m / s, and the water vapor boundary layer damaged by wind erosion should be repaired simultaneously." This content to be optimized directly grasps the physical essence of the problem and provides a precise target for subsequent compensation control.

[0047] refer to Figure 4 In step S13, the specific steps are as follows: S131: Identify food in the preservation area and mark target food and secondary food. Perform inference on the secondary food and combine multiple environmental data of the preservation area to control multiple factors, thereby marking the optimization content of the secondary food in the preservation process. S132: Align the optimization content of secondary food with the optimization content of target food in a four-dimensional spatiotemporal scale to analyze the implicit interference effect in the mixed storage environment of multiple food categories, simultaneously acquire the working data combination of the food preservation room, and further combine the implicit interference effect to perform multi-factor fusion, thereby constructing a multi-fusion content of "food environment - equipment execution end - spatiotemporal interference field". S133: The self-evolution operation of the multi-fusion content in the digital twin space outputs a preservation twin with high fidelity and real-time synchronization; the state offset of the preservation twin is marked, and reinforcement learning is performed on the state offset to determine the preservation compensation content for each actuator in the food preservation room, so as to determine the dynamic balance of multiple foods in the collaborative preservation process.

[0048] In the embodiments of this application, food is identified in the preservation area, and target food and secondary food are marked. The secondary food is extrapolated, and multiple environmental data of the preservation area are combined to control multiple factors, thereby marking the content to be optimized of the secondary food in the preservation process.

[0049] At this point, the system performs a full-area scan of the preservation area where there is cross-storage. Using multi-object instance segmentation and depth information calculation, it accurately identifies all categories within the physical grid and outputs their spatial three-dimensional bounding boxes. After completing the basic identification, the system classifies the identified food entities according to a preset preservation weight strategy or commercial value assessment model. High-priority or high-sensitivity foods that trigger dynamic preservation events are identified as "target foods," while other coexisting food entities in the microenvironment are down-dimension-reduced and labeled as "secondary foods." Simultaneously, the system extracts the spatial adjacency matrix and distance vector between the two to clarify the geometric positional relationship between the secondary foods and the target foods, providing a spatial benchmark for subsequent evaluation of the environmental coupling interference between the two.

[0050] After clarifying the primary and secondary relationships, the system initiates a lightweight digital twin simulation mechanism for secondary foods. The system inputs the category and physical properties of the secondary foods and the real-time status of the current microenvironment into a preset physiological response model to simulate the metabolic trajectory of the secondary foods under the current operating conditions. The system focuses on simulating the dynamic changes of secondary foods in respiration, ethylene release rate, or water transpiration, and predicts the evolution trend of their physiological state within a specific time window in the future. Thus, without performing full-scale in-depth calculations, the system can quickly grasp the potential behavioral patterns of secondary foods as "environmental disturbance sources" or "victim receptors".

[0051] Simultaneously, the system extracts multiple environmental data streams captured by the sensor array within the preservation area, including but not limited to temperature and humidity gradients, gas concentration fields, and wind speed vector fields. The system employs a method that couples multivariate time series analysis with spatial fluid dynamics to perform multi-factor cross-control of these environmental data. This control process aims to analyze the role of the environmental field as a medium for transmission between primary and secondary foods, quantitatively assess how turbulent airflow disturbances accelerate the mixing and diffusion of volatile organic compounds or ripening gases between different foods, and how small fluctuations in temperature gradients superimpose to affect the stability of the local microclimate, thereby constructing a comprehensive environmental interferometry map that includes both physical and biochemical fields.

[0052] The system deeply integrates and detects conflicts between the state evolution projection results of secondary food items and the multi-factor environmental control map. The system assesses whether the current or predicted state of secondary food items will exacerbate the quality degradation of the target food, or whether the secondary food items themselves face preservation risks due to sharing the microenvironment. Based on this two-way environmental game analysis, the system accurately identifies the core variables on the secondary food side that lead to the deterioration of the microenvironment and transforms them into specific control requirements. Finally, it marks the "content to be optimized" of secondary food items in the preservation process. This content clarifies what intervention measures need to be taken for secondary food items to eliminate their negative disturbance to the global microclimate or ensure their own storage safety.

[0053] Specifically, in the current cold storage (3-F-01) grid, the optimization target for blueberries (target food) is "eliminating local high-frequency turbulent shear stress". The system performs a panoramic visual scan of the same shelf shelf where the blueberries are located. Through instance segmentation, it not only identifies the clamshell packaging of the blueberries, but also accurately identifies the other two boxes of food placed next to the blueberries. Based on the preservation weight and event trigger source attributes, the system marks the blueberries as "target food" and the "peaches" placed next to them as "secondary food". At the same time, the spatial topology matrix shows that the peach packaging is only 15 centimeters away from the blueberries and is located downstream of the same strong wind jet path.

[0054] The system initiated a rapid physiological simulation for the secondary food item, "peaches." The model, combined with the varietal attributes of peaches, calculated that under the current abnormal microenvironment of excessively high local wind speed and low humidity, the transpiration rate of peach skin water would increase exponentially. At the same time, the simulation model warned that peaches were in the early stage of respiratory climacteric, and that their ethylene release rate had a probability inflection point of a surge in the next 2 hours, making them a highly active "biochemical interference source."

[0055] The system retrieved multi-dimensional environmental data of the grid for multi-factor control analysis; fluid dynamics control showed that the abnormal wind speed of 2.5 m / s not only peeled off the blueberry bloom, but also created a strong convective mass transfer effect between the blueberries and peaches; gas field control further revealed that due to the strong convection, once the peaches released a surge of ethylene, it would instantly break the original gas balance of the grid, and the high-concentration ethylene airflow would directly wash over the surface of the blueberries.

[0056] The system analyzed the "high transpiration + potential ethylene surge" of peaches and compared it with the environmental control map of "strong convection accelerating ethylene diffusion." The system determined that even if the wind speed was reduced according to the optimization criteria for blueberries, the ethylene produced by the peaches would still cause secondary disasters such as premature ripening and softening of blueberries if no intervention was taken for the peaches. Therefore, the system accurately marked the optimization criteria for peaches: "It is necessary to block the ethylene release and water loss pathways of peaches and establish an airtight isolation between the peaches and blueberries at the microclimate level to eliminate the biochemical disturbance threat to the target microenvironment caused by the peaches as a source of ripening gas diffusion."

[0057] Furthermore, the optimization content of secondary food items is aligned with that of target food items in a four-dimensional spatiotemporal scale to analyze the implicit interference effect in the mixed storage environment of multiple food categories. Simultaneously, the working data combination of the food preservation room is acquired, and multi-factor fusion is further combined with the implicit interference effect to construct a multi-fusion content of "food environment-equipment execution end-spatiotemporal interference field".

[0058] At this point, the system constructs a four-dimensional spatiotemporal tensor field containing three-dimensional spatial coordinates and a one-dimensional time evolution axis, mapping the optimization content of the secondary food and the optimization content of the target food to this spatiotemporal framework respectively. In this process, the system not only aligns the relative positions and distance attenuation coefficients of the two in physical space, but also synchronizes their critical response timelines in the time dimension. Through the convolution operation of the spatiotemporal tensor network, the system deeply analyzes the "hidden interference effect" in the complex microenvironment where multiple categories are mixed, which is jointly generated by the tight spatial coupling and the dynamic evolution of time. This effect aims to reveal the cross-category interaction degradation mechanism that cannot be observed from a static or single-category perspective. For example, how changes in the microenvironmental requirements of a certain food trigger a chain biochemical reaction of another food through the fluid medium under a specific time delay.

[0059] While analyzing implicit interference, the system penetrates down through multiple levels, retrieving in real time the working data collected by the IoT at the bottom layer of the food preservation room. This data includes the dynamic operating conditions of all hardware actuators involved in the construction of the microenvironment, such as the variable frequency operating frequency of the compressor, the air supply speed and deflection angle of the air cooler, and the opening degree of the gas separation valve and the nitrogen and oxygen replacement rate of the controlled atmosphere machine. Through the equipment mechanism model, the system maps these discrete bottom-level control signals in real time into the actual work distribution and energy output topology generated by the equipment actuator in the physical space, clarifying the actual projection state and dynamic response margin of the current hardware resources in three-dimensional space.

[0060] The resolved four-dimensional implicit interference effect is used as a high-dimensional constraint and cross-coupled with the work distribution of the mapped device execution end. This fusion process is essentially solving a complex multi-physics and multi-objective game problem: the system evaluates whether the current energy output distribution of the device is aggravating, mitigating, or failing to reach the resolved implicit interference effect. By introducing fluid dynamics and thermodynamic coupling equations, the system quantifies the comprehensive disturbance degree of each state switch of the device execution end on the "airflow direction-temperature field superposition thermal bridge-volatile gas diffusion path" in the mixed microenvironment, thereby establishing a high-fidelity data mapping channel between the environmental interference mechanism and the physical execution capability of the device.

[0061] Based on the aforementioned cross-coupled calculation results, the system performs information dimensionality enhancement and structured reorganization from a global perspective. The system abandons isolated variable analysis, forcibly placing "food itself and environmental state," "equipment mechanical execution actions," and "category interference field at the spatiotemporal scale" into the same correlation graph. Ultimately, it constructs a multi-layered fusion of "food environment - equipment execution end - spatiotemporal interference field." This fusion forms a holographic digital twin potential perception matrix, which not only characterizes "what interference is currently occurring" within the mixed storage space but also precisely anchors "how the underlying equipment actions participate in or cause this interference," providing complete global decision input for subsequent output of high-dimensional, conflict-free preservation compensation strategies.

[0062] Specifically, the system aligns these two aspects on a four-dimensional spatiotemporal scale. Spatially, the system calculates the coordinate vectors of the blueberry box and the peach box within the (3-F-01) grid and the relative wind direction angle. Temporally, the system predicts that the peach will reach the ethylene release inflection point after 2 hours. Through alignment, the system analyzes the fatal hidden interference effect: if the wind speed is reduced only for the blueberries to meet their needs, although mechanical wind erosion is weakened, the reduced wind speed will lead to a decrease in the airflow's ability to carry ethylene from the peaches. This causes high-concentration ethylene to form an "accumulation pool" in the narrow space between the blueberries and peaches, triggering a more severe ripening and softening effect on the blueberries than wind erosion after the 2-hour inflection point. This is a typical hidden negative effect of "improving the local physical field but exacerbating the interference of the biochemical field".

[0063] The system acquires real-time operational data from the cold storage: the current speed of the No. 2 air cooler is 1450 rpm, corresponding to an abnormal opening of the return air valve, causing jet formation in the (3-F-01) grid; simultaneously, the micro controlled atmosphere membrane separator configured in this area is currently in low-power standby mode. The system performs multi-factor fusion calculations based on the implicit interference effect of "ethylene retention and accumulation" and the data from the equipment execution end; the system evaluation found that: simply reducing the air cooler speed to eliminate blueberry wind erosion can indeed change the physical wind field, but it cannot solve the ethylene accumulation problem; however, if the controlled atmosphere membrane separator is activated and the opening of the nitrogen replacement valve is increased, using high-purity nitrogen to form a micro-positive pressure air curtain in this grid, it can not only provide a buffer to eliminate turbulent wind erosion (satisfying the blueberries), but also push away and dilute the ethylene released by the peaches through positive airflow (satisfying the peaches).

[0064] The system encapsulates the above deductions and constructs a multi-layered fusion of "food environment - equipment execution end - spatiotemporal interference field": the situation matrix clearly indicates that the current (3-F-01) grid's microclimate deterioration is essentially the result of the interplay between "abnormal jet flow from the air cooler" and "weak wind erosion resistance of blueberries / high ethylene release from peaches" within a "specific confined space and a future 2-hour time window"; the global decision input is clear: the single wind speed adjustment strategy must be abandoned, and a path of equipment-side linkage reconstruction of "air cooler frequency reduction combined with large-scale opening of the air conditioning valve" must be adopted.

[0065] Therefore, the self-evolutionary operation of this multi-fusion content in the digital twin space outputs a preservation twin with high fidelity and real-time synchronization; the state offset of the preservation twin is marked, and reinforcement learning is performed on the state offset to determine the preservation compensation content for each actuator in the food preservation room, so as to determine the dynamic balance of multiple foods in the collaborative preservation process, which is compatible with the overall consideration of the state offset of the preservation twin and ensures the accuracy of the preservation compensation content for each actuator in the food preservation room.

[0066] At this point, the system uses the previously constructed "food environment-equipment execution end-spontaneous interference field" multi-fusion content as the initial state tensor and injects it into the parallel computing engine of the digital twin space. In this process, the system uses the embedded partial differential equation system and multi-physics field coupling method as the evolution kernel to drive the multi-fusion content to perform high temporal resolution forward self-evolution calculation in virtual spacetime. This self-evolution process not only synchronizes the underlying data flow of the physical preservation chamber in real time to correct the boundary conditions, but also solves the heat transfer, mass transfer and biochemical reaction dynamics equations inside the microenvironment through high-frequency iteration. Finally, it converges and outputs a high-fidelity "preservation twin" that can map the real physical state of the current mixed storage space with millisecond-level latency and has the ability to infer microscopic mechanisms.

[0067] Throughout the continuous lifecycle of the preservation twin, the system compares its internal state variables with the preset "multi-category collaborative ideal preservation baseline state" in real time. By constructing a Kalman filter or a high-dimensional state observer, the system accurately extracts and marks the "state offset" that characterizes the risk of quality deterioration from complex evolutionary noise. This state offset is not a single temperature or humidity residual, but a multi-dimensional residual vector that integrates the deviation of physiological indicators of the target food and secondary food, the distortion of the microenvironment field, and the lag of equipment execution. It accurately quantifies the absolute distance and offset direction of the current system state from the optimal solution of dynamic equilibrium.

[0068] The system uses the multidimensional state offset as the state space input of the reinforcement learning agent, and defines the controllable parameter range of each actuator in the food preservation room as the action space. The joint reward function is to minimize the state offset and reduce the energy consumption of the equipment. The agent conducts high-concurrency trial and error exploration and policy gradient update in the virtual twin. Through tens of thousands of interactions with the environment model, the reinforcement learning method overcomes the nonlinear physical constraint barrier, strips away the coupling interference caused by the action of a single device, and finally determines a set of "preservation compensation content" that can maximize the global reward function. Optional actuators include fans, compressors, atmosphere control valves, and humidifiers.

[0069] The system logically encapsulates and verifies the feasibility of the preservation compensation content for each actuator output by reinforcement learning, and then sends it to the underlying controller of the physical preservation chamber. This compensation content is not an extreme squeeze for a single food, but a Nash equilibrium solution formed after global game theory. As the actuators coordinate their actions according to this compensation content, the micro-environment field of the physical space is forcibly reshaped, and the marked state offset in the digital twin space decays exponentially and is eventually clamped in a very small dead zone near zero. Thus, the target food and secondary food reach a steady state of non-harming in the microclimate of mutual game and mutual compromise. The system finally establishes and maintains the "dynamic equilibrium" of multiple foods in the collaborative preservation process.

[0070] Specifically, the system has constructed multiple fusion contents pointing to "the frequency reduction of the air cooler combined with the large-scale opening of the controlled atmosphere valve". The system inputs these multiple fusion contents into the digital twin space. The twin engine uses CFD fluid simulation and biochemical dynamics model as its core and performs self-evolution calculations with a very small time step: simulating the turbulent kinetic energy decay process of the grid (3-F-01) after the No. 2 air cooler is reduced in frequency, and simultaneously simulating the flow field reconstruction process of high-purity nitrogen forming a micro-positive pressure air curtain between blueberries and peaches after the controlled atmosphere nitrogen valve is opened; the system provides high-frequency synchronous real sensor feedback and finally outputs a high-fidelity "preservation twin" of blueberry-peach symbiosis that accurately reflects the microscopic physical process of "physical wind erosion elimination + biochemical ethylene being blocked by the air curtain".

[0071] The system monitors the status of the preservation twin in real time; the observer comparison found that the "bloominge wind erosion damage rate" index of blueberries is regressing towards the ideal threshold, but the "local microenvironment humidity" of peaches has residuals due to the frequency reduction of the air cooler. At the same time, although the "ethylene exposure concentration" of blueberries has decreased significantly, there is still a slight offset of +2ppm. The system merges these extracted parameter residuals and marks them as a three-dimensional state offset vector: [-significant wind erosion improvement, humidity offset, +2ppm gas offset].

[0072] The system inputs the offset into the reinforcement learning agent; the agent searches within the action space: it discovers that simply increasing the opening of the atmospheric regulating valve can completely eliminate the +2ppm ethylene offset, but it will cause the nitrogen airflow to carry away more moisture too quickly, exacerbating the humidity offset; after tens of thousands of virtual trial and error optimizations, the agent finally determines a set of non-intuitive multi-mechanism collaborative "preservation compensation content": "Instruct the No. 2 air cooler to reduce its frequency to 900rpm to eliminate wind erosion, instruct the atmospheric regulating separation valve to open to 60% to build a basic nitrogen air curtain to block ethylene, and at the same time call the micro ultrasonic humidification module to intervene at the bottom of the grid in pulse mode to accurately compensate for the moisture carried away by the nitrogen airflow."

[0073] The underlying PLC of the cold storage receives and executes these compensation instructions; within the physical space, the surface of the blueberries returns to a calm, windless state, and the bloom stops fading; the ethylene released by the peaches is pushed towards the return air vent by a directional nitrogen curtain; at the same time, pulse humidification ensures that the peaches do not lose water and shrivel; the state offset in the digital twin quickly converges to zero, and the blueberries and peaches stop influencing each other in this carefully formulated local microclimate. The system successfully establishes the ultimate "dynamic balance" between these two high-value-added categories in the scenario of mixed storage in the same warehouse.

[0074] refer to Figure 5 In step S14, the specific steps are as follows: S141: Monitor the operation of the food preservation room in real time, synchronously acquire the fault signal of the food preservation room, determine the corresponding faulty component based on the tracing of the fault signal, load the faulty component into the preservation twin, and dynamically determine the preservation impact of the faulty component on the target food through the simulation and deduction of the faulty component by the preservation twin. S142: Reorganize the content of the preservation impact and the preservation compensation, and deeply integrate the current preservation status of the target food. Use adaptive fault-tolerant control logic to calculate the backup work content of the food preservation room. The backup work content includes starting the backup phase change cold storage module and switching to the diversion mode of the adjacent healthy cold compartment. S143: The food preservation chamber performs its work according to the backup work content, continuously acquiring the morphological changes of the target food caused by environmental fluctuations, and using this morphological change as a high-priority closed-loop trigger signal to directly trigger the dynamic work content of the food preservation chamber. This dynamic work content includes high-pressure micro-mist water replenishment and air pressure fine adjustment.

[0075] In the embodiments of this application, the operation of the food preservation room is monitored in real time, fault signals of the food preservation room are acquired synchronously, the corresponding faulty component is determined by tracing the fault signal, the faulty component is loaded into the preservation twin, and the preservation impact of the faulty component on the target food is dynamically determined by the simulation and deduction of the faulty component by the preservation twin. This approach takes into account the overall fault signal and ensures the accuracy of the corresponding faulty component.

[0076] At this point, the system establishes a continuous operation monitoring mechanism for the entire electromechanical equipment in the food preservation room. By sampling the data streams of the underlying programmable logic controller (PLC) and various sensor networks at high frequency, it extracts equipment operation characteristics in real time, covering dimensions such as current, voltage, vibration frequency, rotation speed, and pressure difference. Based on this, the system relies on time-frequency domain analysis and anomaly detection methods, such as extracting the frequency domain distortion of non-stationary signals through fast Fourier transform, or calculating the mutation rate of the signal envelope using a sliding window, to accurately isolate abnormal fluctuations from the massive background noise of normal operation, thereby capturing and outputting the original fault signals that characterize the deterioration of the equipment's health status.

[0077] After capturing the fault signal, the system immediately maps it to a pre-constructed topology dependency graph of the fresh-keeping compartment equipment for reverse analysis. This graph uses nodes to represent physical components and directed edges to represent the transmission paths of energy flow, refrigerant flow, or control command flow. Based on the unique multimodal fingerprint characteristics of the fault signal, such as "sudden increase in current accompanied by a drop in speed", the system traces back along the graph edges to check potential failure nodes one by one. By comparing the failure probability distribution of each candidate node with the matching degree of the signal characteristics, the system penetrates the surface-level alarm, accurately eliminates false sources caused by cascading failures, and finally determines the specific physical entity that caused the fault signal, thus completing the absolute location of the "faulty component".

[0078] The system performs forced mutation operations on physical constraints in the digital twin space, injecting the located faulty component as a disturbance source into the high-fidelity preserved twin. The system does not simply delete the faulty component from the twin model, but modifies the corresponding boundary conditions in the multiphysics coupling simulation equations according to the specific failure mode of the faulty component, such as jamming, leakage, and short circuit. For example, the drag coefficient of the component in the fluid network is modified to infinity to represent mechanical stall, or the work term of the component in the thermodynamic equation is forced to zero to represent the interruption of energy transfer. Thus, the system-level topology change and energy flow blockage caused by the failure of the component are realistically reproduced in the virtual space.

[0079] After loading the fault boundary conditions, the system drives the preservation twin to perform forward fault evolution simulation with the current environmental state as the initial value. The system focuses on tracking the collapse process of the microenvironment physical field caused by the failure of faulty components, such as the temperature field deviation trajectory caused by the interruption of cooling and the local concentration accumulation curve of harmful gases caused by airflow stagnation. The system uses these distorted environmental fields as new input excitations, coupled with the biophysical and chemical response model of the target food, to deduce its quality degradation time sequence path under abnormal working conditions. By quantifying the time difference relationship between the environmental field deterioration rate and the physiological tolerance threshold of the food, the system finally dynamically outputs "preservation impact content" covering the degradation dimension, severity and expected time of occurrence.

[0080] Specifically, the system monitors the operation of the cold storage in real time with a sampling rate of 100Hz. Suddenly, the system discovered in the time-frequency domain analysis that the effective value of the three-phase current of the No. 2 evaporator fan responsible for the air duct where the (3-F-01) grid is located experienced a violent non-periodic oscillation within 2 seconds, accompanied by an abnormal surge in the high-frequency harmonic components in the vibration spectrum. The system immediately captured this "current-vibration coupling anomaly" as the original fault signal.

[0081] The system traced the fault signal back through the topology dependency graph of the input device; the signal characteristics matched the "motor overload caused by sudden increase in mechanical resistance" pattern; the system traced back along the control command flow and mechanical transmission chain, ruling out the possibility of inverter failure and belt slippage, because the current was not completely disconnected and there was high-frequency vibration, and finally accurately located the source entity as: "the bearing of the No. 2 evaporator fan was severely mechanically jammed or frozen", and determined that the fan was the specific faulty component.

[0082] The system performs a mutation loading in the blueberry preservation twin; the system modifies the hydrodynamic boundary conditions of the (3-F-01) grid, instantly clamping the outlet velocity boundary of fan No. 2 to "0 m / s" in the simulation model (characterizing the complete interruption of airflow caused by bearing seizure); at the same time, the system corrects the cold convection heat transfer coefficient of the grid to an extremely low value with only natural convection remaining, instantly "killing" the fan's work capacity in the digital space.

[0083] The system performs forward catastrophe simulation based on the twins after the boundary mutation. The simulation shows that after the fan stops, the slightly positive pressure nitrogen air curtain around the blueberries and peaches dissipates within 10 seconds. The ethylene released by the peaches, no longer blocked by the air curtain, begins to diffuse and accumulate towards the blueberries. More fatally, after the forced convection is lost, due to the heat load deep in the storage chamber, the local temperature of the (3-F-01) grid rises rapidly at a slope of 0.8℃ / 10min in the simulation model. The system inputs this abnormal temperature rise and ethylene accumulation into the physiological model of blueberries and dynamically outputs the preservation impact: "Due to the fan jam, the forced cold air and air curtain are interrupted. It is estimated that in the next 18 minutes, the local temperature will exceed the critical point (8℃) for the blueberries to undergo a respiratory jump, and the ethylene concentration will exceed the ripening threshold. The blueberries will face the irreversible preservation impact of 'rapid softening'."

[0084] Furthermore, the preservation impact content and preservation compensation content are reorganized and deeply integrated with the current preservation status of the target food. Adaptive fault-tolerant control logic is used to calculate the backup work content of the food preservation chamber. The backup work content includes the diversion mode of starting the backup phase change cold storage module and switching to the adjacent healthy cold chamber.

[0085] At this point, after confirming the underlying hardware failure, the system performs failure isolation and reorganization on the previously generated conventional preservation compensation content at the logic level; the system identifies the action instructions in the conventional compensation content that depend on the execution of the faulty component (such as adjusting the speed of the faulty fan or valve) and forcibly marks them as unreachable dead zone instructions; the content with serious preservation impact derived from S141 is used as the new highest priority constraint to replace the original compensation target. This reorganization process is essentially a reversal of the compensation strategy originally used for "fine-tuning and optimization" into an emergency intervention baseline for "limited loss control" in the context of the failure, forming a reorganized content set that eliminates failure paths and focuses on core survival indicators.

[0086] The system synchronously retrieves the absolute physical state of the target food at the instant of the failure and deeply integrates it into the reconstructed content set. By analyzing visual perception feedback and microenvironment sensor readings, the system extracts the target food's current initial quality baseline values, such as the current skin integrity, core temperature, and water activity. Combined with the category-specific stress tolerance threshold, the system quantifies the "time margin" and "thermodynamic margin" between the current state and irreversible damage. This integration step ensures that subsequent fault tolerance calculations are no longer based on theoretical deductions of ideal initial states, but are strictly anchored to the target food's actual vulnerability level at the millisecond of the failure.

[0087] The system inputs the recombined content set, which incorporates real vulnerability states, into the adaptive fault-tolerant control engine. This engine abandons traditional linear control methods such as PID that rely on precise models, and instead adopts a fault-tolerant calculation strategy based on model predictive control (MPC) or deep reinforcement learning. The system traverses all non-failed redundant hardware resources and non-standard execution paths in the twin space, with the optimization objective of maximizing the consumption of the previously calculated "time / thermodynamic margin". It solves the optimal control sequence under the dual constraints of resource constraints and topology damage through rolling time-domain optimization. The system automatically downgrades the original fine-grained control logic and calculates a set of extreme survival strategies that can bypass fault nodes and use heterogeneous equipment combinations to offset the impact of preservation content, namely "backup work content".

[0088] The system translates the calculated optimal survival strategy into a specific sequence of physical actions, clearly defining the specific form of backup work content. This content inevitably breaks through the conventional working mode of a single device. The system reconstructs the cold energy transmission network and airflow organization topology, calling on unconventional physical media to replace failed functions. Specifically, the system will precisely trigger the latent heat storage device to provide passive cold energy support, while simultaneously linking the air valve matrix to break the original independent compartment isolation, forcibly introducing the kinetic and cold energy of the adjacent healthy microenvironment, ultimately forming a composite fault-tolerant execution scheme that covers "passive phase change heat absorption compensation" and "cross-space fluid kinetic energy reconstruction".

[0089] Specifically, the system immediately discarded the conventional compensation content derived from stage S133, which was "instructing the No. 2 air cooler to reduce its frequency and cooperate with the gas regulating valve to open at a large ratio," because the No. 2 air cooler had completely failed. The system prioritized the preservation impact content derived from stage S141, which was "the temperature exceeds 8°C and the ethylene threshold exceeds the limit within 18 minutes," and reorganized the current only core requirement: in the absence of mechanical air cooling, the temperature rise of the (3-F-01) grid must be suppressed within 18 minutes, and ethylene diffusion must be blocked.

[0090] The system quickly retrieves the current preservation status of the blueberries at the moment of the malfunction: the multispectral camera shows that the blueberry bloom has just undergone repair, and the current integrity rate is 85%; the infrared thermal imager shows that the current surface temperature of the blueberries is an accurate 2.5℃; combined with the knowledge of blueberry physiology, the system quantifies the current thermodynamic margin: that is, the blueberries can withstand a temperature rise of 5.5℃, and under the harsh conditions of no wind convection and only natural convection, they have a strict "18-minute countdown" margin.

[0091] The adaptive fault-tolerant control engine searches for a way out in the entire library topology; the engine calculation found that waiting for manual repair would not achieve the desired effect; increasing the compressor power would also be ineffective without fan circulation; the engine determined a heterogeneous fault-tolerant path through rolling optimization: a backup phase change cold storage plate for peak shaving and valley filling is hidden directly below the grid (3-F-01) in the library, which is currently in a solid latent heat state, while the fan in the adjacent No. 4 cold compartment (which stores cold-resistant root vegetables) is operating normally and has a slight positive pressure.

[0092] The system converts the calculation path into a defined backup task and immediately issues the following instructions: "First, forcibly trigger the electric heating trigger of the backup phase change cold storage module below the (3-F-01) grid. Utilize the extremely high latent heat absorption capacity of the phase change material as it transforms from solid to liquid to forcibly absorb local heat and stretch the temperature rise curve in the absence of a fan. Second, urgently open the top electric diversion valve between cold chambers 3 and 4, switch to the diversion mode of the adjacent healthy cold chamber, and use the negative pressure suction of the large fan in cold chamber 4 to laterally extract the polyethylene in the blueberry area and guide the cold airflow at the bottom of cold chamber 4 through the diversion channel to form a weak physical air curtain."

[0093] Therefore, the food preservation room executes its work according to the backup work plan, continuously acquiring information on the morphological changes of the target food caused by environmental fluctuations. This morphological change is used as a high-priority closed-loop trigger signal to directly trigger the dynamic work plan of the food preservation room. This dynamic work plan includes high-pressure micro-mist hydration and air pressure fine-tuning. The introduction of high-pressure micro-mist hydration and air pressure fine-tuning, along with further control over the preservation twin, fully considers the preservation impact, preservation compensation, and the current preservation status of the target food, thereby improving the accuracy of the dynamic work plan of the food preservation room.

[0094] At this point, after establishing the backup work content, the system immediately takes over the underlying execution network of the food preservation room, driving the heterogeneous redundant resources in the space to perform forced coordinated action according to the spatiotemporal sequence of fault-tolerant calculation. This execution process breaks the linear rhythm of conventional constant temperature and humidity control and turns into highly dynamic asymmetric energy injection. The system precisely controls the phase change energy release rate of the phase change cold storage device and synchronously drives the cross-zone guide valve to construct a non-standard aerodynamic topology. By forcibly reconstructing the local thermodynamic boundary and hydrodynamic path, a defense barrier against environmental collapse caused by the failure of the main equipment is rigidly erected in the physical space, ensuring that the microenvironment where the target food is located is forcibly anchored within the critical survival range.

[0095] During the transition process with backup mechanism intervention, due to the asynchronous response of heterogeneous equipment and the drastic switching of local flow field caused by phase change heat absorption and cross-regional flow, brief and intense secondary environmental fluctuations will inevitably occur around the target food. The system simultaneously improves the sampling frame rate and feature extraction sensitivity of the visual perception array, and uses multimodal visual tracking to continuously peel off and capture the microscopic physical response of the target food under these secondary micro-perturbations. The system focuses on analyzing the geometric deformation of biological tissues under transient thermal stress or surface tension abrupt changes, and calculates and outputs the spatial geometric morphological features such as the derivation of epidermal micro-folds, the transient change in volume shrinkage rate, and the local attenuation of gloss in real time as quantitative "morphological change content".

[0096] The system directly injects the acquired morphological changes into the lowest level of the control logic, giving it the highest priority closed-loop triggering right, surpassing all conventional environmental parameters. Traditional environmental control usually has a large thermodynamic inertial delay, but the system at this stage abandons the long serial link of "environmental anomaly - equipment adjustment - food response". By using the microscopic morphological changes of the food itself as the direct trigger source, the system establishes an ultra-short causal closed-loop link of "deformation of the food itself - generation of control command - instantaneous response of the equipment". Any tiny distortion judged to exceed the morphological safety threshold will instantly penetrate the dead zone of conventional PID control and be converted into the highest level hardware interrupt control signal without delay.

[0097] In response to this high-priority closed-loop trigger signal, the system completely abandons the repair of the large-scale macro environment and instead adopts a micro-targeted intervention strategy, rapidly reconstructing and issuing "dynamic working content". Targeting the specific physical vulnerabilities exposed by morphological changes, the system precisely schedules end-effectors with extremely low execution inertia. By directionally releasing tiny water vapor molecules or precisely adjusting local spatial pressure differences, it forcibly constructs a micro-physical barrier or compensation mechanism in the boundary layer very close to the surface of the target food. This dynamic working content acts directly on the food body boundary in a "point-to-point" manner, aiming to instantly suppress the trend of morphological deterioration and complete the ultimate physical resolution of secondary environmental fluctuations.

[0098] Specifically, the system forcibly executes the backup content: (3-F-01) The phase change cold storage plate below the grid begins to melt ice in reverse, instantly absorbing a large amount of heat enthalpy; at the same time, the top guide valve is forcibly opened, and the high-pressure cold airflow of the No. 4 cold chamber is introduced, forming a non-directional turbulent cold airflow above the blueberries. The simultaneous intervention of these two heterogeneous cold sources creates an extremely unstable and violently fluctuating microenvironment around the blueberries.

[0099] Due to the heat absorption caused by the phase change, the absolute humidity of the local air dropped sharply. In addition, the non-directional cold air was instantly drawn in by the No. 4 cold chamber, and the water vapor partial pressure on the surface of the blueberries was adjusted within 5 seconds. The high-speed multispectral camera keenly captured that the originally plump tiny waxy layer on the blueberry skin had undergone microscopic "crack deformation" due to transient water loss, and the reflectivity of the skin dropped in a flash. The system calculated in real time the "morphological change content" of this rapid development: the microstructure of the skin underwent 0.1 mm-level brittle deformation due to transient dehydration stress.

[0100] The system directly uses the "0.1 mm-level skin cracking deformation" as a high-priority closed-loop trigger signal, instantly breaking through the conventional temperature and humidity control logic and sending the highest-level hardware trigger pulse to the underlying micro-actuator network. In response to this body deformation trigger signal, the system rapidly issues dynamic work instructions: "Immediately activate the high-pressure micro-mist water replenishment module, bypass the air duct, and directly spray condensed water mist with a particle size of less than 10 micrometers into the air vent area of ​​the clamshell packaging box. Utilize the extremely high specific surface area of ​​the water mist to instantly fill the water vapor partial pressure deficit on the blueberry surface within 0.5 seconds, preventing the cracks from expanding. Simultaneously trigger the cargo pressure fine-tuning device to fine-tune the static pressure box valve of the (3-F-01) grid, increasing the local air pressure by 50 Pascals. Use the micro-positive pressure to forcibly 'press' the extremely fine water mist onto the blueberry surface to form a liquid water film, completely blocking the secondary suction airflow from directly peeling the blueberry skin." Through this dynamic work instruction, the microscopic cracking of the blueberry is forcibly reversed within 1 second after it occurs.

[0101] Please see Figure 6 The intelligent food preservation system based on digital twins is applied to the aforementioned intelligent food preservation method based on digital twins; the intelligent food preservation system based on digital twins includes: The data processing module 21 is used to mark the preservation area of ​​the food preservation room, acquire multi-dimensional preservation data of the preservation area, and determine the preservation data combination of the target food by combining the image of the preservation area, and determine the corresponding dynamic preservation event based on the recognition of the preservation data combination of the target food. The digital twin module 22 is used to determine multiple preservation items of the target food based on the semantic parsing of the dynamic preservation event, and to perform dynamic inference in the digital twin space in combination with the morphological and surface features of the target food, thereby determining the content to be optimized in the dynamic preservation process of the target food. The preservation compensation content module 23 is used to acquire secondary food items in the same preservation area, mark the content to be optimized in the preservation process of secondary food items, and perform spatiotemporal alignment with the content to be optimized of the target food items. It also introduces the working data combination of the food preservation room to perform multi-factor fusion to output the corresponding preservation twin and the corresponding preservation compensation content. The dynamic working module 24 is used to acquire faulty components of the food preservation compartment, determine the corresponding preservation impact content in combination with the preservation twin, further integrate the preservation compensation content and the current preservation status of the target food to determine the backup working content of the food preservation compartment, and trigger the dynamic working content of the food preservation compartment in combination with the morphological changes of the target food.

[0102] It should be noted that although multiple modules are mentioned in the detailed description above, this division is not mandatory; in fact, according to the embodiments of this disclosure, the features and functions of two or more modules or described above can be embodied in one module; conversely, the features and functions of one module described above can be further divided into multiple modules to be embodied.

[0103] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein; this application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein; the specification and embodiments are to be considered exemplary only.

[0104] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A smart food preservation method based on digital twins, characterized in that, include: The preservation areas of the food preservation room are marked, multi-dimensional preservation data of the preservation areas are obtained, and the preservation data combination of the target food is determined by combining the images of the preservation areas. Based on the identification of the preservation data combination of the target food, the corresponding dynamic preservation event is determined. Based on the semantic analysis of the dynamic preservation event, multiple preservation items of the target food are determined, and dynamic inference is carried out in the digital twin space in combination with the morphological and surface features of the target food, so as to determine the content to be optimized in the dynamic preservation process of the target food. Obtain secondary food items from the same preservation area, mark the optimization content of secondary food items during the preservation process, and perform spatiotemporal alignment with the optimization content of target food items. Introduce working data from the food preservation room for multi-factor fusion to output the corresponding preservation twin and the corresponding preservation compensation content. The system identifies faulty components in the food preservation compartment, determines the corresponding preservation impact based on the preservation twin, further integrates preservation compensation content and the current preservation status of the target food to determine backup work content for the food preservation compartment, and triggers dynamic work content for the food preservation compartment based on changes in the morphology of the target food.

2. The intelligent food preservation method based on digital twins according to claim 1, characterized in that, The food preservation room is marked with a preservation area. Multidimensional preservation data of this area is acquired, and combined with images of the preservation area to determine the preservation data combination of the target food. Based on the identification of the preservation data combination of the target food, corresponding dynamic preservation events are determined, including: In the food preservation room, the three-dimensional distribution map of the food preservation room is spatially gridded, and the preservation area is topologically marked. Simultaneously, multi-dimensional preservation data of the preservation area is acquired using a sensor array. The multi-dimensional preservation data covers temperature and humidity gradients, gas field distribution, and wind field distribution. The multi-dimensional preservation data of the preservation area and the real-time image of the preservation area are fused across modes to determine the preservation data combination of the target food under a specific environment.

3. The intelligent food preservation method based on digital twins according to claim 2, characterized in that, The process of marking the preservation area of ​​the food preservation room, acquiring multidimensional preservation data of the preservation area, and determining the preservation data combination of the target food by combining the image of the preservation area, and determining the corresponding dynamic preservation event based on the identification of the preservation data combination of the target food, further includes: The preservation data combination is dynamically identified, and mutation features are marked during the identification process. The food type of the target food is obtained simultaneously, and the corresponding mutation features are further combined for reverse tracing. In this way, multiple dynamic preservation contents at different levels are gradually determined during the tracing process, and the corresponding dynamic preservation events are determined based on the fusion of multiple dynamic preservation contents.

4. The intelligent food preservation method based on digital twins according to claim 1, characterized in that, The process involves semantic parsing of the dynamic preservation event to determine multiple preservation items for the target food, and then dynamically extrapolating these items in a digital twin space based on the morphological and surface features of the target food. This process identifies the aspects of the target food that need optimization during the dynamic preservation process, including: Semantic analysis of the dynamic preservation event was performed, and natural language processing was carried out during the analysis process to identify multiple preservation items of the target food, which include moisture loss, texture depression and color fading.

5. The intelligent food preservation method based on digital twins according to claim 4, characterized in that, The process of determining multiple preservation items for the target food based on semantic parsing of the dynamic preservation event, and dynamically extrapolating the morphological and surface features of the target food in a digital twin space to determine the content to be optimized in the dynamic preservation process, also includes: The morphological and surface characteristics of the target food are acquired and imported into a digital twin space in combination with multiple preservation items of the target food. The corresponding dynamic simulation is carried out in the digital twin space to simulate the evolution of the internal physicochemical indicators of the target food under alternating environment. By comparing the ideal preservation curve in the digital twin space, the core disturbance factors that lead to quality degradation are identified using causal inference logic, thereby determining the content to be optimized in the dynamic preservation process of the target food.

6. The intelligent food preservation method based on digital twins according to claim 1, characterized in that, The process involves acquiring secondary food items from the same preservation area, marking the optimization aspects of these secondary food items during the preservation process, and performing spatiotemporal alignment with the optimization aspects of the target food. This is combined with working data from the food preservation room for multi-factor fusion to output a corresponding preservation twin and corresponding preservation compensation content, including: Food identification is performed in the preservation area, and target food and secondary food are marked. The secondary food is extrapolated, and multiple environmental data of the preservation area are combined to control multiple factors, thereby marking the optimization content of the secondary food in the preservation process.

7. The intelligent food preservation method based on digital twins according to claim 6, characterized in that, The process of acquiring secondary food items from the same preservation area, marking the optimization aspects of these secondary food items during the preservation process, and combining this with the optimization aspects of the target food items for spatiotemporal alignment, incorporating multi-factor fusion using working data from the food preservation room, and outputting corresponding preservation twins and corresponding preservation compensation content, also includes: Align the optimization content of secondary food with that of target food in a four-dimensional spatiotemporal scale to analyze the implicit interference effect in the mixed storage environment of multiple food categories, simultaneously acquire the working data combination of the food preservation room, and further combine the implicit interference effect to perform multi-factor fusion, thereby constructing a multi-fusion content of "food environment - equipment execution end - spatiotemporal interference field". The self-evolutionary computation of the multi-fusion content in the digital twin space outputs a preservation twin with high fidelity and real-time synchronization; the state offset of the preservation twin is marked, and reinforcement learning is performed on the state offset to determine the preservation compensation content for each actuator in the food preservation room, so as to determine the dynamic balance of multiple foods in the collaborative preservation process.

8. The intelligent food preservation method based on digital twins according to claim 1, characterized in that, The process involves acquiring faulty components in the food preservation compartment, determining the corresponding preservation impact based on the preservation twin, further integrating preservation compensation content and the current preservation status of the target food to determine the backup work content for the food preservation compartment, and triggering dynamic work content for the food preservation compartment based on changes in the morphology of the target food, including: The system monitors the operation of the food preservation room in real time, acquires fault signals of the food preservation room synchronously, identifies the corresponding faulty component based on the traceability of the fault signal, loads the faulty component into the preservation twin, and dynamically determines the preservation impact of the faulty component on the target food through simulation and deduction of the faulty component by the preservation twin. The preservation impact content and preservation compensation content are reorganized and deeply integrated with the current preservation status of the target food. Adaptive fault-tolerant control logic is used to calculate the backup work content of the food preservation chamber. The backup work content includes starting the backup phase change cold storage module and switching to the diversion mode of the adjacent healthy cold chamber.

9. The intelligent food preservation method based on digital twins according to claim 8, characterized in that, The process of acquiring faulty components in the food preservation compartment, determining the corresponding preservation impact based on the preservation twin, further integrating preservation compensation content and the current preservation status of the target food to determine the backup work content of the food preservation compartment, and triggering the dynamic work content of the food preservation compartment based on the morphological changes of the target food, also includes: The food preservation chamber executes its work according to the backup work plan, continuously acquiring information on the morphological changes of the target food due to environmental fluctuations. This morphological change is used as a high-priority closed-loop trigger signal to directly trigger the dynamic work plan of the food preservation chamber, which includes high-pressure micro-mist water replenishment and air pressure fine-tuning.

10. A smart food preservation system based on digital twins, characterized in that, The intelligent food preservation system based on digital twins is applied to the intelligent food preservation method based on digital twins as described in any one of claims 1-9; The intelligent food preservation system based on digital twins includes: The data processing module is used to mark the preservation area of ​​the food preservation room, acquire multi-dimensional preservation data of the preservation area, and determine the preservation data combination of the target food by combining the image of the preservation area. Based on the identification of the preservation data combination of the target food, the corresponding dynamic preservation event is determined. The digital twin module is used to determine multiple preservation items of the target food based on the semantic parsing of the dynamic preservation event, and to perform dynamic inference in the digital twin space in combination with the morphological and surface features of the target food, thereby determining the content to be optimized in the dynamic preservation process of the target food. The preservation compensation content module is used to acquire secondary food items in the same preservation area, mark the content to be optimized for secondary food items during the preservation process, and perform spatiotemporal alignment with the content to be optimized for target food items. It also incorporates working data from the food preservation room for multi-factor fusion to output the corresponding preservation twin and the corresponding preservation compensation content. The dynamic working module is used to acquire faulty components of the food preservation compartment, determine the corresponding preservation impact content by combining the preservation twin, further integrate the preservation compensation content and the current preservation status of the target food to determine the backup working content of the food preservation compartment, and trigger the dynamic working content of the food preservation compartment by combining the morphological changes of the target food.

Images