Package management and control method and system based on multi-modal information fusion

Abstract

The invention relates to the technical field of packaging dynamic management and control, in particular to a packaging management and control method and system based on multi-modal information fusion. The packaging management and control method based on multi-modal information fusion comprises the following steps: S1, obtaining inherent parameter data of a packaged product, determining ontology attributes according to the inherent parameter data, obtaining pressure data, temperature data and humidity data of an environment where the packaged product is located, and determining the ontology attributes according to the pressure data, the temperature data and the humidity data; taking the pressure data, the temperature data and the humidity data as external attributes; and obtaining product types corresponding to different types of packaged products, and determining a single change sequence and a single invariant sequence according to the product types in combination with the ontology attributes and the external attributes. According to the method, a multi-modal data fusion system is constructed, dynamic risk monitoring and accurate traceability are realized, an early warning-diagnosis-optimization closed loop is established, and the reliability and the anti-risk capability of a packaging system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of packaging dynamic management and control, and particularly relates to a packaging management and control method and system based on multi-modal information fusion. BACKGROUND

[0002] Multi-modal information fusion technology refers to forming more comprehensive and accurate cognition and decision support for a target object by integrating heterogeneous information from multiple sensors or data sources. In the field of packaging management and control, this technology has gradually been applied to the collaborative monitoring of the physical state of packaging products and the external environment. By combining visual, mechanical and environmental sensor data, real-time perception and risk warning of key indicators such as packaging integrity and sealing are achieved, effectively improving the automation level of packaging safety management.

[0003] However, the existing packaging management and control method based on multi-modal fusion still has obvious limitations. The existing method relies on static thresholds and fixed rules for anomaly judgment, which is difficult to adapt to the nonlinear risk evolution caused by dynamic changes in the environment and the interaction of internal multi-state components in the complex logistics process. At the same time, the existing method lacks the ability to trace the root cause of the risk, and cannot effectively distinguish whether the risk is caused by defects in the packaging properties, the coupling effect between components, or the external environmental stress exceeding the design boundary, resulting in a lack of targetedness of the warning results and difficulty in supporting precise intervention and parameter optimization.

[0004] Therefore, it is urgent to build an intelligent analysis method that can deeply integrate packaging ontology properties, environmental parameters and risk transmission paths, realize the fundamental transformation from passive alarm to active cognition and from isolated detection to system tracing, and improve the foresight and reliability of packaging management and control in complex scenarios. SUMMARY

[0005] In order to overcome the shortcomings of packaging risk dynamic perception and tracing, the present application provides a packaging management and control method and system based on multi-modal information fusion.

[0006] The technical implementation scheme of the present application is: a packaging management and control method based on multi-modal information fusion, comprising the following steps: S1: obtaining inherent parameter data of a packaging product, and determining ontology properties according to the inherent parameter data; obtaining pressure data, temperature data and humidity data of the environment where the packaging product is located, and taking the pressure data, the temperature data and the humidity data as external properties; obtaining product types corresponding to different types of packaging products, and determining a single change sequence and a single invariant sequence according to the product types, the ontology properties and the external properties; S2: obtaining a risk exposure sequence according to the preset safety threshold interval of the single change sequence combined with the single invariant sequence; and determining an early risk signal according to the single change sequence and the single invariant sequence; S3: obtaining a grade classification result according to the risk level classification of the active early warning event according to the risk exposure sequence and the early risk signal; and calculating the correlation degree of different types of packaging products according to the grade classification result; S4: generating an internal joint risk sequence according to the correlation degree; generating an external joint risk sequence according to the internal joint risk sequence; and optimizing the design parameters of the ontology attribute according to the external joint risk sequence.

[0007] Preferably, inherent parameter data of the packaging product is obtained, and the ontology attribute is determined according to the inherent parameter data; pressure data, temperature data and humidity data of the environment in which the packaging product is located are obtained, and the pressure data, the temperature data and the humidity data are taken as external attributes, comprising: The inherent parameter data includes physical and chemical property data, structural design parameters and initial content state, and the physical and chemical property data, the structural design parameters and the initial content state are taken as ontology attributes, which refer to static or quasi-static characteristics inherent to the packaging product and independent of the external environment; The external attribute refers to the environmental characteristics existing outside the packaging product and dynamically interacting with the packaging product to affect the state and behavior of the packaging product.

[0008] Preferably, the product type corresponding to different types of packaging products is obtained, and the single change sequence and the single invariant sequence are determined according to the product type combined with the ontology attribute and the external attribute, comprising: The product type includes solid products, liquid products, gaseous products and mixed products; For each product type, a preset defect attribute state corresponding to the ontology attribute is selected, and a dynamically changing environmental parameter corresponding to the external attribute is extracted; the defect attribute state and the environmental parameter are combined to form a single change sequence; the single change sequence refers to a data sequence used to simulate and represent the state evolution path under a dynamic external environment when a packaging product has a preset defect in a certain ontology attribute; For each product type, all corresponding ontology attributes are in a reference normal state, and a dynamically changing environmental parameter corresponding to the external attribute is extracted; the reference normal state and the environmental parameter are combined to form a single invariant sequence; the single invariant sequence refers to a reference data sequence used to simulate and represent the state evolution path under a dynamic external environment when all ontology attributes of a packaging product are in a reference normal state.

[0009] Preferably, the risk exposure sequence obtained according to the preset safety threshold interval of the combination of the single change sequence and the single invariant sequence comprises: point-by-point comparison of state values of the single change sequence and the single invariant sequence at the same sampling point; when the state value of the single change sequence exceeds the preset safety threshold interval, and the difference between the state value of the single change sequence and the corresponding state value of the single invariant sequence exceeds the preset deviation tolerance, determining that the current data point of the single change sequence is a risk data point; combining all the risk data points in chronological order into a risk exposure sequence, the risk exposure sequence being used to locate high-risk state data points directly caused by a specific ontology attribute defect and having exceeded normal safety behavior.

[0010] Preferably, the early risk signal determined according to the single change sequence and the single invariant sequence comprises: calculating the difference between the single change sequence and the single invariant sequence at the same sampling point, and taking the difference as a state offset, the state offset being used to quantify the instantaneous state deviation caused by the simulation defect; calculating the rate of change of the state offset over time, and taking the rate of change as an offset rate, the offset rate being used to quantify the state deterioration trend and severity caused by the simulation defect; if the absolute value of the state offset continuously exceeds the offset threshold and the offset rate continuously exceeds the rate threshold within a preset sliding time window, determining that there is an early risk signal.

[0011] Preferably, the risk level division of the active early warning event obtained according to the risk exposure sequence and the early risk signal comprises: extracting all continuous and numerical value exceeding risk threshold sequence fragments from the risk exposure sequence, and defining each of the sequence fragments as a risk exposure interval, the risk exposure interval being used to quantify a high-risk state event that represents a persistent state that exceeds normal safety mode; if the early risk signal is true and the risk exposure interval is a non-empty set, confirming that an active early warning event occurs; if the number of data points of the risk exposure interval is greater than the number threshold and the average value of the offset rate is greater than the rate average threshold, dividing into a high-risk level; if only one of the conditions is met, dividing into a medium-risk level; if neither condition is met, dividing into a low-risk level.

[0012] Preferably, the correlation degree of different types of packaging products calculated according to the level division result comprises: Based on the high risk level and the medium risk level, the risk exposure interval corresponding to the packaging product triggering the active early warning event is obtained within the same packaging product; Based on the risk exposure interval, the risk exposure sequence of each component inside the packaging product within the early warning period is extracted; The correlation degree between the risk exposure sequences of the solid product part and the liquid product part is calculated; The correlation degree between the risk exposure sequences of the solid product part and the gaseous product part is calculated; The correlation degree between the risk exposure sequences of the liquid product part and the gaseous product part is calculated.

[0013] Preferably, the internal joint risk sequence is generated according to the correlation degree, comprising: If the correlation degree between any two parts exceeds the internal correlation degree threshold, it is determined that the risk source is the internal interaction of the packaging product; the risk exposure sequences of the two parts with high correlation degree are spliced according to the time axis to generate an internal joint risk sequence, which is used to represent the coupling risk event of cross-type components inside the packaging product; If all internal correlation degrees do not exceed the internal correlation degree threshold, external packaging product correlation analysis is performed: the risk exposure sequences of the same product type of other physically adjacent packaging products within the same period are extracted; the correlation degree between the risk exposure sequence of the initial early warning packaging product and the risk exposure sequence of each adjacent packaging product is calculated.

[0014] Preferably, the external joint risk sequence is generated according to the internal joint risk sequence, and the design parameters of the ontology attribute are optimized according to the external joint risk sequence, comprising: If the correlation degree with a certain adjacent packaging product exceeds the external correlation degree threshold, it is determined that the risk source is the interaction between the packaging products; the risk exposure sequences of the two parties with high correlation degree are spliced according to the time axis to generate an external joint risk sequence, which is used to represent the conduction risk event of cross-influence between the packaging products; If all external correlation degrees do not exceed the external correlation degree threshold, it is determined that the risk source is that the environmental stress in the external attribute has exceeded the common design margin of all related packaging products, and the design parameters of the ontology attribute are optimized.

[0015] The packaging management system based on multi-modal information fusion comprises: a data acquisition and attribute construction module for acquiring inherent parameters and environmental data of the packaging product to determine ontology attributes and external attributes, and generating a single change sequence and a single invariable sequence; a risk quantification and early warning module for obtaining a risk exposure sequence by processing the sequences in combination with a safety threshold, and determining an early risk signal by calculating a state offset and an offset rate, thereby confirming a proactive warning event and a risk level; a correlation analysis and sequence splicing module for calculating a correlation degree of the risk exposure sequence of the product internally or between products based on the risk level, and generating an internal joint risk sequence or an external joint risk sequence; a root cause determination and parameter optimization module for determining that the risk root cause is internal or product interaction based on the joint risk sequence, or determining that the environmental stress exceeds the design margin, and optimizing the design parameters of the ontology attributes.

[0016] Beneficial effects: The present application establishes a comparison and analysis mechanism of the single change sequence and the single invariable sequence by constructing a multi-modal data fusion system of the ontology attributes and external attributes of the packaging product, and realizes the transition of packaging risk from static threshold judgment to dynamic process monitoring. By setting a preset safety threshold interval and a preset deviation tolerance to point-by-point screen the sequences, a risk exposure sequence is constructed, and an early risk signal is determined in combination with a state offset and an offset rate, thereby forming a more rigorous proactive warning event triggering and three-level risk level classification method, which significantly improves the accuracy and early nature of risk identification. By calculating the correlation degree between the internal components of the product and the adjacent packaging outside, an internal joint risk sequence and an external joint risk sequence are generated, which can clearly determine and locate the risk root cause from the coupling of internal components of the packaging, the cross-influence between packaging, or the environmental stress exceeding the design margin, thereby realizing the accurate tracing of the risk root cause. Finally, by optimizing the design parameters of the ontology attributes, a complete management and control closed loop from risk warning to product design iteration and improvement is established, thereby effectively improving the adaptability, reliability and anti-risk ability of the packaging system in the complex flow process. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A packaging management and control method flowchart based on multi-modal information fusion of the present application; Figure 2 A packaging management and control system structure diagram based on multi-modal information fusion of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0019] Example 1: A packaging control method based on multimodal information fusion, such as Figure 1 As shown, it includes the following steps: S1-1: Obtain the inherent parameter data of the packaged product, determine the body attributes based on the inherent parameter data, obtain the pressure data, temperature data and humidity data of the environment in which the packaged product is located, and use the pressure data, temperature data and humidity data as external attributes; The inherent parameter data includes physicochemical property data, structural design parameters, and initial contents state. The physicochemical property data, structural design parameters, and initial contents state are used as intrinsic attributes. The intrinsic attributes refer to the static or quasi-static characteristics inherent to the packaging product itself and existing independently of the external environment. The external attributes refer to the environmental characteristics that exist outside the packaging product and interact dynamically with it, thereby affecting the state and behavior of the packaging product.

[0020] It should be noted that in the field of dynamic packaging management, traditional methods often struggle to address nonlinear risks in complex distribution environments due to their reliance on static thresholds. This solution achieves precise traceability of risk sources by constructing a multi-dimensional attribute system for packaged products. The data acquisition process involves two aspects: firstly, obtaining inherent packaging parameters through product databases and testing reports, including physicochemical properties such as material phase transition temperature and compressive strength, design parameters such as packaging wall thickness and sealing structure, and the initial mass and filling state of the contents; secondly, real-time monitoring of environmental pressure, temperature, and humidity data through a distributed sensor network. Defining inherent packaging parameters as intrinsic attributes and environmental monitoring data as extrinsic attributes clearly defines the boundaries of risk responsibility. Taking high-altitude self-heating food as an example, when the pressure sensor detects a sudden drop in external air pressure, combined with the sealing structure parameters in the packaging's intrinsic attributes, it determines the risk of leakage of the contents, thus overcoming the limitations of traditional isolated detection and laying a data foundation for subsequent risk transmission chain analysis.

[0021] S1-2: Obtain the product type corresponding to different types of packaged products, and determine the single change sequence and the single invariant sequence based on the product type, the ontological attribute and the external attribute; The product types include solid products, liquid products, gaseous products, and mixed products; For each product type, a preset defect attribute state is selected from the corresponding ontological attributes, and dynamically changing environmental parameters are extracted from the corresponding extrinsic attributes; the defect attribute state and the environmental parameters are combined to form a single change sequence; the single change sequence refers to a data sequence used to simulate and characterize the state evolution path under a dynamic external environment when a preset defect exists in a certain ontological attribute of the packaged product. For each product type, all corresponding ontological attributes are brought to a baseline normal state, and dynamically changing environmental parameters are extracted from the corresponding external attributes. The baseline normal state and the environmental parameters are combined to form a single invariant sequence. The single invariant sequence refers to a reference data sequence used to simulate and characterize the state evolution path under a dynamic external environment when all ontological attributes of the packaged product are in a baseline normal state.

[0022] It should be noted that, to overcome the shortcomings of existing packaging control methods that rely on static thresholds and are difficult to predict nonlinear risks, this step constructs a dual-sequence analysis model based on product type and attribute combinations to achieve dynamic simulation of risk evolution. Product types are categorized into solid products, liquid products, gaseous products, and mixed products based on the physical characteristics of their contents. For example, biscuits are solid products, juice is a liquid product, compressed gas canisters are gaseous products, and self-heating hot pots are classified as mixed products. This classification information is automatically obtained by parsing the product specification database and bill of materials.

[0023] Based on the classification results, a single change sequence is formed by combining preset defect attributes with dynamic environmental parameters. This sequence is used to simulate the deterioration path of a specific defect in a real circulation environment. For example, simulating a self-heating hot pot with "sealing ring aging," the internal pressure data sequence during the process of the ambient temperature rising from 25°C to 50°C is [(t1,1.0),(t2,1.1),(t3,1.3),(t4,1.6),(t5,2.0)] MPa. This sequence clearly shows an abnormal surge in pressure over time. The preset defects are derived from historical accident cases and accelerated aging data in the laboratory; for example, the aging coefficient of the packaging sealing ring is set as the defect parameter. Meanwhile, the inherent properties of intact packaging are combined with identical environmental parameters to form a single invariant sequence. This sequence serves as a benchmark for risk assessment. For example, under identical temperature rise conditions, the internal pressure sequence of a "well-sealed" self-heating hot pot is [(t1, 1.0), (t2, 1.05), (t3, 1.08), (t4, 1.1), (t5, 1.12)] MPa, showing stable fluctuations within the normal range. Comparative analysis of these two types of sequences can effectively distinguish between normal fluctuations and actual risks. For instance, when a sealing defect in the heating pack of a self-heating food product combines with an increase in ambient temperature, a single altered sequence provides an early warning of thermal runaway, while the single invariant sequence remains stable. This dual-sequence architecture overcomes the limitations of traditional single-point alarms, providing structured data support for subsequent risk transmission chain analysis.

[0024] S2-1: Obtain a risk exposure sequence by combining the single altered sequence with the single invariant sequence within a preset safety threshold range; The state values ​​of the single changed sequence and the single invariant sequence at the same sampling point are compared point by point. When the state value of the single changed sequence exceeds the preset safety threshold range, and the difference between the state value of the single changed sequence and the state value corresponding to the single unchanged sequence exceeds the preset deviation tolerance, the current data point of the single changed sequence is determined to be a risk data point. All the risk data points are combined in chronological order to form a risk exposure sequence, which is used to locate high-risk state data points that are directly caused by defects in specific ontology attributes and have exceeded normal safety behavior.

[0025] It should be noted that, to accurately pinpoint the unique risks arising from defects in the packaging itself, this step uses dual criteria to filter the data sequence. The preset safety threshold range is set based on product standards and transportation specifications; for example, the threshold for pressure vessels used for self-heating foods is set at 0.8-1.2 MPa. The preset deviation tolerance is used to distinguish between common problems caused by environmental factors and specific problems caused by defects. The preset deviation tolerance value is determined through statistical analysis of historical normal fluctuation data; for example, the preset deviation tolerance is set at 0.05 MPa.

[0026] The dual criteria are designed to effectively distinguish the root cause of risk: if the state values ​​of all packages (whether defective or not) exceed the safety threshold due to excessive environmental stress, the difference between the defective sequence and the non-defective sequence will be small, and the root cause of risk is determined to be a common environmental problem; if only the state value of the defective package deteriorates abnormally and produces a significant difference from the non-defective sequence, the risk is determined to be directly caused by a specific ontological attribute defect.

[0027] In practice, the single-change sequence and the single-invariant sequence are scanned point-by-point. Taking the pressure monitoring of self-heating food packaging as an example, at time t3, the pressure value of the single-change sequence is 1.3 MPa, and the pressure value of the single-invariant sequence is 1.08 MPa. First, it is determined that the pressure value of the single-change sequence (1.3 MPa) is greater than the upper limit of the safety threshold (1.2 MPa). Second, the difference between the single-change sequence and the single-invariant sequence (0.22 MPa) is calculated to be greater than the preset deviation tolerance (0.05 MPa). Therefore, the single-change sequence data point at time t3 is identified as a risk data point. All risk data points that meet both conditions constitute a risk exposure sequence in chronological order. The dual-criteria screening method effectively isolates common problems caused by environmental stress, enabling the system to focus on high-risk states directly caused by defects in the system's inherent properties that have exceeded normal safe behavior, providing accurate input for subsequent risk level classification.

[0028] S2-2: Determine early risk signals based on the single altered sequence and the single invariant sequence; Calculate the difference between the single changed sequence and the single unchanged sequence at the same sampling point, and use the difference as a state offset, which is used to quantify the instantaneous state deviation caused by the simulation defect; Calculate the rate of change of the state offset over time, and use the rate of change as the offset rate. The offset rate is used to quantify the state deterioration trend and severity caused by the simulated defect. If, within a preset sliding time window, the absolute value of the state offset continuously exceeds the offset threshold, and the offset rate continuously exceeds the rate threshold, then an early risk signal is determined to exist.

[0029] It should be noted that, to overcome the limitations of traditional packaging control's delayed response to explicit risks, this step achieves early risk identification by dynamically tracking the state evolution trend. The state offset is calculated by comparing the numerical difference between a single changing sequence and a single unchanged sequence at the same time. This parameter is used to quantify the immediate performance deviation caused by packaging defects. Example: At time t1, the defect simulation sequence shows an internal pressure of 1.15 MPa, while the baseline normal sequence shows 1.05 MPa. Therefore, the state offset at this moment = 1.15 - 1.05 = +0.10 MPa.

[0030] The offset rate is obtained by calculating the rate of change of the aforementioned state offset sequence over time, reflecting the dynamic trend and severity of defect deterioration. Example: At the next sampling point t2, if the state offset increases to 0.12 MPa, and the sampling interval Δt is 1 minute, then the offset rate ≈ (0.12 - 0.10) / 1 = +0.02 MPa / minute. A five-minute sliding time window is used for continuous monitoring. The offset threshold is set based on the material's tolerance limit, and the rate threshold is derived from regression analysis of historical accident data. Within the monitoring window, if the absolute value of the state offset continuously exceeds the offset threshold (e.g., 0.05 MPa), and the offset rate continuously exceeds the rate threshold (e.g., 0 MPa / minute, i.e., positive growth is required), then an early risk signal is identified. Comprehensive example: When the pressure state offset of self-heating food packaging remains above 0.05 MPa for 5 minutes, and its offset rate is simultaneously positive, the system will trigger an early warning. This dual criterion mechanism based on the "amplitude-trend" of the same state offset effectively avoids false alarms caused by instantaneous interference. By quantitatively tracking the evolution of defects, the control nodes can shift from post-event handling to pre-event early warning, providing a basis for decision-making in implementing predictive intervention measures.

[0031] S3-1: Based on the risk exposure sequence and the early risk signal, classify the risk level of the proactive early warning event to obtain the classification result; Extract all consecutive sequence segments whose values ​​exceed the risk threshold from the risk exposure sequence, and define each sequence segment as a risk exposure interval. The risk exposure interval is used to quantitatively characterize a persistent high-risk state event that exceeds the normal security mode. If the early risk signal is true and the risk exposure interval is a non-empty set, then an active early warning event is confirmed to have occurred. If the number of data points in the risk exposure interval is greater than the number threshold and the average value of the offset rate is greater than the rate average threshold, it is classified as a high-risk level; if only one of the conditions is met, it is classified as a medium-risk level; if neither condition is met, it is classified as a low-risk level.

[0032] It should be noted that, addressing the shortcomings of traditional early warning mechanisms, such as their crude risk level classification and lack of quantitative basis, this step achieves refined risk grading by establishing a multi-dimensional criterion system. Risk exposure intervals are obtained by identifying segments in the risk exposure sequence that continuously exceed a risk threshold, which is set according to product safety standards. Taking self-heating food packaging as an example, when pressure monitoring data exceeds 1.15 MPa for five consecutive sampling points, that period is marked as a risk exposure interval.

[0033] The simultaneous occurrence of early risk signals and risk exposure intervals constitutes the triggering condition for proactive early warning events, signifying that the system has entered an early warning state. Risk level classification comprehensively considers both the persistence and severity of risk: the data point quantity threshold is set based on the minimum effective early warning duration, while the average offset rate threshold is derived from safety margin analysis. For example, a risk exposure interval lasting for 10 sampling points with an average offset rate exceeding 0.01 MPa / minute is classified as high risk; meeting only the duration threshold or only exceeding the rate threshold defines medium risk; and neither indicator exceeds the threshold, classifying it as low risk. This grading model overcomes the limitations of the traditional dichotomy, providing quantitative support for the formulation of differentiated control strategies.

[0034] S3-2: Calculate the correlation degree of different types of packaged products based on the grade classification results; Based on the high-risk level and the medium-risk level, within the same packaged product, obtain the risk exposure range corresponding to the packaged product that triggered the proactive warning event; Based on the risk exposure interval, extract the risk exposure sequence of each component inside the packaging product during the warning period; Calculate the correlation between the risk exposure sequences of the solid product portion and the liquid product portion; Calculate the correlation between the risk exposure sequences of the solid product portion and the gaseous product portion; Calculate the correlation between the risk exposure sequences of the liquid product portion and the gaseous product portion.

[0035] It should be noted that, to overcome the shortcomings of traditional packaging control in neglecting the risk transmission between components, this step achieves risk tracing through internal product correlation analysis. When the system confirms a high-risk or medium-risk proactive warning event, it will automatically lock the corresponding risk exposure interval. Example: A risk exposure interval from the 3rd minute to the 5th minute is determined. Subsequently, all risk exposure sequence data recorded within this interval will be extracted for subsequent analysis. Continuing example: The temperature sequence [85, 92, 105]°C of the heating pack and the pressure sequence [1.16, 1.18, 1.22] MPa of the water bag are extracted within this 3-minute interval. For hybrid products containing multiple physical components (such as self-heating foods containing a solid heating pack, liquid water bag, and gaseous steam), the risk exposure sequences of each basic component are extracted separately for the above operations and cross-comparison.

[0036] The correlation coefficient is used to calculate the correlation, and its formula is as follows: ,in, The Pearson correlation coefficient measures the degree of linear correlation between two risk exposure sequences, and its value ranges from [-1, 1]. , This represents the risk exposure sequence of two different components that need to be analyzed for correlation in the first... Data values ​​of each sampling point , Let x and y be the arithmetic mean of the risk exposure sequences x and y respectively during the warning period. This represents the number of sampling points during the warning period. By calculating three sets of correlation coefficients—solid-liquid, solid-gas, and liquid-gas—the risk coupling relationships between components are identified. For example, when the correlation coefficient between the temperature sequence of the heating pack and the pressure sequence of the water bag exceeds 0.8, it confirms the existence of heat conduction risk. This multi-dimensional correlation model overcomes the limitations of traditional single-indicator monitoring and provides a quantitative basis for constructing a risk propagation network.

[0037] S4-1: Generate an internal joint risk sequence based on the aforementioned correlation degree; If the correlation between any two parts exceeds the internal correlation threshold, the root cause of the risk is determined to be the internal interaction of the packaging product; the risk exposure sequences of the two highly correlated parts are spliced ​​together along the time axis to generate an internal joint risk sequence, which is used to characterize the coupled risk events of cross-type components within the packaging product; If none of the internal correlations exceed the internal correlation threshold, then perform external packaging product correlation analysis: extract the risk exposure sequences of other physically adjacent packaging products of the same product type within the same time period; calculate the correlation between the risk exposure sequence of the initial warning packaging product and the risk exposure sequence of each adjacent packaging product.

[0038] It should be noted that, addressing the shortcomings of insufficient risk tracing capabilities in traditional packaging management, this step achieves precise risk root cause localization by constructing an internal and external two-level correlation analysis model. The internal correlation threshold is set based on the statistical significance of historical coupled accident data, typically using a Pearson correlation coefficient of 0.7 or higher, to identify whether there are physicochemical interactions between components. When the correlation between the temperature sequence of the heating pack and the pressure sequence of the water bag in a self-heating food exceeds this threshold, the two sequences are spliced ​​along the time axis to generate an internal joint risk sequence. This sequence fully presents the dynamic process of heat-pressure coupling. For example, the generated internal joint risk sequence is: [(t1, 85°C, 1.16MPa), (t2, 92°C, 1.18MPa), (t3, 105°C, 1.22MPa)], which clearly shows the synchronous increase in temperature and pressure. After generating the joint sequence, the corresponding pattern in the risk transmission chain knowledge base is automatically matched, and specific formation mechanisms and handling suggestions are output, such as "heating pack sealing failure leads to increased heat conduction and water bag vaporization". This knowledge base is pre-built based on historical and experimental data by extracting joint sequence features and mapping them to risk mechanisms and mitigation measures.

[0039] If all internal correlations do not exceed the internal correlation threshold, then external packaging product correlation analysis is performed. Physical proximity refers to spatial distance within the effective range of heat conduction or pressure wave action; causal timing consistency is ensured within the same time period; initial warning refers to the first packaging unit that triggers the alarm. External correlation calculation employs an improved dynamic time warping algorithm. ,in, Normalized dynamic time-warped distance (RTD) is used to measure the morphological similarity between two risk exposure sequences of different packaged products. It has been calculated by dividing by the path length. This eliminates the impact of sequence scaling on distance values, ensuring the comparability of distances between different sequence pairs. The dynamic time-warped original distance is the cumulative minimum Euclidean distance after finding the optimal alignment path between two sequences. The length of the optimal curved path (i.e., the number of steps in the path) is the path that, under the condition of satisfying temporal constraints, maximizes the cumulative distance. Minimize sequence index pairs The set, For sequence In the optimal path Step index The corresponding data value, For sequence In the optimal path Step index The corresponding data value, ,in , This represents a risk exposure sequence for two products with different packaging. , , The time-series alignment weights are used. This hierarchical judgment mechanism breaks through the limitations of traditional single-dimensional diagnosis and provides a theoretical basis for differentiated intervention strategies.

[0040] S4-2: Generate an external joint risk sequence based on the internal joint risk sequence, and optimize the design parameters of the ontology attribute based on the external joint risk sequence.

[0041] If the correlation with a neighboring packaged product exceeds the external correlation threshold, the root cause of the risk is determined to be the interaction between the packaged products. The risk exposure sequences of the two highly correlated parties are spliced ​​together along the time axis to generate an external joint risk sequence, which is used to characterize the transmission risk events of the cross-influence between the packaged products. If none of the external correlations exceed the external correlation threshold, the risk root cause is determined to be that the environmental stress in the external attribute has exceeded the common design margin of all related packaging products, and the design parameters of the ontological attribute are optimized.

[0042] It should be noted that, to address the limitations of traditional packaging control in distinguishing between internal defects and external environmental influences, this step employs a three-tiered judgment mechanism to accurately identify the root causes of risks. The external correlation threshold is set based on industry safety distance standards and experimental data, typically with a dynamic time-normalized distance of less than 0.5. When the correlation exceeds this threshold, physical interaction between packaging units is determined. For example, when the pressure sequences of adjacent self-heating food packaging exhibit synchronous fluctuations and the correlation exceeds the standard, the risk exposure sequences of both are spliced ​​along the time axis to generate an external joint risk sequence. This sequence fully records the transmission path of risk between packaging units. Example: The generated external joint risk sequence is: [(t1, 1.20MPa, 1.18MPa), (t2, 1.25MPa, 1.23MPa), (t3, 1.30MPa, 1.28MPa)], where each data point contains the pressure values ​​of packaging A and packaging B at the same time, visually demonstrating the synchronous fluctuations in their pressures. After generating the external joint sequence, an isolation and disposal protocol is automatically initiated, including adjusting storage layout parameters and updating safety distance standards.

[0043] If all external correlations do not exceed the aforementioned external correlation threshold, the root cause of the risk is determined to be that the environmental stress in the external attributes has exceeded the common design margin of all related packaging products, i.e., the safety redundancy reserved in the standard design of the packaging cluster. In this case, it is necessary to optimize the design parameters of the intrinsic attributes, including enhancing the temperature resistance coefficient of materials, improving the strength of the sealing structure, or adjusting the content ratio. This progressive judgment logic overcomes the ambiguity of traditional experience-based judgments, providing a data-driven decision-making basis for the iterative upgrading of packaging systems.

[0044] Example 2: Based on Example 1, a packaging control system based on multimodal information fusion, such as... Figure 2 As shown, it includes: The data acquisition and attribute construction module is used to acquire inherent parameters and environmental data of the packaged product to determine the intrinsic and extrinsic attributes, and to generate single change sequences and single invariant sequences. The risk quantification and early warning module is used to combine the sequence with the safety threshold to obtain the risk exposure sequence, and to determine the early risk signal by calculating the state offset and offset rate, thereby confirming the proactive warning event and risk level. The correlation analysis and sequence splicing module calculates the correlation degree of risk exposure sequences within or between products based on risk level, and generates internal joint risk sequences or external joint risk sequences. The root cause identification and parameter optimization module is used to determine whether the root cause of risk is internal or between products based on the joint risk sequence; otherwise, it is determined that the environmental stress exceeds the design margin, and the design parameters of the ontological properties are optimized.

[0045] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A packaging control method based on multimodal information fusion, characterized by: Includes the following steps: S1: Obtain the inherent parameter data of the packaged product, and determine the intrinsic attributes based on the inherent parameter data; obtain the pressure data, temperature data, and humidity data of the environment in which the packaged product is located, and use the pressure data, temperature data, and humidity data as extrinsic attributes; obtain the product types corresponding to different types of packaged products, and determine single change sequences and single invariant sequences based on the product types combined with the intrinsic attributes and the extrinsic attributes. S2: Obtain a risk exposure sequence by combining the single changed sequence with the single unchanged sequence within a preset safety threshold range; Early risk signals are determined based on the single altered sequence and the single invariant sequence; S3: Classify the risk level of the proactive warning event based on the risk exposure sequence and the early risk signal to obtain the classification result; calculate the correlation degree of different types of packaged products based on the classification result; S4: Generate an internal joint risk sequence based on the correlation degree; generate an external joint risk sequence based on the internal joint risk sequence, and optimize the design parameters of the ontology attribute based on the external joint risk sequence.

2. The packaging control method based on multimodal information fusion as described in claim 1, characterized in that, The process of acquiring inherent parameter data of the packaged product, determining its intrinsic attributes based on the inherent parameter data, acquiring pressure, temperature, and humidity data of the environment in which the packaged product is located, and using the pressure, temperature, and humidity data as external attributes includes: The inherent parameter data includes physicochemical property data, structural design parameters, and initial contents state. The physicochemical property data, structural design parameters, and initial contents state are used as intrinsic attributes. The intrinsic attributes refer to the static or quasi-static characteristics inherent to the packaging product itself and existing independently of the external environment. The external attributes refer to the environmental characteristics that exist outside the packaging product and interact dynamically with it, thereby affecting the state and behavior of the packaging product.

3. The packaging control method based on multimodal information fusion as described in claim 1, characterized in that, The step of obtaining the product types corresponding to different types of packaged products, and determining single-change sequences and single-invariant sequences based on the product types in combination with the ontological attributes and the extrinsic attributes, includes: The product types include solid products, liquid products, gaseous products, and mixed products; For each product type, a preset defect attribute state is selected from the corresponding ontological attributes, and dynamically changing environmental parameters are extracted from the corresponding extrinsic attributes; the defect attribute state and the environmental parameters are combined to form a single change sequence; the single change sequence refers to a data sequence used to simulate and characterize the state evolution path under a dynamic external environment when a preset defect exists in a certain ontological attribute of the packaged product. For each product type, all corresponding ontological attributes are brought to a baseline normal state, and dynamically changing environmental parameters are extracted from the corresponding external attributes. The baseline normal state and the environmental parameters are combined to form a single invariant sequence. The single invariant sequence refers to a reference data sequence used to simulate and characterize the state evolution path under a dynamic external environment when all ontological attributes of the packaged product are in a baseline normal state.

4. The packaging control method based on multimodal information fusion as described in claim 1, characterized in that, The step of obtaining the risk exposure sequence by combining the single altered sequence with the single invariant sequence within a preset safety threshold range includes: The state values ​​of the single changed sequence and the single invariant sequence at the same sampling point are compared point by point. When the state value of the single changed sequence exceeds the preset safety threshold range, and the difference between the state value of the single changed sequence and the state value corresponding to the single unchanged sequence exceeds the preset deviation tolerance, the current data point of the single changed sequence is determined to be a risk data point. All the risk data points are combined in chronological order to form a risk exposure sequence, which is used to locate high-risk state data points that are directly caused by defects in specific ontology attributes and have exceeded normal safety behavior.

5. The packaging control method based on multimodal information fusion as described in claim 1, characterized in that, The step of determining early risk signals based on the single altered sequence and the single invariant sequence includes: Calculate the difference between the single changed sequence and the single unchanged sequence at the same sampling point, and use the difference as a state offset, which is used to quantify the instantaneous state deviation caused by the simulation defect; Calculate the rate of change of the state offset over time, and use the rate of change as the offset rate. The offset rate is used to quantify the state deterioration trend and severity caused by the simulated defect. If, within a preset sliding time window, the absolute value of the state offset continuously exceeds the offset threshold, and the offset rate continuously exceeds the rate threshold, then an early risk signal is determined to exist.

6. The packaging control method based on multimodal information fusion as described in claim 1, characterized in that, The step of classifying the risk level of proactive early warning events based on the risk exposure sequence and the early risk signal to obtain the level classification result includes: Extract all consecutive sequence segments whose values ​​exceed the risk threshold from the risk exposure sequence, and define each sequence segment as a risk exposure interval. The risk exposure interval is used to quantitatively characterize a persistent high-risk state event that exceeds the normal security mode. If the early risk signal is true and the risk exposure interval is a non-empty set, then an active early warning event is confirmed to have occurred. If the number of data points in the risk exposure interval is greater than the number threshold and the average value of the offset rate is greater than the rate average threshold, it is classified as a high-risk level; if only one of the conditions is met, it is classified as a medium-risk level; if neither condition is met, it is classified as a low-risk level.

7. The packaging control method based on multimodal information fusion as described in claim 6, characterized in that, The calculation of the correlation degree of different types of packaged products based on the classification results includes: Based on the high-risk level and the medium-risk level, within the same packaged product, obtain the risk exposure range corresponding to the packaged product that triggered the proactive warning event; Based on the risk exposure interval, extract the risk exposure sequence of each component inside the packaging product during the warning period; Calculate the correlation between the risk exposure sequences of the solid product portion and the liquid product portion; Calculate the correlation between the risk exposure sequences of the solid product portion and the gaseous product portion; Calculate the correlation between the risk exposure sequences of the liquid product portion and the gaseous product portion.

8. The packaging control method based on multimodal information fusion as described in claim 1, characterized in that, The step of generating an internal joint risk sequence based on the correlation includes: If the correlation between any two parts exceeds the internal correlation threshold, the root cause of the risk is determined to be the internal interaction of the packaging product; the risk exposure sequences of the two highly correlated parts are spliced ​​together along the time axis to generate an internal joint risk sequence, which is used to characterize the coupled risk events of cross-type components within the packaging product; If none of the internal correlations exceed the internal correlation threshold, then perform external packaging product correlation analysis: extract the risk exposure sequences of other physically adjacent packaging products of the same product type within the same time period; calculate the correlation between the risk exposure sequence of the initial warning packaging product and the risk exposure sequence of each adjacent packaging product.

9. The packaging control method based on multimodal information fusion as described in claim 1, characterized in that, The step of generating an external joint risk sequence based on the internal joint risk sequence, and optimizing the design parameters of the ontology attribute based on the external joint risk sequence, includes: If the correlation with a neighboring packaged product exceeds the external correlation threshold, the root cause of the risk is determined to be the interaction between the packaged products. The risk exposure sequences of the two highly correlated parties are spliced ​​together along the time axis to generate an external joint risk sequence, which is used to characterize the transmission risk events of the cross-influence between the packaged products. If none of the external correlations exceed the external correlation threshold, the risk root cause is determined to be that the environmental stress in the external attribute has exceeded the common design margin of all related packaging products, and the design parameters of the ontological attribute are optimized.

10. A packaging control system based on multimodal information fusion, used to implement the packaging control method based on multimodal information fusion as described in any one of claims 1-9, characterized in that, include: The data acquisition and attribute construction module is used to acquire inherent parameters and environmental data of the packaged product to determine the intrinsic and extrinsic attributes, and to generate single change sequences and single invariant sequences. The risk quantification and early warning module is used to combine the sequence with the safety threshold to obtain the risk exposure sequence, and to determine the early risk signal by calculating the state offset and offset rate, thereby confirming the proactive warning event and risk level. The correlation analysis and sequence splicing module calculates the correlation degree of risk exposure sequences within or between products based on risk level, and generates internal joint risk sequences or external joint risk sequences. The root cause identification and parameter optimization module is used to determine whether the root cause of risk is internal or between products based on the joint risk sequence; otherwise, it is determined that the environmental stress exceeds the design margin, and the design parameters of the ontological properties are optimized.

Images