Liquefied gas cylinder full-chain traceability and safety early warning method based on multi-modal fusion

Abstract

The present application relates to liquefied gas cylinder safety management and traceability early warning technical field, specifically relates to a liquefied gas cylinder whole chain traceability and safety early warning method based on multi-modal fusion, the method comprises: establishing a traceability database, defining a business node set and constructing a liquefied gas cylinder whole chain state machine;Link records are formed by collecting multi-modal data and business event data in the flow transfer;Based on link record preprocessing and event segmentation, cumulative damage risk index and thermal pressure overrun preview margin index are calculated;Accordingly, the comprehensive risk level is determined to drive state transition and issue dynamic control instructions, and the execution result is written back to close loop, with the unique identity of the liquefied gas cylinder running through the business node set, and the link record is written into the traceability database;Multi-modal data preprocessing and event segmentation, cumulative damage risk index and thermal pressure overrun preview margin index are calculated;Accordingly, the comprehensive risk level is determined to drive the state machine to issue dynamic control instructions and write back to close loop.

Description

Technical Field

[0001] This invention relates to the field of liquefied petroleum gas (LPG) cylinder safety management and traceability early warning technology, and more specifically, to a method for full-chain traceability and safety early warning of LPG cylinders based on multimodal fusion. Background Technology

[0002] As pressure vessels, liquefied petroleum gas (LPG) cylinders frequently circulate through filling, warehousing, transportation, distribution, on-site safety inspections, and inspection and disposal. Management objectives typically include consistent identification, traceability of circulation responsibility, and closed-loop detection and handling of abnormal states. In current practice, circulation information is commonly recorded using QR codes, RFID tags, or manual ledgers, with a traceability platform centrally storing business event data such as filling information, warehousing information, and transportation handover information. Additionally, in some scenarios, sensor data such as temperature, attitude, or acceleration are used for on-site alarms or post-event analysis.

[0003] The above methods can support basic circulation registration and compliance management, but under complex operating conditions, some aspects may still require further improvement: First, business event data and physical status data often come from different sources and have inconsistent time bases, making it difficult to form continuous and consistent evidence for the records of the same liquefied gas cylinder at different business nodes, resulting in information fragmentation or high review costs; Second, sensor data is easily affected by factors such as noise, gravity component superposition, and temperature zero-point drift. Without unified preprocessing and event segmentation rules, the boundaries of key events such as impact, tipping, and heat exposure are unstable, making it difficult to obtain repeatable wind speed data. Third, existing early warnings are often triggered by a single threshold, reflecting more immediate exceedances than cumulative damage and future trends. When basic parameters such as bottle age, remaining inspection validity period, filling volume, and volume parameters change, the risk sensitivity adjustment is insufficient, which can easily lead to early warning deviations at different specifications and service stages. Fourth, the handling measures are mostly limited to prompts or manual handling, and rarely link the risk assessment results with executable control actions such as transportation and distribution routes, stop windows, and delivery sequences, or write back the instruction execution trajectory to form closed-loop evidence, thus affecting the quantitative verification and continuous optimization of risk reduction effects.

[0004] Based on the above situation, there is an urgent need for a method that can unify identity recognition, link recording, multimodal fusion computing, future hot pressing trend prediction, and risk-driven dynamic control and write-back closed loop across the entire chain, in order to improve traceability consistency, early warning foresight, and actionability. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method for full-chain traceability and safety early warning of liquefied gas cylinders based on multimodal fusion.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A multimodal fusion-based method for end-to-end traceability and safety early warning of liquefied petroleum gas (LPG) cylinders includes: Step 1: Generate and solidify a unique identifier for each LPG cylinder, establish a traceability file corresponding to the unique identifier of the LPG cylinder, the traceability file contains a set of basic parameters, establish a traceability database, define a set of business nodes covering the entire LPG cylinder chain, and construct a state machine for the entire LPG cylinder chain. Step 2: Collect multimodal data and business event data during the flow of liquefied gas cylinders through business nodes, and form link records in the traceability database; Step 3: Perform preprocessing and event segmentation on the multimodal data based on the link records, and then calculate the cumulative damage risk index and the thermal pressure over-limit prediction margin index. Step 4: Determine the comprehensive risk level based on the cumulative damage risk index and the thermal pressure over-limit prediction margin index, drive the state machine of the entire liquefied gas cylinder chain to perform state transitions, and generate and issue dynamic control instructions based on the comprehensive risk level. The comprehensive risk level and the execution results of the dynamic control instructions are written into the traceability database to realize the closed loop of full-chain traceability and safety early warning.

[0007] In one embodiment, the set of basic parameters includes bottle age, remaining inspection expiry date, filling volume, volume parameters, and pressure safety threshold.

[0008] In one embodiment, the liquefied gas cylinder full-chain state machine includes at least a normal flow state, a review state, a predictive scheduling state, and a frozen disposal state.

[0009] In one embodiment, the preprocessing includes at least noise reduction, gravity component separation, and temperature baseline correction; the event segmentation includes at least impact event identification, tipping event identification, and thermal exposure event identification.

[0010] In one embodiment, the cumulative damage risk index is obtained as follows: the impact equivalent energy is calculated for the impact event; the tilting posture deviation is calculated for the tilting event; the heat exposure thermal stress is calculated for the heat exposure event; and the impact equivalent energy, the tilting posture deviation, and the heat exposure thermal stress are combined using a fusion function to obtain the cumulative damage risk index.

[0011] In one embodiment, the fusion function includes a time decay weight and a cross-modal coupling correction term, which are set based on a set of basic parameters and vary at least with the bottle age, remaining validity period, filling amount, and volume parameters.

[0012] In one embodiment, the hot pressure over-limit prediction margin index is calculated as follows: a future temperature sequence within a preset prediction time window is established based on multimodal data; a future pressure sequence is established based on the future temperature sequence, the set of basic parameters, and the correspondence between pressure and temperature; and the hot pressure over-limit prediction margin index is calculated based on the future pressure sequence and the pressure safety threshold.

[0013] In one embodiment, the cumulative damage risk index is compared with a preset set of cumulative damage risk thresholds to obtain the damage risk level, the hot pressing over-limit prediction margin index is compared with a preset set of hot pressing over-limit prediction margin thresholds to obtain the hot pressing risk level, and the comprehensive risk level is determined based on the damage risk level and the hot pressing risk level.

[0014] In one embodiment, when the overall risk level meets the conditions for entering the predictive scheduling state, predictive scheduling dynamic control instructions are issued to the transportation business nodes and delivery business nodes in the business node set; when the overall risk level meets the conditions for entering the review state, a review dynamic control instruction is issued to the next business node corresponding to the link record to enforce review and evidence collection and review result backfilling; when the overall risk level meets the conditions for entering the freeze disposal state, a freeze dynamic control instruction is issued to the filling business nodes, inbound and outbound business nodes and delivery business nodes in the business node set to prohibit continued filling, prohibit outbound and prohibit order dispatch and delivery, triggering inspection or scrap disposal work orders.

[0015] In one embodiment, the predictive scheduling dynamic control instructions include at least changing the transportation and delivery route to reduce heat exposure, adjusting the delivery order to shorten the time spent in the open air, setting a ventilated and shaded area for cooling and parking, and limiting the maximum allowable stay time in heat exposure scenarios, and writing the execution results of the predictive scheduling dynamic control instructions back to the traceability database.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention uses the unique identifier of a liquefied petroleum gas (LPG) cylinder as a central index to establish a traceability archive containing a set of basic parameters. During the flow of LPG cylinders through various business nodes, multimodal data and business event data are collected and linked to form a traceability database. This allows events and data from filling, warehousing, transportation, and distribution to be continuously linked on the same link, improving the consistency, traceability, and verifiability of information across the entire chain. It also facilitates accurate location and review of the responsible party, time and location, node flow, and source of anomalies.

[0017] This invention performs preprocessing and event segmentation on multimodal data based on link records, and then calculates the cumulative damage risk index and the thermal pressure over-limit prediction margin index. Impact, tipping and thermal exposure are stably extracted from the raw data into repeatable event segments and quantified into risk indicators. This improves risk assessment from single-point alarm to joint judgment based on link history and future prediction time windows, enhances the accuracy and robustness of the characterization of the safety status of liquefied gas cylinders, and provides interpretable and reproducible quantitative basis for subsequent risk level determination.

[0018] This invention determines the comprehensive risk level based on the cumulative damage risk index and the thermo-pressure over-limit predictability margin index, drives the state machine of the entire LPG cylinder chain to perform state transitions, and generates and issues dynamic control instructions based on the comprehensive risk level. Furthermore, it writes back the comprehensive risk level and the execution results of the dynamic control instructions to the traceability database, realizing closed-loop management from risk assessment to disposal execution and result tracking. In particular, when entering the predictive scheduling state, it can link changes to transportation and delivery routes, adjust delivery order, set ventilated and shaded cooling and parking windows, and limit the maximum allowable dwell time in heat exposure scenarios, thereby enhancing the foresight of early warning and the feasibility of disposal, and reducing the probability of heat exposure and thermo-pressure over-limit risks. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for full-chain traceability and safety early warning of liquefied gas cylinders based on multimodal fusion. Figure 2 A diagram illustrating the data association between the unique identifier of a liquefied gas cylinder, traceability records, business node sets, multimodal data, business event data, link records, and the traceability database; Figure 3 A schematic diagram of the state transitions of the entire chain state machine for liquefied gas cylinders; Figure 4 This diagram illustrates the preprocessing and event segmentation of multimodal data based on link records, and the calculation of the cumulative damage risk index and the thermal pressure over-limit prediction margin index. Detailed Implementation

[0020] Reference Figure 1 A multimodal fusion-based method for full-chain traceability and safety early warning of liquefied gas cylinders includes: Step 1: Generate and solidify a unique identifier for each LPG cylinder, establish a traceability file corresponding to the unique identifier, including a set of basic parameters, build a traceability database, define a set of business nodes covering the entire LPG cylinder supply chain, and construct a state machine for the entire LPG cylinder supply chain. This completes the basic infrastructure for object uniqueness, data archiving, chain standardization, and machine-readable states. By generating and solidifying a unique identifier for each LPG cylinder, the identity of the LPG cylinder can be uniquely identified and cannot be confused throughout the entire supply chain, and it can be stably associated between different business nodes. Establishing a traceability file corresponding to the unique identifier of the LPG cylinder, and ensuring that the traceability file contains a set of basic parameters, provides a unified and reliable basic input condition for subsequent calculations, judgments, and instruction generation. A traceability database is established to store subsequent link records, risk calculation results, and dynamic control command execution results, enabling centralized data storage and traceable querying. A set of business nodes covering the entire LPG cylinder chain is defined, clearly defining the chain's boundaries and facilitating consistent data collection, recording, and handling standards at each node. A state machine for the entire LPG cylinder chain is constructed, providing a clear set of states and state transition logic for the cylinder's circulation and risk handling states. This provides a unified platform for driving state transitions based on comprehensive risk levels, thereby supporting the realization of a closed-loop system for full-chain traceability and safety early warning.

[0021] Reference Figure 2 In one specific implementation, a unique identification mark is generated and solidified for each liquefied petroleum gas (LPG) cylinder. Specifically, the unique identification mark can be solidified at a fixed position on the cylinder body using a wear-resistant permanent marking method. When the unique identification mark is read for the first time, a traceability file corresponding to the unique identification mark is established. The traceability file contains a set of basic parameters, including cylinder age, remaining inspection validity period, filling volume, volume parameters, and pressure safety threshold. At the same time, a traceability database is established, and the unique identification mark of the LPG cylinder, the set of basic parameters, and the time of their formation are written into the traceability database to form a traceable starting record, thereby ensuring that the identification of the same LPG cylinder is consistent at any subsequent business node and that the set of basic parameters can be continuously referenced. Define a set of business nodes covering the entire LPG cylinder chain, and uniformly encode and sequentially associate these business nodes in the traceability database. This leads to the construction of a full-chain state machine for LPG cylinders. The full-chain state machine includes at least four states: normal circulation, review, predictive scheduling, and frozen disposal. Executable state transition rules are configured for each state, enabling subsequent state transitions to be driven and disposal paths to be constrained based on risk-related results. For example, an LPG cylinder with an age of eight years, insufficient remaining inspection validity, and a volume parameter of 12.5 liters, is in the normal circulation state when entering the inbound / outbound process of the business node set. It transitions to the review state when the review condition is met, to the predictive scheduling state when the predictive scheduling condition is met, and to the frozen disposal state when the frozen disposal condition is met. This ensures that the LPG cylinder maintains a consistent association with the traceability archive and traceability database throughout the entire chain, using its unique LPG cylinder identifier as an index, thus achieving a traceable, constrainable, and backtrackable process foundation.

[0022] Step Two: Collect multimodal data and business event data during the flow of liquefied gas cylinders through the business node set, and create a link record in the traceability database. This data is used to digitize, replay, and trace the actual flow process of liquefied gas cylinders within the business node set. By collecting multimodal data and business event data during the flow of liquefied gas cylinders, it is possible to simultaneously cover information on changes in the cylinder's state, such as stress, attitude, and heat exposure, as well as corresponding business behavior information, thus ensuring that risk analysis is not limited to a single dimension. The collected multimodal data and business event data are linked together in the traceability database to form a chain record. This ensures that every transfer, every event, and every piece of sensor data can be associated with the unique identifier of the liquefied gas cylinder, thus forming a continuous chain of evidence throughout the entire process. The formation of the chain record enables subsequent steps to be based on the same chain record for preprocessing, event segmentation, and indicator calculation, and provides a verifiable data foundation for subsequent comprehensive risk level determination, dynamic control command issuance, and traceability.

[0023] In one specific implementation, multimodal data and business event data are collected during the flow of liquefied gas cylinders through the business node set, and a link record is formed in the traceability database. Specifically, when a liquefied gas cylinder enters any business node in the business node set, the unique identifier of the liquefied gas cylinder is read to determine the traceability file association, and the node identifier, arrival time, departure time, location of occurrence, and execution entity identifier of the business node are used as the basic fields of the business event data. At the same time, multimodal data is continuously or collected on the liquefied gas cylinder during the operation period of the business node, or according to the trigger conditions. The multimodal data may include acceleration change data reflecting impact, posture change data reflecting tilting, and temperature change data reflecting heat exposure. To ensure that the link records are verifiable and reproducible, multimodal data and business event data are timestamped and indexed according to a unified time base. The multimodal data is subjected to collection integrity verification and outlier filtering. The qualified data and business event data are encapsulated together into the same link record and written into the traceability database. The link record contains at least the unique identifier of the liquefied gas cylinder, the business node identifier, the business event data field, and the multimodal data summary or original fragment of the corresponding time period. This allows any business node to trace the behavior and status changes of the liquefied gas cylinder at that node in the traceability database. For example, in a transportation business node, if a liquefied gas cylinder experiences a short-term violent impact and a temperature rise after loading and unloading, the collected acceleration change data and temperature change data, along with the arrival time, departure time, location of the incident, and the identifier of the executing entity of the transportation business node, are combined to form a link record and written into the traceability database. This allows the risk source of the transfer to be located and evidenced based on the link record.

[0024] Reference Figure 4 Step 3: Based on the link records, preprocessing and event segmentation are performed on the multimodal data to calculate the cumulative damage risk index and the thermal pressure over-limit prediction margin index. This transforms the raw multimodal data in the link records into key indicators that can be used for risk assessment and early warning, realizing the core link from data to risk quantification. Preprocessing the multimodal data improves data quality and usability, making data from different sources and with different sampling characteristics comparable and stable, and reducing the errors caused by noise and drift in subsequent calculations. Event segmentation extracts effective risk-related segments from the continuous time series, enabling subsequent calculations to focus on events with safety significance, thereby avoiding the dilution of risk characteristics by invalid data. Based on the preprocessed and event-segmented data, a cumulative damage risk index is calculated, enabling a unified and quantitative expression of the cumulative impact of factors such as impact, tipping, and heat exposure on the safety status of LPG cylinders throughout the entire supply chain. This facilitates comparison with threshold systems and the formation of risk classifications. A thermo-pressure exceedance predictability margin index is also calculated, providing predictive capability for pressure-related risks caused by heat exposure within a predictable time window. This allows for early quantitative judgments before actual exceedances, thus providing a direct basis for subsequent comprehensive risk level assessments and the issuance of dynamic control commands.

[0025] In one specific implementation, preprocessing is performed on the multimodal data in the link record. The preprocessing includes at least denoising, gravity component separation, and temperature baseline correction. Denoising is used to suppress random fluctuations and transient spikes in the multimodal data. It can be combined with time window smoothing and frequency domain filtering to retain the effective changes in impact, tipping, and heat exposure. Gravity component separation is used to remove the static gravity term from the acceleration change data, so that the subsequent characterization of impact intensity focuses on dynamic force changes. The gravity direction can be estimated and separated using attitude change information or static segments in the multimodal data. Temperature baseline correction is used to eliminate zero-point drift and environmental baseline differences in temperature change data. A temperature baseline can be established by selecting relatively stable temperature segments in the link record, and the temperature change data can be normalized relative to the temperature baseline to ensure the comparability of heat exposure data from different service nodes and different time periods. After preprocessing, event segmentation is performed on the multimodal data. Event segmentation includes at least impact event identification, tipping event identification, and thermal exposure event identification. Impact event identification sets amplitude and duration thresholds in the acceleration change data after gravity component separation, combined with hysteresis rules, to avoid noise triggering and ensure event boundary stability. Tipping event identification sets tipping angle and angular velocity thresholds in the attitude change data to identify different forms of continuous tipping and repeated tipping, and records the start and end times and peak characteristics of each tipping. Thermal exposure event identification sets over-temperature thresholds, temperature rise rate thresholds, and duration thresholds in the temperature change data after temperature baseline correction to identify two types of events: short-term high-temperature exposure and long-term medium-temperature exposure, and forms an event index consistent with the link records. To ensure that the threshold determination methods for impact event identification, tipping event identification, and thermal exposure event identification are reproducible, a baseline segment can be selected from the link records during a period of relatively stable temperature and without significant impact or tipping corresponding to the business event data. The mean value of the acceleration change data amplitude sequence after gravity component separation can then be calculated for each baseline segment. with standard deviation The mean of the tilt angle sequence in the attitude change data with standard deviation The mean of the temperature change data series after temperature baseline correction with standard deviation Based on this, the amplitude threshold for impact event recognition is set as follows: And set the lower threshold of hysteresis as and Simultaneously set the duration threshold to This causes the amplitude of the acceleration change data to be continuously higher than And the duration is not less than The time was determined to be the start of the impact event, continuously below And the duration is not less than The impact event is considered over when the time is determined; the tilt angle threshold for tilting event recognition is set to [value missing]. And set the angular velocity threshold to in and Calculated from the tilt angular velocity sequence, the tilt angle exceeds And the tilting angular velocity exceeds If the temperature continues to rise for a preset duration, it is considered a tipping event; the over-temperature threshold for identifying heat exposure events is set as follows. And set the temperature rise rate threshold to in and Calculated from the temperature rise rate sequence of temperature change data, making the temperature change data exceed or the rate of temperature rise exceeds And if it continues for a preset duration, it is determined to be a heat exposure event; among which The minimum threshold value configured according to the basic parameter set is the coefficient. It can be obtained by fitting labeled event samples from calibration experiments or historical link records, and stored together with the basic parameter set in the traceability database for retrieval. The above threshold parameters are updated after the cumulative number of samples reaches the preset number or after an anomaly is confirmed at the inspection and scrapping node. The cumulative damage risk index is calculated based on the event index. Specifically, for impact events, the equivalent impact energy is calculated, which can be obtained by accumulating the dynamic acceleration intensity over time during the duration of the impact event. The equivalent energy can be further characterized by combining the duration difference of the impact pulse. For tipping events, the tipping posture deviation is calculated, which can be obtained by accumulating the excess amount of the tipping angle exceeding the tipping angle threshold over time. The number of repeated tippings and the recovery process of each tipping are included in the deviation accumulation to reflect the degree of instability of the LPG cylinder during handling and placement. For thermal exposure events, the thermal stress is calculated, which can be obtained by accumulating the over-temperature after the temperature exceeds the temperature baseline over time. The degree of thermal shock is characterized by combining the temperature rise rate. Suppose that the acceleration change data is separated into a dynamic acceleration amplitude sequence after gravity component separation. The tilt angle sequence is obtained from the attitude change data. With tilt angular velocity sequence Temperature change data were corrected for temperature baseline to obtain a temperature overheat sequence. ,in The sampling sequence number and the sampling period are And let the impact event index give the start and end sampling sequence numbers of the impact event. The start and end sampling sequence numbers of the dumping event are: The start and end sampling sequence numbers of the heat exposure event are: The impact equivalent energy is defined as follows: It can also introduce duration-consistent characterization items. The corrected impact equivalent energy was obtained. in and These are preset parameters; the tilting posture deviation is defined as follows: Let the number of repeated dumping events corresponding to this dumping event be . This yields the corrected tilting posture deviation. in These are preset parameters; the thermal stress force of heat exposure is defined as... And let the maximum temperature rise rate corresponding to this thermal exposure event be . Then the corrected thermal stress of heat exposure is obtained. in These are preset parameters; the above The reproducible calculation results of impact equivalent energy, tilting posture deviation, and thermal stress from heat exposure are written into the traceability database, and are aligned using timestamps when the sampling frequencies of different sensor data are inconsistent. To ensure consistency in integral accumulation, the sampling period must be unified to the valid sampling period corresponding to the event index. The cumulative damage risk index is obtained by combining the impact equivalent energy, tilting posture deviation, and thermal stress from heat exposure using a fusion function. This fusion function includes a time decay weight and a cross-modal coupling correction term. These terms are set based on a set of fundamental parameters and vary with at least the cylinder age, remaining inspection validity period, filling volume, and volume parameters. For example, when the cylinder age is older and the remaining inspection validity period is shorter, the time decay weight can contribute more to recent events and reduce the impact of long-term events, highlighting the risk clustering near the inspection window. A higher filling volume can increase the contribution of the impact equivalent energy in the fusion function, reflecting the higher sensitivity to stress and consequences under high filling volume conditions. A larger volume parameter can correct the contribution of the tilting posture deviation to adapt to the posture characteristics of different LPG cylinder sizes. The cross-modal coupling correction term can be used to characterize the mutual amplification effect of heat exposure, impact, and tilting. For example, when an impact event occurs within a preset time range after a heat exposure event, the effective contribution of the impact equivalent energy is amplified, thus better reflecting the cumulative damage patterns in complex scenarios. Set at the evaluation time The preceding link records contain several impact events, toppling events, and thermal exposure events, each corresponding to an impact equivalent energy sequence. Tilting posture deviation sequence Thermal exposure thermal stress sequence And let the time of each event be . The time decay weight can then be defined as follows: in Based on a set of fundamental parameters and varying at least with bottle age and remaining inspection validity period, specifically, it can be set as follows: The cross-modal coupling correction term can be used to characterize the amplification effect of thermal exposure on the impact and toppling contributions through a coupling window. Specifically, it can be set as follows: and in To and Recently and earlier than Timing of heat exposure events To preset the coupling window length, and The preset coupling gain can be set, and the modal contribution coefficient can be further configured. It can vary with the filling amount and volume parameters, specifically... Based on this, the fusion function can be defined as the cumulative damage risk index. in The results of historical link records and verification and scrapped nodes can be fitted or preset by the rule table, and stored in the traceability database in the form of parameter version number to support playback review and subsequent updates. While obtaining the cumulative damage risk index, the thermal pressure exceedance prediction margin index is calculated. Specifically, a future temperature sequence within a preset prediction time window is established based on multimodal data. The future temperature sequence can be derived from the temperature change trend and temperature rise rate identified by thermal exposure events and the current temperature level, and upper and lower bounds can be set to characterize the uncertainty of temperature fluctuations. A future pressure sequence is established based on the future temperature sequence, the basic parameter set, and the pressure-temperature correspondence, where the pressure-temperature correspondence is a pressure-temperature calibration curve or a pressure-temperature formula. By mapping the future temperature sequence to the future pressure sequence, the prediction of pressure evolution within the preset prediction time window is achieved. The thermal pressure exceedance prediction margin index is calculated based on the future pressure sequence and the pressure safety threshold, so that the thermal pressure risk is expressed in a quantitative form as the remaining margin between the pressure safety threshold and the future pressure sequence and its trend over time. Among them, the relationship between pressure and temperature The calibration process can be achieved by: collecting pressure values ​​inside liquefied gas cylinders at several stable temperature points under the same set of basic parameters to form temperature-pressure sample pairs; using piecewise linear interpolation or polynomial fitting to obtain pressure-temperature calibration curves; and writing the version number, applicable range, and fitting error statistics of the pressure-temperature calibration curves into the traceability database; when pressure-related anomalies are confirmed at inspection and scrapping nodes or when the cumulative number of newly added temperature-pressure sample pairs reaches a preset threshold, the pressure-temperature calibration curves are refitted and updated, and historical versions are retained to support playback and review. In a feasible forecasting approach, let the evaluation time be... The preset prediction time window length is The predicted step size is The current temperature is obtained from the temperature change data. And in The estimated rate of temperature rise is obtained within the preset review window. With fluctuation standard deviation Then the future temperature sequence can be arranged according to discrete time intervals. Constructed as And construct upper and lower bounds and To characterize the uncertainty of temperature fluctuations; the pressure-temperature relationship is assumed to be a pressure-temperature calibration curve or pressure-temperature formula categorized according to the set of basic parameters. The future pressure sequence can then be constructed as follows: And accordingly obtained and Based on this definition, the minimum residual pressure margin is... And define the expected time of exceeding the limit as And there is no time limit exceeded. Empty; when it is necessary to output the probability of exceeding the limit, the prediction error can be equivalent to the pressure uncertainty. Under the normal approximation, the probability of exceeding the limit at each time step is defined as follows: in Given a standard normal distribution function, the probability of exceeding the limit within the prediction time window is defined as follows: And and The reproducible component of the hot-pressing over-limit prediction margin index is written into the traceability database; Equivalent pressure uncertainty The construction can simultaneously consider the temperature sequence prediction error and the pressure-temperature calibration curve fitting error: Let the standard deviation of the temperature sequence prediction error be... The standard deviation of the fitting residuals of the pressure-temperature calibration curves under the corresponding ranges is: Then the upper and lower bounds of the future temperature sequence can be constructed as follows: and And construct the upper and lower bounds of the future pressure sequence as and Then take Equivalent pressure uncertainty as prediction error; For example, for a liquefied gas cylinder with an older age and a short remaining inspection validity period, the delivery log shows two impact events, one repeated tipping event, and a continuous heat exposure event. After noise reduction, gravity component separation, and temperature baseline correction, impact event identification, tipping event identification, and heat exposure event identification are completed. The impact equivalent energy, tipping posture deviation, and heat exposure thermal stress are calculated respectively. The cumulative damage risk index is obtained through a fusion function that includes time decay weights and cross-modal coupling correction terms. At the same time, the future temperature sequence is extrapolated based on the heat exposure data within a preset prediction time window, and the future pressure sequence is established by combining the pressure-temperature calibration curve. The result is compared with the pressure safety threshold to obtain the thermo-pressure over-limit prediction margin index. This provides a verifiable and reproducible quantitative basis for subsequent risk assessment based on the cumulative damage risk index and the thermo-pressure over-limit prediction margin index.

[0026] Reference Figure 3 Step 4: Determine the comprehensive risk level based on the cumulative damage risk index and the thermo-pressure over-limit predictability margin index. Drive the state machine of the entire LPG cylinder chain to perform state transitions, and generate and issue dynamic control instructions based on the comprehensive risk level. The comprehensive risk level and the execution results of the dynamic control instructions are written into the traceability database to achieve a closed loop of full-chain traceability and safety early warning. Transform the risk quantification results into executable management actions, and write back the actions and results to form a closed loop. Determining the comprehensive risk level based on the cumulative damage risk index and the thermo-pressure over-limit predictability margin index ensures that the current risk of LPG cylinders considers both historical cumulative damage and future thermo-pressure-related trends, thereby forming a comprehensive judgment that more closely reflects the actual risk. The system drives the entire liquefied gas cylinder state machine to perform state transitions, ensuring a unified state management logic for risk management. This guarantees consistent understanding of risk states across different business nodes and enables appropriate actions to be taken according to the state transition path. Dynamic control instructions are generated and issued based on the comprehensive risk level, moving risk management from identification to control. This ensures that risk response is implemented in the actual operational stages of the business node set, guaranteeing that early warnings go beyond mere notifications. The comprehensive risk level and the execution results of dynamic control instructions are written into a traceability database, making risk assessment, action, and execution feedback traceable, verifiable, and replayable. This forms a closed-loop traceability and safety early warning system from data collection, calculation, assessment, and action to data write-back, facilitating continuous rule optimization and improved safety control effectiveness.

[0027] In one specific implementation, the comprehensive risk level is determined based on the cumulative damage risk index and the thermo-pressure over-limit prediction margin index. Specifically, the cumulative damage risk index and the thermo-pressure over-limit prediction margin index corresponding to the link record are read from the traceability database, and the two are jointly judged according to the preset comprehensive risk level judgment rules. This allows the comprehensive risk level to simultaneously reflect the cumulative damage level of the liquefied gas cylinder in the historical circulation process and the thermo-pressure over-limit trend within the preset prediction time window, thereby avoiding misjudgment caused by relying on only a single indicator. The preset set of cumulative damage risk thresholds includes at least a first cumulative damage risk threshold. With the second cumulative damage risk threshold and When the cumulative damage risk index When the damage risk level is determined to be low, When the damage risk level is determined to be medium, The damage risk level is determined to be high; the preset set of hot-pressing over-limit prediction margin thresholds includes at least the first hot-pressing margin threshold. With the second hot pressure margin threshold and It may further include an over-limit probability threshold. Compared with the expected over-limit threshold When the minimum residual pressure margin and When the hot-press risk level is determined to be low, or When the hot-press risk level is determined to be medium, or Not empty and satisfies The risk level of hot pressing is determined to be high. The comprehensive risk level determination rule can be defined as a joint determination rule with damage risk level and hot pressing risk level as input. When the damage risk level is high or the hot pressing risk level is high, the comprehensive risk level is determined to be high. When the damage risk level is medium and the hot pressing risk level is medium or low, the comprehensive risk level is determined to be medium. When the damage risk level is low and the hot pressing risk level is low, the comprehensive risk level is determined to be low. The conditions for entering the predictive scheduling state can be defined as a comprehensive risk level of medium and the results of heat exposure event identification indicating the existence of a continuous heat exposure event or Not empty and satisfies The conditions for entering the review state can be defined as a comprehensive risk level of medium and the link records showing that the impact event identification or the tipping event identification repeatedly occurs within a preset time range and exceeds a preset number of times threshold; the conditions for entering the freeze disposal state can be defined as a comprehensive risk level of high. The joint judgment rule is stored as a configurable rule parameter in the traceability database and bound to the parameter version number. The setting basis may include the statistical quantile of historical link records and inspection and scrapping node results, the risk separation degree of accident backtracking samples, and the safety margin requirements of pressure safety threshold, so as to ensure that different business nodes can achieve consistent risk classification and state transition according to the same rule. The system drives the state machine of the entire LPG cylinder supply chain to perform state transitions and generates and issues dynamic control instructions based on the comprehensive risk level. Specifically, when the comprehensive risk level meets the conditions for entering the predictive scheduling state, predictive scheduling dynamic control instructions are issued to the transportation and delivery business nodes in the business node set. These instructions include at least changing the transportation and delivery routes to reduce heat exposure, adjusting the delivery sequence to shorten the outdoor dwell time, setting ventilated and shaded cooling and parking windows, and limiting the maximum allowable dwell time in heat exposure scenarios. These instructions are then bound together with the unique identifier of the LPG cylinder, the business node identifier, the issuance time, and the instruction parameters. When the overall risk level meets the conditions for entering the review state, a dynamic control instruction for review is issued to the next business node corresponding to the link record to enforce the review and evidence collection and the backfilling of review results, so that the next business node completes the necessary evidence collection before the handover or warehousing and writes the review results into the traceability database to form a traceable basis; when the overall risk level meets the conditions for entering the freeze disposal state, a dynamic control instruction for freeze is issued to the filling business node, the warehousing business node and the delivery business node in the business node set to prohibit the continued filling, prohibit the warehousing and prohibit the dispatch of orders and delivery, trigger the inspection or scrap disposal work order, and block the continued spread of risk from the process. The comprehensive risk level and the execution results of dynamic control instructions are written into the traceability database to achieve a closed loop of full-chain traceability and safety early warning. Specifically, the receipt, execution, review results, freezing of disposal results, and disposal conclusions of the dynamic control instructions and the triggering of inspection or scrap disposal work orders are recorded in a chain and formed a replayable execution trajectory. For example, if the thermal pressure exceedance prediction margin indicator of a liquefied gas cylinder during delivery shows that it is approaching the pressure safety threshold within the preset prediction time window and the cumulative damage risk index continues to rise, then the comprehensive risk level meets the conditions for entering the prediction scheduling state and triggers the prediction scheduling dynamic control instruction. The delivery order is adjusted to priority delivery and the outdoor dwell time is controlled within the maximum allowable dwell time in the limited heat exposure scenario. At the same time, a ventilated and shady area is selected for short-term cooling. After the execution is completed, the execution results are written back to the traceability database so that the risk reduction effect can be traced, verified, reviewed and optimized in the future.

[0028] The above description is merely a specific embodiment of this application, but the scope of protection of this application 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 this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for full-chain traceability and safety early warning of liquefied petroleum gas (LPG) cylinders based on multimodal fusion, characterized in that, include: Step 1: Generate and solidify a unique identifier for each LPG cylinder, establish a traceability file corresponding to the unique identifier, the traceability file includes a set of basic parameters, establish a traceability database, define a set of business nodes covering the entire LPG cylinder chain, and construct a state machine for the entire LPG cylinder chain; the set of basic parameters includes cylinder age, remaining inspection validity period, filling volume, volume parameters, and pressure safety threshold; Step 2: Collect multimodal data and business event data during the flow of liquefied gas cylinders through business nodes, and form a link record in the traceability database; the multimodal data includes acceleration change data reflecting impact, posture change data reflecting tilting, and temperature change data reflecting thermal exposure; Step 3: Perform preprocessing and event segmentation on the multimodal data based on the link records, and then calculate the cumulative damage risk index and the thermal pressure over-limit prediction margin index. The cumulative damage risk index is obtained as follows: the impact equivalent energy is calculated for impact events; the tilting posture deviation is calculated for tilting events; the thermal stress is calculated for thermal exposure events; and the cumulative damage risk index is obtained by combining the impact equivalent energy, tilting posture deviation, and thermal stress based on a fusion function. The fusion function includes time decay weights and cross-modal coupling correction terms. The time decay weights and cross-modal coupling correction terms are set based on the basic parameter set and vary at least with the bottle age, remaining test validity time, filling quantity, and volume parameters. Impact equivalent energy is defined as: in, Let be the dynamic acceleration amplitude at the sampling time of the i-th impact event; The sampling sequence number represents the start and end points of the impact event; The sampling period; The tilting posture deviation is defined as: in, Tilt angle threshold The sampling sequence number is the start and end point of the dumping event; The tilt angle at the sampling time of the j-th tilting event; The thermal stress force of thermal exposure is defined as: in, The sampling sequence number is the start and end point of the heat exposure event; The temperature over-temperature measurement at the sampling time of the m-th exposure event; The over-temperature threshold for identifying thermal exposure events; The calculation method for the hot pressure over-limit prediction margin index is as follows: establish a future temperature sequence within a preset prediction time window based on multimodal data; establish a future pressure sequence based on the future temperature sequence, the basic parameter set, and the pressure-temperature correspondence; calculate the hot pressure over-limit prediction margin index based on the future pressure sequence and the pressure safety threshold. The minimum residual pressure margin is: in, To predict the future pressure value inside the bottle at the nth prediction time within the time window; And define the expected time of exceeding the limit as: in, This refers to the nth prediction time within the prediction time window; and it is assumed that there are no out-of-limit times. Empty; when it is necessary to output the probability of exceeding the limit, the prediction error can be equivalent to the pressure uncertainty: in, This is the upper bound of the predicted pressure inside the bottle at the nth prediction time. Let this be the lower bound of the predicted pressure inside the bottle at the nth prediction time; and under the normal approximation, define the probability of exceeding the limit at each time as: in It is the standard normal distribution function. Let the standard deviation of the pressure forecast be denoted as , and then the probability of exceeding the limit within the forecast time window is defined as: And and The reproducible component of the hot-pressing over-limit prediction margin index is written into the traceability database; Step 4: Determine the comprehensive risk level based on the cumulative damage risk index and the thermal pressure over-limit prediction margin index, drive the state machine of the entire liquefied gas cylinder chain to perform state transitions, and generate and issue dynamic control instructions based on the comprehensive risk level. The comprehensive risk level and the execution results of the dynamic control instructions are written into the traceability database to realize the closed loop of full-chain traceability and safety early warning.

2. The method for full-chain traceability and safety early warning of liquefied gas cylinders based on multimodal fusion according to claim 1, characterized in that, The liquefied gas cylinder full-chain state machine includes at least the normal flow state, the verification state, the predictive scheduling state, and the frozen disposal state.

3. The method for full-chain traceability and safety early warning of liquefied gas cylinders based on multimodal fusion according to claim 2, characterized in that, Preprocessing includes at least noise reduction, gravity component separation, and temperature baseline correction; Event segmentation includes at least impact event identification, tipping event identification, and thermal exposure event identification.

4. The method for full-chain traceability and safety early warning of liquefied gas cylinders based on multimodal fusion according to any one of claims 1-3, characterized in that, The damage risk level is obtained by comparing the cumulative damage risk index with the preset set of cumulative damage risk thresholds. The hot pressing risk level is obtained by comparing the hot pressing over-limit prediction margin index with the preset set of hot pressing over-limit prediction margin thresholds. The comprehensive risk level is determined based on the damage risk level and the hot pressing risk level.

5. The method for full-chain traceability and safety early warning of liquefied gas cylinders based on multimodal fusion according to claim 1, characterized in that, When the overall risk level meets the conditions for entering the predictive scheduling state, predictive scheduling dynamic control instructions are issued to the transportation and delivery business nodes in the business node set; when the overall risk level meets the conditions for entering the review state, review dynamic control instructions are issued to the next business node corresponding to the link record to enforce review and evidence collection and review result backfilling. When the overall risk level meets the conditions for entering the freeze disposal state, a freeze dynamic control instruction is issued to the filling business node, warehousing business node and delivery business node in the business node set to prohibit continued filling, prohibit warehousing and prohibit order dispatching, and trigger inspection or scrap disposal work orders.

6. The method for full-chain traceability and safety early warning of liquefied gas cylinders based on multimodal fusion according to claim 5, characterized in that, The predictive scheduling dynamic control instructions include at least changing the transportation and delivery route to reduce heat exposure, adjusting the delivery order to shorten the time spent in the open air, setting a ventilated and shaded area for cooling and parking, and limiting the maximum allowable stay time in heat exposure scenarios. The execution results of the predictive scheduling dynamic control instructions are written back to the traceability database.

Images