Shared pallet stack petrochemical product comprehensive logistics supervision system based on internet of things

Abstract

The application discloses a shared pallet stack loading petrochemical product comprehensive logistics supervision system based on Internet of Things and relates to the technical field of data sharing. The shared pallet stack loading petrochemical product comprehensive logistics supervision system based on Internet of Things comprises a shared pallet cargo stack data acquisition module, a benchmark construction and data sensing module, a consistency check and abnormal feature extraction module and a module. The shared pallet cargo stack data acquisition module is used for collecting shared pallet supervision data and performing dynamic denoising and normalization processing. The benchmark construction and data sensing module is used for generating a cargo stack state benchmark fingerprint and performing deviation evaluation. The consistency check and abnormal feature extraction module is used for performing consistency check evaluation and screening effective abnormal starting points. The module is used for performing abnormal identification based on an abnormal feature vector and generating a petrochemical product logistics supervision report. The application effectively improves the concealed abnormal identification capability and the automatic scheduling accuracy of shared pallet stack loading petrochemical products in the whole logistics process, and solves the problem that the content of the cargo stack is difficult to be identified in time when the pallet identity and location are normal.

Description

Technical Field

[0001] This invention relates to the field of data sharing technology, specifically to an integrated logistics monitoring system for shared palletized petrochemical products based on the Internet of Things. Background Technology

[0002] With the development of IoT technology, petrochemical warehousing automation equipment, and shared pallet recycling models, petrochemical product logistics is gradually moving towards palletization, intelligence, and unmanned operation. Existing technologies can already achieve pallet identification, location tracking, and environmental limit alarms based on RFID identity binding, UWB positioning, virtual fence permission comparison, and environmental monitoring methods such as temperature and gas.

[0003] For example, invention patent CN120598454B discloses an automated inbound and outbound method for intelligent shared pallets, including: collecting and preprocessing pallet operating status data through a multi-source sensing module, generating task decomposition parameters based on a task decomposition model, constructing a multi-objective programming model for global scheduling, and outputting execution instructions through a hierarchical execution control model. This invention can achieve efficient and accurate inbound and outbound management of shared pallets, optimize circulation efficiency and resource conflicts, and improve the intelligence level and operational efficiency of the logistics system.

[0004] For example, invention patent CN112101854B discloses a logistics system and cargo monitoring method, including: multiple mobile warehouses, an identification system, and a database. At least some of the mobile warehouses include multiple storage units. The identification system is configured to identify goods entering the mobile warehouses and to determine changes in the storage units where goods are located within the mobile warehouses. The database is configured to record the association between the goods stored in the mobile warehouses and their respective storage units. This invention improves the safety of cargo transportation by monitoring goods throughout the entire logistics process and can accurately and promptly provide reasonable and suitable logistics equipment for cargo scheduling, thereby improving the efficiency of cargo logistics.

[0005] In existing technologies, shared pallet IoT monitoring systems typically rely on RFID identity binding, UWB and location tracking, virtual fence permission comparison, and temperature and gas environment threshold monitoring to issue alarms for whether the pallet is in place, whether it has crossed boundaries, or whether it exceeds limits. However, during the palletized storage and turnover of petrochemical materials, issues such as localized leakage inside the pallet, underloading, missing loads, mid-transit replacement, and slow weight loss due to packaging damage may occur. The system can only confirm the existence and compliance of the pallet, but it cannot effectively determine the consistency between the pallet load status and the recorded status, affecting the accuracy of automated loading, unloading, and scheduling decisions.

[0006] Therefore, in order to address the above issues, there is an urgent need for an integrated logistics monitoring system for shared palletized petrochemical products based on the Internet of Things. Summary of the Invention

[0007] Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides an integrated logistics monitoring system for shared palletized petrochemical products based on the Internet of Things, which solves the problem of difficulty in timely identification of slow changes in the contents of the pallet when the pallet's identity and location are normal.

[0009] Technical solution

[0010] To achieve the above objectives, the present invention provides the following technical solution: an IoT-based integrated logistics monitoring system for shared palletized petrochemical products, comprising a shared pallet data acquisition module for acquiring shared pallet monitoring data and performing data supplementation, dynamic noise reduction and normalization processing to generate a unified sorting time; a benchmark construction and data perception module for generating a pallet status benchmark fingerprint based on the preprocessed shared pallet monitoring data, performing deviation assessment, and constructing a perception dataset; a consistency verification and anomaly feature extraction module for identifying candidate anomaly starting points based on the perception dataset, performing consistency verification assessment, screening valid anomaly starting points, and constructing anomaly feature vectors; and an anomaly identification and hierarchical handling module for assessing the degree of deviation of the current pallet status based on the anomaly feature vectors, performing anomaly identification, and generating a petrochemical product logistics monitoring report.

[0011] Furthermore, the specific process of collecting and acquiring shared pallet monitoring data, performing data supplementation, dynamic noise reduction and normalization, and generating a unified sorting time is as follows: Data is collected on shared palletized petrochemical products throughout the entire process of packaging and palletizing, warehousing and shelving, in-warehouse relocation, temporary storage, loading / unloading, and outbound delivery. This data includes the pallet's unique RFID identifier, basic petrochemical product information, UWB positioning coordinates, real-time loading weight of the pallet, load distribution data, thermal field distribution characteristics, thermal field centroid coordinates, ambient gas concentration, temperature and humidity, node entry and exit timestamps, and warehouse dwell time. The system collects identification data by non-contactly reading the built-in UHF-RFID chip using an RFID reader, and supplements the data collection for obscured areas and high-level shelves using a combination of UWB positioning system, 5G thermal imaging equipment, gas sensors, and intelligent inspection robots. It also dynamically denoises the shared pallet monitoring data using a density-based outlier removal method, normalizes the data using Z-Score standardization, and writes three time fields—device time, gateway reception time, and platform entry time—to each collected message, and corrects the timeline based on network RTT to generate a unified sorted time.

[0012] Furthermore, the specific process of generating a pallet status benchmark fingerprint based on preprocessed shared pallet monitoring data is as follows: Based on the preprocessed shared pallet monitoring data, at the completion stage of petrochemical product packaging and palletizing, the ambient gas concentration in the target area is sampled multiple times to obtain a gas benchmark value after packaging and palletizing is completed or when the pallet enters the target logistics node. The initial loading mass and the load distribution at the four corners are obtained through weighing sensors; the initial thermal field distribution is obtained through thermal imaging equipment, and the initial thermal field centroid coordinates are extracted; at the same time, the planned dwell time range of batch products at each logistics node is obtained from the warehouse management system; the above data together constitute the pallet status benchmark fingerprint, which is bound to the pallet RFID identification and stored in the central database, and synchronized to the ERP system and the warehouse management system; the data of the entire logistics process is collected in real time through distributed sensing nodes, as the pallet enters the warehouse, During key logistics nodes such as transshipment, loading / unloading, and outbound, the difference between the real-time loading mass and the initial loading mass of the stack is calculated to obtain the mass change. By calculating the sum of squares of the load differences at each bearing point, the load distribution offset is obtained, which reflects the degree of deviation of the internal force distribution of the stack relative to the initial reference state. The thermal field centroid offset is calculated using Euclidean distance with reference to the initial reference thermal field centroid coordinates to obtain the thermal field centroid offset for each acquisition cycle. Local areas exceeding the threshold fluctuation range are identified from the thermal field distribution characteristics through temperature threshold judgment, and local thermal field temperature anomalies are extracted. The first-order difference operation is performed on the ambient gas concentration to obtain the characteristic gas concentration gradient and concentration change rate. The time difference between the entry and exit timestamps of the nodes is calculated to obtain the actual dwell time, which is compared with the planned dwell time interval to obtain the time deviation value, forming real-time sensing data.

[0013] Further, the specific process for deviation assessment is as follows: the mass deviation is obtained by dividing the mass change by the mass concealment anomaly threshold; the thermal field deviation is obtained by dividing the thermal field centroid offset by the thermal field concealment anomaly threshold; the gas deviation is obtained by dividing the characteristic gas concentration gradient by the gas concealment anomaly threshold; the duration deviation is obtained by dividing the time deviation value by the duration concealment anomaly threshold; the correlation matrix is ​​obtained by performing covariance statistics and normalization on samples of the same type of petrochemical products collected under baseline operating conditions; the mass deviation, thermal field deviation, gas deviation, and duration deviation are sequentially combined into a four-dimensional deviation vector; the four-dimensional deviation vector is transposed and multiplied by the inverse matrix of the correlation matrix plus the regularized identity matrix term, and then multiplied by the four-dimensional deviation vector to obtain the deviation assessment value.

[0014] Furthermore, the specific process of constructing the perception dataset is as follows: real-time comparison of the deviation evaluation value and the deviation threshold. When the deviation evaluation value is greater than the deviation threshold, it is determined that there is an instantaneous disturbance in the data of the current collection cycle, and the recording of batch data is paused. The data will be re-collected and evaluated in the next collection cycle before making a judgment. When the deviation evaluation value is less than or equal to the deviation threshold, it is determined that the current data meets the quality requirements. Starting from the current collection cycle, real-time perception data within a fixed collection cycle length is recorded to construct the perception dataset.

[0015] Furthermore, based on the sensing dataset, the specific process for identifying candidate anomaly initiations and conducting consistency verification and evaluation is as follows: A time series is constructed based on the sensing dataset; time difference calculation is performed on the mass change to obtain the mass change rate within a unit period; time difference calculation is performed on the thermal field centroid offset to obtain the thermal field centroid offset rate; the dwell time deviation value is integrated and accumulated to obtain the dwell time overrun rate; the trend detection of the change rate of each indicator is performed using the sliding window analysis method; when the instantaneous change rate of a certain indicator exceeds ∂ times, it is identified as a change trend abrupt change point, and the time corresponding to the abrupt change point is... The stamp is marked as the candidate anomaly starting point; real-time sensing data of the fixed acquisition period starting from the candidate anomaly starting point is extracted, the load distribution offset is multiplied by the load correction coefficient to obtain the enhancement factor, the ratio of quality deviation to quality threshold is calculated, and the ratio is multiplied by the enhancement factor to obtain the quality anomaly term; the duration deviation is multiplied by the duration amplification factor and then added to obtain the duration anomaly term, and the product of thermal field deviation, gas deviation and duration anomaly term is calculated to obtain the coupling anomaly term; the product of the quality anomaly term and the coupling anomaly term of each period within the fixed acquisition period is accumulated sequentially and then divided by the fixed acquisition period length to obtain the consistency verification evaluation value.

[0016] Furthermore, the specific process for screening valid anomaly starting points is as follows: Real-time comparison of the consistency verification evaluation value and the consistency threshold. When the consistency verification evaluation value is greater than the consistency threshold, the current candidate point is confirmed as a valid anomaly starting point, and the sensing data after the starting point is extracted. When the consistency verification evaluation value is lower than the consistency threshold, the candidate anomaly starting point is slid backward by one acquisition cycle, and the consistency verification evaluation process is repeated until the first valid anomaly starting point is identified. After the valid anomaly starting point is determined, each acquisition cycle is analyzed sequentially from the valid anomaly starting point, and the standard deviations of the mass change rate, thermal field centroid shift rate, residence time overdue rate, and concentration change rate are calculated. For any standard deviation, if it is higher than the anomaly change rate threshold within the acquisition cycle, the current acquisition cycle is determined as the anomaly persistence point.

[0017] Furthermore, the specific process of constructing the abnormal feature vector is as follows: extract the sensing data within the continuous acquisition period from the effective abnormal starting point to the abnormal duration point, calculate the maximum value of the mass change within the abnormal segment to obtain the mass weight loss peak; count the maximum distance of the thermal field center of gravity shift to obtain the thermal field shift extreme value; extract the maximum increase value of the characteristic gas concentration to obtain the gas leakage peak; calculate the maximum duration of the actual residence time exceeding the planned residence time interval to obtain the duration overrun peak; and sequentially concatenate to construct the abnormal feature vector.

[0018] Furthermore, based on the abnormal feature vector, the specific process for assessing the deviation of the current stack status is as follows: multiply the peak value of the time overdue period by the time coupling coefficient and add it to the peak value of gas leakage to obtain the gas anomaly quantity; multiply the peak value of mass loss by the gas anomaly quantity to obtain the joint enhancement quantity; add the extreme value of thermal field offset to the thermal field offset suppression term to obtain the thermal field offset factor; add the gas reference value to the zero-prevention term to obtain the gas factor; multiply the thermal field offset factor and the gas factor and add it to the joint enhancement quantity to obtain the offset term; divide the joint enhancement quantity by the offset term to obtain the comprehensive anomaly assessment value; compare the comprehensive anomaly assessment value with the anomaly classification threshold. When the comprehensive anomaly assessment value is less than or equal to the first-level anomaly threshold, it is determined to be a first-level anomaly; when the comprehensive anomaly assessment value is greater than the first-level anomaly threshold and less than or equal to the second-level anomaly threshold, it is determined to be a second-level anomaly; when the comprehensive anomaly assessment value is greater than the second-level anomaly threshold and less than or equal to the third-level anomaly threshold, it is determined to be a third-level anomaly; when the comprehensive anomaly assessment value is greater than the third-level anomaly threshold, it is determined to be a fourth-level anomaly.

[0019] Further, the specific process for anomaly identification and generation of a petrochemical product logistics supervision report is as follows: The identified abnormal status is structurally bound to the pallet's RFID identification, the logistics node it belongs to, and the time of the anomaly occurrence, generating an abnormal behavior label for the pallet status. Simultaneously, a verification mark is set for pallets identified as abnormal, and the corresponding anomaly type and level are indicated. Based on the anomaly level output from the comprehensive anomaly assessment, a tiered handling mechanism is triggered: Level 1 anomalies send a background alert and are continuously monitored; Level 2 anomalies trigger an audible and visual warning and dispatch an intelligent inspection robot for on-site verification; Level 3 anomalies send an interrupt signal to the automated equipment control system via the industrial Ethernet protocol, freezing the automatic transfer permissions of the abnormal pallet in real time and suspending the dispatch instructions of the automated equipment; Level 4 anomalies trigger the highest level explosion-proof warning, pushing emergency handling instructions to the handheld terminals of on-site personnel via a safe and redundant communication link; after the anomaly is handled, the status data of the repaired pallet is collected, the pallet status baseline fingerprint is updated, the pallet verification mark and transfer restrictions are lifted, and a petrochemical product logistics supervision report is generated.

[0020] Beneficial effects

[0021] The present invention has the following beneficial effects:

[0022] (1) This invention, by embedding RFID chips in shared pallets and binding petrochemical product information, and combining UWB high-precision positioning, 5G thermal imaging, explosion-proof tracking and communication gateway to build a multi-source integrated supervision link, realizes the collaborative perception of pallet identity, location, environment and operation status, solves the problems of information silos and scattered supervision elements in existing systems, and improves the transparency and accuracy of the supervision of the entire process of petrochemical product logistics.

[0023] (2) This invention establishes a pallet status benchmark fingerprint for each shared pallet and extracts real-time sensing data related to quality, thermal field, gas and dwell time at key nodes such as warehousing, transfer, loading and unloading and outbound. It can dynamically compare the book status with the physical status, solving the problem that the existing system can only determine whether the pallet is in place or whether it has crossed the boundary, but cannot identify changes in the status of the contents of the pallet.

[0024] (3) This invention introduces a consistency verification mechanism for mass, thermal field, gas and residence time, and performs coupled analysis on slow weight loss, local thermal anomaly, micro leakage and abnormal retention. It can identify in-situ hidden anomalies under the conditions of normal pallet identity, normal position and stable trajectory, and solves the problem that slow-changing anomalies such as local leakage, missing or missing loading, mid-term replacement and stacking loose sinking are difficult to detect in time.

[0025] (4) This invention, by screening effective abnormal starting points, extracting abnormal duration segments and constructing abnormal feature vectors, comprehensively evaluates the peak value of mass loss, extreme value of thermal field shift, peak value of gas leakage and peak value of duration overdue, thereby improving the sensitivity and stability of identifying continuous and gradual abnormalities and reducing the probability of misjudgment caused by single-cycle occasional fluctuations.

[0026] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0027] Figure 1 This is a diagram illustrating the architecture of the IoT-based integrated logistics monitoring system for shared palletized petrochemical products according to the present invention.

[0028] Figure 2 This is a dynamic perception and deviation trend analysis diagram of shared pallet stacks in this invention;

[0029] Figure 3 This is a comparison and distribution diagram of the deviation of each logistics node of the shared pallet in this invention;

[0030] Figure 4 This is a trend diagram showing the time-series evolution and deviation diagnosis of shared pallet stacks in this invention;

[0031] Figure 5 This is a flowchart for the shared tray anomaly classification determination process of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figures 1-5 This invention provides a technical solution: an IoT-based integrated logistics monitoring system for shared palletized petrochemical products, comprising a shared pallet data acquisition module for acquiring shared pallet monitoring data and performing data supplementation, dynamic noise reduction and normalization processing to generate a unified sorting time; a benchmark construction and data perception module for generating a pallet status benchmark fingerprint based on the preprocessed shared pallet monitoring data, performing deviation assessment, and constructing a perception dataset; a consistency verification and anomaly feature extraction module for identifying candidate anomaly starting points based on the perception dataset, performing consistency verification assessment, screening valid anomaly starting points, and constructing anomaly feature vectors; and an anomaly identification and hierarchical handling module for assessing the degree of deviation of the current pallet status based on the anomaly feature vectors, performing anomaly identification, and generating a petrochemical product logistics monitoring report.

[0034] Specifically, the process of collecting and supplementing shared pallet monitoring data, performing dynamic noise reduction and normalization, and generating a unified sorting time is as follows: Data is collected on shared palletized petrochemical products throughout the entire process of packaging, palletizing, warehousing, in-warehouse relocation, temporary storage, loading / unloading, and outbound delivery. This is achieved through distributed sensing nodes deployed at various logistics nodes, continuously collecting data at a fixed sampling frequency. The shared pallet monitoring data includes the pallet's unique RFID identifier, basic petrochemical product information, ultra-wideband (UWB) positioning coordinates, real-time loading weight of the pallet, load distribution data obtained from the four corner weighing sensors at the bottom of the pallet, thermal field distribution characteristics collected by 5G thermal imaging equipment and processed by the edge computing unit, thermal field centroid coordinates, ambient gas concentration, temperature and humidity, node entry and exit timestamps, and storage dwell time. The data is then read non-contactly by an RFID reader. The system incorporates an ultra-high frequency radio frequency identification (UHF-RFID) chip for identity collection. Combined with a UWB positioning system consisting of a warehouse top-mounted positioning base station and pallet tags, 5G thermal imaging equipment, gas sensors, electrochemical gas sensors, and an intelligent inspection robot that automatically patrols along a track, it collects supplementary data on obscured areas and high-level shelving environments. Dynamic noise reduction of shared pallet monitoring data is achieved using a density-based clustering outlier removal method employing the DBSCAN algorithm. Z-score standardization is used to normalize the shared pallet monitoring data, subtracting the mean from each data point and dividing by the standard deviation. Each collected message is written with three time fields: device time, gateway reception time, and platform entry time. The time offset is estimated based on the network round-trip time (RTT) between the measuring device and the gateway, and the time axis is corrected using the gateway time as a benchmark to generate a unified sorted time.

[0035] This implementation plan continuously collects shared pallet identity, location, load, thermal field, gas and time-series monitoring data at each logistics node, and supplements the data collection for obscured areas and high-level rack scenarios. Combined with dynamic denoising and normalization processing, it improves data stability and comparability, enhances the accuracy and continuity of hidden and slowly changing anomaly identification, and strengthens the credibility and real-time performance of inventory traceability, risk analysis and automated scheduling decisions.

[0036] Specifically, the process of generating a pallet status benchmark fingerprint based on preprocessed shared pallet monitoring data is as follows: Based on the preprocessed shared pallet monitoring data, during the petrochemical product packaging and palletizing process, multiple samples of the ambient gas concentration in the target area are taken to obtain a gas benchmark value after packaging and palletizing or when the pallet enters the target logistics node. The initial loading mass and the load distribution at the four corners are obtained through the four corner weighing sensors embedded at the bottom of the pallet. The initial thermal field distribution is obtained through a 5G thermal imaging device deployed above the palletizing station, and the initial thermal field centroid coordinates are extracted by processing the thermal field image using the centroid method. At the same time, the planned dwell time interval of batch products at each logistics node is periodically obtained from the warehouse management system through database views or intermediate tables. The above data together constitute the pallet status benchmark fingerprint, which is bound to the pallet RFID identification and stored in the central database, and synchronized to the ERP system and the warehouse management system.

[0037] Real-time data collection across the entire logistics process is achieved through distributed sensing nodes. Event-driven data collection tasks are triggered when pallets enter key logistics nodes such as warehousing, transshipment, loading / unloading, and outbound. The difference between the real-time loading quality and the initial loading quality of the pallet is calculated to obtain the weight change. The load distribution offset is obtained by calculating the sum of squares of the load differences at each load-bearing point, reflecting the degree of deviation of the internal force distribution of the pallet relative to the initial baseline state. The thermal field centroid offset is calculated using Euclidean distance with reference to the initial baseline thermal field centroid coordinates, yielding the thermal field centroid offset for each collection cycle. Local areas exceeding the threshold fluctuation range are identified from the thermal field distribution characteristics using temperature threshold judgment, and local thermal field temperature anomalies are extracted. First-order difference calculations are performed on the ambient gas concentration to obtain the characteristic gas concentration gradient and concentration change rate. The time difference between the node entry and exit timestamps is calculated to obtain the actual dwell time, which is compared with the planned dwell time interval to obtain the time deviation value, forming real-time sensing data.

[0038] like Figure 2 The shared pallet stack dynamic perception and deviation trend analysis diagram shown above compares the initial and current thermal field distributions at the palletizing baseline time, and marks the location of the thermal field centroid with markers to intuitively reflect the overall thermal field drift and local hot spot changes. The middle section presents the evolution trend of mass changes, load distribution shifts, and key derived features of characteristic gas concentration gradients and thermal field centroid shifts in a time series manner to depict the continuous accumulation process of slowly changing anomalies. The bottom section shows the time deviation distribution of storage, transshipment, loading / unloading, and outbound nodes to characterize the amplifying effect of logistics time sequence anomalies and abnormal delays on risks, thereby supporting the identification and graded handling of in-situ hidden anomalies where the pallet position is compliant but the contents of the stack are distorted.

[0039] This implementation plan, under the condition that the pallet is legally identified and its location is compliant, can still promptly capture the evolution trend of hidden and slowly changing anomalies such as leakage, underfilling, damaged packaging, and loose or sinking stacks, thereby improving the continuity, interpretability, and accuracy of anomaly identification and graded handling, and enhancing inventory credibility and the reliability of security traceability.

[0040] Specifically, the deviation assessment process is as follows: the quality deviation is obtained by dividing the quality change by the quality concealment anomaly threshold; the thermal field deviation is obtained by dividing the thermal field centroid offset by the thermal field concealment anomaly threshold; the gas deviation is obtained by dividing the characteristic gas concentration gradient by the gas concealment anomaly threshold; the duration deviation is obtained by dividing the time deviation value by the duration concealment anomaly threshold; the correlation matrix is ​​obtained by performing covariance statistics and normalization on samples of the same type of petrochemical products under baseline operating conditions; the quality deviation, thermal field deviation, gas deviation, and duration deviation are sequentially combined into a four-dimensional deviation vector; the four-dimensional deviation vector is transposed and multiplied by the inverse matrix of the correlation matrix plus the regularized identity matrix term, and then multiplied by the four-dimensional deviation vector to obtain the deviation assessment value.

[0041] The specific formula for calculating the deviation from the assessment value is as follows:

[0042] ;

[0043] In the formula, This indicates a deviation from the assessment value, used to quantify the overall degree of deviation between the shared pallet stack status and the baseline fingerprint; The correlation matrix is ​​calculated by collecting deviation vector samples of similar petrochemical products under normal storage, normal transportation and normal loading and unloading conditions. It is used to characterize the statistical coupling relationship between quality, thermal field, gas and time series, so that the deviation assessment value takes into account multidimensional correlation. represents the regularization coefficient, which is preset through numerical stability testing and has a value range of [10-4, 10-1]. It is used to prevent the correlation matrix from becoming invertible and causing computational instability when there are insufficient samples or excessive dimensional correlation. Represents the identity matrix, used in conjunction with the correlation matrix. Construct a regularization matrix; It represents the amount of quality change and is used to identify abnormalities in the physical condition of the stack of goods; This indicates the offset of the center of gravity of the thermal field, used to detect abnormal changes in the temperature distribution inside the stack. It represents the characteristic gas concentration gradient, used to identify micro-leakage of materials; This represents the time deviation value, used to reflect abnormal states in the timing of pallet logistics. The threshold for hidden quality anomalies is obtained through statistical analysis and experimental calibration of the normal weight loss rate of similar petrochemical products during storage and transportation. The value range is [0.5, 5], which is used to eliminate the influence of differences in the quality base of different product categories on the anomaly judgment. The threshold for concealed thermal anomalies is obtained through cluster analysis of the thermal field distribution characteristics of the cargo stack. The value range is [2,10], and it is used to identify local anomalies inside the cargo stack. The threshold value represents the hidden gas anomaly. It is obtained by combining the lower limit of the safe leakage concentration of different petrochemical media with the detection accuracy of the sensor through petrochemical media leakage experiments. The value range is [0.05, 0.5], and it is used to identify hidden leakage risks. The threshold for concealed anomalies is determined by time-series analysis of logistics node operations. It is based on the standard operation time of pallets at each node and combined with scheduling redundancy settings. The value range is [0.5, 4], and it is used to identify the risk of pallet status caused by abnormal delays and transfers.

[0044] like Figure 3 The diagram showing the deviation comparison and deviation assessment distribution of each logistics node of the shared pallet is shown. Figure 3 The upper part is a bar chart comparing the deviation of each node. With warehousing, transshipment, waiting for loading / unloading, and outbound as the horizontal axis, it provides a bar comparison of mass deviation, thermal field deviation, gas deviation, and time deviation. This is used to visually present the structural differences in deviation within the same node, as well as the changes in the strength of deviation in the same dimension between different nodes. Figure 3 The bottom left is a bar chart of deviation assessment values ​​for each node. For each logistics node, the deviation is aggregated into a single deviation assessment value according to the consistency assessment rules, and the overall deviation level between nodes is displayed in bar form. This is used to quantify the overall degree of deviation of the pallet status from the baseline fingerprint. Figure 3 The bottom right is a pie chart showing the proportion of deviation. Taking the deviation of a certain target node as input, the proportion of mass, thermal field, gas and time deviations to the total deviation is calculated and visualized. This is used to identify the dominant contributors to the node deviation assessment value, thereby corresponding the handling actions with the causes of risks.

[0045] This implementation scheme achieves coupled measurement and stable assessment of multidimensional deviations, enabling the deviation assessment results to have better comparability and noise resistance under different pallets, different logistics nodes and different petrochemical media conditions; at the same time, it can effectively suppress false alarms caused by occasional fluctuations in a single dimension and amplify the hidden slow-changing anomaly characteristics caused by multidimensional collaborative deviations, thereby improving the sensitivity and reliability of identifying risks such as leakage, loosening and settling, and abnormal retention under in-situ compliance conditions.

[0046] Specifically, the process of constructing the sensing dataset is as follows: Real-time comparison of deviation evaluation values ​​with deviation thresholds is performed automatically by the central processing unit after each acquisition cycle. When the deviation evaluation value is greater than the deviation threshold, it is determined that the data in the current acquisition cycle has a momentary disturbance. The acquisition cycle refers to the time interval between a complete acquisition of the same tray status data by distributed sensing nodes. The current batch of data is discarded, and a pause command is sent to the data acquisition module to pause the recording of the batch of data. The acquisition module restarts the data acquisition process in the next acquisition cycle, re-acquires and evaluates the data, and then makes a judgment. When the deviation evaluation value is less than or equal to the deviation threshold, it is determined that the current data meets the quality requirements. Starting from the current acquisition cycle, real-time sensing data within a fixed acquisition cycle length is recorded, organized and stored in chronological order, and the sensing dataset is constructed.

[0047] In this implementation scheme, by setting a threshold for the deviation evaluation value after each collection cycle, the system can automatically identify and remove instantaneous disturbance data, thereby improving the stability and consistency of real-time sensing data. This also enables the constructed sensing dataset to have a continuous, comparable, and traceable time structure, reducing the risk of misjudgment caused by noise and occasional jumps.

[0048] Specifically, the process of identifying candidate anomaly initiations and conducting consistency verification and evaluation based on the sensing dataset is as follows: A time series is constructed based on the sensing dataset; time-difference calculation is performed on the mass change to obtain the mass change rate within a unit period; time-difference calculation is performed on the thermal field centroid offset to obtain the thermal field centroid offset rate; the residence time deviation is integrated and accumulated to obtain the residence time overdue rate; the trend detection of the change rate of each indicator is performed using the sliding window analysis method. When the instantaneous change rate of a certain indicator exceeds ∂ times, it is identified as a change trend abrupt change point, and the timestamp corresponding to the abrupt change point is marked as a candidate anomaly initiation. ∂ is obtained by offline calibration of change rate samples of similar petrochemical products under baseline operating conditions and confirmed anomaly samples, and is used to distinguish rate abrupt changes caused by occasional fluctuations and continuous anomaly evolution.

[0049] Real-time sensing data from a fixed acquisition period starting at the candidate anomaly origin is extracted. The load distribution offset is multiplied by the load correction coefficient to obtain the enhancement factor. The ratio of the quality deviation to the quality threshold is calculated, and the ratio is multiplied by the enhancement factor to obtain the quality anomaly term. The duration deviation is multiplied by the duration amplification factor and then added to obtain the duration anomaly term. The product of the thermal field deviation, gas deviation, and duration anomaly term is calculated to obtain the coupling anomaly term. The product of the quality anomaly term and the coupling anomaly term for each period within the fixed acquisition period is accumulated sequentially and then divided by the fixed acquisition period length to obtain the consistency verification evaluation value.

[0050] The specific formula for calculating the consistency verification evaluation value is as follows:

[0051] ;

[0052] In the formula, This represents the consistency verification evaluation value, used to quantify the degree of coupling deviation of shared pallet stacks of indicators within a continuous time window; This represents the fixed acquisition cycle length, which is obtained by statistically analyzing the shortest evolution cycle from the occurrence to the detectability of hidden anomalies in different petrochemical media. The value range is [3,10]. It is used to filter out occasional disturbances in a single cycle, focus on continuous and stable abnormal trends, and avoid misjudgment. The timestamp represents the collection time, which is obtained by aligning the clock synchronization mechanism of the shared tray IoT collection system with the time axis of the UWB positioning and RFID collection nodes, and is used to ensure the temporal consistency of data within the time window. This indicates the degree of quality deviation, used to eliminate the impact of differences in the initial quality base of different stacks on anomaly detection; The quality threshold is determined by statistically analyzing the quality deviation of similar petrochemical products, using the mean plus a multiple of the standard deviation. The value range is [0.05, 0.30], and it is used to characterize the upper limit of allowable quality fluctuation. The load correction coefficient is calculated by taking normal samples and confirmed abnormal samples of the same petrochemical medium at each logistics node, calculating the coupling terms of thermal field deviation and gas deviation, and statistically determining the gain relationship. The value range is [0.2, 1.5], which is used to adjust the enhancement effect of load distribution deviation on the quality anomaly term. This indicates the load distribution offset, which reflects the degree of deviation of the force distribution inside the stack relative to the initial reference state. This indicates the degree of thermal field deviation, used to reflect the extent to which the thermal field distribution deviates from the baseline state; This indicates the degree of gas deviation, reflecting the extent to which the gas concentration deviates from the environmental background value. The duration amplification factor is obtained by collecting load distribution samples after palletizing and during normal turnover, and calibrating them. The value range is [0.1, 1.0], which is used to characterize the degree of amplification of the risk of thermal field and gas coupling caused by abnormal retention. This indicates the duration deviation, used to eliminate the impact of differences in allowed dwell time at different logistics nodes on anomaly detection.

[0053] Table 1 shows the consistency verification and evaluation data table, used to comprehensively evaluate the data consistency level from multiple dimensions of deviation indicators. At timestamp 45:44.4: quality deviation is 0.5063, thermal field deviation is 0.1400, gas deviation is 0.0531, duration deviation is 0.0453, and the consistency verification evaluation value is 6.0077; at timestamp 46:44.4: quality deviation is 0.1490, thermal field deviation is 0.1521, gas deviation is 0.0150, duration deviation is 0.0565, and the consistency verification evaluation value is 18.0393; at timestamp 49:44.4: quality deviation is 0.5... At time stamp 545, the thermal field deviation is 0.3855, the gas deviation is 0.9876, the duration deviation is 0.5268, and the consistency verification evaluation value is 61.9419; at timestamp 50:44.4: the mass deviation is 1.1856, the thermal field deviation is 0.4571, the gas deviation is 1.0468, the duration deviation is 0.6965, and the consistency verification evaluation value is 71.0385; at timestamp 51:44.4: the mass deviation is 0.9381, the thermal field deviation is 0.4823, the gas deviation is 0.4867, the duration deviation is 0.5909, and the consistency verification evaluation value is 65.2082.

[0054] Table 1 Consistency Verification Evaluation Data Table

[0055]

[0056] like Figure 4 The shared pallet stack time-series evolution and deviation diagnosis trend chart shown is used to depict the dynamic process of quality, thermal field, gas, and time-series indicators from stabilization, cumulative deviation, and recovery and handling within a continuous time window after the candidate anomaly initiation point: The upper subplot shows the time series of the original sensed quantities to observe whether the anomaly appears in a continuous rise, step change, or increased fluctuation manner; the middle subplot gives the corresponding rate of change curve to identify the evolution slope and acceleration characteristics of slowly changing anomalies, and to distinguish between occasional disturbances and continuous accumulated risks; the lower subplot converts the indicators into normalized deviations and compares them, intuitively reflecting which dimensions first exceed the threshold within the anomaly window, which dimensions rise together, and the duration of exceeding the threshold, thus providing interpretable time-series evidence for the hierarchical judgment of hidden anomalies, the identification of secondary threshold combinations (such as suspected leakage, stack damage, and safety anomalies), and the freezing of circulation and review and handling.

[0057] In this implementation scheme, by introducing rate of change trend detection into the sensing dataset, candidate anomaly starting points can be automatically identified and continuous observation windows can be locked in the early stage of anomaly evolution, avoiding misjudging occasional noise or short-term fluctuations as anomalies; significantly improving the sensitivity and stability of hidden slowly changing anomalies, and reducing the risk of missed judgment caused by weak single-dimensional signals and the probability of false alarms caused by instantaneous disturbances.

[0058] Specifically, the process for screening valid anomaly starting points is as follows: The consistency verification evaluation value and the consistency threshold are compared in real time. When the consistency verification evaluation value is greater than the consistency threshold, the current candidate point is confirmed as a valid anomaly starting point, and the sensing data after the starting point is extracted. When the consistency verification evaluation value is lower than the consistency threshold, the candidate anomaly starting point is slid backward by one acquisition cycle, and the current candidate starting point time plus one acquisition cycle duration is used as the new candidate starting point. This consistency verification evaluation process is repeated until the first valid anomaly starting point is identified. After the valid anomaly starting point is determined, each acquisition cycle is analyzed sequentially from the valid anomaly starting point. The corresponding mass change rate, thermal field centroid shift rate, dwell time overrun rate, and concentration change rate data are retrieved sequentially according to time. The standard deviations of the mass change rate, thermal field centroid shift rate, dwell time overrun rate, and concentration change rate are calculated. For any standard deviation, if it is higher than the anomaly change rate threshold within the acquisition cycle, the current acquisition cycle is determined as an anomaly persistence point.

[0059] In this implementation scheme, by comparing the consistency verification evaluation value with the consistency threshold in real time, the automatic screening and adaptive shift correction of candidate anomaly starting points can be achieved. This can quickly lock the first effective anomaly starting point that truly has the characteristics of anomaly coupling deviation, avoid false triggering caused by early weak disturbances or candidate point offset, and thus accurately characterize the time boundary from the occurrence to the continuous evolution of anomalies, improving the completeness and credibility of anomaly duration segment interception.

[0060] Specifically, the process of constructing the abnormal feature vector is as follows: Sensing data within a continuous acquisition cycle from the effective abnormal starting point to the abnormal duration point is extracted, and the sensing data is time-aligned and segmented according to the acquisition cycle sequence to form an abnormal segment time-series data block; the maximum value of mass change within the abnormal segment is calculated to obtain the mass weight loss peak value. The mass change is calculated using the initial loading mass as a reference, and the maximum value of mass change in each acquisition cycle within the abnormal segment is scanned cycle by cycle; the maximum distance of the thermal field centroid offset is statistically analyzed to obtain the thermal field offset extreme value. The thermal field centroid offset is calculated using Euclidean distance with the initial reference thermal field centroid coordinates as a reference, and the maximum value is taken; the maximum increase in characteristic gas concentration is extracted to obtain the gas leakage peak value; the maximum duration of the actual dwell time exceeding the planned dwell time interval is calculated to obtain the duration overrun peak value; the abnormal feature vector is sequentially constructed and bound to the pallet RFID identification mark, the effective abnormal starting point timestamp, and the abnormal duration point timestamp to generate the corresponding abnormal feature vector record for the abnormal segment.

[0061] In this implementation scheme, the sensed data is time-aligned and segmented within a continuous acquisition cycle from the effective anomaly starting point to the anomaly duration point. Key extreme value features such as mass loss peak, thermal field offset extreme value, gas leakage peak, and duration overrun peak are extracted from the anomaly segment to form a structured anomaly feature vector, which is then bound to the pallet identity and time sequence boundary. This enables the anomaly information to be stably expressed in a unified and comparable feature form, thereby improving the consistency and traceability of anomaly segment identification and description.

[0062] Specifically, the process of assessing the degree of deviation of the current stack state based on the abnormal feature vector is as follows: Figure 5 The flowchart for classifying shared pallet anomalies shows that the gas anomaly quantity is obtained by multiplying the peak value of the time-overdue period by the time-coupling coefficient and then adding it to the peak value of the gas leakage; the combined enhancement quantity is obtained by multiplying the peak value of mass loss by the gas anomaly quantity; the thermal field offset extreme value is obtained by adding the thermal field offset suppression term; the gas reference value is obtained by adding the zero-prevention term; the gas factor is obtained by multiplying the thermal field offset factor and the gas factor and adding it to the combined enhancement quantity; and the combined enhancement quantity is divided by the offset term to obtain the comprehensive anomaly assessment value.

[0063] The specific formula for calculating the comprehensive anomaly assessment value is as follows:

[0064] ;

[0065] In the formula, This represents the comprehensive anomaly assessment value, used to quantify the degree of coupling between the physical state anomaly and the environmental temporal anomaly of the shared pallet stack within the current detection window; This represents the peak mass weight loss, used to characterize the maximum weight loss anomaly intensity of the stack within the observation window; This indicates the extreme value of the thermal field shift, used to characterize the maximum abnormal intensity of the characteristic gas leaking around the stack; This represents the peak value of gas leakage, used to characterize the maximum abnormal retention intensity of the abnormal pallet at the target logistics node; The duration coupling coefficient is obtained through offline calibration of confirmed abnormal samples, with a value range of [0.1, 1.0], and is used to characterize the amplification effect of the duration exceeding the limit on the risk of gas leakage; This represents the peak duration of the delay, used to characterize the maximum abnormality intensity of the abnormal pallet at the target logistics node; This represents the thermal field offset suppression term, which is determined by statistically analyzing thermal field offset samples of similar petrochemical products, using the mean, quantile, or mean plus standard deviation multiples. The value range is [0.01, 0.20], and it is used to suppress the situation where excessively small extreme values ​​of thermal field offset lead to abnormal amplification of entity anomalies. This represents the gas reference value, used to characterize the background level of characteristic gases under normal conditions; This indicates the zero-prevention term, obtained through statistical analysis of gaseous benchmark values ​​for similar petrochemical products, combined with stability tests on the denominator value. The value range is [range missing]. This is used to prevent the denominator from becoming unstable when the gas reference value is too small or close to zero.

[0066] The comprehensive anomaly assessment value is compared with the anomaly classification threshold. The comprehensive anomaly assessment value is constructed by coupling and normalizing physical anomalies and environmental time-series anomalies. The mass weight loss peak and the equivalent gas anomaly quantity are jointly enhanced, and the thermal field offset factor and the gas reference factor are used to form the normalization denominator, so that the assessment value has higher sensitivity to anomalies that accumulate continuously and evolve slowly.

[0067] When the comprehensive anomaly assessment value is less than or equal to the Level 1 anomaly threshold, it is determined to be a Level 1 anomaly; when the comprehensive anomaly assessment value is greater than the Level 1 anomaly threshold and less than or equal to the Level 2 anomaly threshold, it is determined to be a Level 2 anomaly; when the comprehensive anomaly assessment value is greater than the Level 2 anomaly threshold and less than or equal to the Level 3 anomaly threshold, it is determined to be a Level 3 anomaly; when the comprehensive anomaly assessment value is greater than the Level 3 anomaly threshold, it is determined to be a Level 4 anomaly. The anomaly classification thresholds are obtained based on the statistical calibration of baseline operating condition samples and confirmed anomaly samples: the upper quantile of the distribution of comprehensive anomaly assessment values ​​calculated under baseline operating conditions is set as the first-level anomaly threshold, and the quantiles of the assessment value distributions of suspected leakage anomalies, pallet damage anomalies, and safety anomalies in the confirmed anomaly samples are set as the second-level and third-level anomaly thresholds, respectively, to ensure that the classification rules have repeatable sample basis and interpretable risk boundaries. Compared with the existing classification methods that only rely on a single-dimensional threshold or simple weighted summation, the assessment value construction of this scheme can simultaneously express the coupled characteristics of anomaly intensity and continuous accumulation of anomalies. Especially in scenarios where the pallet is legal and the location is compliant, but the contents of the pallet are experiencing slow weight loss, minor leakage, or abnormal retention, it can achieve more stable classification output without relying on a single over-limit trigger, thereby reducing the probability of early missed judgments and fluctuation misjudgments.

[0068] For different levels of abnormal states, a secondary judgment is made based on feature thresholds: when it is determined to be a Level 1 anomaly, and only a single dimension indicator exceeds the corresponding hidden anomaly threshold, it is identified as a single-dimensional fluctuation anomaly; when it is determined to be a Level 2 anomaly, and both the mass and gas dimensions exceed their respective hidden anomaly thresholds, it is identified as a suspected leakage anomaly; when it is determined to be a Level 3 anomaly, and the mass, thermal field, and gas dimensions all exceed their respective hidden anomaly thresholds, it is identified as a stack damage anomaly; when it is determined to be a Level 4 anomaly, and all four dimensions exceed their respective hidden anomaly thresholds, it is identified as a safety anomaly.

[0069] In this implementation plan, a comprehensive anomaly assessment value is constructed based on anomaly feature vectors. Mass weight loss and gas anomalies are jointly enhanced, and thermal field offset and gas benchmark normalization constraints are introduced to ensure that the assessment results remain bounded and comparable under different pallets, different nodes, and different operating conditions. This achieves stable classification and subdivision of anomaly risks, and under the conditions that the pallet identity is legal and the location is compliant, it can still form a continuous and interpretable judgment output for slow weight loss, trace leakage, abnormal retention and their coupled evolution process.

[0070] Specifically, the process of anomaly identification and generation of petrochemical product logistics supervision reports is as follows: The identified abnormal status is structurally bound to the pallet's RFID identification, the logistics node it belongs to, and the time of the anomaly occurrence, generating an abnormal behavior label for the pallet status; simultaneously, a verification mark is set for pallets identified as abnormal, and the corresponding anomaly type and level are labeled; based on the anomaly level output from the comprehensive anomaly assessment, a graded handling mechanism is triggered: Level 1 anomalies send a background reminder and are continuously monitored; Level 2 anomalies trigger an audible and visual alarm and dispatch an intelligent inspection robot for on-site verification; Level 3 anomalies send an interrupt signal to the automation equipment control system via the industrial Ethernet protocol, freezing the automatic transfer permissions of the abnormal pallet in real time and suspending the dispatching commands of the automation equipment; the control interface includes an industrial Ethernet communication interface and a fieldbus interface. The fieldbus interface supports one of IO-Link, CANopen, or RS485, used to connect to safety access control, emergency stop buttons, audible and visual alarms, isolation baffle actuators, and explosion-proof interlock modules to achieve on-site linkage for anomaly handling. Automated equipment includes stacker cranes, rail shuttles, handling robots, chain or roller conveyors, automated sorting and lifting machines, and forklifts or electric pallet trucks equipped with rotating attachments, used to perform storage, retrieval, transfer, temporary storage, loading, unloading, and isolation operations on abnormal pallets; Level 4 anomalies trigger the highest level of explosion-proof warning, pushing emergency handling instructions to on-site personnel's handheld terminals via a safe and redundant communication link; after the anomaly is handled, the status data of the repaired pallet is collected, the pallet status baseline fingerprint is updated, the pallet verification mark and circulation restrictions are removed; and a petrochemical product logistics supervision report is generated.

[0071] The safety redundancy design includes control link redundancy and alarm link redundancy. Regarding control link redundancy, the central processing unit outputs a freeze command to the automation equipment control system while simultaneously outputting a safety interlock trigger signal. The freeze command is used at the software level to suspend scheduling and task issuance, while the safety interlock trigger signal is used at the hardware level to disconnect the conveying, lifting, and access execution links associated with the abnormal tray, thus maintaining a frozen state even in the event of communication interruption or controller malfunction. Regarding alarm link redundancy, level four anomalies simultaneously send warning information through both the main communication link and the safety redundancy communication link. The main communication link is used for platform push and scheduling coordination, while the safety redundancy communication link preferably uses at least two combinations of cellular networks and short-range wireless or wired private networks. Alarm messages are repeatedly sent and acknowledged to ensure that explosion-proof warnings and emergency response instructions can be reliably delivered to on-site personnel terminals even under conditions of network congestion, partial link loss, or electromagnetic interference.

[0072] In this implementation plan, by structurally binding abnormal states with pallet identity, node location, and timing information to form traceable abnormal behavior tags, the plan achieves unified identification and consistent management of abnormal information throughout the entire process; improves the timeliness and consistency of abnormal handling; reduces the probability of safety risks and inventory chaos caused by the continued circulation of abnormal pallets; and enhances the reliability of handling under conditions of communication interruption, network congestion, or electromagnetic interference by relying on the dual redundancy design of control links and alarm links.

[0073] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0074] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A shared palletized petrochemical product integrated logistics monitoring system based on the Internet of Things, characterized in that, include: The shared pallet pallet data acquisition module is used to collect and acquire shared pallet monitoring data and perform data supplementation, dynamic noise reduction and normalization processing to generate a unified sorted time. The benchmark construction and data awareness module is used to generate a pallet status benchmark fingerprint based on preprocessed shared pallet supervision data, perform deviation assessment, and build a perception dataset. The consistency verification and anomaly feature extraction module is used to identify candidate anomaly starting points based on the perceptual dataset, perform consistency verification evaluation, screen valid anomaly starting points, and construct anomaly feature vectors. The anomaly identification and hierarchical handling module is used to assess the degree of deviation of the current stack status based on the anomaly feature vector, identify anomalies, and generate a petrochemical product logistics supervision report.

2. The IoT-based shared palletized petrochemical product integrated logistics monitoring system according to claim 1, characterized in that: The specific process of collecting shared pallet monitoring data, performing data supplementation, dynamic noise reduction and normalization, and generating a unified sorting time is as follows: Collect data on shared palletized petrochemical products throughout the entire process of packaging, palletizing, warehousing, in-warehouse relocation, temporary storage, loading / unloading, and outbound delivery. This data includes unique RFID identification of the pallet, basic information of the petrochemical products, UWB positioning coordinates, real-time loading quality of the pallet, load distribution data, thermal field distribution characteristics, thermal field center of gravity coordinates, ambient gas concentration, temperature and humidity, node entry and exit timestamps, and storage dwell time. Identification is collected by non-contactly reading the built-in UHF-RFID chip using an RFID reader / writer. This is combined with a UWB positioning system, 5G thermal imaging equipment, gas sensors, and intelligent inspection robots to supplement data on obscured areas and high-level shelving environments. Dynamic noise reduction of shared pallet monitoring data is achieved through density clustering-based outlier removal. Z-Score standardization is used to normalize the shared pallet monitoring data. Each collected message is written to three time fields: device time, gateway reception time, and platform entry time. The timeline is then corrected based on network RTT to generate a unified sorted time.

3. The IoT-based shared palletized petrochemical product integrated logistics monitoring system according to claim 1, characterized in that: The specific process for generating a pallet status baseline fingerprint based on preprocessed shared pallet monitoring data is as follows: Based on the pre-processed shared pallet monitoring data, during the petrochemical product packaging and palletizing process, the ambient gas concentration in the target area is sampled multiple times to obtain a gas baseline value after packaging and palletizing or when the pallet enters the target logistics node. The initial loading mass and the load distribution at the four corners are obtained through weighing sensors; the initial thermal field distribution is obtained through thermal imaging equipment, and the initial thermal field centroid coordinates are extracted; at the same time, the planned dwell time range of batch products at each logistics node is obtained from the warehouse management system. The above data together constitute the pallet status baseline fingerprint, which is bound to the pallet RFID identification and stored in the central database, and synchronized to the ERP system and the warehouse management system. Real-time data collection across the entire logistics process is achieved through distributed sensing nodes. At key logistics nodes such as pallet entry into storage, transfer, loading / unloading, and outbound, the difference between the real-time loading mass and the initial loading mass of the pallet is calculated to obtain the mass change. The load distribution offset is obtained by calculating the sum of squares of the load differences at each load-bearing point, reflecting the degree of deviation of the internal force distribution of the pallet relative to the initial baseline state. The thermal field centroid offset is calculated using Euclidean distance with reference to the initial baseline thermal field centroid coordinates, yielding the thermal field centroid offset for each collection cycle. Local areas exceeding the threshold fluctuation range are identified from the thermal field distribution characteristics using temperature threshold judgment, and local thermal field temperature anomalies are extracted. First-order difference calculations are performed on the ambient gas concentration to obtain the characteristic gas concentration gradient and concentration change rate. The time difference between the node entry and exit timestamps is calculated to obtain the actual dwell time, which is compared with the planned dwell time interval to obtain the time deviation value, forming real-time sensing data.

4. The IoT-based shared palletized petrochemical product integrated logistics monitoring system according to claim 1, characterized in that: The specific process for conducting deviation assessment is as follows: The mass deviation is obtained by dividing the mass change by the mass concealment anomaly threshold; the thermal field deviation is obtained by dividing the thermal field centroid offset by the thermal field concealment anomaly threshold; the gas deviation is obtained by dividing the characteristic gas concentration gradient by the gas concealment anomaly threshold; the duration deviation is obtained by dividing the time deviation value by the duration concealment anomaly threshold; the correlation matrix is ​​obtained by performing covariance statistics and normalization on samples collected from similar petrochemical products under baseline operating conditions; the mass deviation, thermal field deviation, gas deviation, and duration deviation are sequentially combined into a four-dimensional deviation vector; the four-dimensional deviation vector is transposed, multiplied by the inverse matrix of the correlation matrix plus the regularized identity matrix term, and then multiplied by the four-dimensional deviation vector to obtain the deviation assessment value.

5. The IoT-based shared palletized petrochemical product integrated logistics monitoring system according to claim 1, characterized in that: The specific process for constructing the perception dataset is as follows: The deviation evaluation value is compared with the deviation threshold in real time. When the deviation evaluation value is greater than the deviation threshold, it is determined that there is a momentary disturbance in the data of the current collection period. The recording of batch data is paused and the data is re-collected and evaluated in the next collection period before making a judgment. When the deviation evaluation value is less than or equal to the deviation threshold, it is determined that the current data meets the quality requirements. Starting from the current collection period, real-time sensing data within a fixed collection period length is recorded to construct a sensing dataset.

6. The IoT-based shared palletized petrochemical product integrated logistics monitoring system according to claim 1, characterized in that: The specific process of identifying candidate anomaly starting points based on the perception dataset and performing consistency verification and evaluation is as follows: A time series was constructed based on the sensing dataset. Time difference operation was performed on the mass change to obtain the mass change rate within a unit period; time difference operation was performed on the thermal field centroid offset to obtain the thermal field centroid offset rate; the dwell time deviation value was integrated and accumulated to obtain the dwell time overrun rate; the trend detection of the change rate of each indicator was performed by the sliding window analysis method. When the instantaneous change rate of a certain indicator exceeds ∂ times, it is identified as a change trend abrupt change point, and the timestamp corresponding to the abrupt change point is marked as a candidate anomaly starting point. Real-time sensing data from a fixed acquisition period starting at the candidate anomaly origin is extracted. The load distribution offset is multiplied by the load correction coefficient to obtain the enhancement factor. The ratio of the quality deviation to the quality threshold is calculated, and the ratio is multiplied by the enhancement factor to obtain the quality anomaly term. The duration deviation is multiplied by the duration amplification factor and then added to obtain the duration anomaly term. The product of the thermal field deviation, gas deviation, and duration anomaly term is calculated to obtain the coupling anomaly term. The product of the quality anomaly term and the coupling anomaly term for each period within the fixed acquisition period is accumulated sequentially and then divided by the fixed acquisition period length to obtain the consistency verification evaluation value.

7. The IoT-based shared palletized petrochemical product integrated logistics monitoring system according to claim 1, characterized in that: The specific process for screening valid anomaly starting points is as follows: The consistency check evaluation value and the consistency threshold are compared in real time. When the consistency check evaluation value is greater than the consistency threshold, the current candidate point is confirmed as a valid anomaly starting point, and the sensing data after the starting point is captured. When the consistency check evaluation value is lower than the consistency threshold, the candidate anomaly starting point is slid backward by one collection cycle, and the consistency check evaluation process is repeated until the first valid anomaly starting point is identified. Once the effective anomaly starting point is determined, each acquisition cycle is analyzed sequentially from the effective anomaly starting point. The standard deviations of the mass change rate, thermal field centroid shift rate, residence time overrun rate, and concentration change rate are calculated. For any standard deviation, if it exceeds the anomaly change rate threshold within the acquisition cycle, the current acquisition cycle is determined to be the anomaly persistence point.

8. The IoT-based shared palletized petrochemical product integrated logistics monitoring system according to claim 1, characterized in that: The specific process for constructing the abnormal feature vector is as follows: Sensing data within a continuous acquisition period from the effective anomaly initiation point to the anomaly persistence point is extracted, and the maximum value of mass change within the anomaly segment is calculated to obtain the mass loss peak value; the maximum distance of the thermal field center of gravity shift is statistically analyzed to obtain the thermal field shift extreme value; the maximum increase value of the characteristic gas concentration is extracted to obtain the gas leakage peak value; the maximum duration of the actual residence time exceeding the planned residence time is calculated to obtain the duration overrun peak value; and the anomaly feature vector is constructed by sequentially splicing the data.

9. The IoT-based shared palletized petrochemical product integrated logistics monitoring system according to claim 1, characterized in that: The specific process for assessing the degree of deviation of the current stack state based on the abnormal feature vector is as follows: The gas anomaly is obtained by multiplying the peak duration overdue value by the duration coupling coefficient and then adding it to the gas leakage peak value; the combined enhancement value is obtained by multiplying the mass loss peak value by the gas anomaly value; the thermal field offset extreme value is obtained by adding it to the thermal field offset suppression term; the gas reference value is obtained by adding it to the zero-prevention term; the gas factor is obtained by multiplying the thermal field offset factor and the gas factor and then adding it to the combined enhancement value; the combined enhancement value is then divided by the offset term to obtain the comprehensive anomaly assessment value. The comprehensive anomaly assessment value is compared with the anomaly classification threshold. If the comprehensive anomaly assessment value is less than or equal to the first-level anomaly threshold, it is judged as a first-level anomaly; if the comprehensive anomaly assessment value is greater than the first-level anomaly threshold and less than or equal to the second-level anomaly threshold, it is judged as a second-level anomaly. When the comprehensive anomaly assessment value is greater than the level 2 anomaly threshold and less than or equal to the level 3 anomaly threshold, it is judged as a level 3 anomaly; When the comprehensive anomaly assessment value is greater than the level 3 anomaly threshold, it is judged as a level 4 anomaly.

10. The IoT-based shared palletized petrochemical product integrated logistics monitoring system according to claim 1, characterized in that: The specific process for identifying anomalies and generating a petrochemical product logistics monitoring report is as follows: The identified abnormal status is structurally bound to the pallet's RFID identification, the logistics node it is located in, and the time of the abnormality to generate a pallet status abnormal behavior label; at the same time, a review mark is set for the pallets identified as abnormal, and the corresponding abnormality type and abnormality level are marked. An anomaly level triggering and hierarchical handling mechanism based on comprehensive anomaly assessment output: Level 1 anomalies send background reminders and are continuously monitored; Level 2 anomalies trigger audible and visual warnings and dispatch intelligent inspection robots for on-site verification; Level 3 anomalies send instruction interruption signals to the automation equipment control system via the industrial Ethernet protocol, freeze the automatic transfer permissions of the anomaly tray in real time, and suspend the scheduling instructions of the automation equipment. The highest level of explosion-proof warning is triggered by a level 4 anomaly, and emergency handling instructions are pushed to the handheld terminals of on-site personnel through a safe and redundant communication link. After the anomaly is handled, the status data of the repaired pallet is collected, the pallet status benchmark fingerprint is updated, the pallet verification mark and circulation restrictions are lifted, and a petrochemical product logistics supervision report is generated.

Images