Logistics access dynamic management report generation method and system based on Internet of Things lock

Abstract

The invention relates to the technical field of data analysis, provides a logistics access dynamic management report generation method and system based on an Internet of Things lock, and is used for realizing full-process intelligent generation and optimization of a logistics management report. The method comprises the following steps: acquiring an access event data set of a target logistics area; performing cross-modal feature extraction processing on the access event data set to obtain a dynamic access feature set containing space-time dimension features; based on a preset report type template, inputting the dynamic access feature set into a trained report generation model for multi-task analysis processing, and generating an initial logistics management report set; performing semantic integrity verification and logic conflict detection processing on the initial logistics management report set to generate an optimized target logistics management report; and binding the target logistics management report with an authorized access interface of the user terminal, and performing incremental updating operation on the target logistics management report according to the access event triggered in real time.

Description

Technical Field

[0001] The present application belongs to the field of data analysis technology, and specifically relates to a method and system for generating a dynamic management report on logistics access based on an Internet of Things lock. Background Art

[0002] In current logistics management systems, existing technologies usually use static rule engines to generate management reports, which cannot effectively integrate the implicit associations between spatiotemporal trajectories, user behaviors, and package attributes, resulting in the lack of key node analysis. Moreover, traditional report generation methods are often based on a single-task filling mechanism with fixed templates, which makes it difficult to achieve collaborative modeling and cross-validation of multi-dimensional management indicators (such as logistics tracking, authority verification, and anomaly detection). In addition, existing systems lack the ability to verify the inherent logic of report content, and often encounter situations where the causal chain of events is broken or authority conflicts are not identified, requiring manual secondary verification. In terms of dynamic updates, most solutions adopt a periodic batch processing mode, which cannot respond to changes in logistics scenarios in real time, resulting in a significant reduction in the timeliness of reports and the value of decision-making guidance. The above-mentioned defects have seriously restricted the intelligence level of logistics management systems. Therefore, how to achieve full-process intelligent generation and optimization of logistics management reports has become a technical problem that urgently needs to be solved at this stage. Summary of the Invention

[0003] The present application provides a method and system for generating a dynamic management report on logistics access based on an Internet of Things lock, which is used to realize the intelligent generation and optimization of the entire process of logistics management reports.

[0004] In a first aspect, an embodiment of the present application provides a method for generating a dynamic management report on logistics access based on an IoT lock, which is applied to a report generation system, the method comprising: obtaining an access event dataset of a target logistics area, the access event dataset comprising multiple event parameters associated with a smart lock, the multiple event parameters comprising a lock operation type, a user identity, an access timestamp, device location coordinates, and package volume data; performing cross-modal feature extraction processing on the access event dataset to obtain a dynamic access feature set comprising spatiotemporal dimensional features, the cross-modal feature extraction processing comprising jointly encoding the multiple event parameters to generate a feature vector sequence; based on a preset report type template, inputting the dynamic access feature set into a trained report generation model for multi-task parsing processing to generate an initial logistics management report set, the initial logistics management report set comprising a logistics tracking report, a permission verification report, and an anomaly detection report; performing semantic integrity verification and logical conflict detection processing on the initial logistics management report set to generate an optimized target logistics management report; binding the target logistics management report to an authorized access interface of a user terminal, and performing incremental update operations on the target logistics management report based on access events triggered in real time.

[0005] In a second aspect, an embodiment of the present application provides a report generation system, which includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the above method.

[0006] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is run on a report generation system, the computer program is used to enable the report generation system to perform the steps of the above method.

[0007] In the implementation of this application, by introducing cross-modal dynamic analysis, the full-process intelligent generation and optimization of logistics management reports are realized. First, different from traditional single-dimensional log statistics, the embodiment of this application innovatively jointly encodes multi-source heterogeneous data such as spatiotemporal trajectories, user behaviors and package attributes to form a dynamic feature vector sequence with contextual associations, which effectively captures the implicit operating rules between logistics nodes. Secondly, the multi-task parsing mechanism based on deep learning breaks through the static filling mode of traditional report templates, and makes the collaborative analysis of multi-dimensional management indicators possible by generating three types of related reports for logistics tracking, authority verification and anomaly detection in parallel. Then, the semantic integrity verification processing provided by the embodiment of this application adopts a two-way attention mechanism to strengthen the causal reasoning of event links while eliminating logical contradictions, ensuring that the report content has business interpretability. Finally, through the dynamic binding of the authorization interface and real-time events, a closed-loop report update system is constructed, so that management decisions can adapt to the dynamic evolution of logistics scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flowchart of a method for generating a dynamic management report on logistics access based on an Internet of Things lock provided in an embodiment of the present application.

[0009] Figure 2 A structural diagram of a report generation system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of the technical solutions of this application, but not all of them. Based on the embodiments described in this application document, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the technical solutions of this application.

[0011] See also Figure 1, which is a logistics access dynamic management report generation method based on the Internet of Things lock provided in an embodiment of the present application. The method can be applied to the report generation system. The specific process is as follows: Step 101-Step 105.

[0012] Step 101: Obtain an access event dataset of a target logistics area, wherein the access event dataset includes multiple event parameters associated with the smart lock, including lock operation type, user identity, access timestamp, device location coordinates, and package volume data.

[0013] In an embodiment of the present application, an exemplary implementation of obtaining a data set of access event in a target logistics area can be achieved by utilizing a multi-modal smart lock system installed on each floor of a high-rise residential building. The multi-modal smart lock system can continuously run a data acquisition program, and its built-in multi-source sensing unit can capture complete access behavior characteristics by working in collaboration.

[0014] When a courier performs a package deposit or withdrawal operation, the smart lock's electromechanical trigger mechanism first activates the operation type recognition mechanism. The pressure sensor array automatically determines the "deposit" or "retrieval" operation based on the contact area and force pattern. Simultaneously, the user identity verification module reads the encrypted digital identity credential, dynamically generated by the property management platform and containing multiple verification elements such as the courier company registration code, the courier's employee number, and the temporary authorization period. At the moment the operation is completed, the lock's built-in satellite positioning module and inertial navigation unit fuse to generate centimeter-level three-dimensional spatial coordinate data. This device's positioning coordinates not only record the building unit's access control location but also, through the assistance of indoor Bluetooth beacons, pinpoint the specific floor's storage area. A timing recording unit simultaneously generates a strictly sequential sequence of access timestamps. Its time synchronization mechanism utilizes a regional atomic clock timing protocol to ensure millisecond-level consistency in time records across multiple nodes within the building. During the physical feature collection phase of the package, the three-dimensional vision module integrated on the top of the lock initiates a multispectral scanning program, and calculates the package's external dimensions through a combination of structured light projection and infrared imaging technology, generating composite parameters including the maximum projected area and spatial volume.

[0015] It can be understood that the above multimodal data is standardized and encapsulated through the protocol conversion module of the edge gateway to form access event data units that comply with the IoT data specifications, and finally integrated into a complete access event data set covering all buildings at the community-level data aggregation node. This data set can reflect the operating status of each smart lock in the community in real time.

[0016] Step 102: performing cross-modal feature extraction processing on the access event data set to obtain a dynamic access feature set including spatiotemporal dimension features, wherein the cross-modal feature extraction processing includes jointly encoding the multiple event parameters to generate a feature vector sequence.

[0017] In an embodiment of the present application, an exemplary implementation process of performing cross-modal feature extraction processing on the access event dataset can be achieved by launching a multi-stage pipeline processing mechanism through a feature engineering platform deployed in a regional data processing center.

[0018] First, a spatiotemporal benchmark alignment operation is performed to normalize the time axis of heterogeneous data from different lock nodes. The clock drift error between devices is eliminated through a sliding time window algorithm. At the same time, a geographic coordinate system conversion model is used to unify the positioning data of each lock into the benchmark coordinate system of the community building information model.

[0019] During the feature conversion phase, the time series analysis component mines access timestamps for periodic patterns, extracting temporal features such as high-frequency access periods and operation interval distribution. The spatial topology processor constructs a three-dimensional building access heat map based on device location coordinates, quantifying package flow density on each floor. For user identity data, the feature encoder employs differential privacy protection techniques to generate anonymized identity feature vectors while maintaining traceability of authorized relationships. Package volume data is binned and discretized, and combined with historical operational data, multiple capacity tiers are defined.

[0020] In this embodiment of the present application, the cross-modal fusion module uses an attention mechanism to dynamically weight and integrate spatiotemporal features, identity features, and package features, generating a multidimensional feature vector with strong representational capabilities. For example, during peak holiday package delivery periods, the system automatically strengthens the feature weights of nighttime access periods in the temporal dimension, while also increasing the significance of spatial location features in large package operation records. Finally, the output dynamic access feature set fully preserves the spatiotemporal correlations and operational mode characteristics of the original data, providing high-information structured input for subsequent report generation.

[0021] In an exemplary technical implementation, the core of cross-modal feature extraction processing lies in deep correlation modeling of heterogeneous data. Specifically, the joint encoding process adopts a layered fusion strategy: first, the device positioning coordinates are converted into standardized three-dimensional coordinates of the Building Information Model (BIM) through the spatiotemporal reference alignment module. For example, the longitude and latitude of satellite positioning are mapped to a building floor number-grid coordinate pair (such as "5F-A12"). At the same time, the access timestamps of each lock are converted into millisecond-level absolute time series in a unified time zone using an atomic clock synchronization protocol. At the feature conversion layer, the user identity is generated into a 128-dimensional anonymous vector through a hash obfuscation algorithm, while retaining the permission level label (such as the three-level code of "Courier-Enterprise A-Temporary Permission"); the package volume data is divided into 8 discrete capacity levels (such as XS / S / M / L, etc.) according to the size of the building container grid through dynamic binning technology. Cross-modal fusion utilizes a gated attention mechanism, dynamically integrating spatiotemporal, identity, and package features by calculating cross-weights. For example, if the system detects "a user triggers a deposit operation three times within 10 minutes," the system automatically increases the weight of the user's identity within the time window by 1.5 times. It also incorporates the spatiotemporal coordinates of these unusually high-frequency operations into the high-risk flag of the feature vector. The resulting feature vector sequence utilizes a 512-dimensional tensor structure, with the first 128 dimensions encoding the building topology and operation time period. The middle 256 dimensions store parameters related to user behavior patterns and package specifications, and the final 128 dimensions retain the derived feature space required for anomaly detection.

[0022] Step 103: Based on a preset report type template, the dynamic access feature set is input into the trained report generation model for multi-task parsing processing to generate an initial logistics management report set, which includes a logistics tracking report, an authority verification report, and an anomaly detection report.

[0023] In an embodiment of the present application, the exemplary implementation process of generating an initial set of logistics management reports based on pre-set report type templates can be implemented by combining a cloud-based intelligent report generation system with a library of expert-verified templates. Each template in the library corresponds to the structured output requirements of a specific management dimension.

[0024] For example, the logistics tracking report generator uses a spatiotemporal trajectory reconstruction algorithm to correlate and analyze the time series in the dynamic access feature set with spatial coordinates, automatically drawing a flow path map of the package within the community and marking the length of stay at each transfer node. The permission verification report generation module uses a double verification mechanism. First, the identity credential information in the feature vector is batch compared with the property authorization database. Then, the biometric data collected by the smart cat-eye device is correlated to generate a verification conclusion that includes an authorization matching score and abnormal operation records. The anomaly detection component uses an unsupervised learning model to perform cluster analysis on the feature vector and uses an outlier detection algorithm to identify access events that do not conform to the normal pattern. When continuous non-routine operations are detected, an early warning is automatically triggered and a special report containing a risk level assessment is generated.

[0025] For example, in the specialized analysis unit for large packages, the system combines historical feature data to build a capacity prediction model, automatically generating operational optimization recommendations when detecting the access of oversized packages. All of the above report generation processes adhere to pre-set template constraints, ensuring that the output format complies with industry standards and the content is auditable. For example, when generating monthly operational analysis reports, the system automatically aggregates feature data from multiple days and compares the access density and package size distribution of different buildings to produce visual analysis charts with valuable decision-making support.

[0026] In one exemplary implementation, the logistics tracking report achieves full-link visualization through a spatiotemporal trajectory reconstruction engine, whose core technology is a path inference algorithm based on the Hidden Markov Model (HMM). For example, when a package is detected to be transferred from the "3F-C Area Smart Cabinet" to the "1F Unit Access Control Temporary Storage Point," the system automatically links the BIM spatial topology data of the two nodes, generates a three-dimensional movement trajectory diagram with turning angle annotations in the report, and calculates the cargo retention coefficient (retention duration / area capacity threshold). The permission verification report uses a two-factor authentication mechanism: the first level verifies whether the operator is within the property's authorized "enterprise-employee-time period" whitelist by comparing the hash value of the authorization certificate stored on the blockchain; the second level generates a verification conclusion including biometric similarity (0-1 score) by linking the smart lock's NFC card swipe record with the building access control camera's facial feature vector. Anomaly detection reports utilize a dynamic baseline model built using the Isolation Forest algorithm. For example, if a courier attempts to open a smart locker during an unauthorized period (e.g., 2:00 AM) using a deregistered ID, the system combines the spatiotemporal feature vector of the event with the Euclidean distance (>3σ) of historically normal operation clusters. This generates a specialized analysis module within the report, including a risk level (e.g., P3), associated device ID, and recommended actions (e.g., automatically triggering a lock freeze protocol). All three types of reports feature built-in version traceability, utilizing a combined "timestamp-SHA256 (feature vector)" watermarking technology to ensure data immutability.

[0027] Step 104: Perform semantic integrity verification and logic conflict detection on the initial logistics management report set to generate an optimized target logistics management report.

[0028] It can be understood that in the exemplary implementation process of semantic integrity verification and logical conflict detection processing for the initial logistics management report set, the semantic integrity verifier first parses the time narrative logic in the report, verifies the rationality of the sequence of each operation event through the temporal consistency detection algorithm, and automatically triggers the data traceability verification procedure when a contradictory record is found that the pickup time is earlier than the storage time. The logic verification module cross-compares multiple reports, such as correlating the list of authorized personnel in the authority verification report with the actual operator characteristics in the logistics tracking report to detect abnormal situations where identity information does not match. The data completeness assessment unit scans the package volume field in the report. When it is found that some packages lack three-dimensional scanning data, it automatically queries the cached data of the edge node for supplementary extraction.

[0029] During the conflict resolution phase, the system uses a rule-based correction strategy combined with machine learning inference to handle abnormal data. For example, for access records with positioning deviations, the system interpolates and corrects trajectories based on the spatiotemporal characteristics of neighboring nodes. After multiple rounds of iterative optimization, the target logistics management report generated meets industry regulatory standards for data integrity and logical rigor, accurately reflecting the actual operational status of the community logistics access system.

[0030] Step 105: Bind the target logistics management report to the authorized access interface of the user terminal, and perform incremental update operations on the target logistics management report according to the access events triggered in real time.

[0031] In this embodiment of the application, a role-based access control architecture can be established to create differentiated data views for different roles, such as property management personnel, courier companies, and property owners. When a user initiates a report query request through a mobile terminal application, the identity authentication gateway first verifies the legitimacy of the user's digital certificate and then returns the report content with the corresponding confidentiality level according to the pre-set permission rules.

[0032] In practice, the incremental update component continuously monitors real-time data streams. When a new access event is detected, it automatically extracts relevant feature vectors and calculates their impact on the existing report. It then applies differentiated update strategies to partially revise the report content. For example, when a package's status changes from "Deposited" to "Received," the system only needs to update the status marker and timeline node in the logistics tracking report, eliminating the need to regenerate the entire report. The version control module maintains a historical record of report revisions, ensuring that each update operation can be traced back to a specific access event. A secure synchronization mechanism utilizes an end-to-end encrypted transmission protocol to ensure that report data remains under control during transmission between the cloud and the terminal.

[0033] Through the above-mentioned dynamic update method, the system can continuously keep the report content highly synchronized with the actual status of the logistics storage and access site, providing all relevant parties with highly timely decision-making support information.

[0034] In an optional embodiment, the cross-modal feature extraction process is performed on the access event dataset in step 102 to obtain a dynamic access feature set containing spatiotemporal dimension features, including:

[0035] Step 1021: Perform data integrity verification on the access event data set to generate pre-processed access event data; wherein the data integrity verification includes at least two combinations of missing parameter completion, abnormal timestamp correction, positioning coordinate correction, and operation type standardization.

[0036] In specific implementations, the missing parameter completion module detected that satellite positioning coordinates were missing due to network latency during a certain storage operation. It then reconstructed the coordinates using Bluetooth beacon positioning data and displacement trajectories recorded by the inertial navigation unit, generating supplemental positioning information with centimeter-level accuracy through a spatial interpolation algorithm. The abnormal timestamp correction component discovered that smart lock E5 generated access records that deviated from the standard time base due to a clock module failure. Using a sliding time window algorithm, it compared the timestamp sequence of this device with the operation records of adjacent locks. Based on the continuity of the event intervals, the abnormal timestamp was corrected to the correct time point determined by the atomic clock timing protocol. The positioning coordinate correction unit identified that the 3D spatial coordinates of a package access event overlapped with the wall structure in the building information model. Using the spatial constraints of the floor plan, the original coordinates were projected and corrected to accurately map to the actual physical location of the corridor locker. The operation type standardization processor uniformly encoded heterogeneous operation descriptions such as "brute force opening" and "unauthorized opening" reported by different lock nodes into a pre-set "abnormal violent opening" standard type to ensure semantic consistency in subsequent feature processing.

[0037] Step 1022: Call the spatiotemporal feature encoder to perform joint feature mapping processing on the preprocessed access event data to obtain a first-level spatiotemporal feature vector; wherein the spatiotemporal feature encoder uses a multi-head attention mechanism to model the spatiotemporal correlation between the access timestamp and the device positioning coordinates.

[0038] This step constructs a multidimensional spatiotemporal correlation model to convert discrete access timestamps and device location coordinates into continuous feature vectors with strong representational capabilities. The spatiotemporal feature encoder utilizes a layered processing architecture. It first captures the diurnal and monthly patterns of access operations through periodic time encoding, and then combines this with geographic grid encoding to accurately characterize the spatial distribution of package flows. The encoder integrates a temporal-spatial cross-attention mechanism to dynamically capture correlations between access densities in different areas within specific time periods. For example, this mechanism identifies the coordinated use of lockers in Buildings 3 and 5 between 6:00 PM and 8:00 PM on weekday evenings. During processing, a gated recurrent unit (GRU) is used to model time series data, preserving the causal dependencies between operational events along the time axis. Attention-weighted aggregation is also used to enhance the spatiotemporal representation of high-frequency operation areas. The final output, the first-level spatiotemporal feature vector, is a 768-dimensional tensor. The first 256 dimensions encode temporal periodicity, the middle 256 dimensions represent spatial distribution dynamics, and the last 256 dimensions incorporate spatiotemporal interactions, fully capturing the spatiotemporal evolution of logistics access behavior.

[0039] Step 1023: Call the operation mode encoder to perform discrete feature embedding processing on the lock operation type to generate a second-level operation feature vector.

[0040] This step focuses on analyzing the security status and working mode implicit in the operation behavior of smart locks, and converts discrete operation categories into quantifiable feature representations by establishing a mapping relationship between preset operation modes and historical data. The operation mode encoder adopts a technical path that combines feature clustering and embedding mapping. First, based on the density clustering algorithm, feature clusters of typical operation modes such as normal authorized opening and abnormal violent opening are mined from massive historical operation data, and then the real-time operation type is mapped to the vector space of the corresponding feature cluster through the prototype network. The encoding process introduces a feature pyramid network architecture to extract operation mode features at multiple time scales: millisecond timestamps reveal the transient characteristics of operations, minute-level window statistics reflect fluctuations in operation intensity, and hour-level aggregated data characterizes the distribution of operation modes. For special operation types such as abnormal violent opening, the encoder automatically triggers the multi-sensor data fusion mechanism to integrate multi-source signals such as pressure sensor waveforms and inertial navigation unit displacement trajectories to generate composite feature vectors. The generated second-level operation feature vector is a 512-dimensional tensor. The first 128 dimensions represent the basic operation type, the middle 256 dimensions encode the dynamic characteristics of the operation process, and the last 128 dimensions integrate historical pattern matching information to provide fine-grained operation behavior characteristics for subsequent feature fusion.

[0041] Step 1024: Perform feature fusion processing on the first-level spatiotemporal feature vector and the second-level operation feature vector to generate the dynamic access feature set.

[0042] For example, the spatiotemporal feature vector includes dimensions such as [time period encoding, spatial grid encoding, access density], while the operational feature vector includes parameters such as [operation mode category, operation duration, number of verifications]. The fusion process utilizes a channel attention mechanism to dynamically assign weights to the two feature types: when processing data from peak holiday periods, the spatiotemporal feature weight is increased to 65%, highlighting regional parcel flow characteristics; when abnormal operation patterns are detected, the operational feature weight is automatically increased to 70%, enhancing the ability to identify security incidents. The fused dynamic access feature set is dimensionality-compressed using a fully connected layer, generating a 512-dimensional feature vector suitable for input into the subsequent report generation model, fully preserving the spatiotemporal correlations and operational mode characteristics of the original data.

[0043] In a preferred embodiment, the calling of the spatiotemporal feature encoder in step 1022 performs joint feature mapping on the pre-processed access event data to obtain a first-level spatiotemporal feature vector, including:

[0044] Step 10221: Convert the access timestamp into a periodic time coding vector, and perform geographic grid coding on the device positioning coordinates to generate a spatial partition vector.

[0045] During implementation, the system decomposes each operation timestamp into three cyclical dimensions: hourly segment, weekday identifier, and holiday identifier. Hourly segments are encoded using sine-cosine functions to represent the 24-hour cycle. Weekday identifiers are one-hot encoded to distinguish between Monday and Sunday. Holiday identifiers use binary flags to indicate statutory holiday status. Geographic grid encoding projects the device's location coordinates onto a pre-divided 10m x 10m geographic grid and, combined with the building's three-dimensional coordinate system, generates a spatial partition vector containing the floor number, east-west axis partition, and north-south axis partition. For example, an access operation in the east storage area on the 15th floor of Building 3 is encoded as a four-dimensional spatial vector [3, 15, E, 12], accurately representing its orientation characteristics in the three-dimensional space of the community.

[0046] Step 10222: Construct a time-space cross-attention weight matrix, which is used to characterize the correlation strength of access events in different time periods in spatial distribution.

[0047] In specific implementation, the row dimension of the time-space cross-attention weight matrix corresponds to the three time zones of morning, noon, and evening, and the column dimension corresponds to the express storage and access areas of each building. Analysis of historical data revealed that the distribution density of storage and access operations in Building 5 during the evening is significantly higher than at other times. Based on this, the system generates time-space attention weights, giving the characteristics of this area during this time period a higher weight in subsequent processing. The dynamic weight adjustment mechanism monitors the frequency of operations in each spatial zone in real time during the current time period. If a sudden increase in storage and access density in the underground storage room is detected during holidays, the attention coefficient of the corresponding time-space unit is automatically increased.

[0048] Step 10223: Use a gated recurrent unit to perform serial modeling on the periodic time encoding vector and the spatial partition vector to generate initial spatiotemporal features with temporal dependencies.

[0049] For example, the system arranges 24 consecutive hours of access events in chronological order as an input sequence. The input for each time step includes the time and spatial codes for that moment. The gating mechanism dynamically adjusts the information transmission ratio based on the current input and historical status. For example, when processing a period of high access during the afternoon, the reset gate automatically filters out low-correlation historical features, while the update gate strengthens the output of features associated with the current spatial location and operation type. The initial spatiotemporal feature vector generated after multiple layers of repetitive calculations effectively captures the periodic fluctuations of package access operations in the temporal dimension and the clustered distribution patterns in the spatial dimension.

[0050] Step 10224: Perform dynamic weighted aggregation processing on the initial spatiotemporal features based on the time-space cross-attention weight matrix to generate the first-level spatiotemporal feature vector.

[0051] When processing access data from weekday evenings between 6:00 PM and 8:00 PM, the system identified peak usage of the parcel lockers in Building 2 during this time period and automatically increased the weight coefficient of the corresponding spatiotemporal unit by 30%. This resulted in the generated first-level spatiotemporal feature vector highlighting the parcel flow characteristics of this area. A soft attention mechanism was employed during the weighted aggregation process to smoothly transition features between adjacent time periods and spatial partitions, preventing distortion of feature boundaries caused by sudden weight changes.

[0052] In a preferred embodiment, the calling operation mode encoder in step 1023 performs discrete feature embedding processing on the lock operation type to generate a second-level operation feature vector, including:

[0053] Step 10231: Establish a mapping relationship table between lock operation types and preset operation modes, where the preset operation modes include normal authorized opening, abnormal violent opening, timeout failure, and multiple verification failures.

[0054] For example, the system pre-sets four standard operation modes: Normal Authorization Open corresponds to a compliant operation after the courier has passed identity verification; Abnormal Forced Open indicates that the anti-tampering mechanism has been triggered by three or more incorrect passwords; Timeout Failure detects that the cabinet door has been left open for more than 300 seconds; and Multiple Verification Failure records the failure of both biometric and digital certificate authentication. The mapping table is defined by the expert rule engine, for example, marking the combination of "Fingerprint Verification Failure + Dynamic Password Error" as a Multiple Verification Failure mode.

[0055] Step 10232: Perform pattern cluster analysis on the historical event data corresponding to each lock operation type to generate an operation pattern feature cluster.

[0056] For example, the clustering algorithm identified two typical modes of operation for normal authorized opening: the first being a courier completing the deposit and closing the cabinet door within 30 seconds, and the second being an oversized package deposit or withdrawal, resulting in an operation lasting over 120 seconds. Based on this, the system generated two characteristic cluster centers for normal operation, representing routine and special operation modes, respectively. Clustering of abnormal force opening events revealed two characteristic clusters: mechanical damage and electronic attack. The former is accompanied by a peak pressure sensor alarm, while the latter exhibits a high frequency of password attempts.

[0057] Step 10233: Map the lock operation type to the center point of the corresponding operation mode feature cluster to generate an operation mode embedding vector.

[0058] When the system detects an operation such as "door open for 250 seconds, then forced to close," it calculates the Euclidean distance between the operation and each feature cluster, identifies it as belonging to the feature cluster for the timeout-unclosed mode, and maps the operation to a preset four-dimensional embedding vector of [0, 0, 1, 0]. The embedding vector's dimensions correspond to the four preset operation modes, using a one-hot encoding method to represent the mode category to which the current operation belongs. The system also retains the relative distance from the feature cluster center as the vector amplitude parameter.

[0059] Step 10234: Use a feature pyramid network to perform multi-scale feature extraction on the operation mode embedding vector to generate the second-level operation feature vector.

[0060] For example, the bottom-level network extracts atomic features of operation types, such as millisecond-level precision data on operation duration. The middle-level network aggregates operation pattern sequences within adjacent time windows to identify short-term transition patterns. The high-level network analyzes the macroscopic distribution of operation patterns throughout the day and night. Multi-scale features are fused through horizontal connections, resulting in a second-level operation feature vector that combines both instantaneous operation features and long-term pattern features. For example, it can identify the periodic pattern of a courier experiencing a timeout after completing three normal access operations.

[0061] By applying the above-mentioned embodiments, a multi-level feature processing mechanism is used to achieve deep integration and intelligent analysis of multimodal logistics data. From raw data verification to spatiotemporal feature encoding, and then to operation mode mining, each processing link is optimized for the operating characteristics of the smart lock system. In particular, with the synergistic effect of the spatiotemporal cross-attention mechanism and the feature pyramid network, the system can effectively capture the spatiotemporal evolution of package access behavior and operational mode anomalies, laying a data foundation for generating high-precision logistics management reports. The entire processing flow strictly adheres to the spatiotemporal correlation characteristics of IoT data, ensuring that the generated dynamic access feature set not only meets the input requirements of the machine learning model, but also has good business interpretability.

[0062] As an optional implementation, in step 103, based on a preset report type template, the dynamic access feature set is input into a trained report generation model for multi-task parsing to generate an initial logistics management report set, including:

[0063] Step 1031: Perform task relevance analysis on the dynamic access feature set to determine a feature subset division method corresponding to the preset report type template.

[0064] In step 1031, the structured field requirements of the pre-defined report type template are first parsed. For the logistics tracking report, a spatiotemporal trajectory feature subset is extracted, consisting of access timestamp sequences, device location coordinate sequences, and package volume change parameters. For the permission verification report, an identity verification feature subset is selected, encompassing user identity, authorization period validity, and biometric matching parameters. Furthermore, for the anomaly detection report, an operation mode feature subset is extracted, including lock operation type distribution, operation interval statistics, and abnormal operation flags. A multi-label classification algorithm is employed in the feature segmentation process. When a single feature parameter is detected as being associated with multiple report types, the system automatically creates a feature copy and adds a task identifier suffix. For example, device location coordinates are used for both route mapping in the logistics tracking report and location anomaly analysis in the anomaly detection report. In this case, the system generates a feature copy with both the "Logistics Tracking - Coordinates" and "Anomaly Detection - Coordinates" labels to ensure data isolation for subsequent processing.

[0065] Step 1032: Input each feature subset into the corresponding subtask processing module for parsing and processing to generate a preliminary report fragment; wherein, the subtask processing module includes a logistics path restoration module, an authority matching verification module and an abnormal pattern recognition module.

[0066] This step implements multi-dimensional data analysis through a parallel processing mechanism. The logistics path restoration module focuses on reconstructing the spatiotemporal trajectory of package flows, the authority matching and verification module specializes in verifying the legitimacy of operator identities, and the anomaly pattern recognition module deeply explores the characteristics of unconventional operations. The system dynamically allocates computing resources to each subtask module based on the field requirements of the preset report type template: the spatiotemporal trajectory graph generation task is assigned to the graphics processor for accelerated calculations, the authority verification task enables the secure encryption coprocessor, and the anomaly detection task calls the tensor processing unit to perform model inference. Based on the meta-tag information of the feature subset, the data routing controller transmits the logistics tracking feature subset to the memory buffer of the logistics path restoration module, imports the authority verification feature subset into the secure sandbox environment of the authority matching and verification module, and loads the anomaly detection feature subset into the distributed computing nodes of the anomaly pattern recognition module. Each subtask processing module adopts a heterogeneous computing architecture to achieve clock synchronization of the processing process while maintaining data isolation, ensuring that the generated preliminary report fragments have strict time alignment.

[0067] Step 1033: Perform cross-task information fusion processing on the preliminary report segment to eliminate semantic conflicts between different subtask processing modules and obtain a fused report segment.

[0068] In step 1033, the system detects that a package collection operation in the logistics tracking report is displayed as an unauthorized user in the authority verification report. The conflict resolution engine immediately initiates the traceability verification: first, the user identity identification codes in the two reports are compared for consistency, then the timestamp synchronization is checked, and finally the operation logs cached by the edge node are reviewed. When it is confirmed that the identity is misplaced due to a transmission error in the data collection phase, the system automatically corrects the user identification field in the authority verification report and adds a data correction description to the fused report segment. The fusion process simultaneously integrates the spatiotemporal trajectory map and the alarm trigger record. When the device positioning coordinates corresponding to an abnormal operation event stop multiple times in the logistics path, the system generates a joint analysis paragraph to reveal the potential correlation between abnormal operations and package detention.

[0069] Step 1034: According to the structured format requirements of the preset report type template, the merged report fragments are logically sorted and key fields are filled to generate the initial logistics management report set.

[0070] In step 1034, the logistics tracking report adopts a narrative structure based on a timeline, arranging the merged report segments in the order in which the operations occurred. Device location coordinate diagrams and dwell time statistics are inserted at key nodes. The authority verification report uses abnormal user identifiers as an index to create a tree-like hierarchical directory: first-level entries are categorized by express delivery company, second-level entries list abnormal personnel with abnormal work numbers, and third-level entries detail the specific spatiotemporal information of unauthorized operations. The anomaly detection report sorts content based on alarm levels, prioritizing red alarm records and displaying a three-dimensional heat map showing the spatial clustering of abnormal operations. A key field filler automatically extracts core parameters from the report segments, such as total number of operations, average transit time, and maximum package volume, on the summary page of the logistics tracking report. Quantitative analysis results, such as the authorization match rate and the proportion of unauthorized operations, are added to the conclusion page of the authority verification report.

[0071] As an optional implementation, step 1032 inputs each feature subset into the corresponding subtask processing module for parsing and processing to generate a preliminary report segment, including:

[0072] Step 10321: Generate a spatiotemporal trajectory graph based on the access timestamp and device location coordinates in the dynamic access feature set, wherein the spatiotemporal trajectory graph includes device location coordinate nodes connected in chronological order.

[0073] For example, the timestamp sequence of package access operations completed by a courier on the 12th floor of Building 3 is sorted according to millisecond time accuracy and mapped to the three-dimensional coordinate system of the building information model. Each access event is converted into a coordinate node in the spatiotemporal trajectory graph, and the node attributes include the horizontal positioning coordinates and vertical height values accurate to the floor. Connecting edges with directional arrows are generated between adjacent nodes, and the edge attributes record the operation interval duration and the change in package volume. In particular, when continuous operations are detected on smart locks on different floors, the system automatically inserts a floor switching marker node and connects the vertical movement path with a dotted line in the trajectory graph.

[0074] Step 10322: Perform path restoration processing on the space-time trajectory graph, compare the distance between adjacent positioning coordinate nodes with the preset transportation speed threshold, identify the stop nodes and insert the stop period annotations, and generate the logistics transportation path description text.

[0075] For example, the system calculates the Euclidean distance between adjacent positioning coordinate nodes and uses the preset transport speed threshold to determine the package flow status. For example, when a package is transferred from a locker on the 1st floor of Building 5 to the 15th floor of Building 3, the system detects that the straight-line distance between the two points exceeds 50 meters, but the operation interval is only 30 seconds, and automatically marks it as a transport route anomaly. Regarding stop node identification, when a device positioning coordinate node appears three times in a row and the total stop time exceeds 5 minutes, the module inserts a stop period annotation with a yellow highlight mark and generates a detailed record of "stopped at the locker on the east side of the 15th floor of Building 3 from 15:30 to 15:35" in the logistics transport route description text. The path restoration algorithm simultaneously associates the package volume data. When a stop node is detected with the feature of a reduced package volume, a "partial pickup" status annotation is automatically added.

[0076] Step 10323: Extract the user identity identifier from the dynamic access feature set, match it with the area authority field in the pre-stored authorization list step by step, filter out abnormal user identifiers with invalid authority or area out of bounds, and generate a summary of the authority verification results.

[0077] For example, the system matches the company registration code and employee ID information in the courier's digital identity certificate with the access list in the property management authorization database. The verification process uses a multi-level filtering mechanism: first, verifying whether the courier company is on the property management cooperation whitelist, then checking whether the service provider's employee number is valid, and finally comparing whether the operation time period is within the temporary authorization time window. If a courier attempts to open the smart lock in Building 8 during an unauthorized period, the system generates a permission verification result summary, marks the user's identity as "time period unauthorized", and links the device location coordinates of the user's three most recent successful operations to generate an unauthorized area analysis chart.

[0078] Step 10324: Input the lock operation type and access timestamp in the dynamic access feature set into the pre-trained anomaly detection model, extract the historical frequency distribution characteristics of the operation type in the corresponding time period, calculate the feature deviation between the current operation type and the historical frequency distribution, and generate an abnormal operation score.

[0079] For example, the system establishes a historical frequency distribution baseline for storage operations between 8:00 PM and 10:00 PM on weekday evenings. If it detects a 300% increase in storage operations during the same time period on a particular day, the model calculates the Z-score between the current frequency and the historical mean, and combines this with changes in the distribution of operation types to generate a characteristic deviation index. For example, if a smart lock experiences three consecutive "abnormal force opening" operations, the model extracts the device's records of similar operations from the past 30 days, quantifies the difference between the current frequency and the historical distribution using the KL divergence algorithm, and outputs a deviation score.

[0080] Step 10325: Based on the comparison result of the abnormal operation score and the dynamic threshold, an alarm mark is triggered when the abnormal operation score exceeds the dynamic threshold, and an alarm trigger record is generated by associating the corresponding device positioning coordinate node.

[0081] For example, the dynamic threshold adjustment module monitors the overall operational status of the community in real time, automatically increasing the threshold by 20% during peak holiday periods to accommodate normal business volume increases. When the abnormal operation score of a locker exceeds the dynamic threshold, the alarm marker generator creates an alarm trigger record containing the device's location coordinates, operation timestamp, and score details. For example, if multiple "multiple verification failures" were detected in the underground storage area of Building 5 during non-service hours, the system would generate a red alarm marker, link the surveillance camera data in that area, and embed a video retrieval link in the alarm record.

[0082] This design, through a multi-level task analysis and information fusion mechanism, enables the automated generation of multi-dimensional logistics management reports. From feature subset division to preliminary fragment generation, and then to cross-task fusion and structured orchestration, each processing link is deeply coupled with the operational data characteristics of the smart lock system. In particular, in the process of linking spatiotemporal trajectory reconstruction with anomaly scoring, the system effectively reveals the inherent connection between the package flow path and the operational safety situation, providing property management personnel with decision-making support information with both operational guidance and risk warning value. The final generated initial logistics management report set strictly follows the industry standard format, while maintaining data traceability and dynamic update flexibility, fully meeting the refined needs of modern logistics management.

[0083] As an optional implementation, step 10324 inputs the lock operation type and access timestamp in the dynamic access feature set into a pre-trained anomaly detection model, extracts the historical frequency distribution features of the operation type in the corresponding time period, calculates the feature deviation between the current operation type and the historical frequency distribution, and generates an abnormal operation score, including:

[0084] Step 1032401: According to the time continuity of the access timestamp, the lock operation type and the access timestamp are cut into operation sequence units corresponding to the dynamic time window, the starting timestamp of the dynamic time window is the first access timestamp of the current operation sequence unit, and the ending timestamp of the dynamic time window is the last access timestamp of the current operation sequence unit plus the preset delay interval.

[0085] For example, the system detects that the courier performed eight consecutive access operations between 09:00 and 11:30 on October 15, 2023. It sets the first access timestamp 09:00:23 as the start timestamp of the current operation sequence unit, and adds the last access timestamp 11:28:47 to the preset delay interval of 300 seconds to form an end timestamp 11:33:47, thereby generating a dynamic time window covering all operations during this period. During the window cutting process, the system automatically skips access intervals exceeding 5 minutes and divides continuous and intensive operations into independent units. For example, the five intensive access operations between 09:30 and 09:45 are merged into the same operation sequence unit, while the subsequent operations with an interval of 6 minutes are divided into new units.

[0086] Step 1032402: Extract a historical contemporaneous operation sequence that completely matches the time window of each operation sequence unit from the pre-trained anomaly detection model, wherein the historical contemporaneous operation sequence includes a lock operation type sequence chain generated by historical frequency distribution characteristics, a historical time interval threshold set of adjacent operation types, and a historical frequency distribution characteristic of the lock operation type.

[0087] For example, for the period of 09:00-11:30 on October 15th, corresponding to the current dynamic time window, the model retrieves the operation data for the same period over the past three years, extracting a historical sequence containing a sequence of operation types such as normal authorized opening and package storage and retrieval completion. This historical sequence includes the frequency distribution characteristics of each lock operation type during the same period. For example, the historical average number of deposit operations during the period of 09:00-10:00 is 12, and the number of retrieval operations is 8. It also records a set of time interval thresholds between adjacent operation types, such as the average interval between deposit and retrieval operations is 25 minutes ± 5 minutes.

[0088] Step 1032403: compare the current lock operation type sequence chain in the operation sequence unit with the lock operation type sequence chain of the historical concurrent operation sequence step by step, count the number of lock operation types that match in consecutive order, and generate a sequence consistency parameter based on the historical frequency distribution characteristics.

[0089] For example, the sequence chain in the current operation sequence unit is [deposit → deposit → retrieve → deposit → retrieve], and the matching degree is calculated with the typical sequence chain of the historical operation sequence of the same period [deposit → retrieve → deposit → retrieve → deposit]. The system recognizes that the consecutive occurrence of the first two deposit operations breaks the historical pattern and counts the number of lock operation types that match the sequence as three (retrieve → deposit → retrieve), generating a sequential consistency parameter of 60%. A sliding window mechanism is introduced in the parameter calculation process, allowing for some sequence deviations but applying an exponentially decaying weight to consecutive abnormal operation types.

[0090] Step 1032404: perform interval coverage comparison on the current time intervals of adjacent lock operation types in the operation sequence unit and the historical time interval threshold set of the historical concurrent operation sequence, calculate the proportion of current time intervals that meet the historical time interval threshold, and generate time interval matching parameters based on historical frequency distribution characteristics.

[0091] For example, the intervals between adjacent save operations in the current operation sequence are 8 minutes, 15 minutes, and 20 minutes, respectively. These intervals are compared against the 10-18 minute interval threshold set for the historical operation sequence of the same period. The system calculates that the current interval falls within the historical threshold 75% of the time and generates the interval matching parameters. During this comparison, a dynamic boundary adjustment mechanism is employed to automatically relax the interval threshold by 20% during peak hours, for example, adjusting the original 10-18 minute threshold to 8-20 minutes to accommodate fluctuations in business volume.

[0092] Step 1032405: extract the current frequency distribution characteristics of each lock operation type in the operation sequence unit, perform item-by-item difference calculation with the historical frequency distribution characteristics of the lock operation type in the historical operation sequence of the same period, and generate the frequency characteristic deviation of each lock operation type.

[0093] For example, the system calculates the difference between the current operation sequence unit's average frequency of 8 deposit operations and 5 retrieval operations for the same period and the average frequency of 8 deposit operations and 5 retrieval operations. The frequency deviation is -25% for the deposit operation frequency and -20% for the retrieval operation frequency. This is used to generate a frequency characteristic deviation index. A frequency fluctuation correction factor is introduced into the calculation to provide a tolerance for frequency deviations during special periods such as holidays.

[0094] Step 1032406: Extract the sequence feature weight, time interval feature weight and frequency feature weight bound to the lock operation type from the pre-trained anomaly detection model, and the sequence feature weight and time interval feature weight are dynamically adjusted according to the operation type association strength in the historical frequency distribution feature.

[0095] In step 1032406, the dynamic feature weight extraction mechanism is activated. Based on the correlation strength of operation types in the historical frequency distribution features, the pre-trained anomaly detection model assigns a sequence feature weight of 0.4, a time interval feature weight of 0.3, and a frequency feature weight of 0.3 to the current operation sequence unit. The dynamic weight parameter adjustment module, detecting a recent 15% increase in the volatility of the storage operation frequency, automatically increases the frequency feature weight to 0.35 and correspondingly decreases the sequence feature weight to 0.35 to maintain a constant total weight. The weight adjustment process follows a sliding average algorithm to prevent sudden parameter changes from impacting detection stability.

[0096] Step 1032407: perform weighted calculation on the sequence consistency parameter and the sequence feature weight to generate the sequence deviation, perform weighted calculation on the time interval matching parameter and the time interval feature weight to generate the time interval deviation, and perform weighted calculation on the frequency feature deviation and the frequency feature weight to generate the frequency deviation.

[0097] In this step, the weighted deviation calculation engine is activated. The sequence deviation calculation unit multiplies the sequence consistency parameter of 60% by the sequence feature weight of 0.35 to produce a sequence deviation of 21%. The time interval consistency parameter of 75% is combined with the time interval feature weight of 0.3 to generate a time interval deviation of 22.5%. The frequency feature deviation index of -22.5% is weighted by the frequency feature weight of 0.35 to output a frequency deviation of -7.875%. The weighting process uses a direction-aware mechanism to asymmetrically handle positive and negative deviations. For example, the weight coefficient for frequency reduction deviation is adjusted to 0.8 to distinguish between normal fluctuations in traffic volume.

[0098] Step 1032408: Superimpose the sequence deviation, time interval deviation and frequency deviation in a preset ratio to generate an initial feature deviation, and normalize the initial feature deviation according to the total number of lock operation types in the operation sequence unit to generate a standardized feature deviation.

[0099] In step 1032408, the system superimposes the sequence deviation of 21%, the time interval deviation of 22.5%, and the frequency deviation of -7.875% in a preset ratio of 1:1:0.8 to generate an initial characteristic deviation of 35.7%. Based on the total number of lock operation types contained in the current operation sequence unit, a logarithmic function is used to normalize the initial characteristic deviation to a standardized characteristic deviation of 0.68. The normalization parameter matrix is dynamically updated, and the segmented normalization algorithm is automatically activated when the operation sequence unit length exceeds 20% of the historical maximum.

[0100] Step 1032409: Extract the maximum and minimum values of the standardized characteristic deviations of all historical concurrent operation sequences in the historical frequency distribution features, and construct a dynamic threshold interval with the maximum value as the upper limit of the abnormal threshold and the minimum value as the lower limit of the abnormal threshold.

[0101] In step 1032409, the system extracts a standardized feature deviation dataset from historical contemporaneous operation sequences. The maximum value for the three-year data is calculated to be 0.82, and the minimum value is 0.35. This creates a dynamic threshold interval with an upper anomaly threshold of 0.82 and a lower anomaly threshold of 0.35. The threshold adjustment module detects the presence of promotional activity in the current month and automatically increases the upper threshold to 0.85 to accommodate reasonable business fluctuations. The threshold interval utilizes a dynamic buffer mechanism, triggering adaptive calibration if the threshold boundary is reached for three consecutive periods.

[0102] Step 1032410: Compare the standardized feature deviation of the current operation sequence unit with the dynamic threshold interval. When the standardized feature deviation exceeds the upper limit of the abnormal threshold or is lower than the lower limit of the abnormal threshold, mark it as an abnormal operation score, and extract the start access timestamp and end access timestamp corresponding to the dynamic time window as the trigger time range of the abnormal operation score.

[0103] In step 1032410, the standardized characteristic deviation of 0.68 for the current operation sequence unit is compared to the dynamic threshold range of 0.35-0.85 and determined to be within the normal range. When the standardized characteristic deviation of another operation sequence unit reaches 0.91, the system marks it as an abnormal operation and extracts the corresponding access start timestamp of 14:00:00 and the access end timestamp of 14:30:00 as the trigger time range. The scoring and marking process incorporates a confidence grading mechanism, automatically escalating units exceeding the threshold by more than 20% to high-risk alerts.

[0104] Step 1032411: perform a timestamp complete match between the trigger time range and the device location coordinate nodes in the dynamic access feature set, filter out the device location coordinate nodes whose timestamps fall within the trigger time range, and generate an alarm trigger record bound to the abnormal operation score.

[0105] In step 1032411, the system precisely matches the timestamps of the device location coordinate nodes in the dynamic access feature set with the trigger time range of 14:00:00-14:30:00, filtering out five operation records for the lockers on the east side of the 12th floor of Building 3 during this period. A spatial aggregation algorithm identifies that the clustering of these device location coordinate nodes in three-dimensional space exceeds a threshold, generating an alarm trigger record with an abnormal operation score of 0.91, spatial hotspot coordinates, and operation type distribution. During record generation, user identification is simultaneously linked to video surveillance footage, forming a traceable multimodal chain of evidence.

[0106] In a preferred implementation, the step 104 of performing semantic integrity verification and logical conflict detection on the initial logistics management report set to generate an optimized target logistics management report includes:

[0107] Step 1041: extracting the equipment positioning coordinate node sequence and the dwell period annotation from the logistics transportation route description text to generate a spatiotemporal event node set.

[0108] For example, the system analyzes a courier's operation log at a locker on the east side of the 12th floor of Building 3, extracting a sequence of centimeter-accurate three-dimensional coordinates [3-12-E-05, 3-12-E-07, 3-12-W-02] and corresponding timestamps [14:25:30, 14:28:15, 14:35:47]. The dwell time annotation generation module detects that the coordinate 3-12-E-05 appears three times consecutively between 14:30:00 and 14:33:00 and automatically creates a dwell time annotation entry, recording the package's duration and the operator's identity. The spatiotemporal event node collection uses a hierarchical storage structure, with a top-level index sorted by time and a bottom-level index storing each node's coordinate accuracy parameters and sensor data checksums.

[0109] Step 1042: extract the abnormal user identifier and its corresponding device location coordinate node from the permission verification result summary to generate a permission conflict node set.

[0110] For example, when a courier with employee number CN-SF-220715 attempts to open the smart lock in Building 5's underground storage area at 5:45 PM, the permission conflict detection engine records the abnormal user ID and associates it with the coordinate node [5-B1-N-12] of the device they operated. The permission conflict node set is stored in a dual-chain data structure: the main chain is sorted by spatial grid code, and the secondary chain records the anomaly type and authorization expiration time. A data cleaning module runs simultaneously to eliminate coordinate drift points caused by network latency, ensuring that the spatial positioning accuracy of the conflicting nodes is within a 10-centimeter error range.

[0111] Step 1043: Perform spatial overlap analysis on the spatiotemporal event node set and the authority conflict node set to detect the coexistence of spatiotemporal events and authority conflicts under the same device positioning coordinate node.

[0112] For example, during implementation, the system uses the device location coordinate node [3-12-E-05] for a locker on the east side of the 12th floor of Building 3 as the core analysis target. First, the system extracts the corresponding dwell time period (14:30:00-14:33:00) and associated package access and storage records from the spatiotemporal event node set. Simultaneously, the system retrieves the abnormal user ID CN-ZT-210906 and its permission expiration time (14:31:00) recorded for this coordinate node from the permission conflict node set. The spatial overlap analysis engine calculates the inclusion relationship between the permission expiration time and the dwell time period, determining that user CN-ZT-210906 continued operating for 2 minutes and 15 seconds after the permission expired, generating a dual spatial-temporal conflict event record. A layered verification mechanism is employed during conflict detection: the bottom layer verifies the spatial encoding consistency of the device location coordinate node, the middle layer verifies the inclusion relationship of the timestamp, and the upper layer associates the user ID with the permission change log, ensuring the multi-dimensional accuracy of conflict determination.

[0113] Step 1044: When it is detected that there is an authority conflict in the positioning coordinate node marked in the stay period, an authority review mark is inserted into the logistics transportation path description text, and the severity level of the alarm trigger record is updated.

[0114] For example, for the permission conflict event at node 3-12-E-05, a timestamped review mark "[14:30-14:33 requires permission review]" is inserted into the path description text. The video surveillance system is then linked to capture footage from that time period to generate a review evidence package. The severity level of the alarm trigger record is raised from Level 2 to Level 4, and a "P" suffix is added to the alarm code to indicate that manual intervention is required. The level update module simultaneously notifies the property management system to lock the relevant lockers and displays a temporarily disabled red alert on the user terminal.

[0115] Step 1045: traverse the time windows in all alarm trigger records, verify the path continuity of the corresponding time windows in the logistics transportation path description text, and repair the path breakage problem caused by the missing positioning coordinate node.

[0116] For example, the system detected a 32-second data gap in the logistics transport route description between 3:00 PM and 3:05 PM. It automatically used the inertial navigation unit's displacement data to fill in the missing coordinate nodes. The repair algorithm, combined with the package volume change rate at adjacent nodes, inferred the possible path and generated the estimated trajectory segments marked with dashed lines. The data integrity assessment unit assigned a confidence score to the repaired path. If the score fell below a threshold, a manual review process was triggered to ensure the accuracy of the route description.

[0117] Step 1046: Reorganize the updated logistics transport route description text, authority verification result summary, and alarm trigger record according to the paragraph structure of the preset report template to generate a semantically coherent target logistics management report.

[0118] For example, the system reorganizes updated logistics transport route descriptions, permission verification summary results with additional permission review markers, and updated alarm trigger records into a three-level directory structure: the first section presents a timeline analysis with a 3D trajectory diagram, the second section displays a table of permission conflict events sorted by spatial grid, and the third section compiles compound alarm records and their processing status. A template engine automatically populates report metadata, including timestamps, data coverage indicators, and checksums, ultimately outputting documents in the ISO-32000 standard PDF / A-3 format, ensuring long-term readability and tamper-proofing of the report content.

[0119] In a preferred implementation, the step 1043 of performing spatial overlap analysis on the spatiotemporal event node set and the permission conflict node set to detect the coexistence of spatiotemporal events and permission conflicts under the same device location coordinate node includes:

[0120] Step 10431: Establish a spatial grid code for the device positioning coordinate node, and map each coordinate in the spatiotemporal event node set to a corresponding spatial grid.

[0121] For example, the physical space of the lockers in the east section of the 12th floor of Building 3 is divided into a 3D grid with a precision of 0.5 meters. Each grid is assigned a unique code, such as 3-12-E-05-25. The coordinate [3-12-E-05] in the spatiotemporal event node set is mapped to the grid 3-12-E-05-18 via a coordinate conversion engine. The geofence parameters of the grid boundary are also recorded. The spatial grid encoding uses an octree index structure, which enables rapid retrieval of node data from adjacent grids and provides a spatial reference for subsequent conflict analysis.

[0122] Step 10432: Search the permission conflict node set for abnormal user identifiers having the same spatial grid code as the spatiotemporal event node, and generate a candidate conflict node list.

[0123] For example, the system searches for nodes with conflicting permissions within a 1-meter radius, centered around spatiotemporal event node 3-12-E-05-18. It discovers that user CN-ZT-210906 performed an unauthorized operation at grid 3-12-E-05-22 at 14:32:15, generating a candidate entry containing the user's ID and operation details. The spatial matching algorithm uses an R-tree index to accelerate queries, achieving millisecond response times in a database of millions of nodes. A buffer is also established to address ambiguity in the attribution of edge coordinates.

[0124] Step 10433: Verify whether the access timestamp of each node in the candidate conflict node list falls within the stay period range of the corresponding spatiotemporal event node.

[0125] For example, the system compares the timestamp of the abnormal operation under investigation, 14:32:15, with the time period of the spatiotemporal event node, 14:30:00-14:33:00, and determines that there is a 172-second overlap. The time verification engine uses a sliding window mechanism to impose a ±3-second tolerance on boundary timestamps to avoid misjudgments due to clock synchronization errors. The verification results are stored in the conflict analysis log and trigger the multi-source data review process.

[0126] Step 10434: When the access timestamp overlaps with the stay period mark, extract the historical permission change record of the abnormal user ID to determine whether the permission expiration event occurs after the stay period mark.

[0127] For example, the system queried the authorization information for the abnormal user CN-ZT-210906 and discovered that their area access rights were automatically revoked by the property management system at 14:31:00. A timeline comparison revealed that the expiration date was later than the marked start time of the stay period, 14:30:00, generating a delayed expiration event record. The verification process linked the operation logs of the rights management system to verify the issuance time and execution delay parameters of the rights change instruction, ensuring the accuracy of the timeliness judgment.

[0128] Step 10435: If the permission expiration event time is later than the start time marked in the stay period, a retrospective permission exception description is added to the permission verification result summary, and the path risk identifier in the logistics transportation path description text is updated accordingly.

[0129] For example, the system inserts a red warning paragraph into the permission verification summary, detailing the violation: "User CN-ZT-210906 continued operating for 2 minutes and 15 seconds after permission expired." The logistics route description is also updated simultaneously, with shield-shaped warning icons inserted at corresponding spatiotemporal event nodes and the route risk level raised from yellow to orange. The route icon library uses pre-built 3D warning models to generate pulsed light effects within the 3D trajectory diagram, enhancing the warning effect of the visual report.

[0130] Step 10436: Adjust the conflict detection range according to the hierarchical accuracy of the spatial grid coding, establish a buffer area between adjacent spatial grids, detect cross-grid authority conflict events and generate a composite alarm record associated with multiple device positioning coordinate nodes.

[0131] For example, the system detected continuous unauthorized operations by abnormal user CN-ZT-210906 in adjacent grids 3-12-E-05-19 to 3-12-E-05-21, automatically creating a cross-grid composite alarm record. The buffer zone generator established a circular detection zone with a radius of 0.2 meters to capture user wandering behavior at the grid edge. The composite alarm record is represented using a topological association graph. The lines between nodes represent the migration path of the abnormal operation, and the edge weights indicate the time interval and spatial span parameters.

[0132] By applying the above technical solutions, precise optimization of logistics management reports is achieved through multi-level verification and correlation analysis. From basic data extraction to spatial and temporal composite analysis, and then to cross-system data linkage, each processing link deeply integrates the multimodal data characteristics of the smart lock system. In particular, with the synergistic effect of spatial grid coding and time window verification, the system effectively identifies the risks of time and space coupling between permission expiration and business operations, providing property management with decision support that is both real-time and traceable. The final target logistics management report is transformed into intuitive business insights through structured reorganization and visualization enhancement, significantly improving the efficiency of logistics supervision.

[0133] In one non-limiting implementation, step 105 of binding the target logistics management report to the authorized access interface of the user terminal and performing an incremental update operation on the target logistics management report based on a real-time triggered access event includes:

[0134] Step 1051: When the user terminal initiates an access operation request, the current event parameters are captured in real time and an incremental feature vector is generated.

[0135] In this step, the report generation system captures event parameters in real time as the courier deposits a package into a locker on the east side of the 12th floor of Building 3, generating an incremental feature vector. When the courier uses a smart lock to open the locker door and deposit the package, the system simultaneously collects the access timestamp (14:35:23.456), the device's location coordinates ([3-12-E-05]), the user's identity (CN-SF-220715), and the package's volume scan data. Using edge computing nodes, these multidimensional parameters are integrated into a 256-dimensional incremental feature vector. This vector contains structured fields such as a time period code, a spatial grid hash value, and an operation type embedding vector. The package volume data is binned and discretized and then mapped to pre-set capacity level intervals, forming a standardized feature representation.

[0136] Step 1052: Perform feature similarity matching between the incremental feature vector and the dynamically accessed feature set to determine the report section to be updated.

[0137] In this step, the system detected an 87% similarity between the current operation and the storage patterns of Building 3 during the lunchtime peak period in the historical feature set. It automatically linked the information to the "Locker Usage Analysis from 12:00 PM to 2:00 PM" section in the logistics tracking report. The matching process employed a hybrid cosine similarity and Euclidean distance algorithm, assigning a 60% weight to spatiotemporal features, a 30% weight to operation pattern features, and a 10% weight to package features, pinpointing the report sections requiring updates. The dynamic similarity threshold adjustment module lowered the matching threshold by 5 percentage points during peak business hours based on real-time operation density to expand update coverage.

[0138] Step 1053: Calling a difference comparison algorithm to calculate the content difference between the target logistics management report and the incremental event data, and generating an incremental update patch.

[0139] During this step, the system detected that the new operations had increased the average daily usage of lockers on the east side of the 12th floor of Building 3 by 18%. Consequently, an incremental update patch was generated, including updated statistical charts, thermal value corrections, and risk level adjustments. The patch generator used differential block technology to break down the modifications into three independent modules: metadata changes, text paragraph updates, and visual element adjustments. Each module included a version compatibility checksum to ensure the atomicity and traceability of the incremental update.

[0140] Step 1054: A transaction processing mechanism is used to perform an atomic update operation on the target logistics management report to maintain the consistency of the report version during the update process.

[0141] In this step, the system creates a copy of the report in memory and sequentially applies the incremental update patch's metadata changes, text updates, and chart adjustments. After each transaction completes, the system writes the operation log and generates a version snapshot. When a mismatch in the historical data version is detected during the parcel volume distribution map update transaction, the transaction rollback mechanism automatically restores the report to its pre-update state, rechecks data consistency, and resubmits the transaction request, ensuring data integrity throughout the version iteration process.

[0142] Step 1055: After the update is completed, a report change notification is pushed to the user terminal and a version update log is recorded.

[0143] During this step, the property manager's mobile device receives a digitally signed update notification, including the location of the modified section, a summary of the risk level change, and a visual preview thumbnail. The version update log records the incremental update's operation sequence number, timestamp, impact scope, and digital fingerprint. Log entries are encrypted using a Merkle tree structure for tamper-proofing, enabling rapid verification of data integrity during subsequent audits.

[0144] In a non-limiting implementation, after binding the target logistics management report to the authorized access interface of the user terminal, the method further includes the step of performing visual graphic mapping processing on the target logistics management report:

[0145] Step 20101: parse the logistics tracking report, authority verification report and anomaly detection report in the target logistics management report, extract the equipment positioning coordinate node, access timestamp, abnormal operation score and user identity as key fields, and generate structured visual data.

[0146] In this step, the system extracts the device location coordinate node sequence [3-12-E-05, 5-B1-N-12] from the logistics tracking report, associates the access timestamps [14:35:23, 14:38:07] with the package volume parameters; extracts the abnormal user identifier CN-ZT-210906 and its operation records from the permission verification report; and obtains the abnormal operation score [0.68, 0.91] for each coordinate node from the anomaly detection report. The data normalization processor converts heterogeneous fields into a unified unit of measure, converts timestamps to ISO8601 format, and maps spatial coordinates to the WGS84 geographic coordinate system, constructing a structured dataset containing 87 feature dimensions.

[0147] Step 20102: Segment the structured visualization data by geographic grid level according to the spatial distribution density of the device location coordinate nodes and the time span of the access timestamps, and convert and generate an interactive map layer with a time sliding window.

[0148] In this step, the system divides the Building 3 area into a 0.5-meter-precision 3D grid and counts the frequency of operations on each grid using a sliding window between 2:30 PM and 3:00 PM. A spatiotemporal aggregation algorithm automatically identifies the frequently operated grid 3-12-E-05 and highlights it in red. The layer rendering engine uses WebGL technology for dynamic loading, allowing users to adjust the time window range through touch gestures, and updating the grid shading intensity and data tooltip content in real time.

[0149] Step 20103: Map the abnormal operation score to the preset alarm level color scheme, generate a thermal layer based on the spatial clustering result of the device positioning coordinate nodes, and overlay the thermal peak area on the interactive map layer.

[0150] In this step, the system divides the score range of 0.68-0.91 into five alert levels, corresponding to gradients of blue, green, yellow, orange, and red. Spatial cluster analysis identifies the highest density of anomaly scores in the locker cluster on the east side of the 12th floor of Building 3, generating an orange thermal peak region with a radius of 1.2 meters centered at [3-12-E-05]. Thermal rendering uses a Gaussian kernel density estimation algorithm to smooth out the score differences between adjacent grid cells, creating a continuous color transition on the interactive map.

[0151] Step 20104: extract the abnormal user identity identifier and its associated access timestamp in the permission verification report, match the corresponding device location coordinate node in the interactive map layer, and generate a permission abnormality event pop-up window with a timeline mark.

[0152] In this step, when a user clicks on the peak heat area, the system retrieves the device location coordinate node [3-12-E-05] corresponding to the 14:35:23 timestamp, associates it with the permission expiration record for the abnormal user CN-ZT-210906, and pops up a three-layered event details window: the top layer displays user identity information and a timeline of permission status, the middle layer displays thumbnails of the operation video clips, and the bottom layer provides a link to download the original data. The pop-up window's position dynamically adapts to the terminal screen resolution and automatically switches to a full-screen modal dialog on mobile devices.

[0153] Step 20105: Perform path fitting on the device location coordinate nodes in the logistics tracking report in the order of access timestamps to generate a spatiotemporal trajectory line, and mark the stay period mark and the package volume data floating prompt box on the spatiotemporal trajectory line.

[0154] In this step, the system sorts the courier's device location coordinate nodes between 2:30 PM and 2:40 PM by time and uses a cubic spline interpolation algorithm to generate a smooth path curve. The thickness of the path lines dynamically changes with the size of the package. For large packages, the lines between nodes are thickened to 3 pixels and have a flashing effect. A floating tooltip integrates data from multiple sources. When the cursor moves to the 2:35:23 PM node, it displays the package size, the operator's ID, and a timecode hyperlink to the associated video.

[0155] Step 20106: Overlay and fuse the interactive map layer, thermal layer, permission exception event pop-up window and time-space trajectory connection into multiple layers to generate a visual graphic interface with a unified coordinate system, bind the user identity identifier with the pre-stored user portrait icon, and render a dynamic user identification mark in the visual graphic interface.

[0156] In this step, the system sets the interactive map as the base layer, overlays the heatmap with a 40% transparency, and places a pop-up window indicating permission errors as a floating layer on top. The user portrait icon library uses pre-set courier company logos, rendering the CN-SF-220715 logo as a red SF Express icon, which is dynamically marked at the starting point of the movement trajectory. The layer blending mode uses alpha compositing technology to ensure color accuracy and legibility of each visual element in overlapping areas.

[0157] Step 20107: In response to the spatiotemporal range filtering instruction triggered by the authorized access interface of the user terminal, the matching device location coordinate nodes and access timestamps are dynamically intercepted from the structured visualization data, and the time sliding window range of the interactive map layer and the data rendering granularity of the thermal layer are updated.

[0158] In this step, when the property manager selects the 12th floor of Building 3 on their mobile device and sets the time window from 2:30 PM to 2:45 PM, the system dynamically captures the 38 matching device location coordinate nodes and adjusts the thermal rendering granularity from a 0.5-meter grid to a 1-meter grid to improve rendering performance. The data cropping module executes spatiotemporal index queries in parallel, filtering hundreds of thousands of data points within 200 milliseconds, ensuring real-time and smooth interactive operations.

[0159] Step 20108: According to the incremental update operation of the target logistics management report, the newly added equipment positioning coordinate nodes and the updated abnormal operation scores are captured in real time, the incremental data tags are inserted into the visual graphic interface, and the path extension direction of the next time period is predicted based on the historical trajectory fitting results, and a dynamic path layer with a predicted path dotted line is generated.

[0160] In this step, the system detects the newly added device location coordinate node [3-12-W-03] and its anomaly score of 0.85, inserting a flashing blue marker in the visualization interface. The trajectory prediction algorithm, based on historical movement patterns and the current motion vector, calculates the likely path for the next five minutes, generating a gray dashed trajectory extending toward Building 5. The confidence level of the predicted path is visualized using a transparency gradient, with path segments farther from the current time having increasing transparency, up to 60%.

[0161] Step 20109: When it is detected that the spatial distance between the incremental data mark and the historical device positioning coordinate node exceeds a preset threshold, the color gradient transition effect of the thermal layer is triggered, and the time axis mark range of the permission abnormality event pop-up window is updated in conjunction.

[0162] In this step, when the average distance between the newly added device location coordinate node [5-B1-N-12] and the historical nodes exceeds 8 meters, the thermal layer initiates a color gradient transition from orange to dark red, indicating an abnormal movement pattern. The timeline controller also expands the display range of the permission abnormality event pop-up window, extending the time stamp from 2:30 PM to 2:45 PM to 2:50 PM, ensuring the complete presentation of spatiotemporal correlation data.

[0163] Step 20110: Perform a trajectory backtracking operation on the dynamic user identification mark in the visual graphic interface, extract all device positioning coordinate nodes corresponding to the same user identity, generate a trajectory playback animation arranged in reverse order by access timestamp, and synchronously display the permission status change history of the user identity in the permission verification report.

[0164] In this step, after selecting the user CN-ZT-210906 identifier, the system extracts the 214 device location coordinate nodes of the user throughout the day and generates a retrospective animation in reverse order of access timestamps. The animation speed controller supports 0.5-4x speed adjustment. The keyframe automatically inserts a pause mark at 14:31:00, the time when the permission expires, and the information panel on the right synchronizes to display the permission status change record at that moment. The trajectory path presents a color gradient effect during the retrospective process, transitioning from red (abnormal period) to green (compliant period), enhancing the visual recognition of illegal operations.

[0165] As can be seen, applying the above steps, through multi-level visual mapping and real-time interactive mechanisms, complex logistics management data is transformed into an intuitive decision-support interface. From basic data analysis to dynamic prediction and presentation, each processing step deeply integrates spatiotemporal characteristics and business rules, forming an adaptive and scalable visualization system. In particular, through the synergy of incremental updates and historical backtracking, the system achieves full-cycle visualization coverage of logistics supervision, significantly improving the efficiency of anomaly identification and response speed.

[0166] By applying the embodiment of the present application, the full-process intelligent generation and optimization of logistics management reports are realized by introducing cross-modal dynamic analysis. First, different from traditional single-dimensional log statistics, the embodiment of the present application innovatively jointly encodes multi-source heterogeneous data such as spatiotemporal trajectories, user behaviors and package attributes to form a dynamic feature vector sequence with contextual associations, which effectively captures the implicit operating rules between logistics nodes. Secondly, the multi-task parsing mechanism based on deep learning breaks through the static filling mode of traditional report templates, and makes the collaborative analysis of multi-dimensional management indicators possible by generating three types of related reports of logistics tracking, authority verification and anomaly detection in parallel. Then, the semantic integrity verification processing provided by the embodiment of the present application adopts a two-way attention mechanism to strengthen the causal reasoning of event links while eliminating logical contradictions, ensuring that the report content has business interpretability. Finally, through the dynamic binding of the authorization interface and real-time events, a closed-loop report update system is constructed, so that management decisions can adapt to the dynamic evolution of logistics scenarios.

[0167] Based on the same inventive concept, the present application also provides a report generation system. Figure 2 As shown, it is a structural diagram of a possible report generation system provided in an embodiment of the present application. Figure 2 In the embodiment, the report generation system 200 includes a processor 210 and a memory 220. The memory 220 stores a computer program executable by the processor 210. The processor 210 executes the instructions stored in the memory 220 to perform the steps of the above-mentioned method for generating a dynamic management report on logistics access based on an IoT lock.

[0168] Based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is run on a report generation system, the computer program is used to enable the report generation system to execute the steps of the above-mentioned method for generating a dynamic management report on logistics access based on an Internet of Things lock. In some possible implementations, the various aspects of the method for generating a dynamic management report on logistics access based on an Internet of Things lock provided by the present application can also be implemented in the form of a program product, which includes a computer program. When the program product is run on a report generation system, the computer program is used to enable the report generation system to execute the steps of the above-mentioned method for generating a dynamic management report on logistics access based on an Internet of Things lock. For example, the report generation system can execute the following steps: Figure 1 Follow the steps shown in .

Claims

1. A method for generating a dynamic management report on logistics access based on an Internet of Things lock, characterized in that: include: Obtaining an access event dataset for a target logistics area, the access event dataset comprising multiple event parameters associated with the smart lock, including lock operation type, user identification, access timestamp, device location coordinates, and package volume data; Performing cross-modal feature extraction processing on the access event data set to obtain a dynamic access feature set including spatiotemporal dimension features, wherein the cross-modal feature extraction processing includes jointly encoding the multiple event parameters to generate a feature vector sequence; Based on a preset report type template, the dynamic access feature set is input into a trained report generation model for multi-task parsing to generate an initial logistics management report set, which includes a logistics tracking report, an authority verification report, and an anomaly detection report. Performing semantic integrity verification and logical conflict detection on the initial logistics management report set to generate an optimized target logistics management report; The target logistics management report is bound to the authorized access interface of the user terminal, and an incremental update operation is performed on the target logistics management report according to the access event triggered in real time.

2. The method according to claim 1, wherein The cross-modal feature extraction process is performed on the access event data set to obtain a dynamic access feature set containing spatiotemporal dimension features, including: Performing data integrity verification on the access event data set to generate pre-processed access event data; wherein the data integrity verification includes at least two combinations of missing parameter completion, abnormal timestamp correction, positioning coordinate correction, and operation type standardization; Calling a spatiotemporal feature encoder to perform joint feature mapping on the preprocessed access event data to obtain a first-level spatiotemporal feature vector; wherein the spatiotemporal feature encoder uses a multi-head attention mechanism to model the spatiotemporal correlation between the access timestamp and the device location coordinates; Calling an operation mode encoder to perform discrete feature embedding processing on the lock operation type to generate a second-level operation feature vector; The first-level spatiotemporal feature vector and the second-level operational feature vector are subjected to feature fusion processing to generate the dynamic access feature set.

3. The method according to claim 2, wherein The calling of the spatiotemporal feature encoder to perform joint feature mapping processing on the pre-processed access event data to obtain a first-level spatiotemporal feature vector includes: Converting the access timestamp into a periodic time coding vector, and performing geographic grid coding on the device location coordinates to generate a spatial partition vector; Constructing a time-space cross-attention weight matrix, wherein the time-space cross-attention weight matrix is used to characterize the correlation strength of access events in different time periods in spatial distribution; Using a gated recurrent unit to perform serial modeling on the periodic time encoding vector and the spatial partition vector to generate initial spatiotemporal features with temporal dependencies; The initial spatiotemporal features are dynamically weighted and aggregated based on the time-space cross attention weight matrix to generate the first-level spatiotemporal feature vector.

4. The method according to claim 2, wherein The calling operation mode encoder performs discrete feature embedding processing on the lock operation type to generate a second-level operation feature vector, including: Establish a mapping relationship table between lock operation types and preset operation modes, where the preset operation modes include normal authorized opening, abnormal force opening, timeout failure, and multiple verification failures; Perform pattern cluster analysis on the historical event data corresponding to each lock operation type to generate operation pattern feature clusters; Mapping the lock operation type to the center point of the corresponding operation mode feature cluster to generate an operation mode embedding vector; A feature pyramid network is used to perform multi-scale feature extraction on the operation mode embedding vector to generate the second-level operation feature vector.

5. The method according to claim 1, wherein Based on the preset report type template, the dynamic access feature set is input into the trained report generation model for multi-task parsing processing to generate an initial logistics management report set, including: Performing task relevance analysis on the dynamic access feature set to determine a feature subset division method corresponding to the preset report type template; Each feature subset is input into the corresponding subtask processing module for parsing and processing to generate a preliminary report fragment; wherein the subtask processing module includes a logistics path restoration module, an authority matching verification module, and an abnormal pattern recognition module; Performing cross-task information fusion processing on the preliminary report fragments to eliminate semantic conflicts between different subtask processing modules and obtain fused report fragments; According to the structured format requirements of the preset report type template, the merged report fragments are logically sorted and key fields are filled to generate the initial logistics management report set.

6. The method according to claim 5, wherein Each feature subset is input into the corresponding subtask processing module for parsing and processing to generate a preliminary report fragment, including: generating a spatiotemporal trajectory graph based on the access timestamp and the device location coordinates in the dynamic access feature set, wherein the spatiotemporal trajectory graph comprises device location coordinate nodes connected in chronological order; Performing path restoration processing on the spatiotemporal trajectory graph, comparing the distance between adjacent positioning coordinate nodes with a preset transport speed threshold, identifying stop nodes and inserting stop period annotations, and generating a logistics transport path description text; Extract the user identity from the dynamic access feature set, match it with the regional authority field in the pre-stored authorization list, filter out abnormal user identities with invalid authority or out-of-bounds area, and generate a summary of the authority verification results; Input the lock operation type and access timestamp in the dynamic access feature set into a pre-trained anomaly detection model, extract the historical frequency distribution characteristics of the operation type in the corresponding time period, calculate the feature deviation between the current operation type and the historical frequency distribution, and generate an abnormal operation score; According to the comparison result of the abnormal operation score and the dynamic threshold, an alarm mark is triggered when the abnormal operation score exceeds the dynamic threshold, and an alarm trigger record is generated by associating with the corresponding device positioning coordinate node.

7. The method according to claim 6, wherein The method includes inputting the lock operation type and access timestamp in the dynamic access feature set into a pre-trained anomaly detection model, extracting the historical frequency distribution characteristics of the operation type in the corresponding time period, calculating the feature deviation between the current operation type and the historical frequency distribution, and generating an abnormal operation score, including: According to the time continuity of the access timestamps, the lock operation type and the access timestamp are divided into operation sequence units corresponding to the dynamic time window, the starting timestamp of the dynamic time window is the first access timestamp of the current operation sequence unit, and the ending timestamp of the dynamic time window is the last access timestamp of the current operation sequence unit plus a preset delay interval; Extracting a historical contemporaneous operation sequence that completely matches the time window of each operation sequence unit from the pre-trained anomaly detection model, the historical contemporaneous operation sequence comprising a lock operation type sequence chain generated by historical frequency distribution features, a set of historical time interval thresholds for adjacent operation types, and historical frequency distribution features of lock operation types; Compare the current lock operation type sequence chain in the operation sequence unit with the lock operation type sequence chain of the historical operation sequence of the same period step by step, count the number of lock operation types that match in a continuous order, and generate a sequence consistency parameter based on the historical frequency distribution characteristics; Performing interval coverage comparison on the current time intervals of adjacent lock operation types in the operation sequence unit and the historical time interval threshold set of the historical concurrent operation sequence, calculating the proportion of current time intervals that meet the historical time interval threshold, and generating a time interval matching parameter based on the historical frequency distribution characteristics; Extracting the current frequency distribution characteristics of each lock operation type in the operation sequence unit, performing item-by-item difference calculation with the historical frequency distribution characteristics of the lock operation type in the historical operation sequence of the same period, and generating a frequency characteristic deviation of each lock operation type; Extracting sequence feature weights, time interval feature weights, and frequency feature weights bound to the lock operation type from the pre-trained anomaly detection model, wherein the sequence feature weights and time interval feature weights are dynamically adjusted based on the association strength of the operation type in the historical frequency distribution features; The sequence consistency parameter and the sequence feature weight are weighted to generate the sequence deviation, the time interval matching parameter and the time interval feature weight are weighted to generate the time interval deviation, and the frequency feature deviation and the frequency feature weight are weighted to generate the frequency deviation; The sequence deviation, time interval deviation, and frequency deviation are superimposed in a preset ratio to generate an initial characteristic deviation, and the initial characteristic deviation is normalized according to the total number of lock operation types in the operation sequence unit to generate a standardized characteristic deviation; Extracting the maximum and minimum values of the standardized characteristic deviations of all historical concurrent operation sequences in the historical frequency distribution features, and constructing a dynamic threshold interval with the maximum value as the upper limit of the abnormal threshold and the minimum value as the lower limit of the abnormal threshold; Compare the standardized feature deviation of the current operation sequence unit with the dynamic threshold interval. When the standardized feature deviation exceeds the upper limit of the abnormal threshold or is lower than the lower limit of the abnormal threshold, mark it as an abnormal operation score, and extract the start access timestamp and end access timestamp corresponding to the dynamic time window as the trigger time range of the abnormal operation score; The trigger time range is fully matched with the device location coordinate nodes in the dynamic access feature set by timestamp, the device location coordinate nodes whose timestamps fall within the trigger time range are screened out, and an alarm trigger record bound to the abnormal operation score is generated.

8. The method according to claim 6, wherein The performing semantic integrity verification and logic conflict detection on the initial logistics management report set to generate an optimized target logistics management report includes: Extracting equipment location coordinate node sequences and dwell period annotations from the logistics transport route description text to generate a spatiotemporal event node set; Extracting abnormal user identifiers and their corresponding device location coordinate nodes from the authority verification result summary to generate an authority conflict node set; Performing spatial overlap analysis on the spatiotemporal event node set and the authority conflict node set to detect the coexistence of spatiotemporal events and authority conflicts under the same device location coordinate node; When it is detected that there is an authority conflict in the positioning coordinate node marked in the stay period, an authority review mark is inserted in the logistics transportation route description text, and the severity level of the alarm trigger record is updated; Traverse the time windows in all alarm trigger records, verify the path continuity of the corresponding time windows in the logistics transportation path description text, and repair the path breakage problem caused by missing positioning coordinate nodes; The updated logistics transport route description text, authority verification result summary and alarm trigger record are reorganized according to the paragraph structure of the preset report template to generate a semantically coherent target logistics management report.

9. The method according to claim 8, wherein The performing spatial overlap analysis on the spatiotemporal event node set and the permission conflict node set to detect the coexistence of spatiotemporal events and permission conflicts under the same device location coordinate node includes: Establishing a spatial grid encoding of device positioning coordinate nodes, and mapping each coordinate in the spatiotemporal event node set to a corresponding spatial grid; Searching for abnormal user identifiers having the same spatial grid code as the spatiotemporal event node in the permission conflict node set to generate a candidate conflict node list; Verify whether the access timestamp of each node in the candidate conflict node list falls within the stay period range of the corresponding spatiotemporal event node; When the access timestamp overlaps with the stay period mark, extract the historical permission change record of the abnormal user ID to determine whether the permission expiration event occurred after the stay period mark; If the permission expiration event time is later than the start time marked in the stay period, a retrospective permission exception description will be added to the permission verification result summary, and the path risk indicator in the logistics transportation path description text will be updated accordingly; The conflict detection range is adjusted according to the hierarchical accuracy of the spatial grid coding, a buffer area is established between adjacent spatial grids, cross-grid authority conflict events are detected, and a composite alarm record is generated that associates multiple device positioning coordinate nodes.

10. A report generation system, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor is enabled to perform the steps of any one of the methods of claims 1 to 9.

Images