Sorting center data management method and system based on machine learning

Abstract

The invention discloses a sorting center data management method and system based on machine learning, and the method comprises the steps: obtaining a personnel operation frequency, a personnel track, interaction time distribution, a stay area, an operation type, a time window feature, an abnormal motion identifier, a personnel identity tag, a cargo dynamic position, a cargo tag, order data and environment interaction data; environment sensing data is formed, personnel behavior characteristic analysis is carried out, and personnel behavior characteristics are constructed; abnormal behavior recognition is performed in combination with the personnel trajectory and the interaction time distribution; and according to the cargo label, the order data and the abnormal behavior result, calculating a cargo emergency degree score and analyzing cargo abnormity. The verification process is simplified based on the emergency degree. And adjusting the access authority in combination with the verification scheme and the abnormal result. And task allocation and sorting paths are optimized according to the security policy, the abnormal result and the personnel trajectory, and data management optimization configuration of the sorting center based on the cargo position and the personnel label is realized. The method improves the safety and efficiency of the sorting center.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology, and in particular to a sorting center data management method and system based on machine learning. Background Technology

[0002] Currently, logistics sorting centers are the core hubs for the efficient operation of modern supply chains, facing the severe challenge of balancing operational efficiency and data security in a dynamic environment. Their operation involves real-time changes in personnel behavior, cargo priorities (such as urgent orders), and access scenarios (such as night shifts). However, existing systems struggle to perceive and respond to these changes in a timely manner: high-priority goods may be delayed due to rigid security processes (such as fixed password strength); lax controls in low-risk scenarios may lead to security vulnerabilities. Therefore, there is an urgent need for a technology that can analyze multi-dimensional dynamic information such as personnel behavior patterns, access context, and cargo priorities in real time. Introducing machine learning, through continuous learning and pattern recognition of this information, can intelligently drive the adaptive adjustment of multi-layered security control mechanisms, such as password verification strength. This will be the core technological support for ensuring both efficient operation and secure management of sorting centers in complex and dynamic environments.

[0003] In one existing technology, operators must pass through a unified authentication interface before performing any critical operations (such as accessing sensitive data or initiating sorting processes) in a logistics sorting center system. This process requires operators to enter a pre-defined username and password with fixed complexity requirements (e.g., a mandatory fixed-length combination of numbers, letters, and special characters). The system compares the entered password with the hash value of that user's password stored in the database. Regardless of the current operational scenario (e.g., handling urgent goods, routine sorting tasks, low-attendance nighttime hours, or when the system detects an abnormal login attempt), the strength requirements and steps of this password authentication remain exactly the same and cannot be adjusted. The authentication result is a binary decision: a successful match grants the operator the full set of pre-defined fixed permissions for their account; a failed match denies access. This existing authentication method is rigid and uniform, leading to delays for urgent goods due to the fixed steps, and weakened protection during low-risk nighttime hours, making it vulnerable to breaches.

[0004] Therefore, existing technologies suffer from low safety and efficiency issues in sorting centers. Summary of the Invention

[0005] This invention provides a sorting center data management method and system based on machine learning to improve the safety and efficiency of sorting centers.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a sorting center data management method based on machine learning, comprising:

[0007] The environmental perception data consists of personnel operation frequency, personnel trajectory, interaction time distribution, dwell area, operation type, time window characteristics, abnormal action identifier, personnel identity tag, dynamic location of goods, goods tag, order data and environmental interaction data.

[0008] Based on the environmental perception data, perform personnel behavior characteristic analysis to construct personnel behavior characteristics based on the personnel identity tags;

[0009] Based on the personnel behavior characteristics, combined with the personnel trajectory and the interaction time distribution, abnormal behavior is identified to obtain abnormal behavior detection results;

[0010] Based on the cargo label, the order data, and the abnormal behavior detection results, a cargo urgency score is calculated and a dynamic anomaly analysis of the cargo is performed to obtain the cargo handling urgency level.

[0011] Based on the urgency of the cargo handling, the password verification process is simplified and a fast verification process is adopted, and a verification strength adjustment scheme is output.

[0012] Based on the verification strength adjustment scheme and the abnormal behavior detection results, adjust the personnel access permission configuration and output the security control policy;

[0013] Based on the security control strategy, the abnormal behavior detection results, and the personnel trajectory, optimize the task allocation rules and cargo sorting paths, and output the task allocation scheme.

[0014] Based on the task allocation scheme, the sorting paths based on the dynamic location of the goods and the task scheduling based on personnel identification tags within the sorting center are adjusted to obtain the optimized configuration of the sorting center's data management.

[0015] In one optional implementation, the step of performing personnel behavior feature analysis based on the environmental perception data and constructing personnel behavior features based on the personnel identity tags includes:

[0016] Based on dynamic environment perception data, spatiotemporal slicing is performed using the sliding window method to obtain initial behavioral data aggregated by time window;

[0017] Based on the initial behavioral data, trajectory segmentation and region clustering analysis are performed using the DBSCAN clustering algorithm to obtain behavioral grouping data;

[0018] Based on the behavioral grouping data, abnormal behavior is judged using the C4.5 decision tree algorithm, and the behavioral characteristics of the personnel based on the personnel identity tags are output.

[0019] In one optional implementation, the step of identifying abnormal behavior based on the personnel behavior characteristics, combined with the personnel trajectory and the interaction time distribution, to obtain the abnormal behavior detection result includes:

[0020] When the trajectory of a person deviates from the preset normal trajectory threshold, the dwell time in the trajectory is extracted, and abnormal dwell is judged by combining the interaction time distribution analysis and the preset dwell threshold, and an abnormal action identifier is generated.

[0021] Based on the abnormal action identifier and the personnel trajectory, the behavior pattern features are classified using the K-means clustering algorithm to obtain the abnormal behavior detection results.

[0022] In one optional implementation, the step of calculating a cargo urgency score and performing dynamic anomaly analysis based on the cargo label, the order data, and the abnormal behavior detection results to obtain the cargo handling urgency includes:

[0023] Based on the cargo tags, cargo priority is extracted, and based on the order data, delivery time limit and order quantity are extracted. Then, abnormal records of the cargo are extracted through SQL query to obtain cargo attributes including cargo priority, delivery time limit, order quantity and abnormal records.

[0024] When the delivery time limit is lower than the preset time limit threshold, the urgency score of the goods is obtained by weighting the goods priority, the delivery time limit, the preset goods priority weight, and the preset time limit weight.

[0025] Based on the dynamic location of the goods and the abnormal records, the spatial distribution of location points is analyzed by K-means clustering algorithm to identify abnormal location clusters that deviate from the preset standard logistics path and obtain the abnormal goods identification.

[0026] Based on the abnormal behavior detection results and the abnormal cargo identification, the weighted Euclidean distance algorithm is used to calculate the behavior risk coefficient.

[0027] Based on the cargo anomaly identifier and the behavioral risk coefficient, the cargo urgency score is adjusted in a graded manner using the CART decision tree algorithm to generate a cargo handling urgency level that includes cargo priority, delivery time limit, order quantity, anomaly records, and cargo urgency score.

[0028] In one optional implementation, the step of simplifying the password verification process and adopting a fast verification process based on the urgency of cargo handling, and outputting a verification strength adjustment scheme, includes:

[0029] When the priority of the goods is higher than the preset emergency level threshold or the delivery time limit is lower than the time limit threshold, the verification complexity adjustment coefficient is calculated using the CART decision tree algorithm based on the goods label and the order quantity.

[0030] Based on the verification complexity adjustment coefficient and the personnel identity label, the matching identity verification rules are extracted from the preset identity database, and then the verification strength scheme is generated by using a rule engine based on the Rete algorithm and combined with the preset fast verification process to obtain the initial verification strength scheme.

[0031] Based on the initial verification strength scheme, the corresponding verification parameter configuration is obtained from the preset verification rule base, and the Q-learning algorithm is used to optimize the password verification complexity to obtain the verification strength adjustment scheme.

[0032] In one optional implementation, adjusting the access permission configuration of personnel and outputting a security control policy based on the verification strength adjustment scheme and the abnormal behavior detection results includes:

[0033] Based on the personnel identity tags and the abnormal behavior detection results, the initial values ​​of verification frequency and password complexity are set to obtain a preliminary configuration scheme for user access permissions;

[0034] Based on the preliminary configuration scheme, a weighted permission risk coefficient is calculated. When the permission risk coefficient exceeds a preset permission threshold, the access permission level is reduced to obtain permission configuration parameters.

[0035] Based on the permission configuration parameters and the abnormal behavior detection results, the K-nearest neighbor algorithm is used to match abnormal behavior patterns, determine the scope of permission restrictions, and generate an access permission configuration table.

[0036] Based on the access permission configuration table, the permission database is updated, and the permission mapping relationship is reconstructed in conjunction with the role-based access control method, and a security control policy is output.

[0037] In one optional implementation, the step of optimizing task allocation rules and cargo sorting paths based on the security control strategy, the abnormal behavior detection results, and the personnel trajectory, and outputting a task allocation scheme, includes:

[0038] Based on the access permission configuration table and the personnel trajectory, the regional risk level is classified using the C5.0 decision tree algorithm to obtain the regional risk level;

[0039] Based on the regional risk level and the personnel trajectory, a trajectory anomaly coefficient is calculated by weighting. When the trajectory anomaly coefficient exceeds a preset trajectory anomaly threshold, it is marked as an abnormal trajectory, and an abnormal trajectory classification result is obtained.

[0040] Based on the abnormal trajectory classification results, the cargo priority and sorting path data are extracted from the pre-stored sorting task database, and the cargo priority is adjusted in conjunction with the Drools rule engine to obtain the initial task allocation scheme.

[0041] Based on the initial task allocation scheme and the sorting path data, the shortest path is planned using the Dijkstra algorithm, and the permission database records are updated to obtain a task allocation scheme that includes the shortest path planning.

[0042] Secondly, the present invention provides a sorting center data management system based on machine learning, comprising:

[0043] The data acquisition module is used to acquire personnel operation frequency, personnel trajectory, interaction time distribution, dwell area, operation type, time window characteristics, abnormal action identifier, personnel identity tag, dynamic location of goods, goods tag, order data and environmental interaction data, which constitute environmental perception data;

[0044] The personnel behavior analysis module is used to perform personnel behavior feature analysis based on the environmental perception data and construct personnel behavior features based on the personnel identity tags;

[0045] An abnormal behavior detection module is used to identify abnormal behavior based on the personnel behavior characteristics, combined with the personnel trajectory and the interaction time distribution, and to obtain abnormal behavior detection results.

[0046] The cargo urgency analysis module is used to calculate the cargo urgency score and perform dynamic anomaly analysis of the cargo based on the cargo label, the order data, and the abnormal behavior detection results, so as to obtain the cargo handling urgency level.

[0047] The verification strength adjustment module is used to simplify the password verification process and adopt a fast verification process based on the urgency of the cargo handling, and output a verification strength adjustment scheme.

[0048] The security control strategy module is used to adjust the access permission configuration of personnel based on the verification strength adjustment scheme and the abnormal behavior detection results, and output the security control strategy.

[0049] The task allocation analysis module is used to optimize task allocation rules and cargo sorting paths based on the security control strategy, the abnormal behavior detection results, and the personnel trajectory, and output a task allocation scheme.

[0050] The output configuration module is used to adjust the sorting path based on the dynamic location of the goods and the task scheduling based on personnel identification tags within the sorting center based on the task allocation scheme, so as to obtain the data management optimization configuration of the sorting center.

[0051] Thirdly, the present invention also provides an electronic device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the machine learning-based sorting center data management method described in any one of the preceding claims.

[0052] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the sorting center data management method based on machine learning described in any one of the above.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] (1) By integrating cargo labels, order data and abnormal behavior results to calculate cargo urgency scores, the verification process is dynamically triggered (such as matching quick solutions based on rule engines), thereby shortening the processing time of urgent cargo and improving sorting efficiency while ensuring safety.

[0055] (2) By using DBSCAN cluster analysis to segment personnel trajectories and combining interaction time and stay area characteristics, abnormal trajectories and stays are identified through K-means clustering. Compared with static monitoring, this reduces the false alarm rate and improves the accuracy and initiative of abnormal behavior detection.

[0056] (3) Based on the results of abnormal behavior and the urgency of goods, the K-nearest neighbor algorithm is used to match risk patterns and calculate the permission risk coefficient, dynamically adjust the access permission level, achieve accurate matching between permissions and real-time risks, and effectively plug security vulnerabilities.

[0057] (4) Based on the location, priority and regional risk level of goods, the sorting path is optimized in real time by Dijkstra algorithm, and the task is scheduled in combination with personnel tags to shorten the movement distance of high priority goods and avoid risk areas, thereby improving the overall throughput of the sorting center. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the sorting center data management method based on machine learning provided in the first embodiment of the present invention;

[0059] Figure 2 This is a schematic diagram of the data management system structure for a sorting center based on machine learning provided in the second embodiment of the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Reference Figure 1 The first embodiment of the present invention provides a sorting center data management method based on machine learning, including the following steps:

[0062] S11: Acquire personnel operation frequency, personnel trajectory, interaction time distribution, dwell area, operation type, time window characteristics, abnormal action identifier, personnel identity label, dynamic location of goods, goods label, order data and environmental interaction data to form environmental perception data;

[0063] S12, Based on the environmental perception data, perform personnel behavior characteristic analysis and construct personnel behavior characteristics based on the personnel identity tags;

[0064] S13, Based on the personnel behavior characteristics, combined with the personnel trajectory and the interaction time distribution, abnormal behavior identification is performed to obtain abnormal behavior detection results;

[0065] S14. Based on the cargo label, the order data, and the abnormal behavior detection results, calculate the cargo urgency score and perform dynamic abnormal analysis of the cargo to obtain the cargo handling urgency level.

[0066] S15, Based on the urgency of the cargo handling, simplify the password verification process and adopt a fast verification process, and output a verification strength adjustment scheme.

[0067] S16, Based on the verification strength adjustment scheme and the abnormal behavior detection results, adjust the personnel access permission configuration and output the security control strategy;

[0068] S17. Based on the security control strategy, the abnormal behavior detection results and the personnel trajectory, optimize the task allocation rules and the goods sorting path, and output the task allocation scheme.

[0069] S18, based on the task allocation scheme, adjust the sorting path based on the dynamic location of the goods and the task scheduling based on the personnel identity tags within the sorting center to obtain the optimized configuration of the sorting center's data management.

[0070] In step S11, the following data are acquired: personnel operation frequency, personnel trajectory, interaction time distribution, dwell area, operation type, time window characteristics, abnormal action identifier, personnel identity tag, dynamic location of goods, goods tag, order data, and environmental interaction data, which constitute environmental perception data.

[0071] Specifically, the sorting center utilizes an IoT sensor network to collect real-time data on personnel operation frequency (number of operations per unit time counted by RFID scanning devices at workstations), personnel trajectory (coordinate sequences recorded at 0.5-second intervals based on a UWB positioning system), interaction time distribution (ratio of effective operation time to idle time recorded by a timer at the workstation), dwell area (clustering area identifiers generated by clustering trajectory coordinate points using the DBSCAN clustering algorithm), operation type (operation codes in the sorting system logs, such as goods scanning / handling / re-inspection), time window characteristics (time period numbers divided into 1-hour cycles), abnormal action identifiers (unconventional posture codes detected by infrared motion capture devices), personnel identification tags (job number and permission level bound to RFID tags), goods dynamic location (coordinates transmitted in real-time by AGV-mounted GPS), goods tags (priority codes and category codes in electronic waybills), and order data (delivery deadline timestamps and order quantity extracted from the database). Simultaneously, environmental interaction data such as ambient temperature and humidity, and equipment vibration frequency are integrated. After being timestamped, the aforementioned data is integrated into structured environment-aware data by the Kafka data streaming platform. Its sampling frequency is dynamically adjusted by the sorting center's main control server based on real-time system load (the load threshold is set to 70% CPU utilization based on historical peak stress tests). This step, through high-precision, multi-source heterogeneous data acquisition, provides a complete spatiotemporal context for subsequent behavior modeling, addressing the perception blind spots caused by the single data dimension of traditional systems.

[0072] In step S12, based on the environmental perception data, personnel behavior characteristics are analyzed to construct personnel behavior characteristics based on the personnel identity tags.

[0073] In one specific implementation, the step of analyzing personnel behavior characteristics based on the environmental perception data and constructing personnel behavior characteristics based on the personnel identity tags includes:

[0074] Based on dynamic environment perception data, spatiotemporal slicing is performed using the sliding window method to obtain initial behavioral data aggregated by time window.

[0075] Based on the initial behavioral data, trajectory segmentation and region clustering analysis are performed using the DBSCAN clustering algorithm to obtain behavioral grouping data;

[0076] Based on the behavioral grouping data, abnormal behavior is judged using the C4.5 decision tree algorithm, and the behavioral characteristics of the personnel based on the personnel identity tags are output.

[0077] Specifically, firstly, based on the characteristics of the time window (e.g., time window characteristic = 12 represents the period from 12:00 to 13:00), a sliding window method with a fixed step size of 300 seconds is used to spatiotemporally slice the dynamic environment perception data. This operation divides the raw data, such as continuous personnel trajectory coordinate sequences, operation type records, and stay area identifiers, into discrete data blocks according to the time window and performs aggregation calculations: calculating the moving average of personnel operation frequency within the window, counting the occurrence frequency of each stay area, summarizing the distribution ratio of operation types, and generating initial behavioral data containing window number, average operation frequency, area frequency distribution table, and operation type ratio vector.

[0078] Next, based on the personnel trajectory coordinate sequence in the initial behavior data, trajectory segmentation is performed using the DBSCAN clustering algorithm: a neighborhood radius of 1.5 meters is set (determined based on the 85th percentile of the channel width statistically analyzed from historical trajectory data of the sorting center), and a minimum number of neighborhood points of 5 (optimized through experiments on the spatial distribution density of trajectory points). Density clustering is then performed on the trajectory points within the window. The algorithm identifies core points (coordinates of at least 5 trajectory points within a 1.5-meter radius), merges density-connected core points to form trajectory segment clusters, and outputs behavior grouping data including trajectory segment number, coordinate point set, associated dwell area label (bound dwell area field), and corresponding dwell area time (the number of seconds the segment stayed in the area).

[0079] Finally, the behavior grouping data is input into the C4.5 decision tree algorithm: the trajectory segmentation type (normal movement / stagnation), dwell time (numerical), and operation type distribution (vector) are used as feature attributes, and the abnormal action identifier (0 / 1) is used as the target variable. The algorithm calculates the information gain ratio of each feature (such as the information gain ratio of dwell time), selects the optimal split point (such as the split condition of dwell time > 60 seconds, the threshold is set based on the sorting center operation specifications), and recursively builds the decision tree until the purity of the leaf nodes reaches the preset threshold of 95% (algorithm convergence condition), generating a set of behavior judgment rules (such as "dwell time > 60 seconds and the proportion of handling operations in the operation type distribution < 10%" → abnormal stagnation). The final output is a personnel behavior feature table with personnel identity label as the primary key, associated with behavior feature labels and parameters such as dwell time.

[0080] This step maintains the temporal continuity of behavior through a sliding window, overcomes the drift error of positioning devices through DBSCAN density clustering, and constructs interpretable rules through C4.5 decision trees, forming a refined modeling capability for human behavior in sorting scenarios.

[0081] In step S13, based on the personnel behavior characteristics, combined with the personnel trajectory and the interaction time distribution, abnormal behavior is identified to obtain abnormal behavior detection results.

[0082] In one specific implementation, the step of identifying abnormal behavior based on the personnel behavior characteristics, combined with the personnel trajectory and the interaction time distribution, to obtain abnormal behavior detection results includes:

[0083] When the trajectory of a person deviates from the preset normal trajectory threshold, the dwell time in the trajectory is extracted, and abnormal dwell is judged by combining the interaction time distribution analysis and the preset dwell threshold, and an abnormal action identifier is generated.

[0084] Based on the abnormal action identifier and the personnel trajectory, the behavior pattern features are classified using the K-means clustering algorithm to obtain the abnormal behavior detection results.

[0085] Specifically, firstly, based on the trajectory segment type field in the personnel behavior characteristics generated in step S12, it is determined whether the personnel trajectory deviates from the preset normal trajectory threshold. This threshold includes both spatial and temporal constraints: the spatial threshold is set when the offset distance between the trajectory coordinate point and the preset standard path (generated by Gaussian distribution modeling based on historical trajectory data) exceeds 2.5 meters (the threshold is determined based on the positioning accuracy of the sorting center equipment); the temporal threshold is set when the trajectory segment duration exceeds three times the standard deviation of the historical operation duration of the same position (determined through historical operation log statistics). When a deviation in the personnel trajectory is detected, the dwell time associated with the trajectory segment (i.e., the dwell time field in the behavior grouping data) is extracted and analyzed in conjunction with the interaction time distribution data (including the interaction operation duration and idle time sequence recorded by the operation station): if the dwell time exceeds the preset dwell threshold for that area (e.g., the threshold for the packaging area = 90 seconds, set according to the job operation specifications), an abnormal action identifier is generated (value = 1, type = "timeout"). Subsequently, the abnormal action identifiers and personnel trajectory coordinate sequences are merged into a behavior feature vector (containing coordinate point sequences and abnormal identifier values). Behavior pattern classification is then performed using the K-means clustering algorithm: using spatial coordinates of trajectory points, dwell time in the affected area, and abnormal identifier values ​​as three-dimensional features, with the number of clusters set to K=4 (determined according to the elbow rule). After randomly initializing cluster centers, the Euclidean distance between sample points and the center points is iteratively calculated. Samples are reassigned to the nearest cluster, and the cluster centers are updated until the cluster center movement distance is less than 0.01 meters (convergence threshold). Finally, cluster labels (0-3) are output and mapped to four behavior patterns: 0-normal movement, 1-abnormal lingering, 2-path deviation, and 3-high-frequency abnormal operation, forming an abnormal behavior detection result table containing personnel identity labels, time window numbers, and behavior pattern labels. This step reduces the false alarm rate through a spatial-temporal dual-threshold triggering mechanism and achieves refined classification of abnormal patterns through multi-dimensional feature clustering, providing accurate behavioral risk information for emergency cargo handling.

[0086] In step S14, based on the cargo label, the order data, and the abnormal behavior detection results, a cargo urgency score is calculated and a dynamic anomaly analysis of the cargo is performed to obtain the cargo handling urgency level.

[0087] In one specific implementation, the step of calculating a cargo urgency score and performing dynamic anomaly analysis based on the cargo label, the order data, and the abnormal behavior detection results to obtain the cargo handling urgency includes:

[0088] Based on the cargo tags, cargo priority is extracted, and based on the order data, delivery time limit and order quantity are extracted. Then, abnormal records of the cargo are extracted through SQL query to obtain cargo attributes including cargo priority, delivery time limit, order quantity and abnormal records.

[0089] When the delivery time limit is lower than the preset time limit threshold, the urgency score of the goods is obtained by weighting the goods priority, the delivery time limit, the preset goods priority weight, and the preset time limit weight.

[0090] Based on the dynamic location of the goods and the abnormal records, the spatial distribution of location points is analyzed by K-means clustering algorithm to identify abnormal location clusters that deviate from the preset standard logistics path and obtain the abnormal goods identification.

[0091] Based on the abnormal behavior detection results and the abnormal cargo identification, the weighted Euclidean distance algorithm is used to calculate the behavior risk coefficient.

[0092] Based on the cargo anomaly identifier and the behavioral risk coefficient, the cargo urgency score is adjusted in a graded manner using the CART decision tree algorithm to generate a cargo handling urgency level that includes cargo priority, delivery time limit, order quantity, anomaly records, and cargo urgency score.

[0093] Specifically, the process begins by extracting the cargo priority code (e.g., priority = 3 represents high priority) from the cargo label. Delivery deadlines (e.g., delivery deadline = 20230809120000 indicates delivery before 12:00 noon on the same day) and order quantity (e.g., quantity = 50 pieces) are then parsed from the order data. Finally, a structured query is used to retrieve associated anomaly records from the sorting center's historical database (the query conditions are cargo identifier matching and anomaly identifier being true, yielding fields such as anomaly count = 2). This data is then integrated to generate a cargo attribute table containing four core fields: cargo priority, delivery deadline, order quantity, and anomaly records. When the difference between the delivery deadline and the current system time is lower than the preset time limit threshold (e.g., 120 minutes, which is determined based on the 90th percentile of the order fulfillment time data of the sorting center in the past three months), a weighted calculation is used to generate a cargo urgency score: cargo priority weight (weight coefficient of 0.6 obtained by regression analysis of the correlation between historical order processing time and priority) multiplied by the priority value (e.g., 3), plus delivery deadline weight (calculated based on the percentage of remaining time to standard processing time) multiplied by the timeliness coefficient, and finally outputting the score value.

[0094] Synchronous dynamic anomaly analysis of goods is performed: Based on the dynamic location coordinate sequence of goods, a spatial clustering algorithm (the number of clusters is determined by optimization based on the silhouette coefficient) is used to analyze the distribution of location points. The algorithm calculates the distance between each location point and the preset standard logistics path (generated from the coordinates of historical compliant paths using a Gaussian mixture model). Location points with a deviation distance exceeding 2.5 meters (the threshold is determined based on the positioning accuracy of the sorting center equipment) are marked as anomalies, and anomaly location cluster labels are output as goods anomaly identifiers. Subsequently, combining the behavior pattern labels (e.g., pattern number = 2 for high-risk behavior) and cargo anomaly identifiers from the abnormal behavior detection results, a weighted distance algorithm is used to calculate the behavior risk coefficient: The risk level value of the abnormal behavior detection results (specifically, the risk level value of the abnormal behavior detection results is quantified and generated using the Analytic Hierarchy Process (AHP): First, the logistics security expert group assigns pairwise importance scores to the four behavior patterns (0-normal movement, 1-abnormal stagnation, 2-path deviation, 3-high-frequency abnormal operation) based on three dimensions: operational risk level, potential hazard level, and frequency of occurrence (e.g., "high-frequency abnormal operation" can be rated as extremely important (9 points) relative to "normal movement"), constructing a judgment matrix; after calculation using the eigenvector method and satisfying the consistency test (CR < 0.1), the weight allocation is obtained: normal movement 5%, abnormal stagnation 25%, path deviation 35%, high-frequency abnormal operation 35%) as the weight, multiplied by the discrete value (0 or 1) of the cargo anomaly identifier, and then superimposed with the normalized value of the cargo position deviation distance (maximum deviation distance of 10 meters corresponds to a coefficient of 1.0), finally outputting the behavior risk coefficient.

[0095] Finally, a classification regression tree algorithm is used to dynamically adjust the urgency score of the goods: using the goods anomaly identifier, behavioral risk coefficient, and original urgency score as input features, and the urgency adjustment level in historical order processing records as the target variable (the labeling rules are based on customer complaint data and fulfillment failure records). The algorithm selects the optimal split point based on the Gini index (e.g., the split threshold for a behavioral risk coefficient > 0.6 is determined by 10x cross-validation), constructs a binary tree structure of depth 5 (depth optimized by pre-pruning), and finally outputs the graded and adjusted urgency data of the goods, including the original goods attribute fields, the adjusted urgency score, and the adjustment reason classification code (the coding system refers to the national standard for logistics anomaly classification).

[0096] This process achieves dynamic perception and precise priority calibration of cargo emergency status by integrating multi-dimensional analysis of logistics route anomalies and personnel behavior risks.

[0097] In step S15, based on the urgency of the cargo handling, the password verification process is simplified and a fast verification process is adopted, and a verification strength adjustment scheme is output.

[0098] In one specific implementation, the step of simplifying the password verification process and adopting a fast verification process based on the urgency of cargo handling, and outputting a verification strength adjustment scheme, includes:

[0099] When the priority of the goods is higher than the preset emergency level threshold or the delivery time limit is lower than the time limit threshold, the verification complexity adjustment coefficient is calculated using the CART decision tree algorithm based on the goods label and the order quantity.

[0100] Based on the verification complexity adjustment coefficient and the personnel identity label, the matching identity verification rules are extracted from the preset identity database, and then the verification strength scheme is generated by using a rule engine based on the Rete algorithm and combined with the preset fast verification process to obtain the initial verification strength scheme.

[0101] Based on the initial verification strength scheme, the corresponding verification parameter configuration is obtained from the preset verification rule base, and the Q-learning algorithm is used to optimize the password verification complexity to obtain the verification strength adjustment scheme.

[0102] Specifically, in step S15, the specific generation process of the verification intensity adjustment scheme is as follows: First, when the value of the cargo priority field in the cargo handling urgency level is higher than the preset urgency level threshold (e.g., priority ≥ 4, which is determined according to the historical order fulfillment priority classification standard), or the difference between the delivery time limit field and the current time is lower than the preset time limit threshold (e.g., remaining time < 60 minutes, which is derived from the sorting center standard operation time model), the verification process simplification mechanism is triggered. At this time, based on the cargo category code (e.g., fragile item code = FL01) and order quantity field in the cargo label, the verification complexity adjustment coefficient is calculated using the classification regression tree algorithm: taking the cargo category code (discrete variable) and order quantity (continuous variable) as input features, and the verification duration in the historical verification log as the target variable, the split point is selected through the Gini index minimization principle (e.g., splitting when the order quantity > 100), and the decision tree model outputs the adjustment coefficient (e.g., 0.3 indicates a 30% reduction in intensity).

[0103] Subsequently, based on the adjustment coefficient and the permission level field in the personnel identity label (e.g., identity level = 3), matching authentication rules (e.g., rule ID = R103 corresponding to level 3) are extracted from the RBAC (Role-Based Access Control) permission table in the preset identity database. Matching is then performed using a rule engine based on the Rete algorithm: the adjustment coefficient, identity rules, and cargo urgency score are treated as fact objects and matched against a preset rapid verification process rule base to generate an initial verification strength scheme that includes a simplified sequence of verification steps (e.g., skipping secondary verification) and password length requirements. This rule base is built upon in-depth extraction of multi-dimensional historical operational data from the sorting center: firstly, it integrates cargo emergency event handling records from the past three years, job permission change logs, and a verification failure case library; then, it extracts high-frequency scenario combination patterns using association rule mining technology (Apriori algorithm) (e.g., the success rate of simplified verification for the combination of "high-urgency cargo + medium-level identity" reaches 97.3%); forming a rule system with weighted parameters (e.g., rule FP02 has been verified through 2000 sandbox environment stress tests, ensuring a false positive rate of less than 0.1% under high-concurrency scenarios). The rule base includes intelligent strategies such as "triggering the fast process FP02 when the adjustment coefficient is <0.5 and the identity level is ≥2". After matching, an initial verification strength scheme is generated that includes a simplified sequence of verification steps (such as skipping secondary verification) and password length requirements.

[0104] Finally, based on the initial scheme, the corresponding verification parameter configurations (such as the minimum password length of 4 bits for FP02) are extracted from the verification rule base, and the Q-learning algorithm is used to optimize the password verification complexity: a reinforcement learning model is constructed with verification strength level (state S), password complexity parameter (action A), and verification failure rate / efficiency ratio (reward R). The action space is explored through an ε-greedy strategy (ε = 0.1, determined through the convergence test of the reinforcement learning model), and the Q-value table is iteratively updated (learning rate α = 0.01, discount factor γ = 0.9). The final output is a verification strength adjustment scheme that maximizes the reward function (i.e., balances security and efficiency), including a parameter set for password length threshold, biometric activation flag, and verification timeout.

[0105] This step uses a multi-level decision-making mechanism to enable adaptive degradation of the verification process in emergency scenarios, thereby shortening the processing time for high-priority goods while ensuring a safety baseline.

[0106] In step S16, based on the verification strength adjustment scheme and the abnormal behavior detection results, the access permission configuration of personnel is adjusted, and a security control strategy is output.

[0107] In one specific implementation, adjusting the access permission configuration of personnel and outputting a security control policy based on the verification strength adjustment scheme and the abnormal behavior detection results includes:

[0108] Based on the personnel identity tags and the abnormal behavior detection results, the initial values ​​of verification frequency and password complexity are set to obtain a preliminary configuration scheme for user access permissions;

[0109] Based on the preliminary configuration scheme, a weighted permission risk coefficient is calculated. When the permission risk coefficient exceeds a preset permission threshold, the access permission level is reduced to obtain permission configuration parameters.

[0110] Based on the permission configuration parameters and the abnormal behavior detection results, the K-nearest neighbor algorithm is used to match abnormal behavior patterns, determine the scope of permission restrictions, and generate an access permission configuration table.

[0111] Based on the access permission configuration table, the permission database is updated, and the permission mapping relationship is reconstructed in conjunction with the role-based access control method, and a security control policy is output.

[0112] Specifically, firstly, based on the job permission level field in the personnel identity label (e.g., job permission level = 3) and the behavior pattern classification label in the abnormal behavior detection results (e.g., behavior pattern label = 2 represents path deviation), a predefined permission rule mapping table is queried (this table structure contains four core fields: permission level range, behavior pattern number, verification frequency upper limit, and password minimum length value, where the permission level range field stores an interval expression such as "[3, 5]"). By parsing the field matching logic (e.g., when permission level = 3 and behavior pattern number = 2), the corresponding verification frequency upper limit value (e.g., 2 times per hour) and password minimum length value (e.g., 10 characters) are extracted to generate a preliminary configuration scheme containing specific parameter values. Subsequently, based on this scheme, combined with the verification complexity adjustment coefficient in the verification strength adjustment scheme (e.g., coefficient = 0.3) and the behavior risk value in the abnormal behavior detection results (e.g., risk value = 0.8), the permission risk coefficient is calculated using a weighted calculation formula (verification complexity adjustment coefficient × 0.4 + behavior risk value × 0.6), with the weight ratio determined based on the multiple regression analysis of the sorting center's security incident database over the past three years. When the calculation result exceeds the preset permission risk threshold of 0.5 (this threshold is set according to the 95th percentile of the distribution of risk events in the security audit log), the original position permission level will be downgraded by 1 level (e.g., from level 3 to level 2), and permission configuration parameters containing the new permission level and the adjusted verification frequency (e.g., reduced to once per hour) will be generated.

[0113] Next, the permission configuration parameters and behavioral feature vectors (including permission level values, behavioral pattern label numbers, and adjusted verification frequencies) are input into the K-nearest neighbor algorithm to perform permission restriction determination: The algorithm accesses the historical behavioral feature sample database (stores 100,000 records, each containing four fields: permission level value, behavioral pattern number, trajectory anomaly coefficient, and final permission restriction code), and uses the weighted Euclidean distance formula to calculate the distance between the current feature vector and the sample (permission level difference weight 0.6, behavioral pattern difference weight 0.4, weights determined through feature importance analysis), and retrieves K=5 nearest neighbor samples (K value is determined based on silhouette coefficient optimization). If more than 3 of the samples have permission restriction codes that are specific restriction identifiers (e.g., code = 103 represents strict restriction), a restricted operation list is output (e.g., prohibiting goods redistribution, disabling system configuration permissions), generating an access permission configuration table containing permission level values, restricted operation code sets, and effective time ranges.

[0114] Finally, perform a permission database update operation: generate a structured query statement based on the access permission configuration table, modify the role-permission mapping table in the role-based access control model (e.g., downgrade the permission mask of the role "Sorter Level 3" from 0xFF to 0x7F). Synchronously reconstruct the role-user association view (e.g., remap employee 10086 from role 3 to role 2), and output a security control policy containing dynamic policy numbers, change timestamps, and detailed permission change records.

[0115] This process uses a historical behavior pattern matching mechanism to achieve a precise correlation between permission adjustments and risk patterns.

[0116] In step S17, based on the security control strategy, the abnormal behavior detection results, and the personnel trajectory, the task allocation rules and cargo sorting paths are optimized, and a task allocation scheme is output.

[0117] In one specific implementation, the step of optimizing task allocation rules and cargo sorting paths based on the security control strategy, the abnormal behavior detection results, and the personnel trajectory, and outputting a task allocation scheme, includes:

[0118] Based on the access permission configuration table and the personnel trajectory, the regional risk level is classified using the C5.0 decision tree algorithm to obtain the regional risk level;

[0119] Based on the regional risk level and the personnel trajectory, a trajectory anomaly coefficient is calculated by weighting. When the trajectory anomaly coefficient exceeds a preset trajectory anomaly threshold, it is marked as an abnormal trajectory, and an abnormal trajectory classification result is obtained.

[0120] Based on the abnormal trajectory classification results, the cargo priority and sorting path data are extracted from the pre-stored sorting task database, and the cargo priority is adjusted in conjunction with the Drools rule engine to obtain the initial task allocation scheme.

[0121] Based on the initial task allocation scheme and the sorting path data, the shortest path is planned using the Dijkstra algorithm, and the permission database records are updated to obtain a task allocation scheme that includes the shortest path planning.

[0122] Specifically, the algorithm first classifies regional risk levels based on the permission level field in the access permission configuration table (e.g., permission level = 2) and the stay area identifier in the personnel trajectory (e.g., stay area = area A). The algorithm takes into account the discrete values ​​of the permission level (ranging from 1 to 5) and the stay area code (e.g., area A / B / C code values) as input, and extracts the corresponding risk label level (levels 1-5, based on the number of abnormal events occurring in that area within 30 days) from the historical security event database. The algorithm calculates the information gain rate of the stay area field (the specific calculation process involves statistically analyzing the distribution frequency of different risk levels for each area in historical samples, calculating the Gini index based on the frequency distribution, and then dividing by the inherent information content of the field before splitting). When the frequency of abnormal events in a certain area (e.g., area A) exceeds a preset splitting threshold (e.g., >3 times, this threshold is determined through quantile analysis of historical data), the area is classified as high-risk (e.g., level = 4). Finally, it outputs a regional risk level table containing area codes and risk level values.

[0123] Based on the regional risk level table and personnel trajectory data, the trajectory anomaly coefficient is calculated as follows: The risk level value of the current trajectory's region (e.g., risk level of region A = 4) is multiplied by a regional risk weighting coefficient of 0.6 (this coefficient is calculated based on the correlation coefficient between regional risk and trajectory anomalies in historical data). This is then overlaid with the abnormal action marker count in the personnel trajectory (e.g., overstay marker count = 2) multiplied by a behavior weighting coefficient of 0.4 (determined through feature importance analysis). The weighted sum of these two values ​​yields the trajectory anomaly coefficient (example: 4 × 0.6 + 2 × 0.4 = 3.2). When this coefficient exceeds a preset trajectory anomaly threshold of 1.5 (this threshold is set based on the mean of three months of historical abnormal trajectory samples plus twice the standard deviation), the trajectory is marked as an abnormal trajectory (label value = 1), generating an abnormal trajectory classification result table containing the trajectory number, abnormal label value, and risk level value.

[0124] Based on the anomaly label field (e.g., label value = 1) in the anomaly trajectory classification results, the associated cargo priority field (e.g., priority = 4) and sorting path node sequence field (e.g., node sequence = N001→N005) are extracted from the sorting task database. The priority is dynamically adjusted via the Drools rule engine: a predefined rule file is loaded (containing the rule entry: "When [anomaly label = 1] and [cargo priority > 3], execute [priority = original priority + 1]"). The cargo priority field is passed to the rule engine's working memory. The engine performs rule matching using the Rete algorithm (e.g., triggering rule number R102), executes the rule action to modify the priority value (e.g., changing it from 4 to 5), and outputs an initial task allocation scheme containing the new priority value and the original path sequence.

[0125] Finally, based on the sorting path node sequence in the initial task allocation scheme, the shortest path is planned using Dijkstra's algorithm: A sorting center topology network model is constructed (nodes represent functional areas such as receiving area A01; edge weight = standard movement time of the path × regional risk correction coefficient; the generation of this risk correction coefficient is directly related to the regional risk level table—each risk level in the table corresponds to a preset time penalty coefficient (e.g., risk level 1 corresponds to coefficient 1.0 with no effect, level 3 corresponds to coefficient 1.5 indicating a 50% increase in path time, and level 5 corresponds to coefficient 2.0, meaning the time doubles). Specifically, when the risk level of the area associated with the path edge is determined (e.g., the risk level of packaging area B02 is found to be 4 from the regional risk level table), the preset coefficient value for that level is automatically matched (e.g., level 4 corresponds to coefficient 1.8). Starting from the dynamic location node of the goods (e.g., A01), the distance record table (starting point distance = 0, other node distances = maximum value), the priority sorting queue (initially containing only the starting point), and the path backtracking table (the predecessor of the starting point is empty) are initialized. The process iterates through the nodes in a loop: Extract the current node (e.g., A01) with the smallest distance from the queue, traverse its adjacent nodes (e.g., B02), calculate the new distance from the current node to the adjacent nodes (e.g., 0 + 45 = 45 seconds), and if the new distance is less than the existing distance value of the adjacent node (e.g., the original maximum distance value of B02), update the distance value of the adjacent node (B02 is updated to 45 seconds), record the predecessor node (B02's predecessor is set to A01), and add it to the queue. This process is repeated until the target node (e.g., packaging area C03) is reached. An optimal path sequence (e.g., A01→B02→C03) is generated by backtracking the predecessor node. The sorting path field for this task is synchronously updated in the permission database (by executing a structured update statement), and the task allocation scheme, including the optimized path sequence, adjusted priority, and execution time window, is output.

[0126] This process avoids high-risk areas while reducing the length of high-priority cargo processing routes.

[0127] In step S18, the sorting paths based on the dynamic location of the goods and the task scheduling based on personnel identification tags within the sorting center are adjusted based on the task allocation scheme to obtain the optimized configuration of the sorting center's data management.

[0128] Specifically, based on the cargo sorting path field (e.g., path node sequence = N001→N003→N005) and the cargo priority field (e.g., priority = 5) in the cargo processing urgency level, combined with the dynamic location coordinates of the cargo (e.g., real-time location node = N001), the sorting path is updated through a path dynamic adjustment algorithm: First, the path node sequence is parsed, and the deviation distance from the current position to the next node is calculated based on the dynamic position of the cargo (e.g., deviation from the preset path by 2.8 meters). When the deviation value exceeds the preset threshold of 2.5 meters, the real-time topology map of the path re-planning center (including the passage status between each node) is triggered, and an incremental path update strategy is adopted (e.g., deleting the already passed node N001 and taking the current actual position as the new starting point), and a sorting path adjustment scheme adapted to the real-time position of the cargo (e.g., new path = N001_actual→N003→N005) is output.

[0129] Synchronously, based on the job type field in the personnel identity tag (e.g., job type = parcel sorter) and the cargo priority field in the task allocation scheme, the task allocation is executed through the task scheduling rule engine: First, a preset task capability matrix is ​​matched according to the job type (e.g., maximum priority that a sorter can handle = 5), and a set of tasks that meet the capability range (e.g., cargo with priority ≤ 5) is filtered. Then, a weighted round-robin scheduling algorithm is used, with the cargo urgency score (e.g., score = 9.2) as the weight factor (weight calculation formula: scheduling priority = cargo urgency score × 0.7 + task waiting time × 0.3), to generate a task allocation list with the personnel identity tag as the primary key (e.g., employee 10086 assigned task ID = T20230809003). Finally, the sorting path adjustment scheme is merged with the task allocation list and written into the configuration table of the sorting center's main control database (updated fields include real-time cargo path sequence, task-bound personnel identifier, and planned execution time window), forming optimized configuration data containing dynamic path mapping relationships and personnel task binding relationships.

[0130] This step addresses the lag in response of traditional configuration methods in dynamic scenarios by using location-driven real-time path correction and priority-weighted task scheduling, thereby achieving precise matching between sorting resources and cargo status.

[0131] Reference Figure 2 The second embodiment of the present invention provides a sorting center data management system based on machine learning, comprising:

[0132] The data acquisition module is used to acquire personnel operation frequency, personnel trajectory, interaction time distribution, dwell area, operation type, time window characteristics, abnormal action identifier, personnel identity tag, dynamic location of goods, goods tag, order data and environmental interaction data, which constitute environmental perception data;

[0133] The personnel behavior analysis module is used to perform personnel behavior feature analysis based on the environmental perception data and construct personnel behavior features based on the personnel identity tags;

[0134] An abnormal behavior detection module is used to identify abnormal behavior based on the personnel behavior characteristics, combined with the personnel trajectory and the interaction time distribution, and to obtain abnormal behavior detection results.

[0135] The cargo urgency analysis module is used to calculate the cargo urgency score and perform dynamic anomaly analysis of the cargo based on the cargo label, the order data, and the abnormal behavior detection results, so as to obtain the cargo handling urgency level.

[0136] The verification strength adjustment module is used to simplify the password verification process and adopt a fast verification process based on the urgency of the cargo handling, and output a verification strength adjustment scheme.

[0137] The security control strategy module is used to adjust the access permission configuration of personnel based on the verification strength adjustment scheme and the abnormal behavior detection results, and output the security control strategy.

[0138] The task allocation analysis module is used to optimize task allocation rules and cargo sorting paths based on the security control strategy, the abnormal behavior detection results, and the personnel trajectory, and output a task allocation scheme.

[0139] The output configuration module is used to adjust the sorting path based on the dynamic location of the goods and the task scheduling based on personnel identification tags within the sorting center based on the task allocation scheme, so as to obtain the data management optimization configuration of the sorting center.

[0140] It should be noted that the sorting center data management device based on machine learning provided in this embodiment of the invention is used to execute all the process steps of the sorting center data management method based on machine learning in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0141] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a machine learning-based sorting center data management program. When the processor executes the computer program, it implements the steps described in the various machine learning-based sorting center data management method embodiments above, for example... Figure 1The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as a sorting center data management module based on machine learning.

[0142] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0143] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0144] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0145] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0146] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0147] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0148] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A data management method for sorting centers based on machine learning, characterized in that, include: The environmental perception data consists of personnel operation frequency, personnel trajectory, interaction time distribution, dwell area, operation type, time window characteristics, abnormal action identifier, personnel identity tag, dynamic location of goods, goods tag, order data and environmental interaction data. Based on the environmental perception data, perform personnel behavior characteristic analysis to construct personnel behavior characteristics based on the personnel identity tags; Based on the personnel behavior characteristics, combined with the personnel trajectory and the interaction time distribution, abnormal behavior is identified to obtain abnormal behavior detection results; Based on the cargo label, the order data, and the abnormal behavior detection results, a cargo urgency score is calculated and a dynamic anomaly analysis of the cargo is performed to obtain the cargo handling urgency level. Based on the urgency of the cargo handling, the password verification process is simplified and a fast verification process is adopted, and a verification strength adjustment scheme is output. Based on the verification strength adjustment scheme and the abnormal behavior detection results, adjust the personnel access permission configuration and output the security control policy; Based on the security control strategy, the abnormal behavior detection results, and the personnel trajectory, optimize the task allocation rules and cargo sorting paths, and output the task allocation scheme. Based on the task allocation scheme, the sorting paths based on the dynamic location of the goods and the task scheduling based on personnel identification tags within the sorting center are adjusted to obtain the optimized configuration of the sorting center's data management.

2. The sorting center data management method based on machine learning according to claim 1, characterized in that, The step of analyzing personnel behavior characteristics based on the environmental perception data and constructing personnel behavior characteristics based on the personnel identity tags includes: Based on dynamic environment perception data, spatiotemporal slicing is performed using the sliding window method to obtain initial behavioral data aggregated by time window; Based on the initial behavioral data, trajectory segmentation and region clustering analysis are performed using the DBSCAN clustering algorithm to obtain behavioral grouping data; Based on the behavioral grouping data, abnormal behavior is judged using the C4.5 decision tree algorithm, and the behavioral characteristics of the personnel based on the personnel identity tags are output.

3. The sorting center data management method based on machine learning according to claim 1, characterized in that, The abnormal behavior identification based on the personnel behavior characteristics, combined with the personnel trajectory and the interaction time distribution, to obtain abnormal behavior detection results includes: When the trajectory of a person deviates from the preset normal trajectory threshold, the dwell time in the trajectory is extracted, and abnormal dwell is judged by combining the interaction time distribution analysis and the preset dwell threshold, and an abnormal action identifier is generated. Based on the abnormal action identifier and the personnel trajectory, the behavior pattern features are classified using the K-means clustering algorithm to obtain the abnormal behavior detection results.

4. The sorting center data management method based on machine learning according to claim 1, characterized in that, The step of calculating a cargo urgency score and performing dynamic anomaly analysis based on the cargo label, order data, and abnormal behavior detection results to obtain the cargo handling urgency level includes: Based on the cargo tags, cargo priority is extracted, and based on the order data, delivery time limit and order quantity are extracted. Then, abnormal records of the cargo are extracted through SQL query to obtain cargo attributes including cargo priority, delivery time limit, order quantity and abnormal records. When the delivery time limit is lower than the preset time limit threshold, the urgency score of the goods is obtained by weighting the goods priority, the delivery time limit, the preset goods priority weight, and the preset time limit weight. Based on the dynamic location of the goods and the abnormal records, the spatial distribution of location points is analyzed by K-means clustering algorithm to identify abnormal location clusters that deviate from the preset standard logistics path and obtain the abnormal goods identification. Based on the abnormal behavior detection results and the abnormal cargo identification, the weighted Euclidean distance algorithm is used to calculate the behavior risk coefficient. Based on the cargo anomaly identifier and the behavioral risk coefficient, the cargo urgency score is adjusted in a graded manner using the CART decision tree algorithm to generate a cargo handling urgency level that includes cargo priority, delivery time limit, order quantity, anomaly records, and cargo urgency score.

5. The sorting center data management method based on machine learning according to claim 4, characterized in that, The step of simplifying the password verification process and adopting a fast verification process based on the urgency of cargo handling, and outputting a verification strength adjustment scheme, includes: When the priority of the goods is higher than the preset emergency level threshold or the delivery time limit is lower than the time limit threshold, the verification complexity adjustment coefficient is calculated using the CART decision tree algorithm based on the goods label and the order quantity. Based on the verification complexity adjustment coefficient and the personnel identity label, the matching identity verification rules are extracted from the preset identity database, and then the verification strength scheme is generated by using a rule engine based on the Rete algorithm and combined with the preset fast verification process to obtain the initial verification strength scheme. Based on the initial verification strength scheme, the corresponding verification parameter configuration is obtained from the preset verification rule base, and the Q-learning algorithm is used to optimize the password verification complexity to obtain the verification strength adjustment scheme.

6. The sorting center data management method based on machine learning according to claim 1, characterized in that, The step of adjusting personnel access permission configurations and outputting security control policies based on the verification strength adjustment scheme and the abnormal behavior detection results includes: Based on the personnel identity tags and the abnormal behavior detection results, the initial values ​​of verification frequency and password complexity are set to obtain a preliminary configuration scheme for user access permissions; Based on the preliminary configuration scheme, a weighted permission risk coefficient is calculated. When the permission risk coefficient exceeds a preset permission threshold, the access permission level is reduced to obtain permission configuration parameters. Based on the permission configuration parameters and the abnormal behavior detection results, the K-nearest neighbor algorithm is used to match abnormal behavior patterns, determine the scope of permission restrictions, and generate an access permission configuration table. Based on the access permission configuration table, the permission database is updated, and the permission mapping relationship is reconstructed in conjunction with the role-based access control method, and a security control policy is output.

7. The sorting center data management method based on machine learning according to claim 6, characterized in that, The step involves optimizing task allocation rules and cargo sorting paths based on the security control strategy, the abnormal behavior detection results, and the personnel trajectory, and outputting a task allocation scheme, including: Based on the access permission configuration table and the personnel trajectory, the regional risk level is classified using the C5.0 decision tree algorithm to obtain the regional risk level; Based on the regional risk level and the personnel trajectory, a trajectory anomaly coefficient is calculated by weighting. When the trajectory anomaly coefficient exceeds a preset trajectory anomaly threshold, it is marked as an abnormal trajectory, and an abnormal trajectory classification result is obtained. Based on the abnormal trajectory classification results, the cargo priority and sorting path data are extracted from the pre-stored sorting task database, and the cargo priority is adjusted in conjunction with the Drools rule engine to obtain the initial task allocation scheme. Based on the initial task allocation scheme and the sorting path data, the shortest path is planned using the Dijkstra algorithm, and the permission database records are updated to obtain a task allocation scheme that includes the shortest path planning.

8. A machine learning-based data management system for sorting centers, characterized in that, include: The data acquisition module is used to acquire personnel operation frequency, personnel trajectory, interaction time distribution, dwell area, operation type, time window characteristics, abnormal action identifier, personnel identity tag, dynamic location of goods, goods tag, order data and environmental interaction data, which constitute environmental perception data; The personnel behavior analysis module is used to perform personnel behavior feature analysis based on the environmental perception data and construct personnel behavior features based on the personnel identity tags; An abnormal behavior detection module is used to identify abnormal behavior based on the personnel behavior characteristics, combined with the personnel trajectory and the interaction time distribution, and to obtain abnormal behavior detection results. The cargo urgency analysis module is used to calculate the cargo urgency score and perform dynamic anomaly analysis of the cargo based on the cargo label, the order data, and the abnormal behavior detection results, so as to obtain the cargo handling urgency level. The verification strength adjustment module is used to simplify the password verification process and adopt a fast verification process based on the urgency of the cargo handling, and output a verification strength adjustment scheme. The security control strategy module is used to adjust the access permission configuration of personnel based on the verification strength adjustment scheme and the abnormal behavior detection results, and output the security control strategy. The task allocation analysis module is used to optimize task allocation rules and cargo sorting paths based on the security control strategy, the abnormal behavior detection results, and the personnel trajectory, and output a task allocation scheme. The output configuration module is used to adjust the sorting path based on the dynamic location of the goods and the task scheduling based on personnel identification tags within the sorting center based on the task allocation scheme, so as to obtain the data management optimization configuration of the sorting center.

9. An electronic device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the machine learning-based sorting center data management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the machine learning-based sorting center data management method as described in any one of claims 1 to 7.

Images