Intelligent cabinet shelf allocation method and system based on big data collection

Abstract

This invention discloses a method and system for intelligent cabinet allocation based on big data collection, comprising the following steps: collecting task data and preprocessing it to form a standardized feature sequence; establishing a weight matrix in an ART network, marking active nodes if the feature sequence matches, and creating new nodes if they do not match; freezing nodes while keeping their weights unchanged, and only updating the weight matrix for active nodes; calculating a heat factor based on the number of times the cabinet has been opened and the consecutive intervals of the past 24 hours, adjusting the activation weights of high-frequency cabinets, and forming an activation sorting sequence; selecting target cabinets according to the activation sorting sequence, and updating the node-cabinet association and usage count; marking the occupancy status, cold start priority, and cumulative usage count of the selected cabinets; and adjusting the weight matrix and activation sequence parameters according to the weight matrix, frozen status, and usage count. This invention achieves accurate cabinet allocation and load balancing through adaptive clustering of the ART network and inverse adjustment of heat.

Description

Technical Field

[0001] This invention relates to the field of intelligent cabinet compartment allocation and management technology, and in particular to an intelligent cabinet compartment allocation method and system based on big data collection. Background Technology

[0002] In existing smart locker systems, locker allocation typically relies on fixed rules or manual scheduling, resulting in insufficient flexibility and slow response times. As user access behaviors become more diverse and data on item specifications, storage duration, and locker door usage frequency continue to increase, traditional allocation methods struggle to reflect locker usage in real time, easily leading to low locker utilization, overloaded hot lockers, or wasted idle lockers.

[0003] Existing methods lack efficient feature extraction and state representation techniques when processing large amounts of historical access data, making it impossible to perform fine-grained matching of different item categories, cabinet capacities, and usage statuses. Some smart lockers attempt to introduce simple popularity or priority strategies, but these are mostly based on single-indicator control and cannot comprehensively consider multi-dimensional information such as item size, cabinet capacity, usage frequency, and storage duration, resulting in unstable allocation results and a tendency for duplicate scheduling or allocation deviations.

[0004] Furthermore, existing systems lack automated category freezing and heat modulation mechanisms, making it impossible to dynamically adjust allocation strategies based on the stability of cabinet usage, resulting in low management efficiency. Therefore, there is an urgent need for an intelligent cabinet allocation method that can intelligently match and dynamically adjust cabinet activation values ​​and allocation strategies based on big data collection and real-time status analysis, integrating multi-dimensional features, in order to improve cabinet utilization, optimize access efficiency, reduce manual intervention, and achieve refined and intelligent management. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent cabinet allocation method and system based on big data collection. This invention utilizes the adaptive clustering and freeze-active partition update mechanism of the ART network to maintain the stability of the weight matrix. It combines the number of cabinet openings and continuous intervals to calculate the heat inverse adjustment factor, performs dynamic suppression of activation weights on high-frequency dense cabinets, selects target cabinets according to the suppressed sorting sequence, marks the occupancy status, cold start priority and cumulative usage count, and performs closed-loop parameter adjustment based on the weight matrix, freeze status and usage count. It has the advantages of high allocation accuracy, fast response speed and balanced cabinet utilization.

[0006] A smart cabinet compartment allocation method based on big data collection according to an embodiment of the present invention includes the following steps: Collect cabinet task data, perform data preprocessing, and form standardized feature sequences; In the ART network, a weight vector is established for each category node, and the weights are combined to form a category node weight matrix. The similarity of each standardized feature sequence with the category node weight vector is calculated to generate a similarity matrix. Nodes with matching values ​​exceeding a preset threshold in the similarity matrix are marked as active nodes. A list of nodes to be created is generated for unmatched nodes. New category node weight vectors are assigned to the nodes to be created and added to the category node weight matrix. For frozen nodes identified within a continuous time window, the original weights remain unchanged. Only active nodes undergo weight matrix updates, and the corresponding node status and update time are recorded for each update. During the ART network matching process, the number of times the candidate cabinets have been opened and the interval between consecutive openings in the past 24 hours are read. The cabinet popularity reverse adjustment factor is calculated. For cabinets whose number of openings exceeds the threshold and whose consecutive intervals are less than the preset time, the activation weight is adjusted according to the rules to form an activation ranking sequence. Select the target cabinet according to the active sorting sequence, and update the association between the matching category node and the cabinet and the usage count in the ART network; The selected cabinet is marked with status, including occupancy status, cold start priority, and cumulative number of uses, and the allocation history is recorded; Adjust the category node weight matrix and the cabinet activation sequence parameters based on the category node weight matrix, the frozen status, and the cabinet usage count.

[0007] Optionally, the preprocessing of the cabinet task data specifically includes: The cabinet task data, including request content, cabinet status information and timestamps, undergoes digital preprocessing, which includes field unification, time alignment, anomaly removal, and value normalization.

[0008] Optionally, establishing a weight vector for each category node in the ART network, and marking nodes with matching values ​​exceeding a preset threshold in the similarity matrix as active nodes specifically includes: In the ART network, an initial weight vector is assigned to each category node, and the weight vectors of all category nodes are combined to form a category node weight matrix; The standardized feature sequences are matched one by one with the category node weight vector to generate the category node similarity value corresponding to each feature sequence; Arrange all similarity values ​​according to feature sequence index and category node index to form a similarity matrix; Based on the similarity matrix, nodes with matching values ​​exceeding a preset threshold are marked as active nodes, and feature sequences that do not match existing categories are marked as nodes to be created, generating a list of nodes to be created; new category node weight vectors are assigned to the nodes to be created and added to the category node weight matrix.

[0009] Optionally, the frozen nodes retain their weights unchanged, and only active nodes participate in the weight matrix update. Recording the node status and update time specifically includes: For the active nodes identified in the ART network, extract the corresponding category node weight vector and its index in the weight matrix; calculate the weight increment based on the active node matching value, add the increment value to each element of the active node weight vector to form the updated weight vector; perform normalization processing on the updated weight vector to cover the original category node weight vector; freeze the nodes while keeping the original weight vector unchanged. After the weight matrix is ​​updated, the active node number, weight vector index, and update timestamp are written to the node status record table; the frozen node number and frozen status identifier are written to the node status record table, and the most recent update time is recorded; all node statuses are sorted according to the row order of the weight matrix and the node index order to generate a complete node status table. For each active node, calculate the matching error residual and write the residual value into the corresponding field of the node status table; write the update record and timestamp of the active node weight vector into the weight change log. In the frozen node status table, mark the start time and the freezing identifier; add a field to the node status record table to record whether the current node participates in this round of weight update and the corresponding weight increment and normalization adjustment coefficient; according to the node index order, merge the information of active nodes, frozen nodes and nodes to be created to generate a total node status snapshot and write it to the weight matrix operation history database. The update magnitude of active node weights in the node status snapshot is calculated to generate a node weight update statistics table, including maximum, minimum, average and variance. The statistics table is written to the database along with the node status snapshot. After each operation, the node status record table is checked for integrity, and abnormal entries are marked with an error identifier and operation time. The timestamp, operation sequence and weight matrix change are serialized and saved.

[0010] Optionally, the calculation of the cabinet heat reverse adjustment factor, which adjusts the activation weight of cabinets that have been opened more than a threshold and whose consecutive intervals are less than a preset time according to rules to form an activation sorting sequence, specifically includes: For each cabinet in the candidate cabinet list, count the number of times it has been opened in the past 24 hours; The time interval between two consecutive openings is counted to form a sequence of consecutive opening intervals; The number of times each cabinet is opened is compared with two preset thresholds, and cabinets that exceed the thresholds are marked. For cabinets exceeding the threshold, check the number of times the consecutive opening interval is less than the preset interval time, and calculate the number of times the interval limit is exceeded. Based on the number of times the value is exceeded, the activation weight adjustment coefficient is calculated by reducing it by 5% for each instance of exceeding the limit, with the initial weight set at 100%. The adjustment coefficient is applied to the original activation weight of each eligible cabinet to generate the adjusted activation weight; For cabinets that have not exceeded the threshold for the number of times they can be opened or whose consecutive intervals meet the requirements, the original activation weight remains unchanged. After adjusting the activation values ​​of all cabinets, organize them into an activation value list, sort them from high to low, and generate an activation sorting sequence. Record the cabinet number, original activation weight, adjustment coefficient and adjusted activation weight of each cabinet in the activation sorting sequence to form an activation sorting record table; Index and number the active sorted sequences after they have been sorted.

[0011] Optionally, the step of selecting the target cabinet according to the activation sorting sequence and updating the association between the matching category node and the cabinet and the usage count in the ART network specifically includes: The cabinet numbers are sequentially obtained from the activated sorting sequence as candidate target cabinets; Check the current occupancy status of each candidate target cabinet according to the activation order, and remove cabinets that are already occupied; For the remaining available cabinets, select the cabinet with the highest activation weight as the target cabinet; In the ART network, the target cabinet is associated with the matching category node, and the correspondence between the category node number and the cabinet number is recorded. The number of times the target cabinet is allocated is accumulated into the cabinet usage count field, and the usage statistics of the corresponding node in the category node weight matrix are updated. Write the association information between category nodes and cabinets into the node-cabinet association record table, including the category node number, cabinet number, activation weight, and update time; The usage count and associated status of the target cabinet are timestamped to form a complete update history. Organize all updated records into a table that matches category nodes with cabinet associations and usage counts, according to the order of the target cabinets.

[0012] Optionally, the step of marking the selected cabinet with status, including occupancy status, cold start priority, and cumulative usage count, and recording the allocation history specifically includes: Extract the cabinet identifier of the target cabinet from the active sorting sequence; assign the occupancy status field of the target cabinet to 1; retrieve the cumulative total number of allocations from the target cabinet storage area, increment the cumulative total number of allocations by 1, and write it back to the cumulative usage count field; collect year, month, day, hour, minute, and second data from the system clock, combine them into a last allocation timestamp, and write it to the last allocation timestamp field; retrieve the continuous unallocated time count value from the target cabinet. If the count value is greater than 72 hours, assign the cold start priority field to 3; if the count value is greater than 24 hours and less than or equal to 72 hours, assign it to 2; otherwise, assign it to 1; combine the occupancy status field value, cold start priority field value, cumulative usage count field value, and last allocation timestamp field value into an allocation record, and append it to the allocation history log file; clear the continuous unallocated time count value of the target cabinet to zero; for each non-selected candidate cabinet, retrieve the continuous unallocated time count value, increment the count value by 1, and write it back.

[0013] Optionally, adjusting the category node weight matrix and the cabinet activation sequence parameters based on the category node weight matrix, the frozen state, and the cabinet usage count specifically includes: Extract the weight vectors of all category nodes from the category node weight matrix; extract the freeze flag of each category node from the frozen status list, skipping the adjustment operation for nodes with a freeze flag of 1; for each category node with a freeze flag of 0, retrieve the cumulative usage count field value of the associated cabinet; input the retrieved cumulative usage count field value into the normalization function, calculating the output value to be in the range of 0 to 1; perform vector multiplication operation between the category node weight vector and the normalization function output value to generate the adjusted weight vector; replace the original weight vector in the category node weight matrix with the adjusted weight vector; extract the continuous unallocated time count value from the associated cabinet of each category node; reduce the basic weight value of the activation sequence corresponding to the cabinet with a continuous unallocated time count value greater than 72 hours by 20%; write the reduced basic weight value of the activation sequence back to the cabinet activation sequence parameter storage area.

[0014] According to an embodiment of the present invention, an intelligent cabinet compartment allocation system based on big data collection includes the following modules: The data acquisition and preprocessing module collects cabinet task data, performs preprocessing on the data, and outputs a standardized feature sequence. The ART network matching module calculates the similarity between the standardized feature sequence and the category node weight vectors to generate a similarity matrix. Nodes with matching values ​​exceeding a preset threshold are marked as active nodes, and unmatched nodes are assigned new weight vectors and added to the category node weight matrix. The weight update module identifies frozen nodes within a continuous time window, keeps their weights unchanged, updates the weight matrix for active nodes, and records the node status and update time. The cabinet popularity adjustment module reads the number of times candidate cabinets have been opened and the consecutive opening interval in the past 24 hours, calculates the cabinet popularity inverse adjustment factor, adjusts the activation weights of cabinets with opening counts exceeding a threshold and consecutive intervals less than a preset time, and outputs an activation ranking sequence. The target cabinet selection and association update module selects target cabinets according to the activation ranking sequence, updates the association between matching category nodes and cabinets and the usage count in the ART network. The status marking module marks the occupancy status, cold start priority, and cumulative usage count of selected cabinets and records the allocation history. The parameter adjustment module is used to adjust the category node weight matrix and the cabinet activation sequence parameters based on the category node weight matrix, the frozen status, and the cabinet usage count.

[0015] The beneficial effects of this invention are: 1. The invention utilizes the adaptive clustering mechanism of ART networks and adopts a partitioning update strategy for active and frozen nodes to quickly incorporate new task features while maintaining the stability of existing knowledge, significantly reducing the overhead of repeated training of the weight matrix and improving the response speed of cabinet allocation.

[0016] 2. This invention calculates a heat-reverse adjustment factor by collecting the number of times the cabinets have been opened and the interval between consecutive openings in the past 24 hours. It then applies dynamic suppression of activation weights to cabinets that are used frequently and intensively, thereby avoiding excessive concentration of popular cabinets and achieving a balanced distribution of cabinet usage load.

[0017] 3. After allocation, this invention marks the selected cabinet with its occupancy status, cold start priority, and cumulative usage count. Cabinets that have been unallocated for more than 72 consecutive hours are given a high cold start priority, effectively solving the problem of delayed response of long-term idle cabinets in the event of sudden demand.

[0018] 4. Based on the category node weight matrix, frozen state, and cabinet usage count, this invention performs closed-loop adjustment of the weight matrix and activation sequence parameters, continuously optimizing the accuracy and adaptability of the allocation strategy, and improving the system's long-term stable operation capability under dynamic load environment. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of an intelligent cabinet compartment allocation method based on big data collection proposed in this invention; Figure 2 This is a schematic diagram of ART network adaptive clustering and partition update for an intelligent cabinet compartment allocation method based on big data collection proposed in this invention. Figure 3 This is a schematic diagram illustrating the reverse adjustment and activation sorting of cabinet heat in a smart cabinet compartment allocation method based on big data collection proposed in this invention. Figure 4 This is a module structure diagram of an intelligent cabinet compartment allocation system based on big data collection proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figures 1-4 A smart cabinet compartment allocation method based on big data collection includes the following steps: Collect cabinet task data, perform data preprocessing, and form standardized feature sequences; In the ART network, a weight vector is established for each category node, and the weights are combined to form a category node weight matrix. The similarity of each standardized feature sequence with the category node weight vector is calculated to generate a similarity matrix. Nodes with matching values ​​exceeding a preset threshold in the similarity matrix are marked as active nodes. A list of nodes to be created is generated for unmatched nodes. New category node weight vectors are assigned to the nodes to be created and added to the category node weight matrix. For frozen nodes identified within a continuous time window, the original weights remain unchanged. Only active nodes undergo weight matrix updates, and the corresponding node status and update time are recorded for each update. During the ART network matching process, the number of times the candidate cabinets have been opened and the interval between consecutive openings in the past 24 hours are read. The cabinet popularity reverse adjustment factor is calculated. For cabinets whose number of openings exceeds the threshold and whose consecutive intervals are less than the preset time, the activation weight is adjusted according to the rules to form an activation ranking sequence. Select the target cabinet according to the active sorting sequence, and update the association between the matching category node and the cabinet and the usage count in the ART network; The selected cabinet is marked with status, including occupancy status, cold start priority, and cumulative number of uses, and the allocation history is recorded; Adjust the category node weight matrix and the cabinet activation sequence parameters based on the category node weight matrix, the frozen status, and the cabinet usage count.

[0022] In this embodiment, the preprocessing of cabinet task data specifically includes: The cabinet task data, including request content, cabinet status information and timestamps, undergoes digital preprocessing, which includes field unification, time alignment, anomaly removal, and value normalization.

[0023] In this embodiment, establishing a weight vector for each category node in the ART network, and marking nodes with matching values ​​exceeding a preset threshold in the similarity matrix as active nodes specifically includes: In the ART network, an initial weight vector is assigned to each category node, and the weight vectors of all category nodes are combined to form a category node weight matrix; The standardized feature sequences are matched one by one with the category node weight vector to generate the category node similarity value corresponding to each feature sequence; Arrange all similarity values ​​according to feature sequence index and category node index to form a similarity matrix; Based on the similarity matrix, nodes with matching values ​​exceeding a preset threshold are marked as active nodes, and feature sequences that do not match existing categories are marked as nodes to be created, generating a list of nodes to be created; new category node weight vectors are assigned to the nodes to be created and added to the category node weight matrix.

[0024] In this embodiment, frozen nodes retain their weights unchanged, and only active nodes participate in the weight matrix update. Recording node status and update time specifically includes: For the active nodes identified in the ART network, extract the corresponding category node weight vector and its index in the weight matrix; calculate the weight increment based on the active node matching value, add the increment value to each element of the active node weight vector to form the updated weight vector; perform normalization processing on the updated weight vector to cover the original category node weight vector; freeze the nodes while keeping the original weight vector unchanged. After the weight matrix is ​​updated, the active node number, weight vector index, and update timestamp are written to the node status record table; the frozen node number and frozen status identifier are written to the node status record table, and the most recent update time is recorded; all node statuses are sorted according to the row order of the weight matrix and the node index order to generate a complete node status table. For each active node, calculate the matching error residual and write the residual value into the corresponding field of the node status table; write the update record and timestamp of the active node weight vector into the weight change log. In the frozen node status table, mark the start time and the freezing identifier; add a field to the node status record table to record whether the current node participates in this round of weight update and the corresponding weight increment and normalization adjustment coefficient; according to the node index order, merge the information of active nodes, frozen nodes and nodes to be created to generate a total node status snapshot and write it to the weight matrix operation history database. The update magnitude of active node weights in the node status snapshot is calculated to generate a node weight update statistics table, including maximum, minimum, average and variance. The statistics table is written to the database along with the node status snapshot. After each operation, the node status record table is checked for integrity, and abnormal entries are marked with an error identifier and operation time. The timestamp, operation sequence and weight matrix change are serialized and saved.

[0025] Specifically, calculating the weight increment based on the active node matching value includes: Extract similarity values ​​from the matching values ​​of active nodes; multiply the similarity values ​​by the preset learning rate parameter to generate the initial weight increment; read the values ​​of each element in the current weight vector of active nodes; perform addition operations on the initial weight increment and each element in the current weight vector to generate the incremented weight vector; compare each element in the incremented weight vector with the preset upper weight threshold, and assign the upper weight threshold to elements exceeding the upper weight threshold; compare each element in the incremented weight vector with the preset lower weight threshold, and assign the lower weight threshold to elements below the lower weight threshold; output the weight increment vector after upper and lower weight clipping as the final weight increment.

[0026] Specifically, for each active node, calculating the matching error residual includes: Extract the standardized feature sequence vector of active nodes and the updated weight vector of active nodes; perform element-wise subtraction between the standardized feature sequence vector and the updated weight vector to generate a difference vector; perform a square operation on each element in the difference vector to generate a squared difference vector; perform a sum operation on all elements in the squared difference vector to generate a sum of squares; extract the number of dimensions of the weight vector of active nodes; divide the sum of squares by the number of dimensions to generate a mean squared error; output the mean squared error as the matching error residual.

[0027] Specifically, calculating the update magnitude of active node weights in the node status snapshot and generating a node weight update statistics table includes: Extract the weight vector of each active node before the update from the node state snapshot; extract the weight vector of each active node after the update from the node state snapshot; perform element-wise subtraction between the updated weight vector and the weight vector before the update to generate a difference vector; calculate the magnitude of the difference vector as the weight update magnitude of a single node; sort the weight update magnitudes of all active nodes in the node state snapshot and extract the maximum and minimum values; sum the weight update magnitudes of all active nodes and divide by the total number of active nodes to generate the average value; sum the squared differences of the weight update magnitudes of all active nodes and divide by the total number of nodes to generate the variance; write the maximum, minimum, average, and variance into a statistics table, with each row corresponding to an active node number.

[0028] After each operation, an integrity check is performed on the node status record table, and abnormal entries are marked with an error identifier and operation time. The timestamp, operation sequence, and weight matrix changes are serialized and saved. Specifically, this includes: Extract the node ID, weight vector, and timestamp fields from all entries in the node status record table; check if the node ID field of each entry is null; if so, write 1 to the exception flag field of that entry; check if the dimension of the weight vector field of each entry is consistent with the dimension of the category node weight matrix; if not, write 1 to the exception flag field of that entry; check if the timestamp field of each entry is within the system operation start time and current time interval; if not, write 1 to the exception flag field of that entry; for entries with an exception flag field of 1, read the year, month, day, hour, minute, and second data from the system clock and write it to the operation time field; extract the sequence number of this operation as the operation sequence value; extract the difference matrix before and after the category node weight matrix update as the weight matrix change; write the operation sequence value, timestamp, and weight matrix change in sequence to a binary serialized file.

[0029] In this embodiment, calculating the cabinet heat reverse adjustment factor and adjusting the activation weight of cabinets that have been opened more than a threshold and whose consecutive intervals are less than a preset time according to rules to form an activation sorting sequence specifically includes: For each cabinet in the candidate cabinet list, count the number of times it has been opened in the past 24 hours; The time interval between two consecutive openings is counted to form a sequence of consecutive opening intervals; The number of times each cabinet is opened is compared with two preset thresholds, and cabinets that exceed the thresholds are marked. For cabinets exceeding the threshold, check the number of times the consecutive opening interval is less than the preset interval time, and calculate the number of times the interval limit is exceeded. Based on the number of times the value is exceeded, the activation weight adjustment coefficient is calculated by reducing it by 5% for each instance of exceeding the limit, with the initial weight set at 100%. The adjustment coefficient is applied to the original activation weight of each eligible cabinet to generate the adjusted activation weight; For cabinets that have not exceeded the threshold for the number of times they can be opened or whose consecutive intervals meet the requirements, the original activation weight remains unchanged. After adjusting the activation values ​​of all cabinets, organize them into an activation value list, sort them from high to low, and generate an activation sorting sequence. Record the cabinet number, original activation weight, adjustment coefficient and adjusted activation weight of each cabinet in the activation sorting sequence to form an activation sorting record table; Index and number the active sorted sequences after they have been sorted.

[0030] Specifically, based on the number of times the value is exceeded, the activation weight adjustment coefficient is calculated by reducing it by 5% for each instance of exceeding the limit. The initial weight is set to 100%. Extract the number of times the cabinet has exceeded the limit, which is the cumulative number of times the interval between consecutive openings is less than the preset time; set the initial activation weight to 100; multiply the number of times the limit has exceeded the limit by 5 to generate a reduction percentage value; subtract the reduction percentage value from the initial activation weight of 100 to generate the adjusted activation weight; extract the adjusted activation weight and compare it with 0. If the adjusted activation weight is less than 0, set the activation weight adjustment coefficient to 0; if the adjusted activation weight is greater than or equal to 0, set the activation weight adjustment coefficient to the adjusted activation weight divided by 100.

[0031] In this embodiment, the target cabinet is selected according to the activation sorting sequence, and the association between the matching category node and the cabinet, as well as the usage count, are updated in the ART network; specifically, this includes: The cabinet numbers are sequentially obtained from the activated sorting sequence as candidate target cabinets; Check the current occupancy status of each candidate target cabinet according to the activation order, and remove cabinets that are already occupied; For the remaining available cabinets, select the cabinet with the highest activation weight as the target cabinet; In the ART network, the target cabinet is associated with the matching category node, and the correspondence between the category node number and the cabinet number is recorded. The number of times the target cabinet is allocated is accumulated into the cabinet usage count field, and the usage statistics of the corresponding node in the category node weight matrix are updated. Write the association information between category nodes and cabinets into the node-cabinet association record table, including the category node number, cabinet number, activation weight, and update time; The usage count and associated status of the target cabinet are timestamped to form a complete update history. Organize all updated records into a table that matches category nodes with cabinet associations and usage counts, according to the order of the target cabinets.

[0032] In this embodiment, the selected cabinet is marked with a status, including occupancy status, cold start priority, and cumulative usage count. The allocation history is recorded specifically including: Extract the cabinet identifier of the target cabinet from the active sorting sequence; assign the occupancy status field of the target cabinet to 1; retrieve the cumulative total number of allocations from the target cabinet storage area, increment the cumulative total number of allocations by 1, and write it back to the cumulative usage count field; collect year, month, day, hour, minute, and second data from the system clock, combine them into a last allocation timestamp, and write it to the last allocation timestamp field; retrieve the continuous unallocated time count value from the target cabinet. If the count value is greater than 72 hours, assign the cold start priority field to 3; if the count value is greater than 24 hours and less than or equal to 72 hours, assign it to 2; otherwise, assign it to 1; combine the occupancy status field value, cold start priority field value, cumulative usage count field value, and last allocation timestamp field value into an allocation record, and append it to the allocation history log file; clear the continuous unallocated time count value of the target cabinet to zero; for each non-selected candidate cabinet, retrieve the continuous unallocated time count value, increment the count value by 1, and write it back.

[0033] In this embodiment, adjusting the category node weight matrix and the cabinet activation sequence parameters based on the category node weight matrix, the frozen state, and the cabinet usage count specifically includes: Extract the weight vectors of all category nodes from the category node weight matrix; extract the freeze flag of each category node from the frozen status list, skipping the adjustment operation for nodes with a freeze flag of 1; for each category node with a freeze flag of 0, retrieve the cumulative usage count field value of the associated cabinet; input the retrieved cumulative usage count field value into the normalization function, calculating the output value to be in the range of 0 to 1; perform vector multiplication operation between the category node weight vector and the normalization function output value to generate the adjusted weight vector; replace the original weight vector in the category node weight matrix with the adjusted weight vector; extract the continuous unallocated time count value from the associated cabinet of each category node; reduce the basic weight value of the activation sequence corresponding to the cabinet with a continuous unallocated time count value greater than 72 hours by 20%; write the reduced basic weight value of the activation sequence back to the cabinet activation sequence parameter storage area.

[0034] According to an embodiment of the present invention, an intelligent cabinet compartment allocation system based on big data collection includes the following modules: The data acquisition and preprocessing module collects cabinet task data, performs preprocessing on the data, and outputs a standardized feature sequence. The ART network matching module calculates the similarity between the standardized feature sequence and the category node weight vectors to generate a similarity matrix. Nodes with matching values ​​exceeding a preset threshold are marked as active nodes, and unmatched nodes are assigned new weight vectors and added to the category node weight matrix. The weight update module identifies frozen nodes within a continuous time window, keeps their weights unchanged, updates the weight matrix for active nodes, and records the node status and update time. The cabinet popularity adjustment module reads the number of times candidate cabinets have been opened and the consecutive opening interval in the past 24 hours, calculates the cabinet popularity inverse adjustment factor, adjusts the activation weights of cabinets with opening counts exceeding a threshold and consecutive intervals less than a preset time, and outputs an activation ranking sequence. The target cabinet selection and association update module selects target cabinets according to the activation ranking sequence, updates the association between matching category nodes and cabinets and the usage count in the ART network. The status marking module marks the occupancy status, cold start priority, and cumulative usage count of selected cabinets and records the allocation history. The parameter adjustment module is used to adjust the category node weight matrix and the cabinet activation sequence parameters based on the category node weight matrix, the frozen status, and the cabinet usage count.

[0035] Example 1: A logistics transit station handles a large volume of parcels daily, and is equipped with 200 smart lockers. During the afternoon peak hours, couriers simultaneously put packages into the lockers, often resulting in several frequently used lockers being repeatedly opened and closed, leading to long queues and congestion, while lockers in the corners remain unused for extended periods. By the time the evening peak has passed and people come to pick up their packages, the frequently used lockers are already full, and the vacant lockers are often unavailable because no one knows which ones are usable or nearby, causing significant time wasted searching for lockers.

[0036] To address this issue, the method of this invention was deployed into the locker management system of the transfer station. The system collects approximately 12,000 locker request data entries daily. Each entry includes the request time, the current locker status, and a task type flag. The collected raw data suffers from inconsistent field formats; for example, some time records are accurate to the second, while others are only accurate to the minute; some locker statuses are represented in text, while others are represented in numerical codes. The system first performs standardized processing on this data, aligning all time fields to the second level, converting status fields into standardized numerical codes, removing obviously erroneous records, and finally performing numerical normalization to generate a standardized feature sequence.

[0037] When the ART network starts working, the system establishes a weight vector for each category node. When a standardized feature sequence enters the network, the system calculates its similarity to the existing category node weight vectors one by one. For example, a request from 2 PM on a certain day might have similarities of 92%, 67%, and 31% with the three existing category nodes, respectively. The preset matching threshold is 85%, so the node with a 92% similarity is marked as an active node. Another request from late at night has similarities below the threshold with all nodes, so the system creates a new node for it, assigns an initial weight vector, and adds it to the matrix.

[0038] Nodes identified as frozen within 48 consecutive hours retain their weights, allowing only active nodes to participate in weight updates. Each update records the node number and update timestamp. During the matching process, the system counts the number of times each of the 200 candidate cabinets has been opened and the consecutive opening intervals over the past 24 hours. Cabinet A was opened 24 times in the past 24 hours, with 18 of those openings having consecutive intervals of less than 5 minutes. Cabinet B was opened only 3 times, with consecutive intervals all exceeding 2 hours. The threshold for the number of openings is set at 10, and the preset interval for consecutive openings is 10 minutes. Cabinet A exceeds the threshold and has excessively short consecutive intervals; the system calculates its "heat" inverse adjustment factor, decreasing it by 5% for each instance of exceeding the threshold, decreasing by 90% after 18 instances, and lowering its activation weight from 100 to 10. Cabinet B retains its original activation weight of 100.

[0039] The system sorts the cells from highest to lowest according to their adjusted activation weights, forming an activation sorting sequence. Cells at the top of the list are selected first. After a target cell is selected, the system updates the association record between the node of that category and the cell in the ART network and increments the cell's usage count by 1.

[0040] When marking the status of a selected cabinet, the system sets the occupancy status to occupied, increments the cumulative usage count by 1, and writes the current timestamp. Cabinet C has been unassigned for 80 consecutive hours, exceeding 72 hours, therefore its cold start priority is marked as high priority. Cabinet D has been unassigned for 30 consecutive hours and is marked as medium priority. Cabinet E has been unassigned for 10 consecutive hours and is marked as low priority.

[0041] Finally, the system adjusts the weight matrix and cabinet activation sequence parameters based on the category node weight matrix, frozen status, and usage count of each cabinet. The weights of nodes that have been frozen for a long time remain unchanged, while the weights of frequently used nodes gradually decrease, allowing the system to automatically adapt to the usage patterns of the cabinets on site.

[0042] Table 1: Comparison of Intelligent Cabinet Operation Indicators under Different Allocation Methods

[0043] The table above records the results of three allocation methods under the same batch of 2000 locker allocation requests. The traditional round-robin allocation method simply allocates lockers in sequential order of their numbers, without considering usage frequency, leading to popular lockers being repeatedly opened. During peak hours, couriers wait an average of 147 seconds for an available locker, with some waiting over 3 minutes in extreme cases. 32.5% of the 200 lockers are consistently idle, meaning approximately 65 lockers are barely used throughout the day. The standard deviation of the daily average number of locker openings reaches 84.3, meaning the busiest lockers are opened over 70 times a day, while the least busy lockers are opened only a few times—a difference of nearly 10 times. Cold start locker response latency is 125 seconds; lockers that haven't been used for several days suddenly require more than 2 minutes to confirm their availability. The method using only the ART network without adding popularity adjustment showed some improvement, reducing queuing time to 98 seconds, idle rate to 23.7%, standard deviation of activation count to 52.6, and cold start response latency to 89 seconds. However, due to the lack of a mechanism to suppress high-frequency cabinets, the proportion of over-limit usage of popular cabinets was still as high as 29%, meaning that about 29 out of every 100 allocation requests still landed on a few popular cabinets. The activation success rate of long-term idle cabinets was only 68%, with nearly one-third of idle cabinets malfunctioning when needed. The method of this invention reduces queuing time to 34 seconds, about 113 seconds less than the traditional method, and nearly three times faster than the method using only the ART network. The cabinet idle rate dropped to 12.1%, with only about 24 of the 200 cabinets in a long-term idle state, and the usage frequency of the other 176 cabinets remaining within a reasonable range. The standard deviation of activation count dropped to 21.8, and the difference in usage frequency among cabinets was reduced to about one-quarter of the original. The cold start cabinet response latency was reduced to 41 seconds, approximately three times faster than traditional methods and more than twice as fast as methods using only ART networks. The percentage of overused popular cabinets decreased from 41% to 14%, a reduction of 27 percentage points. The activation success rate of long-term idle cabinets increased to 92%, meaning that 92 out of every 100 long-term unused cabinets can be activated and used normally when needed. The system's adaptive convergence requires only 140 iterations, 180 fewer than methods using only ART networks, equivalent to saving more than half the learning time.

[0044] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for allocating smart cabinet compartments based on big data collection, characterized in that, Includes the following steps: Collect cabinet task data, perform data preprocessing, and form standardized feature sequences; In the ART network, a weight vector is established for each category node, and the weights are combined to form a category node weight matrix. The similarity of each standardized feature sequence with the category node weight vector is calculated to generate a similarity matrix. Nodes with matching values ​​exceeding a preset threshold in the similarity matrix are marked as active nodes. A list of nodes to be created is generated for unmatched nodes. New category node weight vectors are assigned to the nodes to be created and added to the category node weight matrix. For frozen nodes identified within a continuous time window, the original weights remain unchanged. Only active nodes undergo weight matrix updates, and the corresponding node status and update time are recorded for each update. During the ART network matching process, the number of times the candidate cabinets have been opened and the interval between consecutive openings in the past 24 hours are read. The cabinet popularity reverse adjustment factor is calculated. For cabinets whose number of openings exceeds the threshold and whose consecutive intervals are less than the preset time, the activation weight is adjusted according to the rules to form an activation ranking sequence. Select the target cabinet according to the active sorting sequence, and update the association between the matching category node and the cabinet and the usage count in the ART network; The selected cabinet is marked with status, including occupancy status, cold start priority, and cumulative number of uses, and the allocation history is recorded; Adjust the category node weight matrix and the cabinet activation sequence parameters based on the category node weight matrix, the frozen status, and the cabinet usage count.

2. The intelligent cabinet compartment allocation method based on big data collection according to claim 1, characterized in that, The preprocessing of the cabinet task data specifically includes: The cabinet task data, including request content, cabinet status information and timestamps, undergoes digital preprocessing, which includes field unification, time alignment, anomaly removal, and value normalization.

3. The intelligent cabinet compartment allocation method based on big data collection according to claim 1, characterized in that, The step of establishing a weight vector for each category node in the ART network and marking nodes with matching values ​​exceeding a preset threshold in the similarity matrix as active nodes specifically includes: In the ART network, an initial weight vector is assigned to each category node, and the weight vectors of all category nodes are combined to form a category node weight matrix; The standardized feature sequences are matched one by one with the category node weight vector to generate the category node similarity value corresponding to each feature sequence; Arrange all similarity values ​​according to feature sequence index and category node index to form a similarity matrix; Based on the similarity matrix, nodes with matching values ​​exceeding a preset threshold are marked as active nodes, feature sequences that do not match existing categories are marked as nodes to be created, a list of nodes to be created is generated, new category node weight vectors are assigned to the nodes to be created, and they are added to the category node weight matrix.

4. The intelligent cabinet compartment allocation method based on big data collection according to claim 1, characterized in that, The frozen nodes retain their weights unchanged, and only active nodes participate in the weight matrix update. Recording the node status and update time specifically includes: For the active nodes identified in the ART network, extract the corresponding category node weight vector and its index in the weight matrix. Calculate the weight increment based on the active node matching value, add the increment value to each element of the active node weight vector to form the updated weight vector, perform normalization processing on the updated weight vector to cover the original category node weight vector, and freeze the nodes while keeping the original weight vector unchanged. After the weight matrix is ​​updated, the active node number, weight vector index, and update timestamp are written into the node status record table, the frozen node number and frozen status identifier are written into the node status record table, the most recent update time is recorded, and all node statuses are sorted according to the row order of the weight matrix and the node index order to generate a complete node status table. For each active node, calculate the matching error residual and write the residual value into the corresponding field of the node status table; write the update record and timestamp of the active node weight vector into the weight change log. In the frozen node status table, mark the start time and the freezing identifier. In the node status record table, add a field to record whether the current node participates in this round of weight update and the corresponding weight increment and normalization adjustment coefficient. According to the node index order, merge the information of active nodes, frozen nodes and nodes to be created to generate a total node status snapshot and write it into the weight matrix operation history database. The update magnitude of active node weights in the node status snapshot is calculated to generate a node weight update statistics table, including maximum, minimum, average and variance. The statistics table is written to the database along with the node status snapshot. After each operation, the node status record table is checked for integrity. Abnormal entries are marked with an error identifier and operation time. The timestamp, operation sequence and weight matrix change are serialized and saved.

5. The intelligent cabinet compartment allocation method based on big data collection according to claim 1, characterized in that, The calculation of the cabinet heat reverse adjustment factor, which adjusts the activation weight of cabinets that have been opened more than a threshold and whose consecutive intervals are less than a preset time according to rules, to form an activation sorting sequence, specifically includes: For each cabinet in the candidate cabinet list, count the number of times it has been opened in the past 24 hours; The time interval between two consecutive openings is counted to form a sequence of consecutive opening intervals; The number of times each cabinet is opened is compared with two preset thresholds, and cabinets that exceed the thresholds are marked. For cabinets exceeding the threshold, check the number of times the consecutive opening interval is less than the preset interval time, and calculate the number of times the interval limit is exceeded. Based on the number of times the value is exceeded, the activation weight adjustment coefficient is calculated by reducing it by 5% for each instance of exceeding the limit, with the initial weight set at 100%. The adjustment coefficient is applied to the original activation weight of each eligible cabinet to generate the adjusted activation weight; For cabinets that have not exceeded the threshold for the number of times they can be opened or whose consecutive intervals meet the requirements, the original activation weight remains unchanged. After adjusting the activation values ​​of all cabinets, organize them into an activation value list, sort them from high to low, and generate an activation sorting sequence. Record the cabinet number, original activation weight, adjustment coefficient and adjusted activation weight of each cabinet in the activation sorting sequence to form an activation sorting record table; Index and number the active sorted sequences after they have been sorted.

6. The intelligent cabinet compartment allocation method based on big data collection according to claim 1, characterized in that, The step of selecting the target cabinet according to the activation sorting sequence and updating the association between the matching category node and the cabinet and the usage count in the ART network specifically includes: The cabinet numbers are sequentially obtained from the activated sorting sequence as candidate target cabinets; Check the current occupancy status of each candidate target cabinet according to the activation order, and remove cabinets that are already occupied; For the remaining available cabinets, select the cabinet with the highest activation weight as the target cabinet; In the ART network, the target cabinet is associated with the matching category node, and the correspondence between the category node number and the cabinet number is recorded. The number of times the target cabinet is allocated is accumulated into the cabinet usage count field, and the usage statistics of the corresponding node in the category node weight matrix are updated. Write the association information between category nodes and cabinets into the node-cabinet association record table, including the category node number, cabinet number, activation weight, and update time; The usage count and associated status of the target cabinet are timestamped to form a complete update history. Organize all updated records into a table that matches category nodes with cabinet associations and usage counts, according to the order of the target cabinets.

7. The intelligent cabinet compartment allocation method based on big data collection according to claim 6, characterized in that, The process of marking the status of selected cabinets, including occupancy status, cold start priority, and cumulative usage count, and recording allocation history specifically includes: Extract the cabinet identifier of the target cabinet from the active sorting sequence; assign the occupancy status field of the target cabinet to 1; retrieve the cumulative total number of allocations from the target cabinet storage area, increment the cumulative total number of allocations by 1, and write it back to the cumulative usage count field; collect year, month, day, hour, minute, and second data from the system clock, combine them into a last allocation timestamp, and write it to the last allocation timestamp field; retrieve the continuous unallocated time count value from the target cabinet. If the count value is greater than 72 hours, assign the cold start priority field to 3; if the count value is greater than 24 hours and less than or equal to 72 hours, assign it to 2; otherwise, assign it to 1; combine the occupancy status field value, cold start priority field value, cumulative usage count field value, and last allocation timestamp field value into an allocation record, and append it to the allocation history log file; clear the continuous unallocated time count value of the target cabinet to zero; for each non-selected candidate cabinet, retrieve the continuous unallocated time count value, increment the count value by 1, and write it back.

8. The intelligent cabinet compartment allocation method based on big data collection according to claim 7, characterized in that, The adjustment of the category node weight matrix and cabinet activation sequence parameters based on the category node weight matrix, frozen status, and cabinet usage count specifically includes: Extract the weight vectors of all category nodes from the category node weight matrix; extract the freeze flag of each category node from the frozen status list, skipping the adjustment operation for nodes with a freeze flag of 1; for each category node with a freeze flag of 0, retrieve the cumulative usage count field value of the associated cabinet; input the retrieved cumulative usage count field value into the normalization function, calculating the output value to be in the range of 0 to 1; perform vector multiplication operation between the category node weight vector and the normalization function output value to generate the adjusted weight vector; replace the original weight vector in the category node weight matrix with the adjusted weight vector; extract the continuous unallocated time count value from the associated cabinet of each category node; reduce the basic weight value of the activation sequence corresponding to the cabinet with a continuous unallocated time count value greater than 72 hours by 20%; write the reduced basic weight value of the activation sequence back to the cabinet activation sequence parameter storage area.

9. A smart cabinet compartment allocation system based on big data collection, applied to the smart cabinet compartment allocation method based on big data collection as described in any one of claims 1 to 8, characterized in that, Includes the following modules: The data acquisition and preprocessing module is used to collect cabinet task data, perform data preprocessing, and output standardized feature sequences; the ART network matching module is used to calculate the similarity between the standardized feature sequences and the category node weight vectors to generate a similarity matrix, mark nodes with matching values ​​exceeding a preset threshold as active nodes, and assign new weight vectors to unmatched nodes and add them to the category node weight matrix. The module includes the following components: a weight update module, a cabinet heat adjustment module, and a parameter adjustment module. The weight update module identifies frozen nodes within a continuous time window, maintains their weights, updates the weight matrix for active nodes, and records node status and update time. The cabinet heat adjustment module reads the number of times candidate cabinets have been opened and the consecutive opening intervals over the past 24 hours, calculates the cabinet heat inverse adjustment factor, adjusts the activation weights of cabinets with opening counts exceeding a threshold and consecutive intervals less than a preset time, and outputs an activation sorting sequence. The target cabinet selection and association update module selects target cabinets according to the activation sorting sequence, updates the association between matching category nodes and cabinets in the ART network, and updates the usage counts. The status marking module marks the occupancy status, cold start priority, and cumulative usage counts of selected cabinets, and records the allocation history. The parameter adjustment module adjusts the category node weight matrix and cabinet activation sequence parameters based on the category node weight matrix, frozen status, and cabinet usage counts.

Images