Knowledge graph-based rental operation management method for dismountable mobile cold storage

Abstract

The application provides a knowledge graph-based rental operation management method for detachable motor-driven cold storage, relates to the field of operation management, and comprises the following steps: determining a plurality of rental cycle units based on historical rental orders of a plurality of rental nodes, and constructing an operation knowledge graph based on the plurality of rental cycle units and the historical rental orders of the plurality of rental nodes; predicting the number of future rental orders of the plurality of rental cycle units based on the operation knowledge graph through a dynamic perception intensity butterfly optimization algorithm; and determining the initial number of detachable motor-driven cold storages of each rental node according to the number of future rental orders of the plurality of rental cycle units, the historical rental orders of the plurality of rental nodes and the operation knowledge graph, and also used for allocating rental nodes for detachable motor-driven cold storage rental requests and return requests initiated by a user end, and has the advantage of improving the quality of rental operation management of detachable motor-driven cold storage.

Description

Technical Field

[0001] This invention relates to the field of operations management, and in particular to a knowledge graph-based method for the leasing operations management of detachable mobile cold storage facilities. Background Technology

[0002] With the rapid development of fresh food e-commerce and cold chain logistics, detachable mobile cold storage units, as flexible and mobile cold chain storage devices, are increasingly in demand for applications in last-mile delivery, temporary warehousing, and event support. Traditional cold storage rental management typically adopts a fixed warehouse model, where cold storage facilities are deployed at fixed locations, requiring users to pick up and return them themselves. This approach suffers from limited coverage, poor scheduling flexibility, and low equipment utilization. In recent years, some companies have begun exploring deploying cold storage facilities at multiple rental nodes (such as community stations, shopping mall parking lots, and subway stations), allowing users to rent and return cold storage units at the nearest node, significantly improving convenience.

[0003] However, this multi-node distributed leasing model also brings new challenges: traditional forecasting methods struggle to accurately capture the interrelationships in demand among multiple nodes, resulting in some nodes having idle cold storage while others are experiencing supply shortages; furthermore, existing leasing management methods are mostly based on rule engines or simple optimization algorithms for cold storage allocation and scheduling, failing to comprehensively consider multi-dimensional information such as the flow relationships between nodes, historical order patterns, and future demand forecasts, leading to a lack of global optimality in scheduling decisions and a tendency to get stuck in local optima; in addition, existing technologies cannot effectively record and infer deep knowledge such as the flow relationships between nodes and the operational characteristics of cyclical units, making it difficult for the system to achieve intelligent demand forecasting and resource allocation.

[0004] Therefore, there is a need to provide a knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities to address issues such as inaccurate demand forecasting and suboptimal resource allocation in existing technologies. Summary of the Invention

[0005] This invention provides a knowledge graph-based method for the leasing operation and management of detachable mobile cold storage facilities, comprising: determining multiple leasing cycle units based on historical leasing orders from multiple leasing nodes; constructing an operation knowledge graph based on the multiple leasing cycle units and the historical leasing orders from multiple leasing nodes, wherein each leasing cycle unit includes at least one leasing node, and the operation knowledge graph is used to record the leasing operation relationships of the multiple leasing cycle units and the multiple leasing nodes; predicting the number of future leasing orders for the multiple leasing cycle units based on the operation knowledge graph using a butterfly optimization algorithm with dynamic sensing intensity; determining the initial number of detachable mobile cold storage units for each leasing node based on the number of future leasing orders for the multiple leasing cycle units, the historical leasing orders from multiple leasing nodes, and the operation knowledge graph; allocating leasing nodes for detachable mobile cold storage leasing requests initiated by users, and also allocating leasing nodes for detachable mobile cold storage return requests initiated by users.

[0006] Furthermore, based on historical rental orders from multiple rental nodes, multiple rental cycle units are determined, including: for any two rental nodes, calculating the interaction coefficient between the rental nodes that rented out and the rental nodes that returned for each historical rental order; and determining multiple rental cycle units based on the interaction coefficient between any two rental nodes using a nested loop clustering algorithm.

[0007] Furthermore, using a nested loop clustering algorithm, multiple rental loop units are determined based on the interaction coefficient between any two rental nodes, including: S101, filtering out isolated rental nodes from the multiple rental nodes based on the interaction coefficient between any two rental nodes; S102, initializing the number of cluster centers N to 3 and the number of iterations T to 1, and determining initial cluster centers from the multiple rental nodes based on the interaction coefficient between any two rental nodes; S103, constructing a cluster list, where the cluster list is used to record rentals that are not clustered and are not cluster centers. Node; S104. From the clustering list, extract a rental node that is not clustered and is not a cluster center. Based on the interaction coefficient between the rental node that is not clustered and is not a cluster center and each rental node included in each rental cycle unit corresponding to the current iteration, determine the interaction coefficient between the rental node that is not clustered and is not a cluster center and each rental cycle unit corresponding to the current iteration. Determine whether the rental node that is not clustered and is not a cluster center should be clustered. If so, assign the rental node that is not clustered and is not a cluster center to the rental cycle unit with the largest interaction coefficient. Remove from the clustering list, proceed to S106; otherwise, proceed to S105. S105: Retain the rental nodes that are not clustered and are not cluster centers in the clustering list, proceed to S104. S106: Determine if all rental nodes that are not clustered and are not cluster centers have been clustered; if yes, proceed to S107; otherwise, proceed to S104. S107: For each rental cycle unit corresponding to the current iteration, determine the center rental node of the rental cycle unit based on the interaction coefficient between any two rental nodes in the rental cycle unit. S108: Root... Based on the central rental node of each rental cycle unit, determine whether the inner loop is complete. If not, proceed to S109; if yes, proceed to S110. S109: Based on the central rental node of each rental cycle unit, update the cluster center and cluster list, and proceed to S103. S110: Determine whether the outer loop is complete. If yes, determine multiple rental cycle units based on the multiple rental cycle units corresponding to multiple consecutive iterations. If not, proceed to S111. S111: Let N=N+1, T=T+1, update the cluster center and cluster list, and proceed to S104.

[0008] Further, determining the central leasing node of the leasing cycle unit includes: for each leasing node included in the leasing cycle unit, calculating the central value of the leasing node based on the interaction coefficient between the leasing node and each other leasing node included in the leasing cycle unit; and determining the central leasing node of the leasing cycle unit based on the central value of each leasing node.

[0009] Furthermore, based on historical rental orders from multiple rental cycle units and multiple rental nodes, an operational knowledge graph is constructed, including: for each rental cycle unit, determining the number of orders for the rental cycle unit in multiple historical periods based on the historical rental orders of the rental nodes included in the rental cycle unit; for any two rental cycle units, calculating the order correlation coefficient between the two rental cycle units based on the order numbers of the two rental cycle units in multiple historical periods; and constructing the operational knowledge graph based on the order correlation coefficient between any two rental cycle units and the interaction coefficient between any two rental nodes.

[0010] Furthermore, using a butterfly optimization algorithm with dynamic perception intensity, based on an operational knowledge graph, the number of future rental orders for multiple rental cycle units is predicted. This includes: for each rental cycle unit, determining the range of future rental orders for the rental cycle unit based on the historical rental orders of the rental nodes included in the rental cycle unit; initializing the population based on the range of future rental orders for each rental cycle unit; constructing a fitness function; and iteratively optimizing the population based on the fitness function to determine the number of future rental orders for multiple rental cycle units. In each iteration, the dynamic perception intensity corresponding to the current iteration is determined.

[0011] Furthermore, the fitness function is distance-dependent on the global correlation coefficient corresponding to the butterfly.

[0012] Further, determining the dynamic perception intensity corresponding to the current iteration includes: for each butterfly in the current iteration, calculating the correlation coefficient of future orders between any two rental cycle units corresponding to the butterfly, and constructing the correlation coefficient vector corresponding to the butterfly; for any two butterflies in the current iteration, calculating the distance between the correlation coefficient vectors corresponding to the two butterflies; calculating the distance between the correlation coefficient vectors corresponding to the globally optimal butterflies in two adjacent iterations within the current iteration window; and determining the dynamic perception intensity corresponding to the current iteration based on the distance between the correlation coefficient vectors corresponding to any two butterflies in the current iteration and the distance between the correlation coefficient vectors corresponding to the globally optimal butterflies in two adjacent iterations within the current iteration window.

[0013] Furthermore, based on the number of future lease orders for multiple lease cycle units, the historical lease orders for multiple lease nodes, and the operational knowledge graph, the initial number of detachable mobile cold storage units for each lease node is determined, including: for each lease cycle unit, based on the operational knowledge graph, determining the interaction weight of each lease node included in the lease cycle unit; based on the historical lease orders of each lease node included in the lease cycle unit, determining the order ratio of each lease node; and based on the number of future lease orders for the lease cycle unit and the interaction weight and order ratio of each lease node included in the lease cycle unit, determining the initial number of detachable mobile cold storage units for each lease node included in the lease cycle unit.

[0014] Further, determining the interaction weights of the rental nodes included in the rental cycle unit includes: for each rental node included in the rental cycle unit, determining the interaction weight of the rental node based on the average of the interaction coefficients between the rental node and each other rental node included in the rental cycle unit.

[0015] Compared with existing technologies, the knowledge graph-based leasing operation and management method for detachable mobile cold storage provided by this invention has at least the following beneficial effects: 1. Based on the pairing relationship between rental nodes and return nodes in historical rental orders, the interaction coefficient of any two rental nodes can be calculated, which can accurately depict the linkage characteristics between nodes in the actual business process. Compared with traditional clustering methods based on geographical distance or simple statistics, the interaction coefficient more realistically reflects the business relationship of cold storage flowing between nodes.

[0016] 2. The nested loop clustering algorithm dynamically adjusts the number of cluster centers in the outer loop and iteratively optimizes the position of cluster centers in the inner loop. In each inner loop, clustering is determined based on the average interaction coefficient between nodes and each loop unit. This effectively avoids the problems of traditional clustering algorithms, such as sensitivity to initial centers and susceptibility to local optima, resulting in a more reasonable and stable division of rental loop units. Simultaneously, by calculating the correlation coefficient of order quantity between any two rental loop units across multiple historical periods and combining it with the interaction coefficients between nodes to construct an operational knowledge graph, a multi-level relationship network covering loop units and nodes is formed, providing effective data support for rental operation management.

[0017] 3. The dynamic sensing intensity is adaptively adjusted based on the ratio of the mean vector distance of the correlation coefficient between butterflies in the population to the mean global optimal movement distance. When the population is dispersed and the global optimum is stable, the sensing intensity is increased to accelerate convergence. When the population is clustered and the global optimum moves quickly, the sensing intensity is decreased for fine-grained search. This effectively overcomes the problem of premature convergence or low search efficiency caused by the fixed sensing intensity of traditional algorithms, and significantly improves the accuracy and speed of future order prediction. 4. The determination of the initial number of cold storage units takes into account the interaction weight of nodes and the order ratio. The interaction weight is obtained by normalizing the mean of the interaction coefficients between nodes in the knowledge graph, which fully reflects the business relationship and spatial linkage effect between nodes. The order ratio reflects the objective distribution of historical demand. The combination of the two makes the resource allocation both in line with historical patterns and take into account the characteristics of node association.

[0018] 5. The order correlation coefficients between cyclic units and the interaction coefficients between nodes recorded in the knowledge graph provide a global optimization basis for the node allocation of leasing and return requests, realize cross-unit load balancing and efficient resource flow between nodes, and significantly improve the turnover efficiency and overall utilization rate of the detachable mobile cold storage. Attached Figure Description

[0019] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is a flowchart illustrating a knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities, according to some embodiments of this specification. Figure 2 This is a flowchart illustrating a nested loop clustering algorithm according to some embodiments of this specification; Figure 3 This is a schematic diagram of an operational knowledge graph according to some embodiments of this specification. Detailed Implementation

[0020] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this specification. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0021] Figure 1 This is a flowchart illustrating a knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities, as shown in some embodiments of this specification. Figure 1 As shown, a knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities may include the following steps.

[0022] S1. Based on historical rental orders from multiple rental nodes, identify multiple rental cycle units. Based on the multiple rental cycle units and historical rental orders from multiple rental nodes, construct an operational knowledge graph.

[0023] The rental cycle unit includes at least one rental node, and the operation knowledge graph is used to record the rental operation relationships of multiple rental cycle units and multiple rental nodes.

[0024] In some embodiments, multiple rental cycle units are determined based on historical rental orders from multiple rental nodes, including: For any two rental nodes, calculate the interaction coefficient between the two rental nodes based on the rental node where the rental was rented and the rental node where the rental was returned for each historical rental order. By using a nested loop clustering algorithm, multiple rental loop units are determined based on the interaction coefficient between any two rental nodes.

[0025] Specifically, the interaction coefficient between two leasing nodes is used to quantify the closeness of bidirectional flow between the two leasing nodes in historical leasing activities.

[0026] For each historical period (e.g., one day, one week, one month, etc.) and two rental nodes, calculate the sum of the number of historical rental orders corresponding to the rental nodes of the two rental nodes within that historical period, which is either one of the two rental nodes, as the first total number. Calculate the sum of the number of historical rental orders corresponding to the rental nodes of the two rental nodes, which is either one of the two rental nodes, and the sum of the number of historical rental orders corresponding to the rental nodes of the other of the two rental nodes, as the second total number. Calculate the ratio of the second total number to the first total number, which is used as the interaction coefficient between the two rental nodes for that historical period.

[0027] For any two rental nodes, calculate the average of the interaction coefficients of the two rental nodes across multiple historical periods, and use this average as the interaction coefficient between the two rental nodes.

[0028] For example, suppose the data for rental nodes A and B over the past three weeks is as follows: In the first week, there were 200 orders with A or B as the rental or return node, of which 80 were two-way orders (A rents, B returns; B rents, A returns), with a periodic interaction coefficient of 80 / 200 = 0.4; in the second week, there were 250 related orders, with 100 two-way orders, and a periodic interaction coefficient of 100 / 250 = 0.4; in the third week, there were 300 related orders, with 120 two-way orders, and a periodic interaction coefficient of 120 / 300 = 0.4. Therefore, the interaction coefficient between rental node A and rental node B is (0.4 + 0.4 + 0.4) / 3 = 0.4.

[0029] Figure 2 This is a flowchart illustrating a nested loop clustering algorithm according to some embodiments of this specification, such as... Figure 2 As shown, in some embodiments, a nested loop clustering algorithm is used to determine multiple rental loop units based on the interaction coefficient between any two rental nodes, including: S101. Based on the interaction coefficient between any two rental nodes, isolate rental nodes are filtered out from multiple rental nodes. The remaining rental nodes are subjected to subsequent clustering processes, and each isolated rental node is treated as a rental cycle unit. For example, if the interaction coefficient between a certain rental node and each other rental node is less than the first interaction coefficient threshold (e.g., 0.1), then the rental node is treated as an isolated rental node. S102. Initialize the number of cluster centers N to 3 and the number of iterations T to 1. Based on the interaction coefficient between any two rental nodes, determine the initial cluster centers from multiple rental nodes. For example, for each rental node, calculate the variance of the interaction coefficient between the rental node and any other rental node. Sort the multiple rental nodes according to the variance from largest to smallest, and select the three rental nodes with the highest ranking as the initial cluster centers. S103. Construct a cluster list, whereby the cluster list is used to record rental nodes that are not clustered and are not cluster centers; S104. From the clustering list, extract a rental node that is not clustered and is not a cluster center. Based on the interaction coefficient between the rental node that is not clustered and is not a cluster center and each rental node included in each rental cycle unit corresponding to the current iteration, determine the interaction coefficient between the rental node that is not clustered and is not a cluster center and each rental cycle unit corresponding to the current iteration. Determine whether the rental node that is not clustered and is not a cluster center should be clustered. If yes, assign the rental node that is not clustered and is not a cluster center to the rental cycle unit with the largest interaction coefficient and delete it from the clustering list, then proceed to S106. If no, proceed to S105. Specifically, for each rental node that is not clustered and is not a cluster center, take the maximum value of the interaction coefficient between the rental cycle unit and each rental node included in the rental cycle unit as the value of the interaction coefficient between the rental cycle unit and each rental node included in the rental cycle unit. The interaction coefficient between the rental node that is not clustered and is not a cluster center and at least one rental cycle unit corresponding to the current iteration is determined to be greater than the second interaction coefficient threshold (e.g., 0.2, 0.3, etc.). For example, if there are two rental cycle units C1 and C2 in the current iteration, and the interaction coefficients between node X and three nodes in C1 are 0.25, 0.15, and 0.35, and the interaction coefficients between node X and three nodes in C2 are 0.10, 0.08, and 0.12, then the interaction coefficient between node X and C1 is max(0.25, 0.15, 0.35) = 0.35, and the interaction coefficient between node X and C2 is max(0.10, 0.08, 0.12) = 0.12. Let the threshold for the second interaction coefficient be 0.3. Since 0.35 > 0.3 and 0.35 > 0.12, we determine that node X should be clustered and assigned to C1. S105. Retain the rental nodes that are not clustered and are not cluster centers in the clustering list, and execute S104; S106. Determine whether all rental nodes that have not been clustered and are not cluster centers have been clustered. If yes, proceed to S107; otherwise, proceed to S104. S107. For each rental cycle unit corresponding to the current iteration, determine the central rental node of the rental cycle unit based on the interaction coefficient between any two rental nodes in the rental cycle unit. S108. Based on the rental node at the center of each rental cycle unit, determine whether the inner loop is completed. If not, proceed to S109. If yes, proceed to S110. Specifically, when the rental node at the center of each rental cycle unit is the cluster center of that rental cycle unit, it is determined that the inner loop is completed. S109. Based on the central rental node of each rental cycle unit, update the cluster center and cluster list, and execute S103. Specifically, update the cluster center of each rental cycle unit to the central rental node, and update the cluster list at the same time, recording each rental node that is not a cluster center to the cluster list. S110. Determine whether the outer loop is complete. If yes, determine multiple rental loop units based on the multiple rental loop units corresponding to multiple consecutive iterations. If no, execute S111. Specifically, when N is greater than the iteration number threshold (e.g., 5, 6, 7, etc.), for each rental loop unit corresponding to each iteration, calculate the average of the interaction coefficients of any two rental nodes included in the rental loop unit as the average of the interaction coefficients of the rental loop unit. For each iteration, calculate the average of the average of the interaction coefficients of each rental loop unit corresponding to the iteration as the average of the interaction coefficients corresponding to the iteration. The iteration with the largest average interaction coefficient is taken as the optimal iteration, and the multiple rental loop units corresponding to the optimal iteration are taken as the final determined multiple rental loop units. S111. Set N=N+1 and T=T+1, update the cluster centers and cluster list, and execute S104.

[0030] Specifically, firstly, quantifying the bidirectional rental flow between rental nodes using interaction coefficients accurately reflects the actual strength of the association between them. Compared to traditional clustering methods based solely on geographical distance or order quantity, this approach better reflects users' actual rental behavior patterns and the dynamic supply and demand relationship between rental nodes, thereby improving the business relevance and practicality of the clustering results. Secondly, pre-screening isolated nodes using a first interaction coefficient threshold avoids interference from low-association rental nodes, ensuring that subsequent clustering focuses on groups of rental nodes with substantial interaction relationships, thus improving the purity and effectiveness of the clustering. Thirdly, the initial cluster centers are selected based on the variance ranking of the interaction coefficients, prioritizing rental nodes with significant differences in interaction with other rental nodes. These nodes are often located at the intersection of multiple rental flows, possessing stronger representativeness, thus providing a high-quality initial clustering foundation for subsequent iterations, accelerating algorithm convergence, and reducing the risk of getting trapped in local optima. Furthermore, the algorithm employs a nested double-loop structure. The outer loop explores clustering schemes of different granularities by gradually increasing the number of cluster centers, while the inner loop continuously adjusts node affiliation and center position through iterative optimization with a fixed number of centers. Working together, the two loops adaptively discover the optimal number and partitioning method in the data, avoiding subjective biases caused by pre-specifying the number of clusters. Simultaneously, the node allocation process uses a dual judgment mechanism based on the maximum interaction coefficient and a second interaction coefficient threshold. This ensures that nodes are assigned to the most relevant cyclic units while preventing forced classification of weakly associated nodes through threshold filtering, enhancing the robustness of the clustering results. Finally, by calculating the average interaction coefficient across multiple iterations and selecting the iteration corresponding to the maximum value as the optimal result, the algorithm effectively avoids unstable clustering that might occur in a single iteration. This ensures that the final output rental cyclic units have the highest internal consistency and external discriminability, providing a more reliable and accurate basis for subsequent resource scheduling, inventory optimization, and operational decisions.

[0031] In some embodiments, determining the rental node at the center of the rental cycle unit includes: For each lease node included in the lease cycle unit, the center value of the lease node is calculated based on the interaction coefficient between the lease node and each other lease node included in the lease cycle unit; Based on the center value of each lease node, determine the center lease node of the lease cycle unit.

[0032] Specifically, for each rental node in a rental cycle unit, the mean of the interaction coefficients between that rental node and every other rental node in the rental cycle unit is calculated. This mean is used as the center value of the rental node. The rental node with the largest center value is designated as the center rental node of the rental cycle unit. For example, a rental cycle unit contains three rental nodes A, B, and C. Their pairwise interaction coefficients are: A to B = 0.3, A to C = 0.5, and B to C = 0.4. The mean value of node A is (0.3 + 0.5) / 2 = 0.4, the center value of rental node B is (0.3 + 0.4) / 2 = 0.35, and the mean value of node C is (0.5 + 0.4) / 2 = 0.45. Since the center value of rental node C is the largest (0.45), rental node C is determined as the center rental node of this rental cycle unit.

[0033] In some embodiments, an operational knowledge graph is constructed based on historical rental orders from multiple rental cycle units and multiple rental nodes, including: For each rental cycle unit, the number of orders for the rental cycle unit in multiple historical periods is determined based on the historical rental orders of the rental nodes included in the rental cycle unit. For any two rental cycle units, calculate the order correlation coefficient between the two rental cycle units based on the order quantity of the two rental cycle units in multiple historical periods; Construct an operational knowledge graph based on the order correlation coefficient between any two rental cycle units and the interaction coefficient between any two rental nodes.

[0034] Specifically, for each rental cycle unit, the number of historical rental orders generated by all rental nodes within that unit in each historical period (e.g., one day, one week, one month) is first counted. This yields the order quantity sequence for that rental cycle unit across multiple historical periods, reflecting the order activity and fluctuation trend of the rental cycle unit over time. For example, if a rental cycle unit contains three rental nodes, and the order numbers for each node in the last four weeks are: 120 orders in the first week, 150 orders in the second week, 130 orders in the third week, and 160 orders in the fourth week, then the order quantity sequence for that rental cycle unit is {120, 150, 130, 160}. Next, for any two rental cycle units, based on their respective order quantity sequences across multiple historical periods, the correlation coefficient between the two sequences is calculated, serving as the order correlation coefficient between the two rental cycle units. The correlation coefficient can be calculated using Pearson correlation coefficient, Spearman rank correlation coefficient, etc., and its value ranges from [-1, 1]. The closer the value is to 1, the more consistent the order volume change trends of the two rental cycle units are, that is, they are at the same order peak or trough in the same cycle, indicating that the two have strong synergy in business. The closer the value is to -1, the more opposite the order volume change trends are, with one peaking and the other troughing. A value close to 0 indicates that there is no obvious correlation between the order volume changes of the two.

[0035] The operational knowledge graph is used to record the order correlation coefficient between any two rental cycle units, the affiliation relationship between rental nodes and rental cycle units, and the interaction coefficient between any two rental nodes included in a rental cycle unit. Figure 3 This is a schematic diagram of an operational knowledge graph based on some embodiments of this specification, such as... Figure 3As shown, in some embodiments, the operational knowledge graph organizes data in the form of triples (subject, relation, object), which include three core triple types. The first type is the node affiliation triple, in the form of (rental node, belongs to, rental cycle unit), for example (rental node A, belongs to, cycle unit C1), which is used to explicitly record the rental cycle unit to which each rental node is currently clustered. This is the basis for all subsequent association analysis, making the affiliation relationship of each rental node clear at a glance, and facilitating quick querying of which nodes are contained in a certain rental cycle unit, or which rental cycle unit a certain rental node belongs to. The second type is the cyclic unit association triple, in the form of (leased cyclic unit C1, order correlation coefficient r, leased cyclic unit C2), for example (cyclic unit C1, order correlation coefficient 0.95, cyclic unit C2). Here, r is the correlation coefficient calculated from the order quantity sequence of the two cyclic units over multiple historical periods. This triple characterizes the degree of synchronization of the order quantity changes of the two cyclic units in the time dimension. The closer r is to 1, the more consistent the order fluctuations of the two cyclic units are, indicating strong synergy. The closer r is to -1, the more opposite the fluctuations are, indicating complementarity. When r is close to 0, it indicates that there is no obvious correlation between the two. This information is of great guiding significance for coordinated scheduling and off-peak operation. The third type is the node interaction triple, in the form of (rental node A, interaction coefficient s, rental node B), where A and B belong to the same rental cycle unit, and s is the average interaction coefficient of the two nodes over multiple historical periods, for example (rental node A, interaction coefficient 0.45, rental node B). This triple characterizes the closeness of bidirectional rental flow between two rental nodes within the same cycle unit. The larger s is, the more frequent the user interactions and the stronger the association between the two nodes. It can be used to identify core node pairs and key flow paths within a unit. Through the organic combination of these three types of triples, the operational knowledge graph not only records the hierarchical affiliation from rental nodes to cycle units, but also preserves the horizontal business relationships between cycle units and the micro-interaction details between nodes within a cycle unit, forming a multi-level, multi-dimensional operational relationship network.

[0036] S2. Using a butterfly optimization algorithm with dynamic intensity perception, based on an operational knowledge graph, predict the number of future rental orders for multiple rental cycle units.

[0037] Specifically, it includes: For each rental cycle unit, the range of future rental orders for the rental cycle unit is determined based on the historical rental orders of the rental nodes included in the rental cycle unit; Based on the range of future rental orders for each rental cycle unit, a population is initialized. Each butterfly in the population represents the number of future rental orders for a set of rental cycle units. When generating each butterfly, samples are taken from the range of future rental orders for each rental cycle unit to form the butterfly. Future rental orders can correspond to at least one future cycle. For example, assuming there are three rental cycle units C1, C2, and C3, their future order quantity ranges are determined as follows: C1 ∈ [120, 180] singles, C2 ∈ [90, 150] singles, and C3 ∈ [200, 260] singles. When initializing the population, the population size is set to 10 butterflies. Each butterfly is a three-dimensional vector, with each dimension corresponding to a predicted future order value for a rental cycle unit. When generating the first butterfly, 145 is randomly sampled from the range [120, 180] in C1, 112 is randomly sampled from the range [90, 150] in C2, and 233 is randomly sampled from the range [200, 260] in C3. The first butterfly is represented as (145, 112, 233). When generating the second butterfly, 167, 138, and 215 are sampled from the three ranges respectively. The second butterfly is represented as (167, 138, 215). This process is repeated to generate the remaining eight butterflies. Construct a fitness function, where the fitness function is related to the distance of the global correlation coefficient corresponding to the butterfly; Based on the fitness function, the population is iteratively optimized to determine the number of future rental orders for multiple rental cycle units. In each iteration, the dynamic sensing intensity corresponding to the current iteration is determined.

[0038] Specifically, for each rental cycle unit, the number of historical rental orders generated by all rental nodes within the rental cycle unit in each historical period (e.g., one day, one week, one month, etc.) is counted, thereby obtaining the order quantity sequence of the rental cycle unit over multiple historical periods. The maximum, minimum, and standard deviation of the order quantity sequence of the rental cycle unit over multiple historical periods are determined. Based on the maximum, minimum, and standard deviation of the order quantity sequence of the rental cycle unit over multiple historical periods, the range of future rental order quantities for the rental cycle unit is determined. For example, the range of future rental order quantities for the rental cycle unit is [,]. ,in, For the minimum quantity, For the maximum quantity, The standard deviation is denoted as .

[0039] For example, the global correlation coefficient distance for a butterfly can be calculated using the following process: For each rental cycle unit, the order quantity sequence of the rental cycle unit in multiple historical periods is concatenated with the number of future rental orders of the rental cycle unit to form the concatenated order quantity sequence corresponding to the rental cycle unit; For any two rental cycle units, calculate the correlation coefficient of the spliced ​​order quantity sequence corresponding to the two rental cycle units, and use it as the future order correlation coefficient of the two rental cycle units; Construct a first correlation coefficient vector based on the order correlation coefficients of any two rental cycle units; Construct the correlation coefficient vector corresponding to the butterfly based on the correlation coefficient of future orders between any two rental cycle units; The Euclidean distance between the first correlation coefficient vector and the correlation coefficient vector corresponding to the butterfly is calculated as the global correlation coefficient distance corresponding to the butterfly.

[0040] The smaller the global correlation coefficient distance for a butterfly, the larger the fitness function value. For example, the fitness function is F=1 / (1+D), where F is the fitness value and D is the global correlation coefficient distance for the butterfly.

[0041] In some embodiments, determining the dynamic sensing intensity corresponding to the current iteration includes: For each butterfly in the current iteration, calculate the correlation coefficient of future orders between any two rental cycle units corresponding to the butterfly, and construct the correlation coefficient vector corresponding to the butterfly; For any two butterflies in the current iteration, calculate the distance between the correlation coefficient vectors of the two butterflies. For example, calculate the Euclidean distance between the correlation coefficient vectors of the two butterflies. Calculate the distance between the correlation coefficient vectors of the globally optimal butterflies in two adjacent iterations within the current iteration window (e.g., 3 iterations, 5 iterations, etc.); The dynamic sensing intensity for the current iteration is determined by the distance between the correlation coefficient vectors of any two butterflies in the current iteration and the distance between the correlation coefficient vectors of the globally optimal butterflies in the two adjacent iterations within the current iteration window.

[0042] Specifically, the mean distance between the correlation coefficient vectors of any two butterflies in the current iteration is used as the mean distance for the current iteration. Similarly, the mean distance between the correlation coefficient vectors of the globally optimal butterflies in two adjacent iterations within the current iteration window is used as the mean distance for the current iteration window. A larger mean distance for the current iteration and a smaller mean distance for the current iteration window indicate a stronger dynamic perception intensity. The mean distance for the current iteration reflects the spatial dispersion of all butterflies in the current population. A larger mean distance indicates greater differences in order allocation schemes among individual butterflies, a dispersed population distribution, and the algorithm still in the global exploration stage, not yet converged to a high-quality region. Stronger perception intensity is needed to guide butterflies to expand their search range and quickly gather towards the potential optimal region. The mean distance for the current iteration window reflects the movement trend of the globally optimal solution in recent iterations. A smaller mean distance indicates that the correlation coefficient vector of the globally optimal butterfly changes slowly within the window period, meaning the optimal solution has basically stabilized near a certain region, and the algorithm is close to convergence. At this point, increasing the perception intensity can encourage butterflies to actively move towards this stable region, accelerating the convergence speed. Combining these two approaches, when the population is dispersed (requiring strong exploration) and the global optimum is stable (with a clear direction), increasing the perception intensity can maintain search breadth while accelerating aggregation towards high-quality areas, achieving an efficient balance between exploration and development. Conversely, when the population is aggregated and the global optimum is moving rapidly, the perception intensity should be reduced to avoid missing the optimal solution due to excessive strides, and a more refined local search should be adopted.

[0043] For example, the formula for calculating the dynamic sensing intensity corresponding to the current iteration is: in, The dynamic sensing intensity corresponding to the current iteration. This is the scaling factor (e.g., 10, etc.). This represents the average distance for the current iteration. The mean distance corresponding to the current iteration window. It is a very small positive number (e.g., 0.01).

[0044] First, by integrating historical rental orders, node information, and business relationships across various rental cycle units through an operational knowledge graph, rich structured knowledge support is provided for demand forecasting. This allows forecasting to move beyond single-dimensional time series analysis and incorporate deep correlation features from multi-source heterogeneous data, significantly improving the comprehensiveness and accuracy of predictions. Second, the fitness function is correlated with the distance to the global correlation coefficient corresponding to the butterfly pattern, transforming the order quantity prediction problem into an optimal solution search problem within the correlation coefficient space. This ensures that the prediction results not only focus on the order quantity itself but also consider the rationality and consistency of order distribution across rental cycle units, avoiding contradictions in prediction results between different rental cycle units. Furthermore, by calculating the mean value of the correlation coefficient vector distance between butterflies within the current iteration population to reflect the degree of population dispersion, and combining it with the mean value of the correlation coefficient vector distance between the globally optimal butterflies within the iteration window to reflect the trend of the optimal solution's movement, the ratio between the two adaptively adjusts the perception intensity: when the population is dispersed and the globally optimal solution is stable, the perception intensity is increased to accelerate the aggregation towards high-quality areas; when the population is aggregated and the globally optimal solution is moving rapidly, the perception intensity is decreased to avoid overtaking the optimal solution. This mechanism enables the algorithm to intelligently balance between global exploration and local development, effectively overcoming the problems of premature convergence or low search efficiency caused by fixed or simple linear decay of perception intensity in traditional butterfly optimization algorithms, and significantly improving the algorithm's convergence speed and prediction accuracy.

[0045] S3. Based on the number of future lease orders for multiple lease cycle units, the historical lease orders for multiple lease nodes, and the operational knowledge graph, determine the initial number of detachable mobile cold storage units for each lease node.

[0046] Specifically, it includes: For each leasing cycle unit, based on the operational knowledge graph, the interaction weight of each leasing node included in the leasing cycle unit is determined. Based on the historical leasing orders of each leasing node included in the leasing cycle unit, the order ratio of each leasing node is determined. Based on the number of future leasing orders of the leasing cycle unit and the interaction weight and order ratio of each leasing node included in the leasing cycle unit, the initial number of detachable mobile cold storage units included in the leasing cycle unit is determined.

[0047] Specifically, the total number of historical rental orders for each rental node within the rental cycle unit is calculated as the total number of historical rental orders for the rental cycle unit. For each rental node, the ratio of its historical rental orders to the total number of historical rental orders for the rental cycle unit is calculated as the order ratio for the rental node.

[0048] In some embodiments, determining the interaction weights of the lease nodes included in the lease cycle unit includes: For each lease node included in the lease cycle unit, the interaction weight of the lease node is determined based on the average of the interaction coefficients between the lease node and each other lease node included in the lease cycle unit.

[0049] Specifically, for each leasing node within a leasing cycle unit, the average interaction coefficient between that leasing node and every other leasing node within the leasing cycle unit is taken as the average interaction coefficient of that leasing node. The ratio of this average interaction coefficient to the sum of the average interaction coefficients of all leasing nodes within the leasing cycle unit is taken as the interaction weight of the leasing node. The introduction of interaction weights allows subsequent order allocation to fully consider the business relationships between nodes. Nodes with higher interaction weights receive more resource allocation in order allocation, thus making the initial configuration of the number of detachable mobile cold storage units for each node more consistent with the spatial distribution characteristics of actual business needs.

[0050] For each leasing node, the sum of the order ratio and interaction weight of the leasing node is used as the comprehensive weight of the leasing node. The ratio of the comprehensive weight of the leasing node to the sum of the comprehensive weights of all leasing nodes included in the leasing cycle unit is used as the proportion coefficient of the leasing node. The initial number of detachable mobile cold storage units for the leasing node is obtained by rounding down the product of the number of future leasing orders in the leasing cycle unit and the proportion coefficient of the leasing node.

[0051] The calculation of interaction weights is based on the average interaction coefficients between nodes in the operational knowledge graph. This fully captures the business relationships and spatial linkages between nodes, ensuring that resource allocation not only relies on historical data but also incorporates the topological relationships between nodes, avoiding configuration biases caused by allocating resources solely based on historical orders. Secondly, the order ratio directly reflects the historical demand contribution share of each node, guaranteeing the data objectivity of the allocation scheme. The two are summed and normalized to obtain the ratio coefficient, achieving an organic integration of historical demand patterns and node association characteristics. This respects the statistical regularities of historical data while also taking into account the business linkage characteristics between nodes, making the initial cold storage configuration more closely aligned with the spatial distribution characteristics of the actual operational scenario.

[0052] S4 is used to allocate rental nodes for user-initiated detachable mobile cold storage rental requests, and also to allocate rental nodes for user-initiated detachable mobile cold storage return requests.

[0053] Specifically, when a user initiates a rental request, the target rental cycle unit is first determined based on the number of requests and timeliness requirements. For example, the rental cycle unit to which the nearest rental node belongs is designated as the target rental cycle unit, and the rental node within the target rental cycle unit that is closest to and has available detachable mobile cold storage is designated as the rental node to be assigned to the detachable mobile cold storage rental request. Similarly, the rental node within the target rental cycle unit that is closest to and has available storage location of the detachable mobile cold storage is designated as the rental node to be assigned to the detachable mobile cold storage return request.

[0054] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A knowledge graph-based method for the leasing, operation, and management of detachable mobile cold storage facilities, characterized in that: include: Based on historical rental orders from multiple rental nodes, multiple rental cycle units are identified. Based on the historical rental orders from multiple rental cycle units and multiple rental nodes, an operational knowledge graph is constructed. Each rental cycle unit includes at least one rental node. The operational knowledge graph is used to record the rental operation relationships of multiple rental cycle units and multiple rental nodes. By using a butterfly optimization algorithm that dynamically senses intensity and based on an operational knowledge graph, the number of future rental orders for multiple rental cycle units can be predicted. Based on the number of future lease orders for multiple lease cycle units, the historical lease orders for multiple lease nodes, and the operational knowledge graph, determine the initial number of detachable mobile cold storage units for each lease node; It allocates rental nodes for user-initiated requests to rent detachable mobile cold storage facilities, and also allocates rental nodes for user-initiated requests to return detachable mobile cold storage facilities.

2. The knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities according to claim 1, characterized in that, Based on historical rental orders from multiple rental nodes, multiple rental cycle units are identified, including: For any two rental nodes, calculate the interaction coefficient between the two rental nodes based on the rental node where the rental was rented and the rental node where the rental was returned for each historical rental order. By using a nested loop clustering algorithm, multiple rental loop units are determined based on the interaction coefficient between any two rental nodes.

3. The knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities according to claim 2, characterized in that, Using a nested loop clustering algorithm, multiple rental loop units are determined based on the interaction coefficient between any two rental nodes, including: S101. Based on the interaction coefficient between any two rental nodes, filter out isolated rental nodes from multiple rental nodes; S102. Initialize the number of cluster centers N to 3 and the number of initial iterations T to 1. Determine the initial cluster centers from multiple lease nodes based on the interaction coefficient between any two lease nodes. S103. Construct a cluster list, whereby the cluster list is used to record rental nodes that are not clustered and are not cluster centers; S104. From the clustering list, extract a rental node that is not clustered and is not a cluster center. Based on the interaction coefficient between the rental node that is not clustered and is not a cluster center and each rental node included in each rental cycle unit corresponding to the current iteration, determine the interaction coefficient between the rental node that is not clustered and is not a cluster center and each rental cycle unit corresponding to the current iteration. Determine whether the rental node that is not clustered and is not a cluster center should be clustered. If yes, assign the rental node that is not clustered and is not a cluster center to the rental cycle unit with the largest interaction coefficient and delete it from the clustering list. Execute S106. If no, execute S105. S105. Retain the rental nodes that are not clustered and are not cluster centers in the clustering list, and execute S104; S106. Determine whether all rental nodes that have not been clustered and are not cluster centers have been clustered. If yes, proceed to S107; otherwise, proceed to S104. S107. For each rental cycle unit corresponding to the current iteration, determine the central rental node of the rental cycle unit based on the interaction coefficient between any two rental nodes in the rental cycle unit. S108. Based on the rental node at the center of each rental cycle unit, determine whether the inner cycle has been completed. If not, proceed to S109; if yes, proceed to S110. S109. Update the cluster center and cluster list based on the rental node at the center of each rental cycle unit, and execute S103; S110. Determine whether the outer loop is complete. If yes, determine multiple rental loop units based on the multiple rental loop units corresponding to the consecutive iterations. If no, execute S111. S111. Set N=N+1 and T=T+1, update the cluster centers and cluster list, and execute S104.

4. The knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities according to claim 3, characterized in that, Determine the central leasing node of the leasing cycle unit, including: For each lease node included in the lease cycle unit, the center value of the lease node is calculated based on the interaction coefficient between the lease node and each other lease node included in the lease cycle unit; Based on the center value of each lease node, determine the center lease node of the lease cycle unit.

5. The knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities according to claim 2, characterized in that, Based on historical rental orders from multiple rental cycle units and multiple rental nodes, an operational knowledge graph is constructed, including: For each rental cycle unit, the number of orders for the rental cycle unit in multiple historical periods is determined based on the historical rental orders of the rental nodes included in the rental cycle unit. For any two rental cycle units, calculate the order correlation coefficient between the two rental cycle units based on the order quantity of the two rental cycle units in multiple historical periods; Construct an operational knowledge graph based on the order correlation coefficient between any two rental cycle units and the interaction coefficient between any two rental nodes.

6. The knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities according to any one of claims 1-5, characterized in that, Using a butterfly optimization algorithm with dynamic intensity perception, based on an operational knowledge graph, the number of future rental orders for multiple rental cycle units is predicted, including: For each rental cycle unit, the range of future rental orders for the rental cycle unit is determined based on the historical rental orders of the rental nodes included in the rental cycle unit; Initialize the population based on the range of future rental orders for each rental cycle unit; Construct the fitness function; Based on the fitness function, the population is iteratively optimized to determine the number of future rental orders for multiple rental cycle units. In each iteration, the dynamic sensing intensity corresponding to the current iteration is determined.

7. The knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities according to claim 6, characterized in that, The fitness function is related to the distance of the global correlation coefficient corresponding to the butterfly.

8. The knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities according to claim 6, characterized in that, Determine the dynamic sensing intensity corresponding to the current iteration, including: For each butterfly in the current iteration, calculate the correlation coefficient of future orders between any two rental cycle units corresponding to the butterfly, and construct the correlation coefficient vector corresponding to the butterfly; For any two butterflies in the current iteration, calculate the distance between the correlation coefficient vectors corresponding to the two butterflies; Calculate the distance between the correlation coefficient vectors of the globally optimal butterflies in two consecutive iterations within the current iteration window; The dynamic sensing intensity for the current iteration is determined by the distance between the correlation coefficient vectors of any two butterflies in the current iteration and the distance between the correlation coefficient vectors of the globally optimal butterflies in the two adjacent iterations within the current iteration window.

9. The knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities according to any one of claims 2-5, characterized in that, Based on the number of future lease orders across multiple lease cycle units, historical lease orders across multiple lease nodes, and the operational knowledge graph, determine the initial number of detachable mobile cold storage units for each lease node, including: For each leasing cycle unit, based on the operational knowledge graph, the interaction weight of each leasing node included in the leasing cycle unit is determined. Based on the historical leasing orders of each leasing node included in the leasing cycle unit, the order ratio of each leasing node is determined. Based on the number of future leasing orders of the leasing cycle unit and the interaction weight and order ratio of each leasing node included in the leasing cycle unit, the initial number of detachable mobile cold storage units included in the leasing cycle unit is determined.

10. The knowledge graph-based leasing operation and management method for detachable mobile cold storage facilities according to claim 9, characterized in that, Determine the interaction weights of the lease nodes included in the lease cycle unit, including: For each lease node included in the lease cycle unit, the interaction weight of the lease node is determined based on the average of the interaction coefficients between the lease node and each other lease node included in the lease cycle unit.

Images