Intelligent logistics optimization method and device, equipment and storage medium

Abstract

The invention provides an intelligent logistics optimization method, device and equipment and a storage medium, and the method comprises the steps: carrying out the multi-source fusion processing of the Internet of Things sensor data and historical business data of a multimodal transportation reloading node in a logistics process, and obtaining a spatial-temporal feature vector; performing operation adaptability evaluation on the reloading nodes by using a P median model according to the spatial-temporal feature vectors to obtain a comprehensive score reflecting the node operation condition; a Tabu search algorithm is adopted to carry out combinational optimization on operation parameters of the reloading nodes based on the spatial-temporal feature vectors and the comprehensive scores, and an initial reloading scheme is obtained; and according to the initial reloading scheme, performing multi-agent cooperative scheduling on the reloading nodes on the digital twin platform to obtain a logistics resource optimal allocation scheme. Through combination of multi-source fusion and an efficient optimization algorithm, dynamic operation analysis and collaborative scheduling of multimodal transport reloading nodes are realized, logistics timeliness and resource utilization rate are improved, and the method is suitable for a complex logistics network and a changeable market environment.

Description

Technical Field

[0001] The present invention relates to the field of logistics technology, and in particular to an intelligent logistics optimization method, device, equipment and storage medium. Background Art

[0002] In the field of modern logistics, multimodal transport has become a key means to improve transportation efficiency and reduce logistics costs. However, in the face of frequent transshipment operations between different modes of transport, traditional logistics optimization methods can often only plan and schedule based on a single data source (such as historical business data) or static conditions (such as fixed node attributes), which makes it difficult to fully consider the dynamic changes of transshipment nodes in real-time operation and the massive real-time data provided by IoT sensors. As the connection and diversion link in multimodal transport, the dynamic attributes of transshipment nodes in dimensions such as geographical location, equipment load, operating efficiency and resource utilization will directly affect the overall transportation timeliness and cost control of multimodal transport. However, the existing technology lacks comprehensive analysis and integration of these multi-dimensional spatiotemporal characteristics, resulting in a lack of accuracy in the evaluation of the operational adaptability of transshipment nodes and the inability to identify bottleneck links in a timely manner. At the same time, when dealing with the resource scheduling and coordination of multimodal transport nodes, traditional optimization algorithms do not take into account the dynamic coupling relationship between nodes and real-time operation data, making it difficult to quickly converge to a high-quality optimization solution in complex scenarios. Summary of the invention

[0003] The main purpose of the present invention is to solve the technical problem that the comprehensive utilization of operation data and historical data of existing multimodal transport transfer nodes is insufficient, resulting in inaccurate node operation adaptability evaluation and poor resource scheduling effect; A first aspect of the present invention provides an intelligent logistics optimization method, the intelligent logistics optimization method comprising: The IoT sensor data and historical business data of the transshipment nodes of multimodal transport in the logistics process are processed by multi-source fusion to obtain the spatiotemporal feature vector containing static and dynamic attributes; Using the P median model to evaluate the operational adaptability of the refitting node according to the spatiotemporal feature vector, a comprehensive score reflecting the node operation status is obtained; The Tabu search algorithm with an adaptive neighborhood structure and a dynamic taboo length is used to perform combined optimization on the operation parameters of the dressing node based on the spatiotemporal feature vector and the comprehensive score to obtain an initial dressing plan; According to the initial refitting plan, multi-agent collaborative scheduling is performed on the refitting nodes on the digital twin platform to obtain an optimal allocation plan for logistics resources.

[0004] Optionally, in a first implementation of the first aspect of the present invention, the multi-source fusion processing of IoT sensor data and historical business data of transshipment nodes of multimodal transport in the logistics process to obtain a spatiotemporal feature vector containing static attributes and dynamic attributes includes: The geographical location, equipment capacity and site area of ​​the transshipment node are encoded to obtain static spatial characteristics, and the real-time operation efficiency, resource utilization and delay risk of the transshipment node are quantified to obtain dynamic time characteristics; Perform time series analysis on the loading and unloading equipment status, personnel configuration and operation duration information in IoT sensor data to obtain dynamic spatiotemporal characteristics; Statistical modeling is performed on the vehicle arrival frequency, cargo type distribution, and weather impact in historical business data to obtain historical spatiotemporal patterns; The static spatial features, dynamic time features, dynamic spatiotemporal features and historical spatiotemporal patterns are multi-dimensionally fused using a tensor decomposition algorithm to obtain a spatiotemporal feature vector.

[0005] Optionally, in a second implementation of the first aspect of the present invention, the P median model is used to perform an operational adaptability assessment on the refitting node according to the spatiotemporal feature vector to obtain a comprehensive score reflecting the node operation status, including: Extracting the location information of the existing distribution centers and the refitting nodes in the spatiotemporal feature vector, and constructing a static network structure according to the location information; According to the real-time operation efficiency and resource utilization in the spatiotemporal feature vector, the dynamic service capability index of each refitting node in the static network structure is calculated; Using the vehicle arrival frequency and cargo type distribution information, dynamic demand weights are assigned to each transshipment node in the static network structure; The P median model is applied to each switching node to calculate the contribution of the corresponding switching node to the static network structure, and a comprehensive score reflecting the node operation status is obtained.

[0006] Optionally, in a third implementation of the first aspect of the present invention, applying the P median model to each refitting node, calculating the contribution of the corresponding refitting node to the static network structure, and obtaining a comprehensive score reflecting the node operation status includes: Traverse each dressing node in the static network structure, calculate the weighted distance matrix between the traversed dressing node and other dressing nodes, and obtain the connection strength between nodes; According to the dynamic service capability index and dynamic demand weight, the objective function of the P median model is constructed to obtain the node importance evaluation function; The objective function is solved using the Lagrangian relaxation method to obtain the initial contribution of each refitting node; Apply the iterative adjustment algorithm to fine-tune the static network structure according to the initial contribution to obtain the optimized node contribution; The optimized node contribution is normalized through the sigmoid function to obtain a comprehensive score of the node operation status.

[0007] Optionally, in a fourth implementation of the first aspect of the present invention, the Tabu search algorithm with an adaptive neighborhood structure and a dynamic taboo length is used to perform combined optimization on the operating parameters of the dressing node based on the spatiotemporal feature vector and the comprehensive score to obtain the initial dressing plan, which includes: Initialize the operation parameter configuration of the refitting node according to the spatiotemporal feature vector and the comprehensive score to obtain an initial solution, and call the adaptive neighborhood structure generation function in the Tabu search algorithm to generate a candidate solution set based on the initial solution; Applying the dynamic taboo table preset in the Tabu search algorithm to evaluate the objective function value of each solution in the candidate solution set to obtain the current optimal solution, wherein the dynamic taboo table automatically adjusts the taboo length according to the current number of iterations; Update the global optimal solution and determine whether it meets the stopping criteria. If not, return to the step of calling the adaptive neighborhood structure generation function in the Tabu search algorithm to generate a candidate solution set based on the initial solution until the stopping criteria are met, and use the global optimal solution as the initial replacement plan.

[0008] Optionally, in a fifth implementation of the first aspect of the present invention, calling the adaptive neighborhood structure generation function in the Tabu search algorithm to generate a candidate solution set based on the initial solution includes: Conduct sensitivity analysis on the operational parameters of the initial solution and obtain the parameter importance ranking; According to the order of parameter importance, the priority and scope of parameter adjustment are determined to obtain the parameter adjustment strategy; Based on the parameter adjustment strategy, a multi-level neighborhood structure is constructed, wherein each layer of the multi-level neighborhood structure corresponds to parameter adjustments of different importance; Perform local search in a multi-level neighborhood structure, adjust different parameters of the initial solution, and obtain an initial candidate solution set; An adaptive mechanism is used to dynamically adjust the search depth and width according to the initial candidate solution set, and the search range is expanded in a multi-level neighborhood structure to obtain a candidate solution set.

[0009] Optionally, in a sixth implementation of the first aspect of the present invention, performing multi-agent collaborative scheduling on the refitting nodes on the digital twin platform according to the initial refitting plan to obtain a logistics resource optimization allocation plan includes: Mapping the initial dressing plan to the digital twin platform, constructing a virtual dressing node network, and obtaining a simulation environment; Assign intelligent agents to each virtual dressing node in the simulation environment and define the communication protocol between agents to obtain a multi-agent system; Executing a distributed constraint optimization algorithm based on the multi-agent system to obtain a preliminary resource allocation plan; Performing dynamic simulation verification on the preliminary resource allocation scheme, identifying potential conflicts, and obtaining conflict resolution strategies; According to the conflict resolution strategy, the resource allocation is adjusted in real time to obtain the optimal allocation plan of logistics resources.

[0010] A second aspect of the present invention provides an intelligent logistics optimization device, the intelligent logistics optimization device comprising: The data fusion module is used to perform multi-source fusion processing on the IoT sensor data and historical business data of the transshipment nodes of multimodal transport in the logistics process to obtain the spatiotemporal feature vector containing static and dynamic attributes; An adaptability evaluation module, used to evaluate the operational adaptability of the refitting node according to the spatiotemporal feature vector using a P median model, and obtain a comprehensive score reflecting the node operation status; A taboo search module, configured to use a Tabu search algorithm with an adaptive neighborhood structure and a dynamic taboo length to perform combined optimization on the operation parameters of the refitting node based on the spatiotemporal feature vector and the comprehensive score to obtain an initial refitting plan; The collaborative scheduling module is used to perform multi-agent collaborative scheduling of the refitting nodes on the digital twin platform according to the initial refitting plan to obtain an optimal allocation plan for logistics resources.

[0011] The third aspect of the present invention provides an intelligent logistics optimization device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected through lines; the at least one processor calls the instructions in the memory so that the intelligent logistics optimization device executes the steps of the above-mentioned intelligent logistics optimization method.

[0012] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the above-mentioned intelligent logistics optimization method.

[0013] The above-mentioned intelligent logistics optimization method, device, equipment and storage medium obtain the spatiotemporal feature vector by multi-source fusion processing of the IoT sensor data and historical business data of the multimodal transshipment nodes in the logistics process; the P median model is used to evaluate the operational adaptability of the transshipment nodes according to the spatiotemporal feature vector to obtain a comprehensive score reflecting the node operation status; the Tabu search algorithm is used to perform combinatorial optimization on the operational parameters of the transshipment nodes based on the spatiotemporal feature vector and the comprehensive score to obtain an initial transshipment plan; according to the initial transshipment plan, the transshipment nodes are coordinated by multiple agents on the digital twin platform to obtain an optimized allocation plan for logistics resources. The present invention combines multi-source fusion with an efficient optimization algorithm to achieve dynamic operational analysis and coordinated scheduling of multimodal transshipment nodes, improve logistics timeliness and resource utilization, and is suitable for complex logistics networks and changing market environments.

[0014] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0015] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of a first embodiment of the intelligent logistics optimization method in an embodiment of the present invention; Figure 2 A schematic diagram of an embodiment of an intelligent logistics optimization device in an embodiment of the present invention; Figure 3 It is a schematic diagram of an embodiment of the intelligent logistics optimization device in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0019] To facilitate understanding of this embodiment, firstly, an intelligent logistics optimization method disclosed in an embodiment of the present invention is described in detail. Figure 1 As shown, the method comprises the following steps: 101. Perform multi-source fusion processing on IoT sensor data and historical business data of transshipment nodes of multimodal transport in the logistics process to obtain spatiotemporal feature vectors containing static and dynamic attributes; In one embodiment of the present invention, the multi-source fusion processing of the IoT sensor data and historical business data of the transshipment nodes of multimodal transport in the logistics process to obtain a spatiotemporal feature vector containing static attributes and dynamic attributes includes: encoding the geographical location, equipment capacity and site area of ​​the transshipment nodes to obtain static spatial characteristics, and quantifying the real-time operating efficiency, resource utilization and delay risk of the transshipment nodes to obtain dynamic time characteristics; performing time series analysis on the loading and unloading equipment status, personnel configuration and operation duration information in the IoT sensor data to obtain dynamic spatiotemporal characteristics; performing statistical modeling on the vehicle arrival frequency, cargo type distribution and weather impact degree in the historical business data to obtain the historical spatiotemporal pattern; and using the tensor decomposition algorithm to multi-dimensionally fuse the static spatial characteristics, dynamic time characteristics, dynamic spatiotemporal characteristics and historical spatiotemporal patterns to obtain a spatiotemporal feature vector.

[0020] Specifically, the geographical location, equipment capacity and site area of ​​the transshipment node are encoded to obtain static spatial features, and the real-time operation efficiency, resource utilization and delay risk of the transshipment node are quantified. When obtaining dynamic time features, it is necessary to convert the geographical location of each node in the multimodal transport network into discrete or continuous coordinate values ​​that can be processed by the algorithm according to the pre-defined coordinate mapping strategy, and combine the total capacity of loading and unloading equipment and site area information inside the node to construct a set of attributes for the spatial dimension. In order to make the encoding results have consistent scales and measurement units, the geographical location information is usually mapped using geo-hashing or an algorithm based on the Gaussian projection coordinate system, and the equipment capacity and site area are normalized so that they can form a static spatial feature vector together with the location features. The quantification process of real-time operation efficiency is based on the cargo flow or loading and unloading success rate of the transshipment node per unit time, and the resource utilization rate is measured in combination with the ratio of in-transit equipment to idle equipment. At the same time, the numerical level of delay risk is calculated by summarizing the delay information in the scheduling log. If the real-time operation efficiency needs to be further subdivided, the average response time and peak load occupancy rate can also be introduced, and the values ​​of these metrics can be converted into the input of a segmented mapping or exponential decay function. Finally, the obtained real-time efficiency, utilization rate and delay risk can be combined into a set of dynamic time characteristics.

[0021] Specifically, when performing time series analysis on the loading and unloading equipment status, personnel configuration, and operation duration information in the IoT sensor data to obtain dynamic spatiotemporal features, it is necessary to segment the data stream reported by the sensor on the time axis and make fine-grained records of the equipment start-up rate, personnel load, and actual operation start and end time in each time period. The loading and unloading equipment status can be evaluated by collecting parameters such as its real-time power consumption, operating temperature, or mechanical operation frequency. The personnel configuration information can be extracted through the electronic work badge or the scheduling system interface to extract the number of people on duty and the distribution of positions. The operation duration can be determined by the start and end time recorded by the barcode scanning log or the automatic recognition system. In order to convert these time series data into effective dynamic spatiotemporal features, the statistics of the equipment and personnel status in each time period are usually mapped into multidimensional sequences through sliding windows or segmented aggregation methods, and then short-term sequence smoothing or anomaly removal strategies are applied to filter out data points that deviate too much from normal operations. On this basis, a two-dimensional or three-dimensional space-time matrix can be defined, where rows or columns represent time segments, and matrix units correspond to information such as equipment status intensity and personnel load level. Combined with the node geographic location coordinates, high-dimensional data oriented to time and space can be formed. This time series analysis helps to compare the operating levels of different time periods, so that the subsequent modeling process can distinguish the differences in node operations between peak and off-peak periods, and consider the resource allocation fluctuations caused by this factor in the overall network scheduling.

[0022] When the vehicle arrival frequency, cargo type distribution and weather impact degree in the historical business data are statistically modeled and the historical spatiotemporal pattern is obtained, it is necessary to first collect the vehicle flow and cargo inflow and output records of each transshipment node in the past multiple time periods, and classify and summarize the historical transportation volume of different categories according to the standards for cargo type classification. The vehicle arrival frequency can be counted according to the hour, day or week period, and the cargo type distribution can correspond to the category code built into the customs or logistics system, thereby generating a set of multi-dimensional cargo flow data. For the measurement of the degree of weather impact, it is usually necessary to introduce meteorological data from previous years, and perform regression analysis on the correlation between different types of weather (such as strong winds, heavy rains or extreme temperatures) and vehicle delays or cargo damage events, so as to extract the mapping relationship of "weather conditions-operation status" in the historical business data. In order to better accommodate the climate differences or transportation mode differences of different nodes, a set of segmented regression functions or multivariate linear models based on weighted coefficients can be designed so that they can output relatively stable impact values ​​when facing different regions and different transportation modes. In the statistical modeling stage, a triple index structure consisting of node identifier, time period identifier and weather type identifier is introduced to incorporate vehicle frequency, cargo type distribution and weather impact into the same data structure, and a constraint relationship is established based on joint probability or conditional probability.

[0023] Specifically, the above-mentioned static spatial features, dynamic time features, dynamic spatiotemporal features and historical spatiotemporal patterns are multi-dimensionally fused using the tensor decomposition algorithm. When the spatiotemporal feature vector is obtained, the four types of data will be first mapped to the same index system, so that they can be aligned in the node dimension, time dimension and business attribute dimension, and a tensor structure containing at least three or more dimensions will be constructed. The element values ​​of this tensor come from the various statistics or indicators mentioned above. In order to ensure that the fused vector has a wider applicability, it is necessary to interpolate or normalize the abnormal or missing data before tensor decomposition. In actual operation, methods such as Tucker decomposition or PARAFAC decomposition can be used to decompose the tensor into several factor matrices and core tensors in terms of dimension, and iteratively update these matrix parameters by minimizing the reconstruction error. After the tensor decomposition is completed, the important features in the core tensor and factor matrix will be mapped back to the index position corresponding to each node, each time period and each business attribute, so as to obtain a comprehensive vector set containing space, time, device status and historical pattern information. This vector set can be regarded as a specific instantiation of the spatiotemporal feature vector, from which the multidimensional attributes of each dressing node at each time segment can be read. 102. Use the P median model to evaluate the operational adaptability of the refitting node based on the spatiotemporal feature vector to obtain a comprehensive score reflecting the node's operational status; In one embodiment of the present invention, the use of the P median model to evaluate the operational adaptability of the transloading node according to the spatiotemporal feature vector to obtain a comprehensive score reflecting the node operation status includes: extracting the location information of the existing distribution center and the transloading node in the spatiotemporal feature vector, and constructing a static network structure according to the location information; calculating the dynamic service capability index of each transloading node in the static network structure according to the real-time operation efficiency and resource utilization in the spatiotemporal feature vector; using the vehicle arrival frequency and cargo type distribution information to assign a dynamic demand weight to each transloading node in the static network structure; applying the P median model to each transloading node, calculating the contribution of the corresponding transloading node to the static network structure, and obtaining a comprehensive score reflecting the node operation status.

[0024] Specifically, the location information of existing distribution centers and transshipment nodes in the spatiotemporal feature vector is extracted. When a static network structure is constructed based on the location information, it is necessary to first retrieve the geographic coordinates, capacity level and connectivity description of each node in the acquired multi-source fusion data, and convert this information into a form suitable for building a graph structure. Specifically, a unique index is assigned to each node, and the distance or comprehensive transportation cost between nodes is calculated based on the geographic coordinate or projection coordinate system. Existing distribution centers in certain areas need to be retained under the same index system so that subsequent algorithms can directly identify these core nodes when building and searching for adjacency relationships. If the multimodal transport scenario involves multiple modes of transportation such as railways, water transport or roads, segmentation coefficients or freight rate additions can be introduced in the distance calculation so that the same geographic distance corresponds to different costs under different modes of transportation, and then all costs are integrated into the adjacency matrix or edge set structure. In order to be compatible with large networks, a sparse matrix form can be used to record only the edges with feasible transportation routes and the freight or distance attributes corresponding to each edge, and retain special identifiers of distribution centers and transshipment nodes in the matrix rows and columns, which will be used as screening or classification conditions when solving the model. If it is necessary to distinguish the loading and unloading functions or special facilities of the nodes, the node function codes can also be stored in the additional dimensions of the adjacency matrix, so that the algorithm can quickly retrieve nodes that can handle specific cargo types or have special equipment when solving. To avoid excessive redundancy of high-dimensional matrices during the calculation process, an acceleration structure can be constructed through hash indexes or spatial search trees, so that when querying any node for its adjacent reachable nodes, only local matching is required in the index tree or hash bucket. In order to further meet the multimodal transport characteristics proposed in the paper, it is necessary to retain the differences in real transportation modes when constructing a static network structure. Different intermodal modes usually correspond to different constraints or priorities in distance calculation or path selection. The element values ​​in the distance matrix or adjacency table can be combined with weighted functions to achieve a unified measurement of transportation mode and distance. Such a static network structure can be regarded as a basic carrier combined with spatiotemporal feature vectors. When the subsequent model evaluates the adaptability of nodes, these distance and connection data will continue to be called, thereby completing the quantitative description of node accessibility and spatial layout.

[0025] When calculating the dynamic service capability index of each transloading node in the static network structure based on the real-time operation efficiency and resource utilization in the spatiotemporal feature vector, it is necessary to expand the previously constructed static network structure and combine it with the real-time performance indicators to obtain a dynamic service characterization at the node level. First, based on the collected spatiotemporal feature vector, the operation efficiency information of each transloading node in different time periods is extracted as a set of distribution values, such as the completion rate of loading and unloading or cargo throughput per unit time, and the equipment utilization rate is statistically analyzed, including the proportion of mechanical operation time and idle time. The operation efficiency and equipment utilization rate can be normalized separately, and the two can be combined into an initial service index according to an adjustable coefficient. If it is necessary to consider the delay time or the downtime of the fault, a function with a penalty factor can be designed to reduce the delay behavior of a longer period of time more significantly in the service index, so that the dynamic service capability of the node is more in line with the actual operation status. In order to make the index comparable between different nodes, segmented or logarithmic mapping is often used for scaling, so that the index value of nodes with extremely high operating efficiency will not be infinitely enlarged, while nodes with extremely low efficiency or frequent failures will show a large difference in index value. Mathematically, we can make Representation Node In time service index, which is regarded as operational efficiency and resource utilization Some combination of functions, such as in is the delay loss factor, is an adjustable parameter used to balance the impact of efficiency, utilization and delay on the final index. After summarizing these time-series service indices, a dynamic capacity curve can be obtained for each node. Then, combined with the constructed static network structure, the service index of each node in different time periods is assigned to the corresponding node attribute in the network. In this way, each transshipment node has a service capacity description that can change over time, which lays the foundation for the calculation input of the multimodal transport operation evaluation described in the paper. When allocating dynamic demand weights to each transshipment node in the static network structure using vehicle arrival frequency and cargo type distribution information, it is necessary to extract vehicle flow and cargo components from the aforementioned spatiotemporal feature vector and perform statistical integration according to the node and time period division. The vehicle arrival frequency can be aggregated hourly or daily based on historical business records or real-time sensor detection data, and the cargo type distribution can be combined with the operation log of the loading and unloading system or the electronic waybill information to obtain the flow proportion of each type of cargo at the node. In order to merge the two types of values ​​into a single demand weight, a common practice is to assign differentiated weights to different types of goods according to the priority or timeliness of the goods, and then multiply them by the vehicle arrival rate to form the demand pressure of the node in a given time interval. If a node has a high traffic density for a long time or has a variety of goods with urgent demand, its demand weight can be amplified upward by designing a set of polynomials or piecewise functions, so as to reflect the demand pull of this node on the entire network in the subsequent model. In order to make this demand information more time-varying, weather effects or seasonal fluctuation coefficients can also be included in the vehicle arrival frequency so that the network model can produce demand fluctuations that are consistent with the actual scenario when dealing with traffic congestion or extreme weather. If some nodes specialize in handling large quantities of cold chain goods or special dangerous goods, their demand weights can be given a higher baseline value in the current step to highlight their position in network operations. Finally, the obtained demand weight matrix is ​​matched with the node index in the static network structure so that each node has a corresponding weight value in different time periods, forming a set of demand constraints or target constraints that can be called by the algorithm, and reflecting the balance between supply and demand in the overall model together with the service capacity index.

[0026] Apply the P median model to each transshipment node, calculate the contribution of the corresponding transshipment node in the static network structure, and obtain a comprehensive score that reflects the operating status of the node. It is necessary to combine the aforementioned dynamic service capability index and demand weight, take the distance or transportation cost matrix as the main input of the model, and superimpose dynamic factors and penalty functions on the traditional framework of P median to balance the distribution of supply and demand. In the solution, the model judges whether the node is selected as the center or transshipment core, determines its satisfaction with the coverage needs of other surrounding nodes, and combines the comprehensive calculation of service efficiency and freight cost to output a numerical score representing the operating status for each node. This value can be regarded as the status and contribution of the node in the network. By introducing time changes and multi-dimensional attributes through the expansion of P median, these scores can present results that are closer to the actual operating status. Furthermore, the method of applying the P median model to each dressing node, calculating the contribution of the corresponding dressing node in the static network structure, and obtaining a comprehensive score reflecting the node operation status includes: traversing each dressing node in the static network structure, calculating the weighted distance matrix between the traversed dressing node and other dressing nodes, and obtaining the connection strength between the nodes; constructing the objective function of the P median model according to the dynamic service capability index and the dynamic demand weight, and obtaining the node importance evaluation function; solving the objective function by the Lagrange relaxation method, and obtaining the initial contribution of each dressing node; applying the iterative adjustment algorithm, fine-tuning the static network structure according to the initial contribution, and obtaining the optimized node contribution; normalizing the optimized node contribution by the sigmoid function, and obtaining a comprehensive score of the node operation status.

[0027] Specifically, when calculating the weighted distance matrix between each transfer node and other transfer nodes in the static network structure, and obtaining the connection strength between nodes, it is necessary to obtain the space or transportation cost of all node pairs from the constructed adjacency matrix or edge set information, and introduce additional weighting factors according to different transportation modes or line capacities, so as to convert the pure geographical distance or freight value into a comprehensive distance value suitable for subsequent calculations. If the static weight of each edge is recorded as , then in the multimodal transport scenario, a set of transport modes can be added as positive coefficients To highlight the differences between different routes, let the final weighted distance matrix element be When traversing a node, you need to retrieve all nodes from the graph structure. Departure reachable nodes , and read them one by one The row vector is formed and then the row vector is summarized into the weighted distance matrix. If there are multiple feasible routes between nodes, the route with the smallest path cost can be selected for recording during the weighting process, or the cost values ​​of multiple routes and their reliability factors can be combined into a combined function to obtain a more expressive This traversal operation is usually used in large-scale networks to batch update the matrix with the help of index or shortest path search algorithm, so that nodes The connection relationship with other nodes is sparse or compact in the data structure, which improves the efficiency of retrieval and writing. The weighted distance submatrix indexed by rows or columns can set filtering rules based on distance thresholds or capacity constraints to retain strongly connected adjacency relationships and downgrade weakly connected edges. The inter-node connection strength output by this step is an important input for improving the evaluation of the P median model, and can provide a quantitative basis for each node to judge the distance cost or freight penalty when linking with other nodes in the subsequent objective function; According to the dynamic service capability index and dynamic demand weight, the objective function of the P median model is constructed. When obtaining the node importance evaluation function, it is necessary to couple the weighted distance matrix obtained in the previous step with the dynamic attributes at the node level, and add multi-dimensional factors to the traditional P median framework. If Represents the set of all nodes, , introduce binary variables Representation Node Whether the node Service and order Representation Node Whether it is selected as a central or key replacement node. Representative Node The dynamic demand weight of Representative Node The dynamic service capability index, Representation Node With Node If you need to highlight the mutual influence of service capacity and demand weight in the objective function, you can define a node importance evaluation function ; in To balance the index parameters of various influences, Representation Node The fixed cost or resource occupation required to be selected as the center. If the model hopes to find the optimal solution between distance, demand and service, Use positive numbers learned through empirical analysis or other methods, and observe the target value changes while solving. In order to ensure that each node has a unique service node and limit the number of central nodes, it is necessary to increase Constraints such as can be used to make the objective function achieve a comprehensive balance among distance, demand, and service capacity at a given number of centers. As a global evaluation indicator of node importance, once one of the nodes of If the node is much higher than other nodes, it will have a greater chance of being selected as a center or taking on more business coverage during the optimization process, thus reflecting the two-way attention to dynamic services and needs in the node evaluation and site selection process: When solving the objective function using the Lagrange relaxation method to obtain the initial contribution of each replacement node, it is necessary to divide the multiple constraints in the above formula into blocks and design corresponding Lagrange multipliers to relax the equations or inequalities. If Representation and The associated multiplier, Representation and The associated multipliers, the Lagrangian function can be written as: ; To reduce computational complexity, the variables and Do linearization and optimize the subproblems separately by decomposing them, each time at a fixed Minimize under the condition , and then update the multiplier rule Make corrections and continue to iterate until the convergence criteria are met. At this time, a set of approximate global optimal solutions can be obtained. and The value of is used to extract the initial node contribution. If the contribution is recorded as , after the solution is completed, Or similar forms of calculation, the node The covered demand and service capacity are concentrated into a single value. If additional penalty factors or upper distance limits are set in the solution process, More in line with the actual transportation conditions. The vector is the initial contribution of each refitting node, which can be used for subsequent fine-tuning. Apply the generation adjustment algorithm to fine-tune the static network structure according to the initial contribution. When the optimized node contribution is obtained, the solution generated by the Lagrangian relaxation method needs to be re-examined, and the service relationship or allocation strategy between nodes should be locally corrected under specific conditions. If the contribution of some nodes in the initial solution is abnormally high or too low, these nodes and their associated edges can be scanned in the static network structure to find the distance or transportation cost items that cause the imbalance, and redistribute matches with high demand weights and too far distances to make the service relationship within the node community more reasonable. At this time, local search or simulated annealing and other methods can be used to find the optimal node in the node cluster. The service range is contracted or expanded around the node to balance the coverage requirements of adjacent nodes. After each local correction, the And summarize the new contribution ,If the overall target value continues to decrease, the correction is accepted and the new imbalance point is inspected again in the next iteration. This iterative process is often carried out in large-scale networks through partitioning or layering strategies, that is, fine-tuning is first completed within the local sub-network, and then the solutions between sub-networks are gradually merged in a larger range. If the weighted distance of some edges in the static network structure changes frequently due to transportation methods or routes, the weight of the edge can be dynamically updated during the iteration, and a new weighted distance matrix or additional penalty parameters can be included in the next local search to keep the node contribution consistent with the actual transportation environment. After such iterative adjustments, a set of optimized node contributions will be output, which usually better reflect the actual operation status than the initial solution. Finally, the optimized node contribution is normalized by the sigmoid function to obtain a comprehensive score of the node operation status. .

[0028] 103. The Tabu search algorithm with adaptive neighborhood structure and dynamic taboo length is used to optimize the operation parameters of the refitting nodes based on the spatiotemporal feature vector and comprehensive score to obtain the initial refitting plan; In one embodiment of the present invention, the Tabu search algorithm with an adaptive neighborhood structure and a dynamic taboo length is used to perform combinatorial optimization on the operating parameters of the dressing node based on the spatiotemporal feature vector and the comprehensive score to obtain an initial dressing plan, which includes: initializing the operating parameter configuration of the dressing node according to the spatiotemporal feature vector and the comprehensive score to obtain an initial solution, and calling the adaptive neighborhood structure generation function in the Tabu search algorithm to generate a candidate solution set based on the initial solution; applying the dynamic taboo table preset in the Tabu search algorithm to evaluate the objective function value for each solution in the candidate solution set to obtain the current optimal solution, wherein the dynamic taboo table automatically adjusts the taboo length according to the current number of iterations; updating the global optimal solution, and determining whether the stopping criterion is met, if not, returning to the step of calling the adaptive neighborhood structure generation function in the Tabu search algorithm to generate a candidate solution set based on the initial solution, until the stopping criterion is met, and taking the global optimal solution as the initial dressing plan.

[0029] Specifically, according to the spatiotemporal feature vector and the comprehensive score, the operation parameter configuration of the changeover node is initialized to obtain the initial solution, and the adaptive neighborhood structure function backbone initial solution in the Tabu search algorithm is called to generate the candidate solution set. It is necessary to encode the adjustable parameters such as resource allocation, operation period allocation and equipment start-stop strategy for each changeover node, and include the comprehensive score of the node operation status into the weight factor of the initial solution. If the node set is recorded as , the parameter combination of each node is represented as a vector , then the overall solution can be regarded as In order to make the initialization closer to the actual needs, Comprehensive score of China Fusion Constraints are formed, such as giving priority to resource caps or time allocations to nodes with higher scores during encoding. After initialization, an initial solution that meets business constraints and node differentiation requirements is generated. In the design of adaptive neighborhood structured function, it is necessary to stratify the solution space. The node parameters can be divided into two categories: high priority and low priority. When generating candidate solutions, the parameters of different priorities can be adjusted differently through multi-layer neighborhood search. If the search radius is defined in the high priority subspace , define the search radius in the low priority subspace , the neighborhood generating function can be set in Or the logic of segmented mapping, a batch of candidate solutions are constructed through random perturbations and guided scanning strategies. . When a candidate solution set is formed, The disturbance amplitude is adjusted dynamically through an adaptive mechanism for different parameter dimensions to avoid falling into a too narrow solution domain. After the generation function is completed, the output candidate solution set can meet complex requirements such as equipment switching and time allocation in multimodal transport scenarios, so that the next evaluation link can effectively retrieve excellent solutions. The dynamic taboo table preset in the Tabu search algorithm is used to evaluate the objective function value for each solution in the candidate solution set to obtain the current optimal solution. When the dynamic taboo table automatically adjusts the taboo length according to the current number of iterations, it is necessary to design a taboo mechanism with time series constraints for the Tabu search to limit the algorithm from repeatedly accessing solutions or operations marked as taboos within a number of steps. The objective function can be defined as ; in and It can represent the evaluation of operational efficiency and delay risk respectively. is the weighting coefficient, is the cost function for the connection between nodes in the multimodal transport environment, Representation Node and The objective function is to calculate the distance or cost information of each candidate solution. When performing calculations, it is necessary to retrieve all node parameters and adjacency relationships in the solution vector, comprehensively quantify its network coverage, resource scheduling balance, and logistics timeliness, and compare the obtained values ​​with the existing global optimal values. The dynamic taboo table will automatically update the taboo length according to the current number of iterations and the change speed of the solution during the iteration process. ,If the iteration speed is slow or the solution is highly repetitive in a local area, will increase accordingly to prevent returning to the previous solution; if the solution jumps frequently in consecutive iterations, It can be shortened adaptively to allow more attempts to find a better global solution. After each evaluation, the local operations that cause the solution to degrade significantly or the inferior solutions that have been visited will be placed in the taboo table and marked with their taboo validity period. Through this dynamic adjustment mechanism, the algorithm can search in a larger range without repeatedly falling into the same local extreme value. At the end of the evaluation phase, according to The minimum value of determines the current optimal solution of the iteration, and records the relevant information in the global storage for query and comparison in subsequent steps. Update the global optimal solution and determine whether it meets the stopping criterion. If not, return to the step of calling the adaptive neighborhood structure generation function in the Tabu search algorithm to generate a candidate solution set based on the initial solution until the stopping criterion is met. When the global optimal solution is used as the initial replacement solution, it is necessary to compare the current optimal solution after each generation of iteration. and the global optimal solution The objective function value of , then the global optimal solution is updated and some taboo records related to the previous solution are cleared or reset, so that subsequent solutions have a greater chance to perform refined searches near the new local area. The design of the stopping criterion can integrate multiple factors such as the number of iterations, the convergence rate of the objective function, and the stagnation length of the algorithm. For example, The algorithm is terminated when no improvement is achieved in multiple iterations or the total number of iterations exceeds a given limit. If the stopping condition is not met, the adaptive neighborhood structure generation function is called again. or A new round of perturbations and taboo evaluations is performed to continuously move toward a lower objective function value or a more balanced solution distribution. This cycle is repeated until the preset convergence or runtime criteria are met, and the final converged As the initial transshipment plan, it is executed or scheduled in the multimodal transport system. The operating parameters of each transshipment node in the plan will be retained in In the form of a scheduling system, subsequent resource configuration modules or simulation tools can directly map the solution to the digital twin platform or other scheduling systems.

[0030] Furthermore, the calling of the adaptive neighborhood structure generation function in the Tabu search algorithm to generate a candidate solution set based on the initial solution includes: performing a sensitivity analysis on the operating parameters of the initial solution to obtain a parameter importance ranking; determining the priority and range of parameter adjustment based on the parameter importance ranking to obtain a parameter adjustment strategy; constructing a multi-level neighborhood structure based on the parameter adjustment strategy, wherein each layer of the multi-level neighborhood structure corresponds to a parameter adjustment of different importance; performing a local search in the multi-level neighborhood structure to adjust different parameters of the initial solution to obtain an initial candidate solution set; and utilizing an adaptive mechanism to dynamically adjust the search depth and width based on the initial candidate solution set, expanding the search range in the multi-level neighborhood structure to obtain a candidate solution set.

[0031] Specifically, when performing sensitivity analysis on the operating parameters of the initial solution and obtaining the parameter importance ranking, it is necessary to first split all the operating parameters in the initial solution by category or function, and define a measurement system to measure the degree of parameter influence. If the initial solution is composed of variables of several dimensions, such as the start and stop strategy of the equipment at the changeover node, the shift schedule of personnel, the batches of material flow, and the threshold of site resource allocation, the variables of these dimensions can be recorded separately. In the measurement system, a set of partial derivatives or sensitivity indicators can be defined for a specific objective function (such as comprehensive transportation cost or delay rate). For example, for each parameter, And numerical approximation is performed on several simulation or sampling points. By normalizing or sorting, all the parameters are arranged from large to small to obtain a list of parameter importance. If some parameters have strong nonlinear correlation, they can be calculated by segmented gradient or spline interpolation to find the inflection point of change on the sensitivity curve. This sensitivity analysis can be combined with the multimodal transport simulation method mentioned in the paper: each parameter is fine-tuned positively and negatively in different scenarios (such as local peak period, off-peak period, and bad weather), and the impact on the objective function is recorded. If the sensitivity index of a set of parameters is high for a long time, it means that such variables have a significant impact on the overall logistics efficiency or operating cost; if some parameters have a low impact in multi-scenario tests, their importance ranking will be relatively low. Finally, a ranking table will be output, listed in order from the most influential to the least influential, forming the prerequisite information for the subsequent construction of the multi-level neighborhood structure. According to the ranking of parameter importance, the priority and range of parameter adjustment are determined. When obtaining the parameter adjustment strategy, it is necessary to compare the ranking results obtained in the previous step with the actual demand for resource or time allocation of the transshipment node during operation, and define the acceptable range of change for each important parameter. If the top rankings are equipment start-stop factors that directly affect transportation costs, the upper and lower limits of the start-stop duration or start-stop batches can be set by statistically analyzing historical operation data or vehicle arrival frequency; if the middle rankings mainly involve personnel scheduling or site resource scheduling, the node comprehensive score can be combined to define the adjustment range of the number of people scheduled or the area of ​​the site that can be allocated. In order to make the adjustment strategy more targeted, inconsistent step lengths and step widths can be designed according to the different levels of influence of high-priority parameters and low-priority parameters. Some high-priority parameters can be assigned a larger search radius so that the algorithm can more actively try different combinations in the neighborhood search; while low-priority parameters may only need to be fine-tuned within a smaller range. By introducing a multi-level penalty or reward mechanism within the above interval, differentiated constraints can be imposed on different parameter dimensions. If a parameter can significantly optimize the objective function by fluctuating only within a very small range, a wider moving step can be assigned to it; if a parameter exceeds the specified time and affects the actual business feasibility, a penalty item is set to limit its crossing the boundary. Finally, these priorities and range information are combined into a parameter adjustment strategy. When the algorithm searches in the neighborhood, if it detects that the high-priority parameter is in a range that can be moved significantly, a larger-scale perturbation is performed, and if a low-priority parameter is detected, only a slight modification is made. This mechanism can reflect the focus on key resources in the multimodal transport scheduling model described in the paper, and enhance the search efficiency during the operation. Based on the parameter adjustment strategy, a multi-level neighborhood structure is constructed, wherein each layer of the multi-level neighborhood structure corresponds to a parameter adjustment of different importance, and the strategy refined in the previous step needs to be mapped to a specific neighborhood definition. The parameters can be divided into several layers, such as a high-priority layer, a medium-priority layer, and a low-priority layer, and a set of operation operators are defined in each layer to describe the way to incrementally adjust, exchange, or other composite transformations of the parameters.If a large search step is decided for a certain type of parameter in the high-priority layer, the operator can be used to represent a significant increase or decrease in the magnitude of the parameter; if a moderate transformation is defined in the medium-priority layer, the relevant parameters are moderately moved; in the low-priority layer, fine-tuning or sequential small-scale modifications are used. To ensure that there is no conflict between layers, a set of merging rules needs to be designed: when the algorithm decides to apply a layer operation to a parameter in the solution vector, the parameter is not superimposed and modified at the same time, unless the merge operation has a positive benefit on the objective function after detection. The multi-neighborhood search algorithms mentioned in some literature or papers usually regard each neighborhood as a parallel candidate path. Here, the multi-level neighborhoods can be uniformly scheduled so that different levels can be switched freely in the same round of search. If the objective function is detected to be less deteriorated or a feasibility problem occurs after the high-priority layer operation is completed, the medium-priority or low-priority layer can be used in the same round of search to make compensatory adjustments to other parameters so that the solution vector remains feasible and diverse within an acceptable range. Through this multi-level structure, the tuning order and amplitude of different operating parameters can be systematically divided, allowing the algorithm to flexibly allocate search resources in complex multimodal transport scenarios.

[0032] 104. According to the initial refitting plan, multi-agent collaborative scheduling of the refitting nodes is carried out on the digital twin platform to obtain the optimal allocation plan for logistics resources.

[0033] In one embodiment of the present invention, the multi-agent collaborative scheduling of the dressing nodes on the digital twin platform according to the initial dressing plan to obtain the logistics resource optimization allocation plan includes: mapping the initial dressing plan to the digital twin platform, building a virtual dressing node network, and obtaining a simulation environment; assigning intelligent agents to each virtual dressing node in the simulation environment, and defining the communication protocol between agents to obtain a multi-agent system; executing a distributed constraint optimization algorithm based on the multi-agent system to obtain a preliminary resource allocation plan; dynamically simulating and verifying the preliminary resource allocation plan, identifying potential conflicts, and obtaining a conflict resolution strategy; adjusting resource allocation in real time according to the conflict resolution strategy to obtain a logistics resource optimization allocation plan.

[0034] Specifically, when mapping the initial reloading scheme to the digital twin platform, constructing a virtual reloading node network, and obtaining a simulation environment, it is necessary to first enter the existing multimodal transport node information, line distribution, and equipment configuration data in digital form, and combine the topological relationship with the spatiotemporal feature mapping rules to integrate the connection mode and distance cost between nodes into the digital twin system. If the node assumes the functions of loading and unloading, transshipment, or warehousing in the real environment, the corresponding functional module and operation parameters can be loaded for each node in the virtual world, so that the digital twin platform can simulate the line capacity, operation efficiency, and resource occupancy of the physical world. In order to retain the differences and dynamic changes in spatiotemporal features during the mapping process, a reference framework of the time dimension can be set for each line or hub node on the digital twin platform, and variables such as transportation demand during peak and off-peak periods, weather impact, and real-time operation efficiency can be implanted into the parameter library. During the mapping period, the equipment start-stop strategy, personnel scheduling, and cargo loading and unloading rhythm involved in the initial reloading scheme need to be virtualized and configured so that each node corresponds to the specific scheduling strategy in the scheme. If a high-fidelity model is used, multiple layers can be integrated into the digital twin, including the physical layer (such as hardware and facilities), the network communication layer (such as data transmission and control), and the logical decision-making layer (such as scheduling and optimization). After the virtual dressing node network is built, it is necessary to check the connectivity and data feasibility of the network, and correct some input deviations by comparing with the actual perception data, so as to obtain a simulation environment consistent with the real scene. This environment can carry different real-time information flows in the subsequent multi-agent scheduling link, and demonstrate the impact of various operation strategies on the overall network performance and local node load in the virtual space, so that the subsequent multi-agent system can better connect with the real business.

[0035] Assign intelligent agents to each virtual transshipment node in the simulation environment, and define the communication protocol between agents. When a multi-agent system is obtained, it is necessary to set the corresponding agent type and behavior logic according to the role of different nodes in the multimodal transport scenario. If a node has high throughput or a key geographical location, its agent can have richer decision-making modules, such as dynamic scheduling, emergency transfer strategy or accident emergency response; if a node is an ordinary transfer station, its agent only needs to have basic shipping, receiving and inventory management functions. In order to achieve real-time data exchange between agents, it is necessary to select a scalable and robust communication protocol. A common solution is to introduce a standard interface based on message middleware or event triggering mechanism in the simulation environment, so that each agent can obtain available resources or time windows by registering its own identity and demand type. If the concept of multi-agent collaborative scheduling or distributed control is mentioned in the paper, its method can be integrated into the communication protocol design, such as defining a subscription and publishing mechanism so that the node can actively initiate an announcement when a sudden change in cargo flow or equipment failure is detected, and the neighboring agent will immediately update its job schedule or transfer channel after receiving the message. It is necessary to formulate conflict management and priority transfer rules between agents within the system so that when resource competition or line congestion occurs, each agent can reach a consistent scheduling conclusion through negotiation or constraint solving. After all agents are deployed and have a unified communication protocol, the multi-agent system can observe and respond to the state changes of each node and line on the digital twin platform, simulating the collaborative decision-making process of different roles in the real logistics network to a certain extent.

[0036] When executing the distributed constraint optimization algorithm based on the multi-agent system and obtaining a preliminary resource allocation plan, each agent needs to collect the resource requirements, equipment capabilities and available operation time periods of the node within the range of the node, and broadcast or share this information to other agents in the message channel defined by the communication protocol. If each agent maintains the parallel management of local constraints and global constraints, a model of distributed constraint satisfaction problem (DCOP) or distributed constraint optimization problem (DCOP) can be constructed. This model will establish constraint functions between each node, describe the constraints of resource allocation, timing connection and logistics capacity, and continuously update feasible solutions during the interaction between nodes. If the multi-agent cooperation mechanism mentioned in the paper is introduced, the node agents can be organized into several subgroups, and a temporary allocation plan can be reached in the local subgroup through a consensus algorithm or iterative optimization, and then the boundary information or coupling cost can be exchanged with other subgroups, and finally merged to form a network-level plan. In order to improve the efficiency of the algorithm, heuristic strategies can be built into each agent, such as giving priority to meeting the needs of high-priority goods or critical time periods, or making temporary corrections to resource allocation when line congestion is detected, so as to ensure that the distributed optimization process continues to move forward. When the iteration converges, a preliminary resource allocation plan will be formed. The agent of each node can obtain the local transportation schedule, equipment activation plan and personnel allocation table, and achieve time and capacity docking with the corresponding allocation plan of the adjacent node at the junction. The preliminary plan is then submitted to the simulation environment for feasibility verification and performance evaluation to determine whether there is conflict or resource waste between nodes.

[0037] When dynamically simulating and verifying the preliminary resource allocation plan, identifying potential conflicts, and obtaining conflict resolution strategies, the digital twin platform needs to load the allocation plan just generated and gradually calculate it on the virtual timeline, and track the cargo flow, equipment load, and personnel scheduling of each node in different time periods in real time. If there are abnormal values ​​such as queuing, overloading, or delays in a certain period, or if multiple nodes are detected competing for limited transportation capacity or yard space in the same period, the system will mark the area as having a scheduling conflict and record the relevant parameter status. In combination with the dynamic event response mechanism discussed in the paper, the conflict information can be returned to the communication layer where each agent of the multi-agent system is located, so that the agent with scheduling decision-making power or conflict arbitration power can initiate a negotiation process at the local or global level. If the scale of the conflict is small, the conflict can be resolved by transferring some resources or delaying some low-priority operations through neighboring nodes; if the conflict is large or involves connecting lines of multiple nodes, the global coordination algorithm needs to adopt a priority-based feasibility correction plan to rearrange part of the transportation process while ensuring core needs. During this dynamic simulation phase, a visual dashboard can be set up to intuitively present the load curves and resource occupancy levels of nodes and lines during critical periods, and to issue an alarm when resource saturation or congestion trends are detected. After multiple rounds of simulation and arbitration, a set of strategies that can alleviate conflicts and maintain operational stability will be converged, and packaged and output as a conflict resolution strategy, including suggestions for modifying specific node and line parameters, such as whether it is necessary to extend the equipment opening time, merge certain delivery batches, or adjust certain loading and unloading periods. The conflict resolution strategy will also be saved in the operation record of the digital twin platform to provide a reference for subsequent iterations.

[0038] According to the conflict resolution strategy, the resource allocation is adjusted in real time. When the optimal allocation plan of logistics resources is obtained, the updated results need to be written into the state synchronization module of the multi-agent system and the digital twin platform based on the arbitration or correction in the previous step, so that each agent can immediately adapt its own scheduling plan and equipment instruction set after receiving the new resource scheduling instruction. If there is a node that has invested some manpower or material resources in the previous allocation, it is necessary to perform a return or rescheduling operation during the adjustment process to ensure system consistency. If it is necessary to perform changes on multiple mutually coupled nodes simultaneously, a synchronization window can be set in the multi-agent collaborative protocol to allow the relevant nodes to freeze the operations of certain time periods or lines first, and then restore the system clock to flow after all agents have completed the acceptance of the conflict resolution strategy and the modification of the local schedule. This can avoid parameter confusion or repeated scheduling in the middle. After real-time adjustment, the transportation resources, work team allocation and transportation schedule of each node will be more in line with actual needs, and the utilization rate of paths and equipment will be further balanced. Finally, the dynamic simulation will be re-executed for a period of time in the digital twin platform to verify the new plan. If no new conflicts or resource waste points are found, the plan will be marked as the final output, that is, the logistics resource optimization allocation plan.

[0039] In this embodiment, the IoT sensor data and historical business data of the multimodal transshipment nodes in the logistics process are multi-source fused to obtain the spatiotemporal feature vector; the P median model is used to evaluate the operational adaptability of the transshipment nodes based on the spatiotemporal feature vector to obtain a comprehensive score reflecting the node operation status; the Tabu search algorithm is used to combine and optimize the operational parameters of the transshipment nodes based on the spatiotemporal feature vector and the comprehensive score to obtain an initial transshipment plan; based on the initial transshipment plan, the transshipment nodes are multi-agent collaboratively scheduled on the digital twin platform to obtain an optimal allocation plan for logistics resources. The present invention combines multi-source fusion with an efficient optimization algorithm to achieve dynamic operational analysis and collaborative scheduling of multimodal transshipment nodes, improve logistics timeliness and resource utilization, and is suitable for complex logistics networks and changing market environments.

[0040] The above describes the intelligent logistics optimization method in the embodiment of the present invention. The following describes the intelligent logistics optimization device in the embodiment of the present invention. Figure 2 , an embodiment of the intelligent logistics optimization device in the embodiment of the present invention includes: The data fusion module 201 is used to perform multi-source fusion processing on the IoT sensor data and historical business data of the transshipment nodes of the multimodal transport in the logistics process to obtain a spatiotemporal feature vector containing static attributes and dynamic attributes; Adaptability evaluation module 202, used to evaluate the operational adaptability of the refitting node according to the spatiotemporal feature vector using the P median model, and obtain a comprehensive score reflecting the node operation status; The taboo search module 203 is used to use a Tabu search algorithm with an adaptive neighborhood structure and a dynamic taboo length to perform combined optimization on the operation parameters of the dressing node based on the spatiotemporal feature vector and the comprehensive score to obtain an initial dressing plan; The collaborative scheduling module 204 is used to perform multi-agent collaborative scheduling of the refitting nodes on the digital twin platform according to the initial refitting plan to obtain an optimal allocation plan for logistics resources. In the embodiment of the present invention, the intelligent logistics optimization device runs the above intelligent logistics optimization method, and the intelligent logistics optimization device obtains the spatiotemporal feature vector by multi-source fusion processing of the IoT sensor data and historical business data of the multimodal transshipment node in the logistics process; uses the P median model to evaluate the operational adaptability of the transshipment node according to the spatiotemporal feature vector to obtain a comprehensive score reflecting the node operation status; uses the Tabu search algorithm to combine and optimize the operating parameters of the transshipment node based on the spatiotemporal feature vector and the comprehensive score to obtain an initial transshipment plan; according to the initial transshipment plan, multi-agent collaborative scheduling is performed on the transshipment node on the digital twin platform to obtain a logistics resource optimization allocation plan. The present invention combines multi-source fusion with an efficient optimization algorithm to achieve dynamic operational analysis and collaborative scheduling of multimodal transshipment nodes, improve logistics timeliness and resource utilization, and is suitable for complex logistics networks and changing market environments.

[0041] above Figure 2 The intelligent logistics optimization device in the embodiment of the present invention is described in detail from the perspective of modular functional entities, and the intelligent logistics optimization device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0042] Figure 3 It is a structural diagram of an intelligent logistics optimization device provided by an embodiment of the present invention. The intelligent logistics optimization device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the intelligent logistics optimization device 300. Furthermore, the processor 310 can be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the intelligent logistics optimization device 300 to implement the steps of the above-mentioned intelligent logistics optimization method.

[0043] The intelligent logistics optimization device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3The structure of the intelligent logistics optimization device shown does not constitute a limitation on the intelligent logistics optimization device provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0044] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the intelligent logistics optimization method.

[0045] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0046] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0047] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent logistics optimization method, characterized in that: The intelligent logistics optimization method comprises: The IoT sensor data and historical business data of the transshipment nodes of multimodal transport in the logistics process are processed by multi-source fusion to obtain the spatiotemporal feature vector containing static and dynamic attributes; Using the P median model to evaluate the operational adaptability of the refitting node according to the spatiotemporal feature vector, a comprehensive score reflecting the node operation status is obtained; The Tabu search algorithm with an adaptive neighborhood structure and a dynamic taboo length is used to perform combined optimization on the operation parameters of the refitting node based on the spatiotemporal feature vector and the comprehensive score to obtain an initial refitting plan; According to the initial refitting plan, multi-agent collaborative scheduling is performed on the refitting nodes on the digital twin platform to obtain an optimal allocation plan for logistics resources.

2. The intelligent logistics optimization method according to claim 1, characterized in that: The multi-source fusion processing of IoT sensor data and historical business data of transshipment nodes of multimodal transport in the logistics process to obtain a spatiotemporal feature vector containing static attributes and dynamic attributes includes: The geographical location, equipment capacity and site area of ​​the transshipment node are encoded to obtain static spatial characteristics, and the real-time operation efficiency, resource utilization and delay risk of the transshipment node are quantified to obtain dynamic time characteristics; Perform time series analysis on the loading and unloading equipment status, personnel configuration and operation duration information in IoT sensor data to obtain dynamic spatiotemporal characteristics; Statistical modeling is performed on the vehicle arrival frequency, cargo type distribution, and weather impact in historical business data to obtain historical spatiotemporal patterns; The static spatial features, dynamic time features, dynamic spatiotemporal features and historical spatiotemporal patterns are multi-dimensionally fused using a tensor decomposition algorithm to obtain a spatiotemporal feature vector.

3. The intelligent logistics optimization method according to claim 1, characterized in that: The P median model is used to evaluate the operational adaptability of the refitting node according to the spatiotemporal feature vector to obtain a comprehensive score reflecting the node operation status, including: Extracting the location information of the existing distribution centers and the refitting nodes in the spatiotemporal feature vector, and constructing a static network structure according to the location information; According to the real-time operation efficiency and resource utilization in the spatiotemporal feature vector, the dynamic service capability index of each refitting node in the static network structure is calculated; Using the vehicle arrival frequency and cargo type distribution information, dynamic demand weights are assigned to each transshipment node in the static network structure; The P median model is applied to each switching node to calculate the contribution of the corresponding switching node to the static network structure, and a comprehensive score reflecting the node operation status is obtained.

4. The intelligent logistics optimization method according to claim 3 is characterized in that: The P median model is applied to each refitting node to calculate the contribution of the corresponding refitting node to the static network structure, and the comprehensive score reflecting the node operation status is obtained, including: Traverse each dressing node in the static network structure, calculate the weighted distance matrix between the traversed dressing node and other dressing nodes, and obtain the connection strength between nodes; According to the dynamic service capability index and dynamic demand weight, the objective function of the P median model is constructed to obtain the node importance evaluation function; The objective function is solved using the Lagrangian relaxation method to obtain the initial contribution of each refitting node; Apply the iterative adjustment algorithm to fine-tune the static network structure according to the initial contribution to obtain the optimized node contribution; The optimized node contribution is normalized through the sigmoid function to obtain a comprehensive score of the node operation status.

5. The intelligent logistics optimization method according to claim 1, characterized in that: The Tabu search algorithm with an adaptive neighborhood structure and a dynamic taboo length is used to perform combined optimization on the operation parameters of the refitting node based on the spatiotemporal feature vector and the comprehensive score, and the initial refitting scheme is obtained, including: Initialize the operation parameter configuration of the refitting node according to the spatiotemporal feature vector and the comprehensive score to obtain an initial solution, and call the adaptive neighborhood structure generation function in the Tabu search algorithm to generate a candidate solution set based on the initial solution; Applying the dynamic taboo table preset in the Tabu search algorithm to evaluate the objective function value of each solution in the candidate solution set to obtain the current optimal solution, wherein the dynamic taboo table automatically adjusts the taboo length according to the current number of iterations; Update the global optimal solution and determine whether it meets the stopping criteria. If not, return to the step of calling the adaptive neighborhood structure generation function in the Tabu search algorithm to generate a candidate solution set based on the initial solution until the stopping criteria are met, and use the global optimal solution as the initial replacement plan.

6. The intelligent logistics optimization method according to claim 5, characterized in that: The calling of the adaptive neighborhood structure generation function in the Tabu search algorithm to generate a candidate solution set based on the initial solution includes: Conduct sensitivity analysis on the operational parameters of the initial solution and obtain the parameter importance ranking; According to the order of parameter importance, the priority and scope of parameter adjustment are determined to obtain the parameter adjustment strategy; Based on the parameter adjustment strategy, a multi-level neighborhood structure is constructed, wherein each layer of the multi-level neighborhood structure corresponds to parameter adjustments of different importance; Perform local search in a multi-level neighborhood structure, adjust different parameters of the initial solution, and obtain an initial candidate solution set; An adaptive mechanism is used to dynamically adjust the search depth and width according to the initial candidate solution set, and the search range is expanded in a multi-level neighborhood structure to obtain a candidate solution set.

7. The intelligent logistics optimization method according to claim 1, characterized in that: According to the initial refitting plan, multi-agent collaborative scheduling is performed on the refitting nodes on the digital twin platform to obtain the optimal allocation plan for logistics resources, including: Mapping the initial dressing plan to the digital twin platform, constructing a virtual dressing node network, and obtaining a simulation environment; Assign intelligent agents to each virtual dressing node in the simulation environment and define the communication protocol between agents to obtain a multi-agent system; Executing a distributed constraint optimization algorithm based on the multi-agent system to obtain a preliminary resource allocation plan; Performing dynamic simulation verification on the preliminary resource allocation scheme, identifying potential conflicts, and obtaining conflict resolution strategies; According to the conflict resolution strategy, the resource allocation is adjusted in real time to obtain the optimal allocation plan of logistics resources.

8. An intelligent logistics optimization device, characterized in that: The intelligent logistics optimization device comprises: The data fusion module is used to perform multi-source fusion processing on the IoT sensor data and historical business data of the transshipment nodes of multimodal transport in the logistics process to obtain the spatiotemporal feature vector containing static and dynamic attributes; An adaptability evaluation module, used to evaluate the operational adaptability of the refitting node according to the spatiotemporal feature vector using a P median model, and obtain a comprehensive score reflecting the node operation status; A taboo search module, configured to use a Tabu search algorithm with an adaptive neighborhood structure and a dynamic taboo length to perform combined optimization on the operation parameters of the refitting node based on the spatiotemporal feature vector and the comprehensive score to obtain an initial refitting plan; The collaborative scheduling module is used to perform multi-agent collaborative scheduling of the refitting nodes on the digital twin platform according to the initial refitting plan to obtain an optimal allocation plan for logistics resources.

9. An intelligent logistics optimization device, characterized in that: The intelligent logistics optimization device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the intelligent logistics optimization device executes the steps of the intelligent logistics optimization method as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the intelligent logistics optimization method as described in any one of claims 1 to 7 are implemented.

Images