Logistics scheduling decision optimization method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of logistics, and provides a logistics scheduling decision optimization method and device, electronic equipment and a storage medium. Multi-source heterogeneous data are acquired in real time, space-time tensor fusion processing is performed on the multi-source heterogeneous data to obtain a dynamic space-time feature tensor, and dynamic service capability evaluation is performed on logistics nodes according to the dynamic space-time feature tensor to obtain a node contribution degree matrix. Constructing a multi-target constraint optimization model according to the node contribution degree matrix to obtain an initial scheduling strategy, performing dynamic tabu search optimization processing on the initial scheduling strategy to obtain a candidate solution set, and performing dynamic simulation verification on the candidate solution set through a digital twin model to obtain a conflict resolution strategy, and finally, adjusting the scheduling scheme in real time according to the conflict solving strategy and the optimization mode to obtain a logistics scheduling strategy optimization instruction. Through multi-source data fusion, optimization modeling and simulation verification, the logistics scheduling decision is optimized to improve the logistics scheduling optimization efficiency.

Description

Technical Field

[0001] The present application relates to the field of logistics technology, and in particular, to a method, device, electronic device, and storage medium for optimizing logistics scheduling decisions. Background Art

[0002] With the growth of global trade and the popularity of e-commerce, the demand for efficient, flexible, and cost-effective logistics transportation methods has been increasing continuously. Multimodal logistics can meet these demands and achieve the rapid flow of goods by integrating multiple transportation modes. Secondly, the application of modern information technologies, such as electronic bills of lading and cargo tracking systems, has improved the operational efficiency and transparency of multimodal logistics and reduced logistics costs.

[0003] Existing multimodal logistics scheduling usually relies only on a single data source or static conditions for planning and scheduling, and it is difficult to fully consider the dynamic changes in transshipment nodes during real-time operation and the massive real-time data provided by Internet of Things sensors. This makes the scheduling decision lack timeliness and accuracy. At the same time, traditional optimization algorithms have insufficient consideration of the dynamic coupling relationship and real-time operation data between nodes when dealing with resource scheduling and coordination problems in multimodal transport nodes, resulting in difficulty in quickly converging to a high-quality optimization solution in complex scenarios. Summary of the Invention

[0004] In view of this, the present application provides a method, device, electronic device, and storage medium for optimizing logistics scheduling decisions to solve the problem of low efficiency of multimodal logistics scheduling.

[0005] The first aspect of the present application provides a method for optimizing logistics scheduling decisions, the method comprising: Obtaining multi-source heterogeneous data in real time, and performing spatio-temporal tensor fusion processing on the multi-source heterogeneous data to obtain a dynamic spatio-temporal feature tensor; Performing dynamic service capacity evaluation on logistics nodes according to the dynamic spatio-temporal feature tensor to obtain a node contribution matrix; Constructing a multi-objective constrained optimization model according to the node contribution matrix to obtain an initial scheduling strategy; Performing dynamic tabu search optimization processing on the initial scheduling strategy to obtain a candidate solution set; Performing dynamic simulation verification on the candidate solution set through a preset digital twin model to obtain a conflict resolution strategy; Performing real-time adjustment of the scheduling plan according to the conflict resolution strategy and a preset optimization method to obtain an optimized instruction for the logistics scheduling strategy.

[0006] In an alternative embodiment, the performing spatio-temporal tensor fusion processing on the multi-source heterogeneous data to obtain a dynamic spatio-temporal feature tensor includes: Perform One - Hot encoding on the static spatial features of each logistics node in the multi - source heterogeneous data to obtain static feature vectors; Extract the dynamic time - series data of each logistics node in the multi - source heterogeneous data through a preset sliding time window, and construct a time - feature matrix based on the dynamic time - series data; Perform tightly - coupled spatio - temporal alignment on the GPS trajectory data and IMU attitude data in the multi - source heterogeneous data to obtain a motion - trajectory tensor; Perform tensor decomposition on the static feature vectors, the time - feature matrix, and the motion - trajectory tensor to obtain the dynamic spatio - temporal feature tensor.

[0007] In an alternative embodiment, the evaluating the dynamic service capabilities of logistics nodes based on the dynamic spatio - temporal feature tensor to obtain a node contribution matrix includes: Extract the static spatial features in the dynamic spatio - temporal feature tensor to obtain the static position features of each logistics node; Construct a weighted distance matrix between the logistics nodes according to the static position features; Calculate the resource - fluctuation variance of each logistics node according to the time - series data and resource - occupancy - rate data in the dynamic spatio - temporal feature tensor; Evaluate the service capabilities of the logistics nodes according to the weighted distance matrix and the resource - fluctuation variance to obtain the service - capability index of each logistics node; Calculate the service - capability index through a preset weighted - distance attenuation - resource - fluctuation correction function to obtain a dynamic service - capability matrix; Calculate the synergy - effect data between the logistics nodes according to the dynamic service - capability matrix, and perform propagation processing on the node contribution according to the synergy - effect data and a preset service - capability diffusion model to obtain the node contribution matrix.

[0008] In an alternative embodiment, the constructing a multi - objective constraint optimization model according to the node contribution matrix to obtain an initial scheduling strategy includes: Perform weighting on the node contribution matrix to obtain the weighted contribution of each logistics node; Sort the task - execution priorities of each logistics node according to the weighted contribution to obtain a priority matrix; Construct a task - allocation model according to the priority matrix and the resource - capacity constraint between preset logistics nodes; Perform time - window constraint processing on the task - allocation model to obtain a time - window constraint matrix; Solve the first objective function for the time window constraint matrix, the transportation mode connection constraint between preset logistics nodes, and the node contribution degree matrix through a preset multi-objective optimization model to obtain a multi-objective constraint optimization model; Solve the multi-objective constraint optimization model to obtain the initial scheduling strategy.

[0009] In an alternative embodiment, the dynamic tabu search optimization process for the initial scheduling strategy to obtain a candidate solution set includes: Step S41: Process the initial scheduling strategy for neighborhood structure design to obtain a first candidate solution set; Step S42: Process the first candidate solution set for second objective function evaluation to calculate the second objective function value of each candidate solution and obtain the first objective function value matrix of the first candidate solution set; Step S43: Sort the candidate solutions in the first candidate solution set according to the first objective function value matrix to obtain the first priority list of the first candidate solutions; Step S44: Perform dynamic tabu list management on the candidate solutions in the first priority list according to a preset first tabu list to obtain a second candidate solution set and a tabu length; Step S45: Update the first tabu list according to a preset tabu length adjustment strategy and the tabu length to obtain a second tabu list; Step S46: Repeat steps S42 to S43 for the second candidate solution set to obtain a second priority list; Step S47: Perform dynamic tabu list management on the candidate solutions in the second priority list according to the second tabu list to obtain the candidate solution set.

[0010] In an alternative embodiment, the dynamic simulation verification of the candidate solution set through a preset digital twin model to obtain a conflict resolution strategy includes: Construct a virtual logistics network model according to a preset digital twin model and the candidate solution set; Perform dynamic simulation processing on the logistics network model to obtain a conflict simulation data set for each candidate solution in the candidate solution set; Calculate the conflict index of each candidate solution for the conflict simulation data set to obtain a conflict index matrix; Screen the matrix elements in the conflict index matrix to obtain the optimal candidate solution corresponding to the minimum conflict index; Generate the conflict resolution strategy according to the optimal candidate solution.

[0011] In an optional embodiment, the real-time adjustment of the scheduling plan according to the conflict resolution strategy and a preset optimization method to obtain an optimized instruction for the logistics scheduling strategy includes: Performing resource reallocation processing on the optimal candidate solution according to the conflict resolution strategy to obtain a resource optimized allocation matrix; Performing path optimization processing on the resource optimized allocation matrix to obtain a transportation path optimized matrix; Adjusting the task scheduling of the logistics nodes according to the transportation path optimized matrix to obtain a task scheduling optimized matrix; Performing real-time dynamic correction processing on the task scheduling optimized matrix to obtain the optimized instruction for the logistics scheduling strategy.

[0012] The second aspect of the present application provides a logistics scheduling decision optimization device, and the device includes: A tensor fusion module, configured to obtain multi-source heterogeneous data in real time and perform spatio-temporal tensor fusion processing on the multi-source heterogeneous data to obtain a dynamic spatio-temporal feature tensor; A capability evaluation module, configured to perform dynamic service capability evaluation on logistics nodes according to the dynamic spatio-temporal feature tensor to obtain a node contribution degree matrix; An initial strategy module, configured to construct a multi-objective constraint optimization model according to the node contribution degree matrix to obtain an initial scheduling strategy; A tabu optimization module, configured to perform dynamic tabu search optimization processing on the initial scheduling strategy to obtain a candidate solution set; A simulation verification module, configured to perform dynamic simulation verification on the candidate solution set through a preset digital twin model to obtain a conflict resolution strategy; A strategy optimization module, configured to perform real-time adjustment of the scheduling plan according to the conflict resolution strategy and a preset optimization method to obtain an optimized instruction for the logistics scheduling strategy.

[0013] The third aspect of the present application provides an electronic device, and the electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the logistics scheduling decision optimization method described above are implemented.

[0014] The fourth aspect of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the logistics scheduling decision optimization method described above are implemented.

[0015] In summary, the present application at least includes the following beneficial technical effects: 1. By obtaining multi-source heterogeneous data in real time and performing spatio-temporal tensor fusion processing, changes in the logistics environment are dynamically captured, enabling scheduling decisions to be adjusted according to real-time logistics status, resource requirements, node capabilities, etc., thereby improving the timeliness of logistics scheduling decisions.

[0016] 2. Through dynamic tabu search optimization, the possible solution space is effectively explored, ensuring the accuracy and diversity of scheduling strategies. At the same time, the candidate solution set is simulated and verified through a digital twin model, further screening out the optimal solution to avoid unforeseen conflicts or inadaptability problems.

[0017] 3. Through the collaborative effect data and service capacity diffusion model, the collaboration between logistics nodes is optimized, improving the collaborative ability and scalability of the overall logistics system.

[0018] 4. Through conflict simulation and calculation of the conflict index, conflicts in the scheduling plan are effectively avoided in practical applications, providing the optimal candidate solution, and being able to perform real-time dynamic correction based on actual operation data, thereby enhancing the adaptability and stability of the logistics scheduling strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0020] Figure 1 is a flowchart of a method for optimizing logistics scheduling decisions provided by an embodiment of the present application; Figure 2 is a functional module diagram of a device for optimizing logistics scheduling decisions provided by an embodiment of the present application; Figure 3 is a schematic structural diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0022] Such as Figure 1As shown in the figure, it is a flowchart of the logistics scheduling decision optimization method provided by the embodiments of the present application. The logistics scheduling decision optimization method provided by the embodiments of the present application includes the following steps.

[0023] Step S1: Obtain multi-source heterogeneous data in real time, and perform spatio-temporal tensor fusion processing on the multi-source heterogeneous data to obtain a dynamic spatio-temporal feature tensor.

[0024] Among them, the multi-source heterogeneous data includes but is not limited to static spatial features, dynamic time series data, and cargo position and attitude information, etc. Static spatial features usually come from geographic information systems, warehouse management systems, and other external sources, and are used to record the geographical location, facility size, equipment capacity, etc. of logistics nodes; dynamic time series data is usually collected in real time through Internet of Things sensors (such as temperature and humidity sensors, vehicle-mounted GPS, and vehicle sensors, etc.), and is used to record dynamic data such as the operation efficiency, resource occupancy, transportation status, and delay information of logistics nodes; cargo position and attitude information is usually provided by global positioning system (Global Positioning System, GPS) devices and inertial measurement units (Inertial Measurement Unit, IMU) on vehicles, and is used to record and track the real-time position and attitude information of transportation tools or goods.

[0025] For the static spatial features of each logistics node, for example, the geographical coordinates, facility types, equipment capacity, etc. of the node, convert the static spatial features into static feature vectors of a fixed length, where each dimension in the static feature vector represents a specific category or feature state. For example, for the facility type of a node, several boolean values (0 or 1) can be used to indicate whether the node has certain facilities. The static feature vector is specifically represented by the following formula: Among them, represents the static feature vector of node , represents the value of the -th dimension in the static feature vector of node (for example, the category value of the geographical location), is the dimension of the static feature vector. One-Hot encoding converts discrete categorical data into digital vectors, which is convenient for spatio-temporal analysis and fusion.

[0026] Meanwhile, using a preset sliding time window (e.g., a 5-minute window), the dynamic time series data of each logistics node is extracted. Among them, the time series data includes, but is not limited to, the operation efficiency, resource occupancy rate, transportation status, etc. of the logistics node, which are data that change over time. For each logistics node, a time feature matrix is constructed using the time series data extracted by the sliding time window. Among them, each row of the time feature matrix represents the data of a time segment, and each column represents different dynamic features (e.g., efficiency, occupancy rate, etc.). The time feature matrix is specifically expressed as the following formula: Wherein, represents the node at time corresponding time feature matrix, with a dimension of , where is the time window length (e.g., 5 minutes), is the dimension of the dynamic feature. represents the value of the th dimension in the time feature matrix at time (e.g., the resource occupancy rate of the node). The time feature matrix is used to store the dynamic feature values of the logistics node in each time period, providing time series data for spatio-temporal tensor fusion.

[0027] Meanwhile, through a tightly coupled method, the trajectory data and attitude data obtained from the GPS device and the IMU device are spatio-temporally aligned to ensure the accuracy of the trajectory. Among them, the GPS data provides position coordinates (e.g., longitude and latitude), while the IMU data provides attitude information (e.g., angle change). Thus, the aligned trajectory data and attitude data are organized into a motion trajectory tensor, and each element in the motion trajectory tensor represents the three-dimensional motion trajectory information (i.e., position and attitude) of a moment node. The motion trajectory tensor is specifically expressed as the following formula: Wherein, represents the motion trajectory tensor of the node , with a dimension of , where 3 represents the position and attitude( three dimensions. represents the number of data points. …… represents the numerical values of the position and attitude. The tightly coupled spatio-temporal alignment method ensures the accuracy of the motion trajectory data and provides unified spatio-temporal data for tensor decomposition.

[0028] Finally, through the following tensor decomposition algorithm, the static feature vector, the time feature matrix, and the motion trajectory tensor are fused, so as to integrate data of different dimensions into a unified spatio-temporal feature tensor.

[0029] Among them, is denoted as the spatio-temporal feature tensor, which contains all spatio-temporal data. is denoted as the decomposition rank number. is denoted as the weight coefficient of each rank, representing the contribution of each decomposition rank. o is denoted as the outer product operator. are respectively denoted as the decomposition factors of the static feature, the time feature, and the motion trajectory. Tensor decomposition integrates feature data from different sources into a high-dimensional spatio-temporal feature tensor, which is convenient for multi-dimensional analysis, pattern discovery, and optimized decision-making.

[0030] Step S2: According to the dynamic spatio-temporal feature tensor, perform a dynamic service capacity evaluation on the logistics nodes to obtain a node contribution matrix.

[0031] The static spatial feature is the fixed attribute data of the logistics nodes and is the basis for calculating the distance between nodes and resource allocation. By extracting the static spatial feature in the dynamic spatio-temporal feature tensor, the static position feature of each logistics node is obtained. Among them, the static position feature includes, but is not limited to, information such as the geographical location, facility type, and equipment capacity of the logistics node. For example, the location of the logistics node will affect its service capacity and the distance from other nodes.

[0032] The distance between each pair of logistics nodes is determined by their geographical location or facility characteristics. By weighted processing of the distance matrix, the importance of the service capacity and distance of each node to other nodes is integrated. For example, some nodes may be far from other nodes, but their service capacity is strong, so the distance needs to be weighted and calculated according to the service capacity. Specifically, calculate the Euclidean distance or other applicable distance metrics (such as the spherical distance of geographical coordinates) between each pair of logistics nodes according to the static position feature, and perform weighted adjustment on the calculated distance according to the facility capacity of the logistics node (such as equipment capacity, storage space), so as to reflect the relative position and facility capacity between the logistics nodes and provide important data for the service capacity evaluation of the logistics nodes. The weighted distance calculation formula is as follows: Among them, is denoted as the weighted distance between logistics node i and logistics node j. is denoted as the static position feature vectors of node i and node j. is denoted as the Euclidean distance calculation. Denoted as the maximum value of all eigenvalues in the static feature vector, which is used to normalize the distance.

[0033] The resource fluctuation variance is used to measure the stability of resource allocation and utilization at a logistics node, which affects the evaluation of the service capacity of the logistics node in scheduling. For example, high fluctuations may indicate unstable resource utilization at the node, which may affect its service capacity. By using the time series data in the dynamic spatio-temporal feature tensor (such as the operation efficiency and resource occupancy rate of the node), the resource fluctuation variance of each node can be calculated. Specifically, the resource occupancy rate data of each logistics node is processed with a sliding time window to obtain the resource fluctuations, and the variance of the resource occupancy rate of the logistics node is calculated to measure the resource fluctuations. The formula for calculating the resource fluctuation variance is as follows: where denotes the resource fluctuation variance of node at time . denotes the resource occupancy rate of node at time . denotes the average value of the resource occupancy rate of node . denotes the length of the sliding time window.

[0034] Furthermore, the service capacity of each logistics node is evaluated through the weighted distance matrix and the resource fluctuation variance. It should be understood that the service capacity evaluation combines multiple factors such as the geographical location, facility capacity, and resource usage of the logistics node. A logistics node with a higher service capacity can handle more tasks or undertake more complex operations. Specifically, the service capacity index of each logistics node is calculated using the weighted distance matrix and the resource fluctuation variance, where the service capacity index is used to evaluate the processing capacity of the logistics node at a given time, considering distance, resource utilization, and the stability of the logistics node. The formula for calculating the service capacity index is as follows: where denotes the service capacity index of node at time , denotes the weighted distance from node to node , denotes the resource fluctuation variance of node at time , denotes the resource occupancy rate of node at time , denotes the resource occupancy rate of node j at time The weight of the requirement point is the experience adjustment parameter.

[0035] After obtaining the service capacity index, through the weighted distance decay-resource fluctuation correction function, calculate the final service capacity of each logistics node, and form a dynamic service capacity matrix. The dynamic service capacity matrix represents the service capacity of all logistics nodes at different times. Specifically, the dynamic service capacity matrix is obtained through the following formula: where is expressed as the service capacity expansion coefficient, which is used to adjust the service capacity of the node. is expressed as the weighted distance decay coefficient. is expressed as the resource fluctuation correction coefficient. Through the correction function, further adjust the service capacity of the logistics node, and obtain a dynamic matrix containing the service capacity of all logistics nodes at different times, so as to quantify the service potential of the logistics node in the spatio-temporal dimension.

[0036] Finally, according to the dynamic service capacity matrix, calculate the synergy effect between nodes, and perform contribution degree propagation processing through the service capacity diffusion model. Thus, considering the mutual influence and cooperation effect between nodes, obtain the contribution degree matrix of logistics nodes. Among them, the contribution degree matrix is used to describe that through the service capacity diffusion model, the node contribution degree is propagated to the entire logistics network, so as to obtain the synergy effect between logistics nodes. The contribution degree propagation process is specifically expressed by the following formula: where is expressed as the diffusion coefficient, which is determined by the transportation mode. In the embodiment of the present application, for road , railway , air . is expressed as the node coupling strength between node i and node j. Among them, is obtained through calculation, is expressed as the historical transportation frequency. is expressed as the diffusion term of the service capacity of node i at time t.

[0037] Step S3: Construct a multi-objective constraint optimization model according to the node contribution degree matrix to obtain an initial scheduling strategy.

[0038] After obtaining the node contribution degree matrix, by constructing a multi-objective constraint optimization model, an initial scheduling strategy is obtained on the premise of balancing economy, timeliness and sustainability.

[0039] It should be understood that the node contribution degree matrix contains the service capabilities of each logistics node within a specific time period. To make the multi-objective constraint optimization model more accurately reflect the importance of each node, it is necessary to weight the contribution degree according to the role and requirements of the node in the actual task. Specifically, each element in the node contribution matrix (i.e., the service capability of each logistics node) is multiplied by the weight of the logistics node in the entire scheduling process to obtain the weighted contribution degree of each node. Among them, the weights are usually preset considering factors such as business requirements, the facility capabilities of the nodes, and their roles in the logistics network. The weighting process is used to strengthen the dependence on key nodes in the multi-objective constraint optimization model, so that key nodes receive more attention during task assignment and scheduling.

[0040] After obtaining the weighted contribution degrees of each logistics node, the task execution of the logistics nodes is prioritized according to these weighted contribution degrees. The task execution priority determines which nodes should be considered first during task scheduling. Generally, nodes with a high contribution degree are considered to be able to handle tasks better, so they should have a higher priority. Specifically, all logistics nodes are sorted according to the weighted contribution degree of each logistics node to form a priority matrix. Each row in the priority matrix represents the priority of a node at different time points. The priority sorting can be carried out according to the magnitude of the weighted contribution degree, and nodes with a higher priority will be ranked in the front of the priority matrix. Through priority sorting, the processing order of each node during task scheduling can be clarified, so as to ensure that key nodes execute tasks first to avoid waste of logistics system resources.

[0041] As one of the core models in the scheduling process, the task assignment model can reasonably allocate tasks according to the priority matrix of logistics nodes and the preset resource capacity constraints between logistics nodes. Among them, the resource capacity constraints include, but are not limited to, objective conditions such as the processing capacity, storage capacity, and transportation capacity of logistics nodes that limit the amount of tasks that each logistics node can handle within a certain time. Specifically, according to the priority matrix of nodes, the task processing order of each node is determined. At the same time, tasks are allocated according to the resource capacity constraints of the nodes (such as processing speed, storage capacity, etc.) to ensure that there will be no resource overloading. Finally, the logistics node priority and resource capacity information are combined to generate the task assignment model. The task assignment model will allocate scheduling tasks according to the resource capacity and priority of each logistics node. Through the task assignment model, it is ensured that each logistics node does not exceed its resource capacity when executing tasks, thus avoiding resource conflicts during task execution.

[0042] Time window constraint is also one of the common constraints in logistics scheduling. Time window constraint means that a task must be completed within a specific time interval, which affects task arrangement and resource allocation at logistics nodes. Therefore, adding time window constraint to the obtained task allocation model can ensure that each task is completed within the specified time. By applying the time window constraint to the task allocation model, a new time window constraint matrix is generated, where the time window constraint matrix records the start and end times of each task and ensures that they comply with their respective time window constraints. The time window matrix ensures that each task is executed within an appropriate time range to avoid timeliness issues in task execution.

[0043] To integrate the time window constraint matrix, the predefined transportation mode connection constraint between logistics nodes, and the node contribution degree matrix, an optimization model (i.e., a multi-objective constraint optimization model) needs to be constructed. The multi-objective constraint optimization model includes multiple optimization objectives and corresponding constraint conditions. Each optimization objective is usually to optimize a certain aspect of the effect, such as reducing transportation costs, reducing task completion time, minimizing carbon emissions, etc., and the constraint conditions include the resource capacity of nodes, time window requirements, etc. The multi-objective constraint optimization model includes but is not limited to the first objective function and multiple constraint conditions, where the first objective function can be expressed as the following formula: where t k is the completion time of task k, c k is the execution cost of task k, is the carbon emission of transportation mode m, w 1 , w 2 , w 3 are the weight coefficients of different objectives, which are determined by methods such as the Analytic Hierarchy Process (AHP).

[0044] The constraint conditions included in the multi-objective constraint optimization model are as follows: Resource capacity constraint: where r ik is the resource requirement of task k at node i, and R i is the resource capacity of node .

[0045] Time window constraint: where a k and b k are the start and end times of the time window of task k, is the actual start time of task .

[0046] Transport mode connection constraint: Among them, and are the end time and start time of the transport modes and respectively, and is the time interval between them.

[0047] It should be understood that multiple logistics nodes (such as warehouses, transportation routes, etc.) can be regarded as multiple agents. Each agent selects an action according to the current state and adjusts its strategy according to the environmental feedback. The Multi-Agent Deep Reinforcement Learning (MADRL) framework can be applied to the environment of multiple agents, where each agent may affect the decisions of other agents when performing actions, so it is necessary to coordinate the actions of multiple agents.

[0048] The MADRL framework structure includes states, actions, and rewards. States are used to represent the state of each agent, which usually consists of information such as the current task requirements, service capabilities, and resource usage of the node. For example, the state can include the node contribution matrix and real-time demand. Actions are used to represent the actions selected by each agent, that is, the scheduling decisions taken in the current state, such as task allocation, resource scheduling, etc. Actions will affect the resource allocation and task execution order of the logistics node. Rewards are metrics used to measure the effect of each action, usually related to the optimization goal. For example, task completion time, cost reduction, etc. The complete MADRL framework consists of an Actor network and a Critic network. Through the collaborative work of the Actor network and the Critic network, the decision-making process is jointly optimized.

[0049] Among them, the high-level Actor network is responsible for generating global scheduling instructions, that is, outputting a macro scheduling decision in a given state. The high-level Actor network will output specific scheduling actions according to the contribution degree of the node (such as the service capacity and demand of the node). The input of the high-level Actor network is the state of the entire logistics network, including the node contribution matrix and real-time demand, and the output is the probability distribution of the scheduling action, that is, the selection probability of each action. The low-level Critic network is responsible for evaluating the value of the actions generated by the Actor network. The task of the Critic network is to calculate the state-action value function, that is, the value of taking a certain action in a given state. The Critic guides the Actor's decision-making through the reward signal to help it improve its strategy.

[0050] The high-level Actor network needs to be trained through the policy gradient method. The core of the policy gradient method is to update the parameters of the Actor network by calculating the gradient of the policy, promoting the network to produce better results in future decisions. The policy gradient update formula is as follows: where, is the probability of the high-level Actor network selecting action given state , is the value evaluation of the low-level Critic network for action in state , is the gradient update of the policy parameter . By updating the policy gradient, the Actor network continuously improves its scheduling decision, and ultimately can maximize the expected return.

[0051] Through multiple rounds of iterative updates of the policy gradient, the Actor and Critic networks learn according to the feedback signals of state-action pairs. In each iteration, the Actor network generates a new scheduling instruction, the Critic network evaluates its value, and updates the Actor network through the policy gradient method. This process continues until the network converges, that is, the policy no longer changes significantly. During this process, the agent (i.e., the logistics node) continuously adjusts its policy according to the environmental feedback, and finally obtains a set of optimal scheduling policies, which can minimize task completion time, cost and other objectives in complex multi-objective optimization problems. The scheduling policy (i.e., the initial scheduling policy) finally output by the trained high-level Actor network. The initial scheduling policy takes into account multiple objectives (such as task completion time, cost, etc.) and constraints (such as time windows, resource limitations, etc.), and can dynamically adapt to different logistics network states.

[0052] Step S4: Perform dynamic tabu search optimization on the initial scheduling policy to obtain a candidate solution set.

[0053] It should be understood that tabu search is a heuristic optimization algorithm based on local search. By dynamically adjusting the management mechanism of the tabu list, it avoids local optimal solutions and searches for more potential excellent solutions. Through the tabu search method, explore the neighborhood of the solution space, so as to find a more suitable scheduling policy and solve potential problems existing in the original initial policy, such as resource bottlenecks, scheduling conflicts, etc.

[0054] Design the neighborhood structure for the initial scheduling strategy, that is, define which operations can be used to generate candidate solutions. The neighborhood structure is the most critical step in tabu search, which determines the way to expand the search space. The design of the neighborhood structure should fully consider the changes of variables such as resource allocation, task allocation, and path selection in the scheduling decision. Specifically, the design of the neighborhood structure usually includes local changes to the current solution. For example, swap the paths between different tasks, reassign a task from one node to another, change the transportation mode of a task (e.g., switch from land transportation to air transportation), etc. The design of the neighborhood structure generates multiple new candidate solutions (i.e., the first candidate solution set) through small-scale adjustments, so as to explore the feasible solution space around the current scheduling strategy.

[0055] After obtaining the first candidate solution set, use the preset second objective function to evaluate the objective function of the candidate solutions in the first candidate solution set. The second objective function is usually an evaluation index related to other objectives in the initial optimization model (such as transportation cost, time, carbon emissions, etc.). Specifically, for each candidate solution, calculate its value under the second objective through the following formula. For example, for each candidate solution, calculate its comprehensive score in terms of time, cost, carbon emissions, etc.

[0056] Among them, represents the candidate solution The second objective function value under the second objective function, is the weight of the th objective function, is the candidate solution The value under the kth objective function. After obtaining the second objective function value corresponding to each first candidate solution, store the second objective function value in the first objective function value matrix. Each row of the first objective function value matrix represents the second objective function value of a first candidate solution, and each column of the matrix represents the value of the corresponding objective (such as time, cost, etc.).

[0057] Sort the candidate solutions in the first objective function value matrix according to the magnitude of the second objective function value to determine which candidate solutions best meet the optimization objectives. The sorted solutions form the first priority list. Among them, in the first priority list, the candidate solutions with lower objective function values are ranked in the front, indicating that they are the most prioritized in the optimization process.

[0058] Furthermore, use the preset first tabu list to manage the tabu solutions. The tabu list records the solutions that have been visited during the tabu search process. According to the candidate solutions in the first priority list, check whether each solution is in the tabu list one by one. If the current solution is already in the tabu list, skip it; otherwise, add it to the second candidate solution set.

[0059] While obtaining the second candidate solution set, a tabu length is also obtained by dynamically adjusting the tabu list and managing tabu solutions. The tabu length represents a measure of the expiration period or quantity of solutions recorded in the tabu list, and is usually related to the progress of the search and the number of iterations. The tabu length can be expressed by the following formula: Wherein, is the time at which the tabu length is, is the initial tabu length, is the maximum number of iterations, is the decay exponent. After obtaining the tabu length, according to the preset tabu length adjustment strategy and the tabu length, by increasing the length of the first tabu list, it is ensured that the tabu list can record solutions for a longer time to avoid the occurrence of duplicate solutions. The updated second tabu list is used to control the access of solutions in the subsequent search process to ensure the diversity of the search.

[0060] After obtaining the second candidate solution set and the second tabu list, the candidate solutions in the second candidate solution set are again evaluated for the objective function and sorted. By recalculating the objective function values and sorting, a new priority list (i.e., the second priority list) is obtained. And the second tabu list is used to manage the candidate solutions in the second priority list to obtain an optimized final candidate solution set.

[0061] Step S5, dynamically simulate and verify the candidate solution set through a preset digital twin model to obtain a conflict resolution strategy.

[0062] It should be understood that each solution in the candidate solution set usually includes information such as task assignment, path selection, resource scheduling, etc. By mapping the scheduling decisions of each candidate solution into a preset digital twin model, a virtual logistics network model is constructed. The mapping operations include transportation nodes, task execution times, path connections, etc. The construction of the logistics network model also needs to consider limiting factors such as resources, task flows, and time windows between logistics nodes. Among them, the digital twin model is a real-time synchronous virtual environment that can simulate the operation of the physical logistics network. Through the digital twin model, the virtual environment can be connected to the data and behaviors of the real world, and the dynamic changes in the network can be reflected in real time. The constructed logistics network model is used to simulate the candidate solutions, evaluate the execution effects of each solution, check for conflicts, and provide a basis for subsequent optimization.

[0063] After constructing the virtual logistics network, next, dynamic simulation is carried out for each solution in the candidate solution set to simulate the task execution process and collect the simulation data of each candidate solution. The candidate solutions are loaded through the digital twin platform to perform simulations according to the logistics network model and the candidate solutions, so as to simulate task execution, resource occupancy, path selection, etc. During the simulation process, the following conflict data are collected in real time, and the collected conflict data are stored in the corresponding conflict simulation data set: Number of resource conflicts: Record the number of times tasks conflict on the same resource during the simulation process.

[0064] Time window violation amount: Calculate the total amount by which task execution exceeds the predetermined time window through the following formula: where is the actual completion time of task , is the end time of the time window of task .

[0065] Path crossing probability: Calculate the probability of path crossing through the following formula: where is the indicator function indicating whether paths and cross.

[0066] Use the digital twin platform for simulation to simulate the execution process of candidate solutions in a virtual environment. The simulation process includes: task scheduling and allocation, transportation path selection, use of node resources, time window constraints, etc. Through simulation, the conflict simulation data set of each candidate solution can be obtained. These data include information such as task delays, resource conflicts, and path crossings, recording the execution of candidate solutions in the virtual environment. Simulation is used to identify possible problems that may occur in candidate solutions during actual operations, such as resource conflicts and path crossings. Through dynamic simulation, these problems can be discovered and corrected in advance to ensure that the finally selected candidate solutions can operate efficiently in the real environment.

[0067] For the conflict simulation data set of each candidate solution, calculate the conflict index through the following formula. Among them, the conflict index is an indicator used to measure the degree of conflict between different scheduling decisions in the candidate solution. For example, resource conflicts, time conflicts, path conflicts, etc. will all affect the overall effect of the candidate solution.

[0068] where is the conflict index of the th candidate solution, is the The conflict metric between a candidate solution and the th task, is the weight of the conflict type, indicating the impact of different conflict types on the total conflict. Among them, the conflict types include but are not limited to resource conflicts, time conflicts, and path conflicts. A resource conflict means that a resource occupancy conflict occurs when tasks are allocated on the same node. A time conflict means that the execution time of a task conflicts with the time window of other tasks or resources. A path conflict means that two tasks conflict on the same path, resulting in traffic bottlenecks or delays. Finally, all the calculated conflict indices are aggregated and stored as a conflict index matrix.

[0069] It should be understood that each row of the conflict index matrix represents a candidate solution, and each column represents the conflict value of the candidate solution under different conflict types. The candidate solution corresponding to the minimum value in the conflict index matrix is the optimal solution. By screening the candidate solution with the smallest conflict index from the conflict index matrix, it can be ensured that the finally selected optimal candidate solution will generate the least conflicts during actual execution, and has good feasibility and execution effects.

[0070] Finally, based on the selected optimal candidate solution, a corresponding conflict resolution strategy is generated. The conflict resolution strategy will include how to resolve resource conflicts, path conflicts, time conflicts, etc., to ensure that the optimal solution can be smoothly executed during actual scheduling. Among them, the content of the conflict resolution strategy usually includes rescheduling the time of conflicting tasks, adjusting the resource allocation, and modifying the selection of transportation paths, etc. Specifically, by analyzing the specific conflicts in the optimal candidate solution, a detailed conflict handling plan is generated. For example, if a path conflict cannot be avoided, an alternative path can be selected for scheduling; if a resource conflict cannot be resolved, the priority of the resource can be adjusted.

[0071] Step S6: According to the conflict resolution strategy and the preset optimization method, perform real-time adjustment of the scheduling plan to obtain an optimized instruction for the logistics scheduling strategy.

[0072] Since the candidate solution has gone through the conflict simulation and screening process, but there may still be potential conflicts in resource allocation, it is necessary to adjust the conflict resolution strategy to ensure that various resources can be reasonably allocated. By reviewing the scheduling decisions in the optimal candidate solution, identify the set of conflicting tasks in resource allocation. For example, some tasks may generate resource conflicts on the same node or device, or the resource utilization rate of some nodes exceeds the maximum available capacity. According to the conflict resolution strategy, adjust the conflicting resources. For example, if two tasks use the same resource on the same node, the time window of the task can be adjusted, or the task can be transferred to other idle resources. For the set of conflicting tasks, reallocate resources according to the task priority. Adjust the resources through the following formula: Among them, represents the task after resource reallocation, is the original resource allocation, is the total amount of allocable resources, is the task priority.

[0073] Furthermore, after obtaining the reallocated resources, a resource optimization allocation matrix is generated according to the reallocated resources. Among them, the resource optimization allocation matrix records the resource allocation situations of all nodes and tasks, and ensures the minimization of resource conflicts through optimization. The resource optimization allocation matrix can be expressed by the following formula: Among them, represents the resource requirement of task at node , represents the task set of node , represents the resource capacity of node . The resource optimization allocation matrix indicates that the total resource requirement of the node needs to be less than or equal to the resource capacity of the node to ensure the rationality of resource allocation.

[0074] After resource reallocation, next, path optimization needs to be performed on the resource optimization allocation matrix. Since the reallocation of tasks and resources may lead to redundancy or irrationality in the original transportation paths, it is necessary to adjust the transportation paths according to the current resource allocation situation (i.e., the resource optimization allocation matrix). According to the new resource allocation situation, analyze which transportation paths are inefficient or conflict. For example, the traffic volume on some paths is too large, resulting in delays, while other paths can carry more tasks and are not fully utilized. Adjust the transportation paths through algorithms to reduce the transportation time and cost of tasks. Path optimization should not only consider transportation costs, but also consider transportation timeliness and the connection situation between nodes. After adjustment through algorithms, multiple transportation path optimization matrices can be obtained. The transportation path optimization matrix describes the adjusted task transportation paths, including new path allocations and the transportation time, cost, etc. of each path. The transportation path optimization matrix can be expressed by the following formula: Among them, represents the transportation time or cost from node to node , represents the transportation time or cost of task on path , Total number of tasks. The transportation path optimization matrix represents calculating the total transportation time or cost on each path in the transportation path matrix.

[0075] After obtaining the transportation path optimization matrix, an improved clonal selection algorithm is adopted to evaluate the path quality and conflict situation in the transportation path optimization matrix: where, is the quality evaluation of the path, is the transportation distance or time on the path, is the penalty coefficient for path conflict, is the minimum distance between the path and obstacle j. According to select the transportation path optimization matrix that minimizes the delay and cost of each path after path optimization.

[0076] Path optimization will affect the task execution order and resource scheduling of each node. Therefore, it is necessary to reorder and adjust the task scheduling globally. According to the optimized path matrix and resource allocation in the transportation path optimization matrix, adjust the scheduling order of tasks. The key is to ensure that tasks can be executed according to the shortest path and the most reasonable resource allocation plan, avoiding resource idleness or long task waiting times. At the same time, according to the path optimization, rearrange the time window of tasks to ensure that tasks will not be delayed due to path optimization, especially the scheduling of cross-node tasks. The task scheduling adjustment aims to ensure that after path optimization, the task execution order is reasonable and the resource allocation is more efficient. The generated task scheduling optimization matrix after adjustment records the adjusted task execution order and time window to ensure that tasks can be executed according to the optimal path and resources.

[0077] Finally, after resource allocation, path optimization, and task scheduling optimization, it is necessary to perform real-time dynamic correction on the final task scheduling optimization matrix to dynamically adjust the scheduling plan according to possible changes during the actual execution process, ensuring that it can adapt to emergencies or environmental changes (such as equipment failures, traffic jams, etc.) during actual execution. Use real-time data to monitor the task execution situation and dynamically adjust the task scheduling according to feedback information (such as task completion time, resource occupancy, etc.). According to the dynamically corrected task scheduling plan, update the scheduling instructions and send them to each logistics node. The finally generated optimized logistics scheduling strategy instructions include the execution details of task scheduling, such as resource allocation, path selection, time arrangement, etc., to ensure the efficient execution of logistics scheduling.

[0078] This application is applied to the field of logistics technology. By obtaining multi-source heterogeneous data in real time, performing spatio-temporal tensor fusion processing on the multi-source heterogeneous data to obtain a dynamic spatio-temporal feature tensor, evaluating the dynamic service capabilities of logistics nodes based on the dynamic spatio-temporal feature tensor to obtain a node contribution matrix, constructing a multi-objective constrained optimization model based on the node contribution matrix to obtain an initial scheduling strategy, performing dynamic tabu search optimization processing on the initial scheduling strategy to obtain a candidate solution set, dynamically simulating and verifying the candidate solution set through a digital twin model to obtain a conflict resolution strategy, and finally making real-time adjustments to the scheduling plan according to the conflict resolution strategy and the optimization method to obtain an optimized instruction for the logistics scheduling strategy. This application optimizes logistics scheduling decisions through multi-source data fusion, optimization modeling, and simulation verification to improve the efficiency of logistics scheduling optimization.

[0079] As Figure 2 shown, it is a functional module diagram of a logistics scheduling decision optimization device provided by an embodiment of this application.

[0080] In some embodiments, the logistics scheduling decision optimization device 2 may include multiple functional modules composed of computer program segments. The computer programs of each program segment in the logistics scheduling decision optimization device 2 can be stored in the memory of the server and executed by at least one processor to execute (see details in Figure 1 the description) the functions of the logistics scheduling decision optimization method.

[0081] In this embodiment, according to the functions it executes, the logistics scheduling decision optimization device 2 can be divided into multiple functional modules. The functional modules may include: a tensor fusion module 21, a capability evaluation module 22, an initial strategy module 23, a tabu optimization module 24, a simulation verification module 25, and a strategy optimization module 26. The module referred to in the present invention means a series of computer program segments that can be executed by at least one processor and can complete fixed functions, and are stored in the memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0082] The tensor fusion module 21 is used to obtain multi-source heterogeneous data in real time and perform spatio-temporal tensor fusion processing on the multi-source heterogeneous data to obtain a dynamic spatio-temporal feature tensor.

[0083] In an optional implementation manner, the tensor fusion module 21 is specifically used for: Performing One-Hot encoding processing on the static spatial features of each logistics node in the multi-source heterogeneous data to obtain a static feature vector; Extracting the dynamic time series data of each logistics node in the multi-source heterogeneous data through a preset sliding time window and constructing a time feature matrix according to the dynamic time series data; Perform a tightly coupled spatio-temporal alignment process on the GPS trajectory data and IMU attitude data in the multi-source heterogeneous data to obtain a motion trajectory tensor; Perform a tensor decomposition process on the static feature vector, the time feature matrix, and the motion trajectory tensor to obtain the dynamic spatio-temporal feature tensor.

[0084] The capability evaluation module 22 is used to perform a dynamic service capability evaluation on the logistics nodes according to the dynamic spatio-temporal feature tensor to obtain a node contribution degree matrix.

[0085] In an alternative embodiment, the capability evaluation module 22 is specifically configured to: Extract the static spatial features in the dynamic spatio-temporal feature tensor to obtain the static position features of each logistics node; Construct a weighted distance matrix between the logistics nodes according to the static position features; Calculate the resource fluctuation variance of each logistics node according to the time series data and resource occupancy rate data in the dynamic spatio-temporal feature tensor; Evaluate the service capabilities of the logistics nodes according to the weighted distance matrix and the resource fluctuation variance to obtain the service capability index of each logistics node; Calculate the service capability index through a preset weighted distance attenuation-resource fluctuation correction function to obtain a dynamic service capability matrix; Calculate the synergy effect data between the logistics nodes according to the dynamic service capability matrix, and perform a propagation process on the node contribution degree according to the synergy effect data and a preset service capability diffusion model to obtain the node contribution degree matrix.

[0086] The initial strategy module 23 is used to construct a multi-objective constraint optimization model according to the node contribution degree matrix to obtain an initial scheduling strategy.

[0087] In an alternative embodiment, the initial strategy module 23 is specifically configured to: Perform a weighting process on the node contribution degree matrix to obtain the weighted contribution degrees of each logistics node; Sort the task execution priorities of each logistics node according to the weighted contribution degrees to obtain a priority matrix; Construct a task allocation model according to the priority matrix and the resource capacity constraints between the preset logistics nodes; Perform a time window constraint process on the task allocation model to obtain a time window constraint matrix; Solve the first objective function for the time window constraint matrix, the pre-set transportation mode connection constraint between logistics nodes, and the node contribution matrix through a pre-set multi-objective optimization model to obtain a multi-objective constraint optimization model; Solve the multi-objective constraint optimization model to obtain the initial scheduling strategy.

[0088] The tabu optimization module 24 is used to perform dynamic tabu search optimization processing on the initial scheduling strategy to obtain a candidate solution set.

[0089] In an optional embodiment, the tabu optimization module 24 is specifically used for: Step S41: Perform neighborhood structure design processing on the initial scheduling strategy to obtain a first candidate solution set; Step S42: Perform second objective function evaluation processing on the first candidate solution set to calculate the second objective function value of each candidate solution and obtain the first objective function value matrix of the first candidate solution set; Step S43: Sort the candidate solutions in the first candidate solution set according to the first objective function value matrix to obtain the first priority list of the first candidate solutions; Step S44: Perform dynamic tabu list management processing on the candidate solutions in the first priority list according to a pre-set first tabu list to obtain a second candidate solution set and a tabu length; Step S45: Update the first tabu list according to a pre-set tabu length adjustment strategy and the tabu length to obtain a second tabu list; Step S46: Repeat steps S42 to S43 for the second candidate solution set to obtain a second priority list; Step S47: Perform dynamic tabu list management processing on the candidate solutions in the second priority list according to the second tabu list to obtain the candidate solution set.

[0090] The simulation verification module 25 is used to perform dynamic simulation verification on the candidate solution set through a pre-set digital twin model to obtain a conflict resolution strategy.

[0091] In an optional embodiment, the simulation verification module 25 is specifically used for: Construct a virtual logistics network model according to a pre-set digital twin model and the candidate solution set; Perform dynamic simulation processing on the logistics network model to obtain a conflict simulation data set for each candidate solution in the candidate solution set; Calculate the conflict index of each candidate solution for the conflict simulation data set to obtain a conflict index matrix; Screen the matrix elements in the conflict index matrix to obtain the optimal candidate solution corresponding to the minimum conflict index; Generate the conflict resolution strategy according to the optimal candidate solution.

[0092] The policy optimization module 26 is configured to perform real-time adjustment of the scheduling plan according to the conflict resolution strategy and a preset optimization method to obtain a logistics scheduling strategy optimization instruction.

[0093] In an alternative embodiment, the policy optimization module 26 is specifically configured to: Perform resource reallocation processing on the optimal candidate solution according to the conflict resolution strategy to obtain a resource optimization allocation matrix; Perform path optimization processing on the resource optimization allocation matrix to obtain a transportation path optimization matrix; Adjust the task scheduling of the logistics nodes according to the transportation path optimization matrix to obtain a task scheduling optimization matrix; Perform real-time dynamic correction processing on the task scheduling optimization matrix to obtain the logistics scheduling strategy optimization instruction.

[0094] It should be understood that the various variations and specific embodiments of the methods provided in the above embodiments are equally applicable to the logistics scheduling decision optimization device of this embodiment. Through the foregoing detailed description of the logistics scheduling decision optimization method, those skilled in the art can clearly know the implementation method of the logistics scheduling decision optimization device in this embodiment. For the sake of brevity of the specification, it will not be elaborated here.

[0095] As Figure 3 shown, it is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

[0096] In a preferred embodiment of the present invention, the electronic device 3 may include, but is not limited to: a memory 31, at least one processor 32, and at least one communication bus 33.

[0097] Those skilled in the art should understand that Figure 3 the structure of the electronic device 3 shown does not constitute a limitation on the embodiments of the present invention. The electronic device 3 may further include more or fewer other hardware or software than shown, or different component arrangements.

[0098] In some embodiments, the electronic device 3 is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits, programmable gate arrays, digital signal processors, and embedded devices, etc.

[0099] It should be noted that the electronic device 3 is only an example. Other existing or future possible electronic products that can be adapted to this application should also be included within the protection scope of this application and are hereby incorporated by reference.

[0100] In some embodiments, a computer program is stored in the memory 31. When the computer program is executed by the at least one processor 32, all or part of the steps in the described logistics scheduling decision optimization method are implemented. The memory 31 includes a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electrically-erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc memories, magnetic disk memories, tape memories, or any other computer-readable medium capable of carrying or storing data. Further, the computer-readable storage medium may mainly include a program storage area and a data storage area. Among them, the program storage area may store an operating system, application programs required for at least one function, and the like.

[0101] In some embodiments, the at least one processor 32 is the control core (Control Unit) of the electronic device 3, connecting various components of the entire electronic device 3 through various interfaces and circuits. By running or executing the programs or modules stored in the memory 31, and by calling the data stored in the memory 31, it executes various functions of the electronic device 3 and processes data. For example, when the at least one processor 32 executes the computer program stored in the memory 31, all or part of the steps in the logistics scheduling decision optimization method described in the embodiments of this application are implemented; or all or part of the functions of the logistics scheduling decision optimization device are implemented. The at least one processor 32 may be composed of integrated circuits. For example, it may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same or different functions packaged, including a combination of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips.

[0102] In some embodiments, the at least one communication bus 33 is arranged to enable connection communication between the memory 31, the at least one processor 32, and the like. Although not shown, the electronic device 3 may further include a power source (such as a battery) for supplying power to each component. Preferably, the power source can be logically connected to the at least one processor 32 through a power management device, so as to implement functions such as management of charging, discharging, and power consumption management through the power management device. The power source may further include any components such as one or more DC or AC power sources, a recharge device, a power failure detection circuit, a power converter or inverter, and a power status indicator. The electronic device 3 may further include a variety of sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated herein.

[0103] The integrated unit implemented in the form of a software functional module as described above can be stored in a computer-readable storage medium. The above software functional module stored in a storage medium includes several instructions for causing an electronic device (which may be a personal computer, an electronic device, or a network device, etc.) or a processor to execute a part of the methods described in various embodiments of the present application.

[0104] In several embodiments provided by the present application, it should be understood that the disclosed apparatus and method can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0105] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical units, and may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0106] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present application should be covered within the protection scope of the present application.

Claims

1. A logistics scheduling decision optimization method, characterized in that: The method comprises: Acquire multi-source heterogeneous data in real time, and perform spatiotemporal tensor fusion processing on the multi-source heterogeneous data to obtain a dynamic spatiotemporal feature tensor; Performing a dynamic service capability evaluation on the logistics nodes according to the dynamic spatiotemporal feature tensor to obtain a node contribution matrix; Constructing a multi-objective constraint optimization model according to the node contribution matrix to obtain an initial scheduling strategy; Performing dynamic tabu search optimization processing on the initial scheduling strategy to obtain a candidate solution set; Dynamically simulating and verifying the candidate solution set through a preset digital twin model to obtain a conflict resolution strategy; The scheduling plan is adjusted in real time according to the conflict resolution strategy and the preset optimization method to obtain the logistics scheduling strategy optimization instruction.

2. The logistics scheduling decision optimization method according to claim 1 is characterized in that: The performing spatiotemporal tensor fusion processing on the multi-source heterogeneous data to obtain a dynamic spatiotemporal feature tensor comprises: One-Hot encoding is performed on the static spatial features of each logistics node in the multi-source heterogeneous data to obtain a static feature vector; Extracting and processing the dynamic time series data of each logistics node in the multi-source heterogeneous data through a preset sliding time window, and constructing a time feature matrix according to the dynamic time series data; Performing tight coupling spatiotemporal alignment processing on the GPS trajectory data and the IMU attitude data in the multi-source heterogeneous data to obtain a motion trajectory tensor; The static feature vector, the time feature matrix and the motion trajectory tensor are subjected to tensor decomposition processing to obtain the dynamic spatiotemporal feature tensor.

3. The logistics scheduling decision optimization method according to claim 2 is characterized in that: The dynamic service capability evaluation of the logistics node according to the dynamic spatiotemporal feature tensor to obtain the node contribution matrix includes: Extracting and processing the static spatial features in the dynamic spatiotemporal feature tensor to obtain the static position features of each logistics node; Constructing a weighted distance matrix between the logistics nodes according to the static location features; Calculate the resource fluctuation variance of each logistics node according to the time series data and resource occupancy rate data in the dynamic spatiotemporal feature tensor; Evaluating the service capacity of the logistics node according to the weighted distance matrix and the resource fluctuation variance to obtain a service capacity index of each logistics node; The service capability index is calculated by a preset weighted distance decay-resource fluctuation correction function to obtain a dynamic service capability matrix; The synergy data between the logistics nodes are calculated according to the dynamic service capability matrix, and the node contributions are propagated according to the synergy data and a preset service capability diffusion model to obtain the node contribution matrix.

4. The logistics scheduling decision optimization method according to claim 1 is characterized in that: The constructing of a multi-objective constraint optimization model according to the node contribution matrix to obtain an initial scheduling strategy includes: Performing weighted processing on the node contribution matrix to obtain the weighted contribution of each logistics node; Sorting the task execution priority of each logistics node according to the weighted contribution to obtain a priority matrix; Constructing a task allocation model according to the priority matrix and the resource capacity constraints between preset logistics nodes; Performing time window constraint processing on the task allocation model to obtain a time window constraint matrix; The first objective function is solved for the time window constraint matrix, the transportation mode connection constraints between the preset logistics nodes, and the node contribution matrix through a preset multi-objective optimization model to obtain a multi-objective constraint optimization model; The multi-objective constraint optimization model is solved to obtain the initial scheduling strategy.

5. The logistics scheduling decision optimization method according to claim 1 is characterized in that: The performing dynamic tabu search optimization processing on the initial scheduling strategy to obtain a candidate solution set includes: Step S41, performing neighborhood structure design processing on the initial scheduling strategy to obtain a first candidate solution set; Step S42: performing a second objective function evaluation process on the first candidate solution set to calculate the second objective function value of each candidate solution and obtain a first objective function value matrix of the first candidate solution set; Step S43: sorting the candidate solutions in the first candidate solution set according to the first objective function value matrix to obtain a first priority list of the first candidate solutions; Step S44: performing dynamic taboo table management processing on the candidate solutions in the first priority list according to the preset first taboo table to obtain a second candidate solution set and a taboo length; Step S45: updating the first taboo table according to a preset taboo length adjustment strategy and the taboo length to obtain a second taboo table; Step S46, repeatedly performing steps S42 to S43 on the second candidate solution set to obtain a second priority list; Step S47: Perform dynamic taboo table management processing on the candidate solutions in the second priority list according to the second taboo table to obtain the candidate solution set.

6. The logistics scheduling decision optimization method according to claim 1 is characterized in that: The dynamically simulating and verifying the candidate solution set by using a preset digital twin model to obtain a conflict resolution strategy includes: Constructing a virtual logistics network model according to the preset digital twin model and the candidate solution set; Performing dynamic simulation processing on the logistics network model to obtain a conflict simulation data set for each candidate solution in the candidate solution set; Calculating the conflict index of each candidate solution for the conflict simulation data set to obtain a conflict index matrix; Screening the matrix elements in the conflict index matrix to obtain an optimal candidate solution corresponding to a minimum conflict index; The conflict resolution strategy is generated according to the optimal candidate solution.

7. The logistics scheduling decision optimization method according to claim 6 is characterized in that: The real-time adjustment of the scheduling scheme according to the conflict resolution strategy and the preset optimization method to obtain the logistics scheduling strategy optimization instruction includes: Perform resource reallocation processing on the optimal candidate solution according to the conflict resolution strategy to obtain a resource optimization allocation matrix; Performing path optimization processing on the resource optimization allocation matrix to obtain a transportation path optimization matrix; Adjusting the task scheduling of the logistics node according to the transportation path optimization matrix to obtain a task scheduling optimization matrix; The task scheduling optimization matrix is ​​subjected to real-time dynamic correction processing to obtain the logistics scheduling strategy optimization instruction.

8. A logistics scheduling decision optimization device, characterized in that: The device comprises: A tensor fusion module is used to acquire multi-source heterogeneous data in real time and perform spatiotemporal tensor fusion processing on the multi-source heterogeneous data to obtain a dynamic spatiotemporal feature tensor; A capability evaluation module, used to perform dynamic service capability evaluation on logistics nodes according to the dynamic spatiotemporal feature tensor to obtain a node contribution matrix; An initial strategy module, used to construct a multi-objective constraint optimization model according to the node contribution matrix to obtain an initial scheduling strategy; A taboo optimization module, used for performing dynamic taboo search optimization processing on the initial scheduling strategy to obtain a candidate solution set; A simulation verification module, used to dynamically simulate and verify the candidate solution set through a preset digital twin model to obtain a conflict resolution strategy; The strategy optimization module is used to adjust the scheduling plan in real time according to the conflict resolution strategy and the preset optimization method to obtain the logistics scheduling strategy optimization instruction.

9. An electronic device, characterized in that: The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the logistics scheduling decision optimization method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the logistics scheduling decision optimization method according to any one of claims 1 to 7 are implemented.

Images