Logistics scheduling method and system

Abstract

The invention relates to the field of logistics information processing, in particular to a logistics scheduling method and system.The method comprises the steps that multi-source data such as traffic flow, weather, orders and logistics network states are collected in real time, and dynamic logistics state data are generated through a data fusion model; on the basis of the data, performing fine-grained modeling of path fragments by using distributed simulation nodes, performing simulation of the state of the path fragments in combination with a cellular automaton model, and performing adaptive adjustment by using a reinforcement learning algorithm to generate optimized path decision data; global state modeling and resource scheduling optimization of a multi-layer feedback interaction mechanism are carried out through a multi-agent system, and finally dynamic prediction and automatic adjustment of a global logistics state are realized through model prediction control and a multi-layer self-correction mechanism. According to the invention, the flexibility and response speed of logistics scheduling are effectively improved, and efficient execution of logistics tasks is ensured.

Description

Technical Field

[0001] The present invention relates to the field of logistics information processing, and in particular to a logistics scheduling method and system. Background Art

[0002] Logistics scheduling is a key link in the logistics system, which aims to ensure the rational allocation of logistics resources and the smooth execution of tasks. The existing logistics scheduling scheme (Chinese invention patent, publication number: CN118278715B, name: logistics scheduling method, device, system and medium) is mainly predicted and scheduled through periodic tables or historical data. For example, the existing technology uses a logistics periodic table to record the transportation time between outlets, and uses this to estimate the time for the goods to arrive at each outlet. Although the existing technology can make a certain degree of prediction in combination with historical data, its scheduling scheme mainly relies on static data, and lacks the ability to respond in real time and flexibly adjust when facing dynamically changing logistics scenarios. In addition, the existing technology lacks modeling and real-time feedback adjustment for dynamic traffic conditions. Especially when the amount of logistics tasks fluctuates greatly or the external conditions change dramatically, it is often difficult to adapt in time, resulting in problems such as waste of resources and backlog of goods. Summary of the invention

[0003] In view of the many problems existing in the above-mentioned prior art, the present invention provides a logistics scheduling method and system. The present invention collects multi-source data in real time, generates dynamic logistics status data using a data fusion model, and performs path segment modeling and optimization through distributed simulation nodes combined with a cellular automaton model. The present invention realizes dynamic scheduling and adaptive optimization of logistics network resources through a multi-agent system to improve overall logistics efficiency and reduce costs.

[0004] A logistics scheduling method comprises the following steps: Collect traffic flow data, meteorological data, order data and logistics network status data, pre-process and standardize the collected data, and generate dynamic logistics status data through data fusion model; Based on dynamic logistics status data, distributed simulation nodes are used to perform fine-grained modeling of path segments, the status of path segments is simulated through a cellular automaton model, and a reinforcement learning algorithm is used to adaptively adjust the simulation results to generate optimized path decision data; Based on the route decision data, a multi-agent system is constructed to optimize the resource scheduling of each logistics network through global state modeling and multi-layer feedback interaction mechanism to generate agent scheduling data; Based on the agent scheduling data, the global logistics status is dynamically predicted using model predictive control. The system parameters are automatically adjusted through the multi-layer self-correction mechanism of the real-time correction layer, trend correction layer and global correction layer to generate global optimization data. Based on global optimization data and agent scheduling data, logistics tasks are executed and the status of logistics tasks is monitored in real time, and feedback data is obtained to optimize the logistics scheduling plan.

[0005] Preferably, when simulating the state of a path segment by a cellular automaton model, the cellular automaton model is composed of a plurality of cellular units, each cellular unit represents a minimum unit length in the path segment, the state of the cellular unit includes traffic flow, road occupancy rate and average vehicle speed, and the state of each cellular unit is updated according to a preset state transition rule, and the state transition rule is set by traffic flow prediction based on historical data and congestion propagation characteristics, and is used to simulate changes in roads under different traffic conditions.

[0006] Preferably, the reward function of the reinforcement learning algorithm is defined as follows: in, Represents the reward value; Represents the shortest travel time of a route segment; represents the traffic density of the path segment; Indicates the congestion level of the path segment; , and is an adjustable weight coefficient used to adjust the influence of each factor in the reward function.

[0007] Preferably, when constructing a multi-agent system, the resource status, path information, task load and interaction relationship of each logistics point are taken as input, and a global state graph is constructed through a graph convolutional network. The nodes of the graph convolutional network represent each logistics point, and the edges of the graph convolutional network represent the path information between adjacent logistics points. The parameters of the graph convolutional network are iteratively optimized through the status data of each logistics point to establish a global dependency relationship between logistics points.

[0008] Preferably, the multi-layer feedback interaction mechanism consists of a global optimization layer and a local feedback layer. The global optimization layer is responsible for the allocation of global resources. The local feedback layer monitors the task execution status of each logistics network point in real time and regularly sends the monitoring results as update signals to the global optimization layer. The monitoring indicators of the local feedback layer include resource utilization, task completion rate and task execution time.

[0009] Preferably, the objective function of the model predictive control is defined as follows: in, represents the target value of global optimization; Indicates Estimated travel time for each route; Indicates Resource utilization of each path; Indicates The travel risk of the route; , and They are weighted coefficients of the path segments, respectively, and are used to adjust the influence of each factor in the objective function.

[0010] Preferably, the real-time correction layer, the trend correction layer and the global correction layer dynamically update their respective control parameters through automatic differentiation technology. The real-time correction layer calculates the difference between the real-time monitoring data and the global prediction data through automatic differentiation, and adjusts the control parameters based on the difference; the trend correction layer updates the trend adjustment parameters according to the medium-term trend deviation calculated by automatic differentiation, and the global correction layer performs global optimization adjustment according to the long-term deviation of the system.

[0011] Preferably, when the execution status of the logistics task is monitored in real time through the feedback and monitoring mechanism, the content of the real-time monitoring includes the transportation time of the vehicle, the loading status of the goods, the real-time congestion of the road and the location information of the vehicle. The real-time monitoring data is obtained through vehicle-mounted sensors, GPS positioning system and road monitoring equipment, and transmitted to the central dispatching system for analysis in real time.

[0012] Preferably, the path simulation model and global optimization parameters are periodically improved based on the cumulative analysis of historical monitoring data, and the cumulative analysis optimizes the path selection strategy and resource scheduling parameters by identifying high-congestion areas, fluctuation characteristics of resource utilization and changing trends in vehicle driving efficiency in various time periods.

[0013] A system for implementing the logistics scheduling method, the system comprising: The data collection module is used to collect traffic flow data, meteorological data, order data and logistics network status data, and pre-process and standardize the collected data to generate dynamic logistics status data through the data fusion model; A simulation node module is used to perform fine-grained modeling of path segments using distributed simulation nodes based on the dynamic logistics state data, simulate the path segment states through a cellular automaton model, and adaptively adjust the simulation results using a reinforcement learning algorithm to generate optimized path decision data; A multi-agent scheduling module is used to build a multi-agent system based on the path decision data, optimize the resource scheduling of each logistics network point through global state modeling and multi-layer feedback interaction mechanism, and generate agent scheduling data; A global prediction control module is used to dynamically predict the global logistics status based on the intelligent agent scheduling data using model prediction control, automatically adjust system parameters through a multi-layer self-correction mechanism of a real-time correction layer, a trend correction layer, and a global correction layer, and generate global optimization data; The task execution and monitoring module is used to execute logistics tasks and monitor the status of logistics tasks in real time based on the global optimization data and the intelligent agent scheduling data, and obtain feedback data to optimize the logistics scheduling plan.

[0014] Compared with the prior art, the advantages and beneficial effects of the present invention are: The present invention collects traffic flow data, meteorological data, order data and logistics network status data in real time, and performs dynamic fusion analysis to generate dynamic logistics status data, thereby ensuring the real-time and flexibility of logistics scheduling; The present invention uses distributed simulation nodes to perform fine-grained modeling of path segments, combines a cellular automaton model to simulate dynamic traffic conditions, and uses a reinforcement learning algorithm to perform adaptive adjustments, thereby achieving collaborative optimization of path optimization and resource scheduling. The present invention optimizes the resource scheduling of each logistics network point through global state modeling and multi-layer feedback interaction mechanism, and enhances the system's adaptability to changes in logistics task volume; the present invention effectively compensates for the deficiencies of the prior art in dynamic changes and real-time response capabilities through innovative modeling and optimization strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 It is a schematic diagram of the real-time acquisition and data fusion model in the present invention; Figure 3 A schematic diagram of the global state modeling and feedback interaction mechanism of the multi-agent system in the present invention; Figure 4 Schematic diagram of model predictive control and multi-layer self-correction mechanism in the present invention; Figure 5 It is a structural block diagram of the system in the present invention. DETAILED DESCRIPTION

[0016] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0017] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0018] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0019] like Figure 1 As shown, a logistics scheduling method includes the following steps: Collect traffic flow data, meteorological data, order data and logistics network status data, pre-process and standardize the collected data, and generate dynamic logistics status data through data fusion model; The present invention collects traffic flow data, meteorological data, order data and logistics network status data from multiple data sources to achieve real-time, accurate prediction and decision-making of logistics scheduling. The data collection here is completed by sensor equipment, vehicle-mounted systems and external data interfaces deployed in various logistics network points, transport vehicles and road monitoring. Traffic flow data usually comes from road traffic monitoring systems, vehicle-mounted GPS devices, etc., which record the real-time flow, average speed and congestion of each road; meteorological data is provided by meteorological stations and third-party meteorological interfaces to obtain real-time weather information such as temperature, precipitation, wind speed, etc.; order data is generated by the order management system of the logistics company, including information such as customer order time, destination, cargo type, delivery time, etc.; logistics network status data includes storage status, inventory status, personnel allocation, vehicle idle rate, etc.

[0020] After collecting these raw data, necessary preprocessing and standardization are performed. Preprocessing includes data cleaning, denoising, filtering of redundant information and other operations to ensure the accuracy and consistency of the data. Standardization refers to converting data from different sources into the same dimension and unit to facilitate subsequent data fusion processing. Data preprocessing and standardization help eliminate noise, reduce deviations between different data sources, and improve data availability.

[0021] Dynamic logistics status data is generated through a data fusion model. In the present invention, the data fusion model adopts a time series prediction model based on a long short-term memory (LSTM) network and a multi-source data fusion algorithm of a Bayesian network. The main function of the LSTM network is to model and predict time series data, and is particularly suitable for processing dynamic data with strong time series characteristics such as traffic flow and meteorological data. The Bayesian network is used to model the dependency relationship of multi-source heterogeneous data (such as order data and logistics network status data), and fuse the information of these data sources through reasoning and association analysis.

[0022] In actual operation, the application of data fusion model includes two specific steps: first, the time series prediction of traffic flow and meteorological data is carried out based on LSTM model to generate short-term prediction data of road flow and meteorological change trend data; then, these prediction data are combined with order data and logistics network status data through Bayesian network to deduce the overall status of the current logistics system and generate "dynamic logistics status data". For example, when LSTM network predicts that a major transportation route will be seriously congested within the next hour, the data fusion model combines the inventory status, order priority and other information of logistics network through Bayesian network to calculate the logistics pressure of alternative routes, thereby forming more accurate status data.

[0023] Through the combination of the above data collection, preprocessing and standardization processing, and the efficient application of the data fusion model, the present invention can generate highly consistent and real-time updated dynamic logistics status data. These data reflect the current operation status of the logistics system, including the real-time situation of road traffic, the potential impact of meteorological changes, the resource status of each logistics network point, and the urgency of order processing.

[0024] The generation of this dynamic logistics status data enables the present invention to grasp the overall operation status of the logistics system in real time and provide accurate basic information for subsequent path simulation and intelligent scheduling. This not only improves the flexibility of logistics scheduling, but also enhances the ability to respond to emergencies (such as bad weather, traffic accidents, etc.), ultimately making logistics scheduling more efficient and intelligent.

[0025] For example, in a logistics dispatch during a peak period, the traffic flow data collected by the system indicated that a trunk road was about to be congested; at the same time, the meteorological data forecast showed that there would be heavy rainfall in the next two hours, and there were a number of high-priority fresh food orders in the order data that needed to be delivered as soon as possible. Through the time series prediction of traffic and meteorological data by the LSTM network, the system identified that the trunk road might be more congested due to rainfall in the future. Subsequently, the data fusion model combined the predicted congestion information with the urgency of the order, the inventory and resource status of each logistics outlet through Bayesian network analysis, and finally generated dynamic logistics status data, reflecting the load pressure of multiple alternative routes. Based on this dynamic logistics status data, the system can adjust the dispatch plan in time and select an alternative route with lower load and less impact of rainfall, so as to ensure the timely delivery of orders.

[0026] Based on dynamic logistics status data, distributed simulation nodes are used to perform fine-grained modeling of path segments, the status of path segments is simulated through a cellular automaton model, and a reinforcement learning algorithm is used to adaptively adjust the simulation results to generate optimized path decision data; like Figure 2 As shown, in the process of logistics scheduling, the present invention uses dynamic logistics status data to guide path planning and resource scheduling. First, the current road traffic, weather conditions, logistics network status, vehicle status and other information are obtained through dynamic logistics status data. These data reflect the global operation of the logistics system and the local status of each path segment. Then, the path segments are fine-grained modeled using distributed simulation nodes. In the present invention, the main purpose of the design of distributed simulation nodes is to distribute the state modeling and simulation of each path segment to improve the parallel computing efficiency of the system.

[0027] Specifically, each distributed simulation node models a path segment, which refers to a subdivision unit of a specific road or path in the entire logistics network. In order to better capture the dynamic changes of the path segment, the present invention uses a cellular automaton model for simulation. The characteristic of the cellular automaton model is that it can divide a path segment into multiple minimum cellular units, each of which represents a fixed length on the path. By defining the initial state of these cellular units and the state transition rules, the change process of traffic flow and road congestion on the path can be simulated.

[0028] In actual operation, the state transition rules of the cellular automaton model are mainly set by the following parameters: the occupancy state of the cell unit (indicating whether there is a vehicle occupying it), vehicle speed changes, road congestion propagation, etc. In each distributed simulation node, the operation of the cellular automaton is controlled by these state transition rules, thereby simulating the dynamic state of each path segment. Multiple distributed simulation nodes work together to simulate all path segments of the entire logistics network through efficient parallel processing. The generated simulation data includes real-time status information of the path segments (such as traffic density, congestion level, average driving speed, etc.).

[0029] The present invention uses a reinforcement learning algorithm to adaptively adjust the simulation results. The core of reinforcement learning is to optimize the system strategy by setting a reward function. Here, the task of the reinforcement learning algorithm is to continuously adjust the path selection strategy according to the simulation data generated by the cellular automaton to obtain the optimal path planning. Specifically, the reinforcement learning algorithm collects the state information of each path segment from the simulation node and optimizes the path selection strategy based on the set reward function. The definition of the reward function usually includes the following key factors: the average travel time of the path, the degree of congestion of the road, the risk of path passage, etc. The reinforcement learning algorithm selects the path segment combination with the largest reward value by continuously updating the strategy.

[0030] Through the joint optimization of simulation modeling and reinforcement learning, the present invention generates optimized path decision data, which includes recommended optimal path combinations, estimated travel time, resource scheduling solutions, etc.

[0031] By using distributed simulation nodes to perform fine-grained modeling of path segments and combining state simulation with cellular automaton models, the present invention can accurately capture the dynamic state of each path in the logistics network, thereby improving the accuracy and real-time performance of path planning. In addition, the use of reinforcement learning algorithms to adaptively adjust the simulation results can effectively cope with complex changes in the logistics network and dynamically optimize the path selection strategy. The generated optimized path decision data provides a more scientific path planning and resource scheduling solution for the logistics system, thereby improving the efficiency of logistics scheduling and reducing the risk of delays and congestion.

[0032] For example, during a peak period of logistics distribution, the information obtained by the logistics system from the dynamic logistics status data shows that the traffic flow on the main road A is high and congested. Through distributed simulation nodes, each path segment is modeled in a fine-grained manner. The system simulates the state of multiple subdivided segments of the main road A, and uses the cellular automaton model to predict the congestion propagation trend of the road, and obtains the traffic density and average driving speed of each path segment. At the same time, the reinforcement learning algorithm dynamically adjusts the path selection strategy based on the simulation data, and the reward function focuses on reducing delivery time and avoiding risks. Through continuous optimization, the reinforcement learning algorithm finally recommends a secondary road B as an alternative route, and generates optimized path decision data including estimated driving time and resource scheduling. Based on the optimized path decision data, the system adjusts the scheduling of logistics vehicles in real time, avoids congestion on the main road A, and improves distribution efficiency.

[0033] Preferably, when simulating the state of a path segment by a cellular automaton model, the cellular automaton model is composed of a plurality of cellular units, each cellular unit represents a minimum unit length in the path segment, the state of the cellular unit includes traffic flow, road occupancy rate and average vehicle speed, and the state of each cellular unit is updated according to a preset state transition rule, and the state transition rule is set by traffic flow prediction based on historical data and congestion propagation characteristics, and is used to simulate changes in roads under different traffic conditions.

[0034] In the present invention, in order to accurately simulate the dynamic changes of each path segment in the logistics network, the Cellular Automaton Model (CA model) is used to model and simulate the path segment state. The core idea of ​​the cellular automaton model is to decompose complex traffic flow, congestion propagation and other problems into many small "cell units" and update the state of these units according to the preset state transition rules.

[0035] Specifically, a path segment is a fine-grained division of a section of road in a logistics network, and a cell unit represents the minimum unit length in the path segment. Each cell unit can be regarded as a microscopic road "block", and its state includes information such as traffic flow, road occupancy rate, and average vehicle speed. By defining the state of these cell units, the cellular automaton model can efficiently simulate complex road state changes at a low computational cost.

[0036] In the cellular automaton model, the state of each cell unit is controlled by a preset state transition rule. The core of the state transition rule is to determine the state change of the cell unit at the next moment based on the state of the current cell unit and its adjacent cells. Here, the setting of the state transition rule is realized by combining the traffic flow prediction and congestion propagation characteristics of historical data. Traffic flow prediction refers to the use of historical road flow data to infer the traffic conditions of each road in the future period; while the congestion propagation characteristics indicate how the congestion on the road spreads from one path segment to the adjacent segment under specific traffic conditions.

[0037] In order to better realize the dynamic simulation of road status, the present invention uses multiple traffic characteristic data to constrain the state transition rules. For example, traffic flow data determines the traffic density of each cell unit in different time periods, the road occupancy rate reflects the usage of the cell unit, and the average vehicle speed is used to represent the traffic efficiency of the path segment. During the simulation process, the state of each cell unit is continuously updated according to its state transition rules, and finally forms the dynamic change process of the entire path segment.

[0038] Through this simulation based on the cellular automaton model, the present invention can accurately capture the microscopic changes of each path segment in the logistics network. Since the states between cellular units are highly correlated, under certain conditions (such as when congestion occurs), the cellular automaton model can effectively simulate the propagation path and congestion degree of road congestion. This allows the solution to obtain the refined state of each road in real time during path planning, thereby making a more scientific path selection. In addition, due to the parallel computing characteristics of the cellular automaton model, the solution can simulate a large number of path segments in a relatively short period of time, thereby improving the real-time performance and response speed of the entire logistics scheduling system.

[0039] For example, in a city distribution scenario during peak hours, a main road in the logistics network may show signs of congestion due to excessive traffic density. Through the cellular automaton model, the system divides this main road into several cellular units, each of which represents a road block of a specific length. The state information of the cellular unit includes: the current traffic density (i.e., traffic flow), the proportion of vehicles occupying the road (i.e., road occupancy rate), and the average speed of vehicles on this road. Next, the system dynamically updates the states of these cellular units according to the set state transition rules (for example, when the traffic density exceeds a certain threshold, the average speed of the vehicle will decrease, resulting in an increase in the road occupancy rate of the cellular unit).

[0040] At the same time, by analyzing the state changes of multiple adjacent cell units, the cellular automaton model can simulate the propagation path of congestion. For example, when a cell unit on the main road is congested, the state of the adjacent cell units will change accordingly according to the state transition rules, forming a gradual diffusion process of congestion. The system combines historical traffic data and congestion propagation characteristics to predict which path segments the congestion will spread to, thereby providing real-time basis for path planning. Finally, based on these simulation results, the system selects an alternative route with lower traffic density and expected less congestion to ensure the smooth completion of the delivery task.

[0041] Preferably, the reward function of the reinforcement learning algorithm is defined as follows: in, Indicates the reward value; Represents the shortest travel time of a route segment; represents the traffic density of the path segment; Indicates the congestion level of the path segment; , and is an adjustable weight coefficient used to adjust the influence of each factor in the reward function.

[0042] The present invention optimizes logistics path decisions through a reinforcement learning algorithm, and the design of the reward function is a key element in the reinforcement learning algorithm. The reward function is used to measure the positive or negative impact of the system's choice at each step, thereby providing a basis for the strategy update of the reinforcement learning algorithm. In the scenario of logistics scheduling, in order to achieve efficient path selection, the reward function needs to comprehensively consider multiple factors, such as the travel time of the path segment, traffic density, and congestion level.

[0043] The goal of the reinforcement learning algorithm is to continuously adjust the strategy so that the reward value Maximize to find the optimal path selection solution.

[0044] Shortest driving time : In route selection, travel time is a core indicator for measuring logistics efficiency. The shortest travel time of a route segment represents the optimal travel time to complete the segment under ideal conditions without severe traffic interference. Selecting a route with a shorter travel time can improve delivery efficiency, so in the reward function, the contribution of travel time to the reward value is set to a positive value.

[0045] Traffic density Traffic density is an important parameter to measure road traffic conditions. It indicates the number of vehicles on a road per unit length. An increase in traffic density is usually accompanied by a decrease in traffic speed and potential congestion risks. Therefore, in order to avoid selecting path segments with high traffic density, traffic density is used as a negative weight term in the reward function.

[0046] Congestion level : The congestion level reflects the current congestion status of the road. The congestion level is usually determined by the ratio between the traffic volume and the driving speed of the road. For example, when the traffic volume of the road increases significantly and the average driving speed decreases, the congestion level will increase significantly. In the reward function, the increase in congestion level means that the traffic efficiency of the road decreases, so it is also used as a negative weight term.

[0047] Weight coefficient , and Used to adjust the impact of each factor on the reward value. These weights can be set according to the actual logistics scenario. For example, in some scenarios with high requirements for delivery time, you can appropriately increase The weight of the traffic volume can be increased in urban roads with dense traffic. and The weights are used to better avoid high-density or high-congestion paths.

[0048] Through the design of the above-mentioned reward function, the present invention can balance the travel time, traffic density and congestion level of the path segments to achieve accurate optimization of the logistics path. In the continuous iterative training, the reinforcement learning algorithm gradually selects the optimal path segment combination by maximizing the reward function value, thereby generating optimized path decision data. The optimized path decision data not only improves the transportation efficiency of logistics, but also reduces delays caused by congestion, ensuring the smooth progress of logistics tasks.

[0049] For example, in a logistics distribution system, the system is selecting the best path from warehouse A to distribution point B. The dynamic logistics status data collected by the system shows that road C is the shortest direct path, but the actual driving time during peak hours is longer, the traffic density is higher, and the average speed is lower. In order to find a better path combination, the system optimizes through reinforcement learning algorithms and reward functions.

[0050] In each route selection, the system calculates a reward value based on the current travel time, traffic density and congestion level. In the early stages of optimization, due to the imperfect adjustment of the algorithm to the weight of the reward function, some path segments with short travel time but high traffic density may be selected, resulting in low reward values. However, as training continues, the reinforcement learning algorithm gradually discovers the negative impact of traffic density and congestion on driving efficiency, and finds a better path combination by continuously adjusting the path selection strategy.

[0051] Ultimately, the system chose an alternative route that was slightly longer but had lower traffic density. By optimizing the reward function and updating the strategy, the system generated path decision data that was more in line with actual traffic conditions, significantly improving delivery efficiency and reducing waiting time in congested sections.

[0052] Preferably, when constructing a multi-agent system, the resource status, path information, task load and interaction relationship of each logistics point are taken as input, and a global state graph is constructed through a graph convolutional network. The nodes of the graph convolutional network represent each logistics point, and the edges of the graph convolutional network represent the path information between adjacent logistics points. The parameters of the graph convolutional network are iteratively optimized through the status data of each logistics point to establish a global dependency relationship between logistics points.

[0053] In modern logistics scheduling, in order to better handle the complex resource scheduling and task allocation problems between logistics outlets, the present invention introduces the concept of multi-agent system. The multi-agent system can regard each logistics outlet as an independent "agent" and complete logistics tasks through mutual collaboration and interaction. In order to achieve global modeling and optimization of the multi-agent system, the present invention uses a graph convolutional network (GCN) to construct a global state diagram that reflects the relationship between each logistics outlet.

[0054] In the present invention, the resource status, path information, task load and interaction relationship of each logistics network point are used as the input of the graph convolutional network. Specifically, the resource status includes the available inventory, number of vehicles, staffing, etc. of the logistics network point; the path information includes the traffic conditions, path length, transportation time, etc. between the logistics network points; the task load refers to the number of orders currently processed by the logistics network point, the order priority and the urgency of the task; the interaction relationship describes the collaboration of each logistics network point in task allocation and resource sharing. By integrating these input data into the graph convolutional network, a global state graph can be constructed.

[0055] The nodes of the graph convolutional network represent each logistics network point, and the feature data of the nodes are composed of information such as the resource status and task load of each network point. The edges of the graph convolutional network represent the path information between adjacent logistics networks, and the weights of the edges can be set based on parameters such as the length of the path, travel time, and road conditions. In the process of constructing the global state graph, the graph convolutional network optimizes parameters through multiple iterations. During each iteration, the graph convolutional network merges the features of each node with the features of its adjacent nodes through a convolution operation and updates the state of each node. This convolution operation can capture the global dependencies between logistics networks, that is, how the state of a certain network point is affected by other adjacent networks.

[0056] In this way, the graph convolutional network builds a global state graph in the multi-agent system, reflecting the relationships between resources, paths, and tasks among all logistics nodes.

[0057] By constructing a global state graph, the present invention can fully grasp the state of each node in the logistics network and their mutual relationship, providing a reliable basis for subsequent resource scheduling and path optimization. The introduction of graph convolutional networks enables each node (i.e., logistics node) in the multi-agent system to make scheduling decisions from a global perspective, avoiding suboptimal paths or resource allocation problems caused by relying only on local information. At the same time, through the optimization iteration of the global state graph, the parameters of each node can be dynamically adjusted, making the system more adaptable when facing resource changes or task adjustments.

[0058] For example, suppose that in a nationwide logistics network, a logistics company has multiple logistics outlets distributed in different cities. Each logistics outlet is responsible for receiving and processing a certain number of orders, and allocating resources and selecting routes based on the destination of the orders. During peak hours, due to the surge in order volume, the pressure on resource scheduling between logistics outlets increases. In order to ensure efficient task execution and resource allocation, the system first collects the current resource status of each logistics outlet (such as the number of idle vehicles, inventory quantity, etc.), task load (such as the number of orders and priority), route information (such as traffic conditions between outlets), and the interaction relationship between outlets (such as collaborative processing of orders or resource sharing).

[0059] The system then uses this data as input to construct a global state graph through a graph convolutional network. The nodes of the graph convolutional network represent each logistics point, and the edges represent the path information between adjacent logistics points. During multiple iterations of the graph convolutional network, each layer of the network will perform convolution fusion on the features of each node and edge and update the state of the node. In this process, the interaction between logistics points is optimized, and the state information of the node gradually integrates the state of other adjacent points, thereby constructing a globally optimized state graph.

[0060] Finally, the system uses the global state graph to generate an optimized resource scheduling plan. For certain urgent and high-priority orders, the system will prioritize the deployment of outlets with sufficient resources and relatively smooth routes to ensure the timely completion of logistics tasks. This global state-based resource scheduling and route selection avoids the waste of resources and delays caused by local decisions in traditional logistics scheduling.

[0061] Based on the route decision data, a multi-agent system is constructed to optimize the resource scheduling of each logistics network through global state modeling and multi-layer feedback interaction mechanism to generate agent scheduling data; Preferably, the multi-layer feedback interaction mechanism consists of a global optimization layer and a local feedback layer. The global optimization layer is responsible for the allocation of global resources. The local feedback layer monitors the task execution status of each logistics network point in real time and regularly sends the monitoring results as update signals to the global optimization layer. The monitoring indicators of the local feedback layer include resource utilization, task completion rate and task execution time.

[0062] In the present invention, if Figure 3 As shown in the figure, in order to optimize the resource scheduling in the logistics network, a multi-agent system (MAS) is constructed and optimized through global state modeling and multi-layer feedback interaction mechanism. First, the multi-agent system is initialized based on the path decision data, and each logistics network point is regarded as an independent "agent". The path decision data is used as input, including the travel time, traffic conditions, risk assessment of path selection and other information of each path. These data are used to determine the role and corresponding load of each logistics network point in task allocation. By combining these agents together, the system forms a collaborative, multi-point linkage scheduling network to ensure efficient resource utilization in logistics tasks.

[0063] In order to globally grasp the status of each logistics network point, the present invention adopts the method of global state modeling. Global state modeling is based on path decision data, comprehensively considering the resource status, task load, path information and mutual dependencies of each logistics network point, so as to build an overall global view. This view reflects the state changes of all intelligent agents in real time, so that the system can monitor and adjust the operating status of each logistics network point in a global scope, ensuring coordination and efficiency in a changing logistics environment.

[0064] The multi-layer feedback interaction mechanism is the core part of the present invention, which is composed of a global optimization layer and a local feedback layer. This multi-level feedback mechanism enables the system to allocate and optimize resources from a global perspective, and flexibly adjust through local real-time feedback. The global optimization layer is responsible for uniformly allocating resources for the entire logistics network based on global status information. For example, the system will adjust the resource allocation of different outlets based on information such as the number of vehicles, order priority, and traffic conditions of each logistics outlet to ensure that the optimal resource utilization state is achieved at the global level. The decision-making process of the global optimization layer takes into account multiple factors, including the travel time of the path, traffic density, the urgency of the task, etc., to form a global optimization plan for resource allocation.

[0065] The role of the local feedback layer is to monitor the task execution of each logistics network in real time and pass these monitoring data as feedback signals to the global optimization layer. The monitoring indicators of the local feedback layer include resource utilization, task completion rate and task execution time. Through these real-time monitoring data, the system can understand the task execution status of each logistics network at the local level, such as the current resource usage, task completion progress and distribution efficiency. When the task execution of a logistics network deviates from expectations, the local feedback layer will pass this information as an update signal to the global optimization layer, and the global optimization layer will readjust the resource allocation strategy accordingly, thereby achieving global and local coordinated optimization.

[0066] For example, in urban logistics dispatch during peak periods, the system uses global state modeling and path decision data to understand the resource situation and current task load of each logistics network. Based on this information, the global optimization layer of the system uniformly allocates resources and prioritizes the allocation of resources to networks with urgent tasks. However, as tasks are executed, the resource utilization rate or task completion progress of some networks may be abnormal. At this time, the local feedback layer will monitor these anomalies in real time and promptly feedback the abnormal information to the global optimization layer. The global optimization layer adjusts the resource allocation plan based on the latest monitoring data, thereby avoiding problems such as uneven resource allocation or task delays.

[0067] Through the above-mentioned global state modeling and multi-layer feedback interaction mechanism, the present invention can maintain a high degree of flexibility and global optimization capabilities in a dynamically changing logistics network. The unified allocation of the global optimization layer and the real-time adjustment of the local feedback layer enable the system to respond quickly and flexibly in different task scenarios, thereby improving the overall resource utilization efficiency and task completion rate of the logistics network.

[0068] Based on the agent scheduling data, the global logistics status is dynamically predicted using model predictive control. The system parameters are automatically adjusted through the multi-layer self-correction mechanism of the real-time correction layer, trend correction layer and global correction layer to generate global optimization data. In the present invention, if Figure 4 As shown in the figure, in order to dynamically predict and adaptively adjust the global logistics status, Model Predictive Control (MPC) is introduced based on the agent scheduling data, and the multi-layer self-correction mechanism of real-time correction layer, trend correction layer and global correction layer is combined to automatically adjust the system parameters and generate global optimization data. The key to the solution is to cope with the complex changes in the logistics network through multi-level dynamic prediction and feedback regulation to ensure the continuity and accuracy of resource allocation and path optimization.

[0069] After obtaining the agent scheduling data, the system uses model predictive control (MPC) to dynamically predict the global logistics status. Model predictive control is a control method based on dynamic models. By constructing a mathematical model of the global logistics status, MPC can predict the changing trend of the logistics system in the future and dynamically adjust the logistics scheduling strategy based on the prediction results. Specifically, the MPC prediction model uses the agent scheduling data as input, combined with global data such as resource status, task load, and path information in the logistics network, to simulate the possible changes of the logistics network in the future. This prediction process is achieved by solving the optimization problem. MPC will solve the optimal control strategy in each scheduling process, so that the future system state is as close to the expected optimal state as possible.

[0070] In order to further improve the robustness and adaptability of the system, the present invention introduces a multi-layer self-correction mechanism, including a real-time correction layer, a trend correction layer, and a global correction layer. These three correction layers each undertake different correction tasks and perform adaptive adjustments through different time scales and data levels. First, the real-time correction layer is used to make subtle adjustments to system parameters in the short term. Its core task is to identify the deviation between the current system and the expected state by real-time monitoring of various indicators in the logistics network (such as resource utilization, road traffic status, etc.), and quickly correct the control parameters based on this deviation. Through the adjustment of the real-time correction layer, the system can quickly respond to emergencies (such as sudden congestion on a road or temporary shortage of resources at a logistics outlet) and make flexible control strategy adjustments.

[0071] The main task of the trend correction layer is to monitor and correct the mid-term system change trend. Some changes in the logistics network have long-term cumulative effects, such as traffic pressure during certain peak hours, periodic changes in the number of orders, etc. These trend changes may have a greater impact on global scheduling. The trend correction layer updates parameters by analyzing historical data and trend deviations to ensure that system parameters can be dynamically adjusted during the prediction process as the trend of logistics status changes, thereby improving the prediction accuracy of MPC.

[0072] The global correction layer is responsible for adjusting the long-term global optimization of the entire system. The role of the global correction layer is to optimize the system parameters globally on a longer time scale. It analyzes long-term logistics data, identifies potential problems in global scheduling, and periodically adjusts the global optimization strategy. For example, when the system finds that certain specific paths always have bottlenecks due to uneven resource allocation or excessive task load in long-term monitoring, the global correction layer will reallocate resources or adjust the path planning strategy based on these problems, thereby achieving continuous optimization on a global scale.

[0073] Through the above multi-layer self-correction mechanism, the present invention can not only respond to short-term changes in real time, but also adapt to mid- and long-term trend changes and global optimization needs, thus forming a highly flexible scheduling system. The final generated global optimization data includes resource allocation plans, path selection strategies, scheduling adjustment suggestions, etc. These data can guide the further optimization of logistics scheduling and improve the overall logistics efficiency and accuracy.

[0074] For example, in a logistics dispatch task in a large city, the system uses MPC to dynamically predict traffic and resource changes in the next two hours based on the real-time dispatch data of intelligent agents, and identifies that some main roads may be congested during the upcoming peak hours. The real-time correction layer adjusts the strategy of some path selections based on the monitored traffic conditions and optimizes the allocation of vehicle resources. The trend correction layer updates the priority weights of certain path segments based on historical data of previous peak periods to reduce errors in the prediction process. After analyzing long-term historical data, the global correction layer found that a certain logistics network was often in a state of resource shortage, which affected the overall dispatch efficiency, so the resource allocation of the network was replanned. Finally, the system generated global optimization data including resource scheduling and path optimization, and implemented new dispatch strategies based on this, significantly improving the efficiency of completing logistics tasks.

[0075] Preferably, the objective function of the model predictive control is defined as follows: in, represents the target value of global optimization; Indicates Estimated travel time for each route; Indicates Resource utilization of each path; Indicates The travel risk of the route; , and They are weighted coefficients of the path segments, respectively, and are used to adjust the influence of each factor in the objective function.

[0076] In logistics scheduling, the model predictive control (MPC) method dynamically adjusts the control parameters of the system by solving a set of optimization problems to achieve global optimization of the entire logistics network. In order to quantify the effect of global optimization, this paper defines an objective function , which is used to comprehensively evaluate multiple factors of each path segment to provide an overall optimization criterion. The design of the objective function mainly considers three core indicators: estimated travel time , Resource Utilization and passage risks .

[0077] Estimated driving time :In logistics scheduling, travel time is an important factor in measuring the quality of a route. Short travel time means that goods can reach their destination faster, thereby improving distribution efficiency. The calculation of travel time is based on the current status of the route segment (such as traffic flow, road congestion, etc.) and the system's dynamic prediction model. The weighted coefficient of this factor Adjustments can be made based on the urgency of the task and time-sensitive needs, allowing the solution to flexibly respond to different delivery needs.

[0078] Resource Utilization :Each path in the logistics network not only undertakes the transportation task, but also needs to consider the effective use of resources along the way. Resource utilization refers to the effectiveness of each path segment in allocating resources (such as vehicles, storage, etc.). Improving resource utilization helps reduce waste and reduce operating costs. Therefore, in the objective function, resource utilization is As an independent optimization factor. Its weighted coefficient Depending on the relative importance of the resources, adjustments can be made based on the emphasis of the scheduling strategy.

[0079] Passage Risk :In logistics scheduling, safety is a crucial indicator. Traffic risk mainly refers to the traffic risk of each path segment in a specific period of time, such as the probability of road accidents, the impact of weather changes on the path, etc. In this invention, the traffic risk is evaluated through a comprehensive analysis of path historical data, traffic forecasts and external factors (such as weather forecasts). In order to minimize safety hazards and unexpected situations, the risk term is introduced into the objective function. , and through the weighting coefficient To control the impact of this factor on the optimization results.

[0080] Preferably, the real-time correction layer, the trend correction layer and the global correction layer dynamically update their respective control parameters through automatic differentiation technology. The real-time correction layer calculates the difference between the real-time monitoring data and the global prediction data through automatic differentiation, and adjusts the control parameters based on the difference; the trend correction layer updates the trend adjustment parameters according to the medium-term trend deviation calculated by automatic differentiation, and the global correction layer performs global optimization adjustment according to the long-term deviation of the system.

[0081] In the present invention, in order to achieve precise control and dynamic optimization of the logistics system, each control parameter is dynamically updated through automatic differentiation technology, which plays a role in the real-time correction layer, trend correction layer and global correction layer respectively. Specifically, automatic differentiation technology is a numerical method that can efficiently calculate the gradient of the function, so that the system can make subtle adjustments and optimizations to the parameters in each layer of the self-correction mechanism. The introduction of automatic differentiation enables the entire logistics scheduling system to flexibly respond to real-time changes, medium-term trends and long-term optimization needs, thereby realizing dynamic adjustment of scheduling strategies at multiple levels.

[0082] In the present invention, the real-time correction layer, the trend correction layer and the global correction layer respectively undertake control tasks under different time scales. In order to achieve refined control and adjustment, these three layers dynamically update their respective control parameters through automatic differentiation technology. The core of automatic differentiation is to accurately calculate the gradient information of the function in a complex calculation process, and the gradient information plays a vital role in the optimization problem. By calculating the gradient information, the system can clearly know the sensitivity of the scheduling objective function to each parameter under the current control parameters, so as to make targeted parameter adjustments.

[0083] The main task of the real-time correction layer is to quickly correct system deviations in the short term. It calculates the difference between real-time monitoring data and global forecast data through automatic differentiation technology, and dynamically adjusts control parameters based on this difference. Specifically, after obtaining the real-time monitoring data of the logistics network, the system will compare these data with the global forecast data. Automatic differentiation technology can automatically calculate the degree of influence of the difference between the two on the current control parameters and form gradient information. By using this gradient information, the real-time correction layer can timely adjust the key control parameters in the scheduling strategy, such as the number of logistics vehicles allocated, the priority of path selection, etc., so as to eliminate deviations in the short term and improve the real-time and response speed of scheduling.

[0084] The task of the trend correction layer is to dynamically adjust the medium-term trends in the logistics network. Some changes in the logistics network have trend and periodic characteristics, such as peak traffic flow in certain time periods, seasonal changes in order volume, etc. The trend correction layer calculates the medium-term trend deviation based on automatic differentiation technology, and uses this deviation information to update the trend adjustment parameters. The role of automatic differentiation here is to accurately calculate the impact of medium-term deviations on trend adjustment parameters, and to ensure that the system can continue to adapt to trend changes by continuously iterating and correcting trend adjustment parameters. In specific implementation, the system will regularly perform trend analysis on historical data, and update trend adjustment parameters in combination with current medium-term change forecasts, such as timely increasing the proportion of vehicle dispatch during peak periods or adjusting the distribution of storage resources to better cope with cyclical changes in the logistics network.

[0085] The global correction layer is responsible for making comprehensive adjustments to the long-term optimization of the entire logistics network. Long-term changes in logistics networks usually involve the overall allocation of resources, the optimization of path strategies over a long period of time, etc. The global correction layer needs to make comprehensive adjustments to the system over a longer time scale. Through automatic differentiation technology, the global correction layer can calculate the impact of long-term deviations on global optimization parameters, thereby achieving long-term strategy adjustments. For example, when the system's long-term monitoring finds that a certain path has become a bottleneck due to low resource utilization efficiency or excessively high risk factors, the global correction layer will re-optimize the global resource allocation strategy based on the automatic differentiation results of the long-term deviation to achieve the overall optimal state.

[0086] By introducing automatic differentiation technology, the present invention can accurately calculate the gradient information of parameters at various time scales, so that the real-time correction layer, trend correction layer and global correction layer can more efficiently and accurately update parameters and adjust strategies in their respective control tasks. This multi-level self-correction mechanism ensures that the logistics network can maintain good adaptability and scheduling effects when facing changes at different time scales. The real-time correction layer can respond quickly to sudden changes, the trend correction layer ensures that the system remains stable in mid-term changes, and the global correction layer provides guarantees for long-term resource optimization, so that the system can continue to operate stably in a complex logistics environment.

[0087] For example, in the logistics distribution of a large city, the real-time correction layer quickly adjusted the path selection and allocation quantity of logistics vehicles based on the difference between the current traffic flow data calculated by automatic differentiation and the global prediction, thereby avoiding serious congestion caused by traffic accidents on a certain main road. At the same time, the trend correction layer analyzed the frequent resource shortage problems of some logistics outlets during holidays through automatic differentiation, so the resource scheduling ratio and distribution strategy during peak periods were adjusted. After analyzing long-term data, the global correction layer found that some path segments have long become system bottlenecks due to high traffic risks and unbalanced resource allocation, so the global optimization strategy was re-planned, the priority and resource allocation of each path were adjusted, and finally the overall optimal state was achieved.

[0088] Based on global optimization data and agent scheduling data, logistics tasks are executed and the status of logistics tasks is monitored in real time, and feedback data is obtained to optimize the logistics scheduling plan.

[0089] Preferably, when the execution status of the logistics task is monitored in real time through the feedback and monitoring mechanism, the content of the real-time monitoring includes the transportation time of the vehicle, the loading status of the goods, the real-time congestion of the road and the location information of the vehicle. The real-time monitoring data is obtained through vehicle-mounted sensors, GPS positioning system and road monitoring equipment, and transmitted to the central dispatching system for analysis in real time.

[0090] In the present invention, in order to ensure the smooth execution and efficient management of logistics tasks, the execution status of logistics tasks is monitored in real time through feedback and monitoring mechanisms. The core content of real-time monitoring includes the transportation time of vehicles, the loading status of goods, the real-time congestion of roads and the location information of vehicles. These monitoring contents are indispensable basic data in logistics scheduling. Through these data, the central scheduling system can understand and grasp the operating status of vehicles, road conditions and the status of goods in real time, so as to make flexible and accurate scheduling decisions in the dynamically changing logistics network.

[0091] In the present invention, the collection and transmission of real-time monitoring data mainly rely on information collection tools such as vehicle-mounted sensors, GPS positioning systems and road monitoring equipment. First, the vehicle-mounted sensors are responsible for monitoring the transportation time of the vehicle and the loading status of the goods. The transportation time data is obtained by the vehicle-mounted sensors through the driving recorder and dynamically calculated in combination with the vehicle start time and arrival time; the loading status of the goods is monitored in real time by pressure sensors, temperature and humidity sensors and other equipment installed in the vehicle cargo hold. These sensors can accurately measure the weight changes and environmental conditions of the goods, and upload the relevant data in real time.

[0092] Secondly, the GPS positioning system can continuously track the location information of the vehicle. Through satellite positioning, the system can obtain the specific location, driving speed, driving direction and other data of the vehicle in real time. These location data are combined with road information, so that the central dispatching system can accurately grasp the real-time location of each logistics vehicle and the current traffic flow distribution. For the real-time congestion of the road, the solution uses road monitoring equipment to monitor. Road monitoring equipment mainly includes road cameras, traffic flow sensors, etc. These devices can obtain the traffic conditions of the road in real time, such as the current traffic density, vehicle speed, distribution of congested areas and other information.

[0093] These collected real-time monitoring data are transmitted to the central dispatching system via a wireless network. The central dispatching system centrally processes and analyzes these data, and integrates various types of information to evaluate the operating status of the vehicle, the traffic conditions of the road, and the safety of the goods. During the data analysis process, the system will compare the real-time data with the preset path decision data to identify potential deviations and problems. For example, if the transportation time of a truck exceeds expectations, the system will determine the cause of the delay based on the location information and road congestion, and make dynamic adjustments based on the specific situation. In addition, if the loading status of a vehicle is detected to be abnormal (such as overloading or cargo displacement), the system will send an alarm signal in time, and re-plan the route or arrange for the nearest logistics outlets to adjust the cargo according to the specific situation.

[0094] Through the above-mentioned real-time monitoring and feedback mechanism, the present invention can fully grasp the vehicle operation status and road traffic conditions in the logistics network, and improve the response capability to emergencies. Real-time vehicle location information and road congestion data enable the central dispatching system to dynamically optimize route selection, avoid vehicles getting stuck in congested areas, and thus improve overall transportation efficiency. Real-time monitoring of cargo loading status ensures the safety of logistics transportation and avoids damage to cargo due to abnormal cargo location or changes in the external environment. In addition, the centralized processing and intelligent analysis capabilities of the central dispatching system enable the system to make flexible dispatches in the face of the variability of the logistics network, reducing delivery delays and waste of resources.

[0095] Preferably, the path simulation model and global optimization parameters are periodically improved based on the cumulative analysis of historical monitoring data, and the cumulative analysis optimizes the path selection strategy and resource scheduling parameters by identifying high-congestion areas, fluctuation characteristics of resource utilization and changing trends in vehicle driving efficiency in various time periods.

[0096] In the complex scenario of logistics scheduling, factors such as traffic conditions, resource utilization and driving efficiency are often affected by the external environment, seasonal fluctuations and emergencies. Therefore, single real-time monitoring and short-term optimization may not be sufficient to cope with long-term trends and global optimization needs. In order to make up for this deficiency, the present invention periodically improves the path simulation model and global optimization parameters through cumulative analysis of historical monitoring data.

[0097] The cumulative analysis of historical monitoring data mainly includes three core contents: identification of high-incidence areas of congestion, analysis of fluctuation characteristics of resource utilization, and analysis of changing trends in vehicle driving efficiency. First, by analyzing the traffic flow and road conditions in each time period in the historical monitoring data, the system can identify high-incidence areas where traffic congestion often occurs in the road network. High-incidence areas of congestion are usually caused by reasons such as limited carrying capacity of certain specific road sections, unreasonable traffic signal control, or a surge in vehicle traffic during peak hours. After identifying these areas, the system can feed their information back to the path simulation model to adjust the weight setting and priority sorting of the relevant path segments, so as to more accurately predict future traffic conditions.

[0098] The analysis of resource utilization fluctuation characteristics mainly focuses on the use of key resources such as vehicles, storage and personnel in the logistics network. Through the cumulative analysis of historical data, the system can identify significant fluctuation characteristics of resource utilization in certain time periods, such as during promotional activities, the storage demand of certain logistics outlets surges or the frequency of vehicle dispatch increases significantly. By incorporating these fluctuation characteristics into the adjustment of global optimization parameters, the system can predict and respond to possible resource shortages or overconsumption in the future in advance, thereby optimizing resource scheduling strategies.

[0099] The trend analysis of vehicle driving efficiency is designed to evaluate the driving performance of vehicles under different road and traffic conditions over a long period of time. By analyzing data such as vehicle speed, average travel time, and waiting time, the system can determine changes in the efficiency of different roads during a specific period of time. For example, some roads may have a long-term decline in efficiency during a specific period of time, which may be related to road maintenance, traffic signal adjustments, or changes in infrastructure. After identifying these trends, the system can adjust the path selection strategy accordingly and prioritize high-efficiency paths in the scheduling plan.

[0100] Through cumulative analysis based on historical monitoring data, the present invention can perform more accurate and comprehensive global optimization of the logistics network. The path simulation model and global optimization parameters are periodically improved so that the solution can not only respond to short-term changes in real time, but also maintain good adaptability in long-term trend changes. This improvement not only improves the accuracy of path selection, but also improves the stability of resource scheduling, thereby significantly reducing congestion and resource waste in the logistics network and ensuring the efficient execution of logistics tasks.

[0101] like Figure 5 As shown, a system for implementing the logistics scheduling method, the system comprising: The data acquisition module is used to collect traffic flow data, meteorological data, order data and logistics network status data, and pre-process and standardize the collected data, and generate dynamic logistics status data through the data fusion model; in actual implementation, the data acquisition module is mainly composed of a variety of sensors and data acquisition equipment, which is used to collect traffic flow data, meteorological data, order data and logistics network status data in real time. The collection of traffic flow data usually relies on road cameras, traffic flow sensors and signal light detectors, and meteorological data is obtained through weather station sensors, such as temperature, humidity and wind speed. Order data and network status data are extracted in real time through the company's database and logistics management system. After the data collection is completed, the edge computing device or data processing server in the module will pre-process and standardize the collected data to eliminate noise and redundant information, thereby generating standardized data suitable for further analysis and calculation.

[0102] The simulation node module is used to perform fine-grained modeling of path segments using distributed simulation nodes based on the dynamic logistics state data, simulate the state of the path segments through the cellular automaton model, and adaptively adjust the simulation results using the reinforcement learning algorithm to generate optimized path decision data; the hardware form of the simulation node module usually includes multiple distributed servers or high-performance computing clusters. These simulation nodes perform fine-grained modeling of path segments based on dynamic logistics state data through a distributed architecture. The fine-grained modeling and simulation of the path adopts a cellular automaton model, which represents the state of each segment by dividing each path segment into several minimum cellular units. The simulation node also has parallel computing capabilities, and can efficiently run the reinforcement learning algorithm to adaptively adjust the simulation results and generate optimized path decision data.

[0103] The multi-agent scheduling module is used to build a multi-agent system based on the path decision data, optimize the resource scheduling of each logistics network through global state modeling and multi-layer feedback interaction mechanism, and generate agent scheduling data; the hardware form of the multi-agent scheduling module mainly includes a group of distributed intelligent nodes or scheduling servers with computing and communication capabilities. In practical applications, these intelligent nodes are responsible for building a multi-agent system based on path decision data and performing global state modeling. In order to achieve efficient resource scheduling, the intelligent node optimizes the resource allocation of each logistics network through global state modeling and multi-layer feedback interaction mechanism. The scheduling algorithm running inside the intelligent node can quickly adjust the scheduling plan according to the state changes of different networks and generate agent scheduling data.

[0104] The global predictive control module is used to dynamically predict the global logistics state based on the intelligent agent scheduling data using model predictive control, and automatically adjust the system parameters through the multi-layer self-correction mechanism of the real-time correction layer, the trend correction layer and the global correction layer to generate global optimization data; the hardware form of the global predictive control module is usually a central control server or cloud computing platform with high-performance computing capabilities. The model predictive control algorithm running in the module can dynamically predict the global logistics state based on the intelligent agent scheduling data. In practical applications, the module also includes a multi-layer self-correction mechanism, which is a real-time correction layer, a trend correction layer and a global correction layer. Each layer automatically adjusts the control parameters through automatic differentiation and parameter optimization technology. Efficient computing hardware ensures the dynamic prediction and real-time adjustment capabilities of the module, thereby generating global optimization data.

[0105] The task execution and monitoring module is used to execute logistics tasks and monitor the status of logistics tasks in real time based on the global optimization data and the intelligent agent scheduling data, and obtain feedback data to optimize the logistics scheduling plan. The hardware form of the task execution and monitoring module includes the on-board terminal of each logistics vehicle, the GPS positioning device and the central monitoring server. The on-board terminal of the logistics vehicle can receive and execute logistics tasks in real time, and collect the vehicle's operating status, transportation time, loading status and location information. The monitoring data is transmitted to the central monitoring server through the communication module of the on-board terminal, and the server optimizes the logistics scheduling plan through real-time analysis and feedback mechanism. The hardware configuration of this module ensures the comprehensive execution and accurate monitoring of logistics tasks.

[0106] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0107] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A logistics scheduling method, characterized in that: The following steps are involved: Collect traffic flow data, meteorological data, order data and logistics network status data, pre-process and standardize the collected data, and generate dynamic logistics status data through data fusion model; Based on dynamic logistics status data, distributed simulation nodes are used to perform fine-grained modeling of path segments, the status of path segments is simulated through a cellular automaton model, and a reinforcement learning algorithm is used to adaptively adjust the simulation results to generate optimized path decision data; Based on the route decision data, a multi-agent system is constructed to optimize the resource scheduling of each logistics network through global state modeling and multi-layer feedback interaction mechanism to generate agent scheduling data; Based on the agent scheduling data, the global logistics status is dynamically predicted using model predictive control. The system parameters are automatically adjusted through the multi-layer self-correction mechanism of the real-time correction layer, trend correction layer and global correction layer to generate global optimization data. Based on global optimization data and agent scheduling data, logistics tasks are executed and the status of logistics tasks is monitored in real time, and feedback data is obtained to optimize the logistics scheduling plan.

2. The logistics scheduling method according to claim 1, characterized in that: When simulating the state of a path segment through a cellular automaton model, the cellular automaton model is composed of multiple cellular units, each of which represents a minimum unit length in the path segment. The state of the cellular unit includes traffic flow, road occupancy rate and average vehicle speed. The state of each cellular unit is updated according to a preset state transition rule. The state transition rule is set by traffic flow prediction based on historical data and congestion propagation characteristics, and is used to simulate changes in roads under different traffic conditions.

3. The logistics scheduling method according to claim 2, characterized in that: The reward function of the reinforcement learning algorithm is defined as follows: in, Indicates the reward value; Represents the shortest travel time of a route segment; represents the traffic density of the path segment; Indicates the congestion level of the path segment; , and is an adjustable weight coefficient used to adjust the influence of each factor in the reward function.

4. The logistics scheduling method according to claim 1, characterized in that: When constructing a multi-agent system, the resource status, path information, task load and interaction relationship of each logistics point are taken as input, and a global state graph is constructed through a graph convolutional network. The nodes of the graph convolutional network represent each logistics point, and the edges of the graph convolutional network represent the path information between adjacent logistics points. The parameters of the graph convolutional network are iteratively optimized through the status data of each logistics point to establish a global dependency relationship between logistics points.

5. The logistics scheduling method according to claim 1, characterized in that: The multi-layer feedback interaction mechanism consists of a global optimization layer and a local feedback layer. The global optimization layer is responsible for the allocation of global resources. The local feedback layer monitors the task execution status of each logistics network in real time and regularly sends the monitoring results as update signals to the global optimization layer. The monitoring indicators of the local feedback layer include resource utilization, task completion rate and task execution time.

6. The logistics scheduling method according to claim 1, characterized in that: The objective function of the model predictive control is defined as follows: in, represents the target value of global optimization; Indicates Estimated travel time for each route; Indicates Resource utilization of each path; Indicates The travel risk of the route; , and They are weighted coefficients of the path segments, respectively, and are used to adjust the influence of each factor in the objective function.

7. The logistics scheduling method according to claim 1, characterized in that: The real-time correction layer, the trend correction layer and the global correction layer dynamically update their respective control parameters through automatic differentiation technology. The real-time correction layer calculates the difference between the real-time monitoring data and the global prediction data through automatic differentiation, and adjusts the control parameters based on the difference; the trend correction layer updates the trend adjustment parameters according to the medium-term trend deviation calculated by automatic differentiation, and the global correction layer performs global optimization adjustment according to the long-term deviation of the system.

8. The logistics scheduling method according to claim 1, characterized in that: When the execution status of logistics tasks is monitored in real time through feedback and monitoring mechanisms, the content of the real-time monitoring includes the transportation time of the vehicle, the loading status of the goods, the real-time congestion of the road and the location information of the vehicle. The real-time monitoring data is obtained through on-board sensors, GPS positioning systems and road monitoring equipment, and transmitted to the central dispatching system for analysis in real time.

9. The logistics scheduling method according to claim 1, characterized in that: Based on the cumulative analysis of historical monitoring data, the path simulation model and global optimization parameters are periodically improved. The cumulative analysis optimizes the path selection strategy and resource scheduling parameters by identifying the high-congestion areas, the fluctuation characteristics of resource utilization and the changing trends of vehicle driving efficiency in various time periods.

10. A system for implementing the logistics scheduling method according to any one of claims 1 to 9, characterized in that: The system comprises: The data collection module is used to collect traffic flow data, meteorological data, order data and logistics network status data, and pre-process and standardize the collected data to generate dynamic logistics status data through the data fusion model; A simulation node module is used to perform fine-grained modeling of path segments using distributed simulation nodes based on the dynamic logistics state data, simulate the path segment states through a cellular automaton model, and adaptively adjust the simulation results using a reinforcement learning algorithm to generate optimized path decision data; A multi-agent scheduling module is used to build a multi-agent system based on the path decision data, optimize the resource scheduling of each logistics network point through global state modeling and multi-layer feedback interaction mechanism, and generate agent scheduling data; A global prediction control module is used to dynamically predict the global logistics status based on the intelligent agent scheduling data using model prediction control, automatically adjust system parameters through a multi-layer self-correction mechanism of a real-time correction layer, a trend correction layer, and a global correction layer, and generate global optimization data; The task execution and monitoring module is used to execute logistics tasks and monitor the status of logistics tasks in real time based on the global optimization data and the intelligent agent scheduling data, and obtain feedback data to optimize the logistics scheduling plan.

Images