Logistics rapid distribution analysis system based on artificial intelligence

Abstract

The invention discloses a logistics rapid distribution analysis system based on artificial intelligence, and relates to the technical field of logistics distribution analysis, in which orders, vehicle positions and traffic flow data are collected in real time, a distribution environment state is constructed, an intelligent agent is used as an independent decision unit deployed on a distribution vehicle, and the intelligent agent is used as an independent decision unit deployed on the distribution vehicle. The local planning module is used for generating a path and a resource allocation scheme, and a local scheme is shared through the communication module, so that collaborative optimization is realized, and the disadvantage that traditional static planning is difficult to deal with order changes and emergencies is solved; in the delivery process, the intelligent agent dynamically adjusts the path and task priority according to traffic and order data updated in real time, so that the loss caused by traffic congestion or order delay is remarkably reduced, and meanwhile, continuous improvement of path planning is realized through feedback adjustment; and the analysis center combines an NSGA-II algorithm to bring the distribution time, cost and task balance degree into a target function, optimizes an MADDPG strategy based on actual execution data, and improves global benefits.

Description

Technical Field

[0001] The present invention relates to the technical field of logistics distribution analysis, and in particular to a logistics rapid distribution analysis system based on artificial intelligence. Background Art

[0002] With the rapid development of e-commerce and instant logistics, logistics and delivery tasks are becoming increasingly complex. Current logistics rapid delivery methods mainly rely on path planning algorithms and resource scheduling models, which perform well in single static tasks, but have poor results when faced with real-time, changeable and complex delivery scenarios.

[0003] The orders in actual delivery tasks are highly random, such as new orders, canceled orders or priority adjustments. The existing path planning algorithms lack dynamic adjustment capabilities. At the same time, most optimization strategies only optimize the efficiency of a single delivery vehicle or a single delivery node, ignoring the importance of multi-node collaboration in the distribution network, resulting in the overall delivery efficiency failing to meet expectations. In addition, in delivery tasks, multiple objectives such as delivery time, operating costs and energy consumption need to be optimized at the same time. Current distribution analysis mostly uses fixed weights for single optimization, and is unable to weigh the priorities of different objectives in real time, resulting in the optimization results being difficult to meet dynamic actual needs. Therefore, there is an urgent need for an artificial intelligence-based logistics rapid delivery analysis system to solve such problems. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] The present invention provides an artificial intelligence-based logistics rapid delivery analysis system to solve the problem that existing path planning algorithms lack dynamic adjustment capabilities. At the same time, most optimization strategies only optimize the efficiency of a single delivery vehicle or a single delivery node, ignoring the importance of multi-node collaboration in the distribution network; and are unable to balance the priorities between different goals in real time.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The embodiment of the present invention provides a logistics rapid delivery analysis system based on artificial intelligence, which includes:

[0008] The distribution network data collection unit collects distribution data in real time through IoT devices, including order type, order address, vehicle location, and traffic flow, and extracts features to construct the distribution environment status as input for the intelligent agent;

[0009] The multi-agent unit is responsible for receiving real-time environmental status data during the delivery process, dynamically updating the delivery route and priority, and feeding back the adjusted results to the analysis center; the multi-agent unit contains multiple agents;

[0010] Intelligent agents are independent decision-making units in the distribution network, deployed on distribution vehicles. They are responsible for autonomously planning routes and allocating resources based on environmental conditions, and interacting and co-optimizing among agents. Intelligent agents refer to devices with autonomous decision-making capabilities, deployed on distribution vehicles, and capable of completing distribution tasks through perception, calculation, and execution in the distribution environment.

[0011] The agent includes a local planning module and a communication module.

[0012] In the local planning module, each agent generates path planning and vehicle allocation solutions based on the received delivery environment status using the multi-agent deep deterministic policy gradient MADDPG algorithm.

[0013] The communication module, through which agents share local solutions and dynamically adjust paths and task allocations based on a collaborative interaction mechanism;

[0014] The analysis center collects feedback data from multi-agent units and uses the NSGA-II algorithm to iteratively adjust the MADDPG strategy.

[0015] As a preferred solution of the artificial intelligence-based logistics rapid delivery analysis system of the present invention, the analysis method of the logistics rapid delivery analysis system includes:

[0016] Step S1, collecting delivery data in the delivery network through IoT devices, including order type, order address, vehicle location and traffic flow data; extracting features from the delivery data, constructing a representation of the delivery environment state based on the extracted features, and distributing it to each agent as input;

[0017] Step S2, each delivery agent generates a local plan based on the received delivery environment status using the MADDPG algorithm strategy, including the path planning and vehicle allocation plan of each agent;

[0018] Step S3: Agents share local solutions through communication modules and perform collaborative optimization based on the environment status.

[0019] Step S4: During the delivery process, each agent obtains the traffic flow data and order urgency collected by the delivery network data collection unit in real time, dynamically updates the delivery route and delivery priority, and feeds back the adjusted execution results to the analysis center.

[0020] In step S5, the analysis center collects the execution result data of each intelligent agent, and based on the execution results of the overall delivery, uses the multi-objective optimization NSGA-II algorithm to re-evaluate the MADDPG algorithm strategy in step S2, incorporating delivery time, cost, and task balance into the objective function and performing iterative adjustments.

[0021] As a preferred solution of the artificial intelligence-based logistics rapid delivery analysis system described in the present invention, the order type is used to mark the urgency of the order.

[0022] As a preferred solution of the artificial intelligence-based logistics rapid delivery analysis system of the present invention, the step of extracting features from the delivery data and constructing a representation of the delivery environment state based on the extracted features is as follows:

[0023] The distribution data is converted into a feature vector, and the feature vector formula is defined as:

[0024] F=[f1,f2,f3,…,f n ],

[0025] Among them, F represents the characteristic vector of the distribution environment, f1 represents the order distribution density, and the calculation formula is: f1=N o / A r , where N o is the total number of orders in the region, A r is the area of ​​the region, f2 represents the traffic flow characteristics, defined as the flow density of a certain road section, and the calculation formula is: f2 = N v / L r , where N v is the number of vehicles passing through the road section per unit time, L r is the length of the road section, f3 represents the vehicle status characteristic, which is defined as the current load rate, and the calculation formula is: f3=W c / W m , where W c is the current load, W m is the maximum load of the vehicle, f4,f5,…,f n Indicates other extended features, which are defined according to actual delivery requirements;

[0026] The distribution environment state is constructed based on the extracted feature vectors. The expression formula of the environment state is:

[0027] S e =g(f1,f2,…,f n ),

[0028] Among them, S e is the global state of the distribution environment, g is the feature mapping function, which is normalized and defined as:

[0029]

[0030] Among them, min(f i ) and max(f i ) are the features f i The minimum and maximum values ​​of

[0031] The method of marking the urgency of an order is:

[0032] For each order, the urgency weight is calculated based on the order type. The urgency weight formula is:

[0033] u j =h(d j ,t j ),

[0034] Among them, u j represents the urgency weight of order j, d j represents the delivery distance of order j, t j represents the deadline of order j, h(d j ,t j ) is the urgency calculation function, defined as:

[0035]

[0036] Among them, τ represents the current time, and k represents the weight coefficient of the delivery distance;

[0037] The feature vector and the urgency mark are combined to form a delivery status vector. The combination formula is:

[0038] S c =[S e ,u1,u2,…,u m ],

[0039] Among them, S c is the comprehensive state vector, u1,u2,…,u m Represents the set of urgency weights for all orders.

[0040] As a preferred solution of the artificial intelligence-based logistics rapid delivery analysis system described in the present invention, wherein: based on the received delivery environment state, the MADDPG algorithm strategy is used to generate a local solution, including the path planning and vehicle allocation solution of each agent, as follows:

[0041] The global state S of the distribution environment e Divided into multiple local states, the local state formula is:

[0042] o i,t =f i (S e ),

[0043] Among them, i,t is the local observation of agent i at time t, f i Extract the local state function, which is segmented according to the agent's geographical location or task assignment.

[0044] Each agent generates actions based on local state input through the policy network. The policy function is defined as:

[0045] a i,t =π i (o i,t θ i ),

[0046] Among them, a i,t is the action performed by agent i at time t, including path planning and vehicle allocation scheme, π i is the strategy function of agent i, with parameter θ i ,θ i is the trainable parameter of the policy network;

[0047] The local delivery plan is formed by the action set of all agents, which is expressed as:

[0048] A t =[a 1,t ,a 2,t ,…,a m,t ],

[0049] Among them, A t is the set of actions of all agents at time t, m is the total number of agents,

[0050] Define the reward function r i,t , the formula is:

[0051]

[0052] Among them, T i is the time for agent i to complete the current task, u i is the urgency weight of the order currently processed by the agent, α, β, γ are the weight coefficients of the reward function, R g is the global efficiency reward, defined as:

[0053]

[0054] Using the centralized training and distributed execution mechanism of the MADDPG algorithm, the strategy parameters are updated by gradient descent. The update formula is:

[0055]

[0056] in,

[0057] J(θ i ) is the objective function of agent i, defined as:

[0058]

[0059] Among them, η is the learning rate; through multiple rounds of iterative optimization, each agent gradually adjusts the path planning and vehicle allocation plan to improve the overall efficiency.

[0060] As a preferred solution of the artificial intelligence-based logistics rapid delivery analysis system described in the present invention, the collaborative optimization method is:

[0061] Utilize the interaction mechanism in multi-agent reinforcement learning to adjust their respective delivery paths.

[0062] As a preferred solution of the artificial intelligence-based logistics rapid delivery analysis system described in the present invention, the steps of sharing local solutions between the intelligent agents through the communication module and performing collaborative optimization in combination with the environmental state are as follows:

[0063] After the agent's action is executed, the global state of the delivery environment is updated, and the update formula is:

[0064] S e,t+1 =f(S e,t ,A t ),

[0065] Among them, S e,t+1 is the global state at time t+1, S e,t is the global state at time t, A t is the set of actions performed by all agents at time t, f is the state transition function, and the state is dynamically updated based on traffic flow and order completion.

[0066] Each agent shares part of the global reward and exchanges local states through a communication mechanism. The collaboration formula is:

[0067]

[0068] in, is the collaboration reward of agent i, r j,t is the individual reward of agent j, γ is the collaboration coefficient, which adjusts the degree of reward sharing;

[0069] The objective function is designed by integrating individual and collaborative rewards. The objective function formula is:

[0070]

[0071] By optimizing the policy parameters through gradient descent, the update formula is:

[0072]

[0073] in,

[0074] η is the learning rate, is the gradient of the objective function,

[0075] Combining local states and collaborative rewards, the agent generates optimized path planning and vehicle allocation solutions. The path planning formula is:

[0076]

[0077] Among them, P i 新 is the new path of agent i, C(P i ,S e ) is the path P i Cost function, combining time and urgency;

[0078] The vehicle allocation formula is:

[0079]

[0080] Among them, V i 新 is the vehicle allocation plan of agent i, U(V i ,S e ) is the benefit function of vehicle allocation, combining task balance and transportation capacity,

[0081] Define the overall benefit function to comprehensively evaluate the effect of agent collaborative optimization. The overall benefit function is:

[0082]

[0083] Among them, E g is the global benefit, α, β are weight coefficients.

[0084] As a preferred solution of the artificial intelligence-based logistics rapid delivery analysis system described in the present invention, the adjusted execution results include:

[0085] The agent’s adjusted path nodes and estimated arrival time each time,

[0086] Priority changes based on order urgency,

[0087] As well as the number of delivery tasks and mileage delivered for each vehicle.

[0088] As a preferred solution of the artificial intelligence-based logistics rapid delivery analysis system described in the present invention, the step of dynamically updating the delivery path and delivery priority is:

[0089] The agent calculates the priority of each order based on the remaining delivery time of the order and the urgency of the current task.

[0090] Each agent replans its path based on the current traffic conditions and order priorities, using dynamic programming and heuristic algorithms for planning.

[0091] After completing the route adjustment, each agent generates an updated delivery action based on the new priority information and route plan, and sends it to the delivery vehicle in real time.

[0092] As a preferred solution of the artificial intelligence-based logistics rapid delivery analysis system described in the present invention, the step of re-evaluating the MADDPG algorithm strategy in step S2 using the multi-objective optimization NSGA-II algorithm based on the execution result of the overall delivery is:

[0093] Collect the execution data of all agents after completing the delivery task, including actual delivery time, path cost and task allocation.

[0094] Based on the execution data, define the objective function, the objectives include:

[0095] Minimize the total delivery time,

[0096] Minimize delivery costs,

[0097] and maximize the balance of task allocation,

[0098] The NSGA-II algorithm is used to generate the initial population based on the objective function, and the Pareto frontier solution set is selected through crossover, mutation and non-dominated sorting.

[0099] During the optimization process, each solution balances multiple objectives, and a solution is selected from the frontier solution set to readjust the MADDPG strategy.

[0100] The beneficial effects of the present invention are as follows: the present invention collects order, vehicle location and traffic flow data in real time, constructs the distribution environment state, and the intelligent agent, as an independent decision-making unit deployed on the distribution vehicle, generates a path and resource allocation plan using a local planning module, and shares the local plan through a communication module to achieve collaborative optimization, thereby solving the disadvantage that traditional static planning is difficult to cope with order changes and emergencies; during the distribution process, the intelligent agent dynamically adjusts the path and task priority according to the real-time updated traffic and order data, significantly reduces the losses caused by traffic congestion or order delays, and realizes continuous improvement of path planning through feedback adjustment; the analysis center incorporates the delivery time, cost and task balance into the objective function in combination with the NSGA-II algorithm, optimizes the MADDPG strategy based on actual execution data, and improves the overall benefit.

[0101] The present invention introduces a collaborative reward mechanism and a global benefit function among multiple agents to avoid waste of resources. At the same time, it ensures dynamics and real-time response capabilities by combining comprehensive optimization with time and task priority. While improving the overall delivery efficiency, it reduces operating costs and effectively solves the problem of imbalance in multi-objective optimization.

[0102] In summary, the present invention effectively improves the distribution efficiency and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0104] Figure 1 It is a schematic diagram of the framework of the artificial intelligence-based rapid logistics distribution analysis system of the present invention. DETAILED DESCRIPTION

[0105] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0106] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0107] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0108] Example 1, reference Figure 1 , this embodiment provides a logistics rapid delivery analysis system based on artificial intelligence, including:

[0109] The distribution network data collection unit collects distribution data in real time through IoT devices, including order type, order address, vehicle location, and traffic flow, and extracts features to construct the distribution environment status as input for the intelligent agent;

[0110] The multi-agent unit is responsible for receiving real-time environmental status data during the delivery process, dynamically updating the delivery route and priority, and feeding back the adjusted results to the analysis center; the multi-agent unit contains multiple agents;

[0111] Intelligent agents are independent decision-making units in the distribution network, deployed on distribution vehicles. They are responsible for autonomously planning routes and allocating resources based on environmental conditions, and interacting and co-optimizing among agents. Intelligent agents refer to devices with autonomous decision-making capabilities, deployed on distribution vehicles, and capable of completing distribution tasks through perception, calculation, and execution in the distribution environment.

[0112] The agent includes a local planning module and a communication module.

[0113] In the local planning module, each agent generates path planning and vehicle allocation solutions based on the received delivery environment status using the multi-agent deep deterministic policy gradient MADDPG algorithm.

[0114] The communication module, through which agents share local solutions and dynamically adjust paths and task allocations based on a collaborative interaction mechanism;

[0115] The analysis center collects feedback data from multi-agent units and uses the NSGA-II algorithm to iteratively adjust the MADDPG strategy;

[0116] This embodiment also provides an analysis method for the above-mentioned artificial intelligence-based logistics rapid delivery analysis system, comprising the following steps:

[0117] Step S1, collecting delivery data in the delivery network through IoT devices, including order type, order address, vehicle location and traffic flow data; extracting features from the delivery data, constructing a representation of the delivery environment state based on the extracted features, and distributing it to each agent as input;

[0118] Order type is used to mark the urgency of the order;

[0119] The steps of extracting features from the delivery data and constructing a representation of the delivery environment state based on the extracted features are as follows:

[0120] The distribution data is converted into a feature vector, and the feature vector formula is defined as:

[0121] F=[f1,f2,f3,…,f n ],

[0122] Among them, F represents the characteristic vector of the distribution environment, f1 represents the order distribution density, and the calculation formula is: f1=N o / A r , where N o is the total number of orders in the region, Ar is the area of ​​the region, f2 represents the traffic flow characteristics, defined as the flow density of a certain road section, and the calculation formula is: f2 = N v / L r , where N v is the number of vehicles passing through the road section per unit time, L r is the length of the road section, f3 represents the vehicle status characteristic, which is defined as the current load rate, and the calculation formula is: f3=W c / W m , where W c is the current load, W m is the maximum load of the vehicle, f4,f5,…,f n Indicates other extended features, which are defined according to actual delivery requirements;

[0123] The distribution environment state is constructed based on the extracted feature vectors. The expression formula of the environment state is:

[0124] S e =g(f1,f2,…,f n ),

[0125] Among them, S e is the global state of the distribution environment, g is the feature mapping function, which is normalized and defined as:

[0126]

[0127] Among them, min(f i ) and max(f i ) are the features f i The minimum and maximum values ​​of

[0128] The way to mark the urgency of an order is:

[0129] For each order, the urgency weight is calculated based on the order type. The urgency weight formula is:

[0130] u j =h(d j ,t j ),

[0131] Among them, u j represents the urgency weight of order j, d j represents the delivery distance of order j, t j represents the deadline of order j, h(d j ,t j ) is the urgency calculation function, defined as:

[0132]

[0133] Among them, τ represents the current time, and k represents the weight coefficient of the delivery distance;

[0134] The feature vector and the urgency mark are combined to form a delivery status vector. The combination formula is:

[0135] S c =[S e ,u1,u2,…,u m ],

[0136] Among them, S c is the comprehensive state vector, u1,u2,…,u m represents the set of urgency weights of all orders;

[0137] Specifically, by i ) is normalized to map the features to a uniform scale so that the state vector S e More suitable for agent policy network input; in addition, the urgency u j Considering time and distance factors, the orders are given reasonable priority, taking into account both expression ability and computing efficiency. In actual deployment, the weight k or the number of features n can be adjusted according to the data characteristics.

[0138] Step S2, each delivery agent generates a local plan based on the received delivery environment status using the MADDPG algorithm strategy, including the path planning and vehicle allocation plan of each agent;

[0139] Based on the received distribution environment status, the MADDPG algorithm strategy is used to generate a local solution, including the path planning and vehicle allocation steps of each agent.

[0140] The global state S of the distribution environment e Divided into multiple local states, the local state formula is:

[0141] o i,t =f i (S e ),

[0142] Among them, i,t is the local observation of agent i at time t, f i Extract the local state function, which is segmented according to the agent's geographical location or task assignment.

[0143] Each agent generates actions based on local state input through the policy network. The policy function is defined as:

[0144] a i,t =π i (o i,t θ i ),

[0145] Among them, a i,t is the action performed by agent i at time t, including path planning and vehicle allocation scheme, π i is the strategy function of agent i, with parameter θ i ,θ i is the trainable parameter of the policy network;

[0146] The local delivery plan is formed by the action set of all agents, which is expressed as:

[0147] A t =[a 1,t ,a 2,t ,…,a m,t ],

[0148] Among them, A t is the set of actions of all agents at time t, m is the total number of agents,

[0149] Define the reward function r i,t , the formula is:

[0150]

[0151] Among them, T i is the time for agent i to complete the current task, u i is the urgency weight of the order currently processed by the agent, α, β, γ are the weight coefficients of the reward function, R g is the global efficiency reward, defined as:

[0152]

[0153] Using the centralized training and distributed execution mechanism of the MADDPG algorithm, the strategy parameters are updated by gradient descent. The update formula is:

[0154]

[0155] in,

[0156] J(θ i ) is the objective function of agent i, defined as:

[0157]

[0158] Among them, η is the learning rate; through multiple rounds of iterative optimization, each agent gradually adjusts the path planning and vehicle allocation plan to improve the overall efficiency;

[0159] Specifically, step S2 generates a local delivery plan based on the MADDPG algorithm. In this process, each agent generates actions through the policy network according to the local state, and combines the multi-dimensional reward function to guide the agent to maintain a balance between local and global goals. Through centralized training and distributed execution, MADDPG effectively solves the collaboration problem among multiple agents, making path planning and vehicle allocation more dynamic and efficient. In practice, the weights α, β, and γ of the reward function are adjusted as needed.

[0160] Step S3: Agents share local solutions through communication modules and perform collaborative optimization based on the environment status.

[0161] The collaborative optimization method is:

[0162] Use the interaction mechanism in multi-agent reinforcement learning to adjust their respective delivery paths;

[0163] The agents share local solutions through the communication module and perform collaborative optimization based on the environment state.

[0164] After the agent's action is executed, the global state of the delivery environment is updated, and the update formula is:

[0165] S e,t+1 =f(S e,t ,A t ),

[0166] Among them, S e,t+1 is the global state at time t+1, S e,t is the global state at time t, A t is the set of actions performed by all agents at time t, f is the state transition function, and the state is dynamically updated based on traffic flow and order completion.

[0167] Each agent shares part of the global reward and exchanges local states through a communication mechanism. The collaboration formula is:

[0168]

[0169] in, is the collaboration reward of agent i, r j,t is the individual reward of agent j, γ is the collaboration coefficient, which adjusts the degree of reward sharing;

[0170] The objective function is designed by integrating individual and collaborative rewards. The objective function formula is:

[0171]

[0172] By optimizing the policy parameters through gradient descent, the update formula is:

[0173]

[0174] in,

[0175] η is the learning rate, is the gradient of the objective function,

[0176] Combining local states and collaborative rewards, the agent generates optimized path planning and vehicle allocation solutions. The path planning formula is:

[0177]

[0178] Among them, P i 新 is the new path of agent i, C(P i ,S e ) is the path P i Cost function, combining time and urgency;

[0179] The vehicle allocation formula is:

[0180]

[0181] Among them, V i 新 is the vehicle allocation plan of agent i, U(V i ,S e ) is the benefit function of vehicle allocation, combining task balance and transportation capacity,

[0182] Define the overall benefit function to comprehensively evaluate the effect of agent collaborative optimization. The overall benefit function is:

[0183]

[0184] Among them, E g is the global benefit, α, β are weight coefficients,

[0185] Specifically, step S3 sets the collaboration reward and performs global status update to improve the overall optimization capability of the multi-agent system. While maintaining local benefits, each agent improves the global benefits through the collaboration mechanism, avoiding resource waste or path conflicts caused by independent optimization. In addition, path planning and vehicle allocation are used to ensure the system's responsiveness in a dynamic environment.

[0186] Step S4: During the delivery process, each agent obtains the traffic flow data and order urgency collected by the delivery network data collection unit in real time, dynamically updates the delivery route and delivery priority, and feeds back the adjusted execution results to the analysis center.

[0187] The adjusted execution results include:

[0188] The agent’s adjusted path nodes and estimated arrival time each time,

[0189] Priority changes based on order urgency,

[0190] As well as the number of delivery tasks and mileage delivered by each vehicle;

[0191] The steps to dynamically update the delivery route and delivery priority are:

[0192] The agent calculates the priority of each order based on the remaining delivery time of the order and the urgency of the current task.

[0193] Each agent replans its path based on the current traffic conditions and order priorities, using dynamic programming and heuristic algorithms for planning.

[0194] After completing the route adjustment, each agent generates updated delivery actions based on the new priority information and route plan, and sends them to the delivery vehicles in real time.

[0195] Specifically, the dynamic adjustment mechanism can significantly improve delivery efficiency and reduce losses caused by traffic jams or order delays.

[0196] Step S5, the analysis center collects the execution result data of each agent, and based on the execution result of the overall delivery, uses the multi-objective optimization NSGA-II algorithm to re-evaluate the MADDPG algorithm strategy in step S2, incorporates the delivery time, cost, and task balance into the objective function, and performs iterative adjustments;

[0197] Based on the execution results of the overall distribution, the steps of re-evaluating the MADDPG algorithm strategy in step S2 using the multi-objective optimization NSGA-II algorithm are as follows:

[0198] Collect the execution data of all agents after completing the delivery task, including actual delivery time, path cost and task allocation.

[0199] Based on the execution data, define the objective function, the objectives include:

[0200] Minimize the total delivery time,

[0201] Minimize delivery costs,

[0202] and maximize the balance of task allocation,

[0203] The NSGA-II algorithm is used to generate the initial population based on the objective function, and the Pareto frontier solution set is selected through crossover, mutation and non-dominated sorting.

[0204] During the optimization process, each solution balances multiple objectives, and a solution is selected from the frontier solution set to readjust the MADDPG strategy.

[0205] Specifically, this feedback mechanism enables continuous improvement of the strategy and achieves global optimization effects among time, cost, and balance. Combined with the optimization results of NSGA-II, the distribution system can dynamically respond to changes in demand and improve overall distribution efficiency and resource utilization.

[0206] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A logistics rapid delivery analysis system based on artificial intelligence, characterized by: include, The distribution network data collection unit collects distribution data in real time through IoT devices, including order type, order address, vehicle location, and traffic flow, and extracts features to construct the distribution environment status as input for the intelligent agent; The multi-agent unit is responsible for receiving real-time environmental status data during the delivery process, dynamically updating the delivery route and priority, and feeding back the adjusted results to the analysis center; the multi-agent unit contains multiple agents; Intelligent agents are independent decision-making units in the distribution network and are deployed on distribution vehicles. They are responsible for autonomously planning routes and allocating resources based on environmental conditions, and for interaction and collaborative optimization among agents. The agent includes a local planning module and a communication module. In the local planning module, each agent generates path planning and vehicle allocation solutions based on the received delivery environment status using the multi-agent deep deterministic policy gradient MADDPG algorithm. The communication module, through which agents share local solutions and dynamically adjust paths and task allocations based on a collaborative interaction mechanism; The analysis center collects feedback data from multi-agent units and uses the NSGA-II algorithm to iteratively adjust the MADDPG strategy.

2. The artificial intelligence-based logistics rapid delivery analysis system according to claim 1, characterized in that: The analysis method of the logistics rapid delivery analysis system includes: Step S1, collecting delivery data in the delivery network through IoT devices, including order type, order address, vehicle location and traffic flow data; extracting features from the delivery data, constructing a representation of the delivery environment state based on the extracted features, and distributing it to each agent as input; Step S2, each delivery agent generates a local plan based on the received delivery environment status using the MADDPG algorithm strategy, including the path planning and vehicle allocation plan of each agent; Step S3: Agents share local solutions through communication modules and perform collaborative optimization based on the environment status. Step S4: During the delivery process, each agent obtains the traffic flow data and order urgency collected by the delivery network data collection unit in real time, dynamically updates the delivery route and delivery priority, and feeds back the adjusted execution results to the analysis center. In step S5, the analysis center collects the execution result data of each intelligent agent, and based on the execution results of the overall delivery, uses the multi-objective optimization NSGA-II algorithm to re-evaluate the MADDPG algorithm strategy in step S2, incorporating delivery time, cost, and task balance into the objective function and performing iterative adjustments.

3. The artificial intelligence-based logistics rapid delivery analysis system according to claim 2, characterized in that: The order type is used to mark the urgency of the order.

4. The artificial intelligence-based logistics rapid delivery analysis system as claimed in claim 3, characterized in that: The step of extracting features from the delivery data and constructing a representation of the delivery environment state based on the extracted features is: The distribution data is converted into a feature vector, and the feature vector formula is defined as: <h2 style=";text-align:left;direction:ltr">F=[f1,f2,f3,…,f<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ], Among them, F represents the characteristic vector of the distribution environment, f1 represents the order distribution density, and the calculation formula is: f1=N o / A r , where N o is the total number of orders in the region, A r is the area of ​​the region, f2 represents the traffic flow characteristics, defined as the flow density of a certain road section, and the calculation formula is: f2 = N v / L r , where N v is the number of vehicles passing through the road section per unit time, L r is the length of the road section, f3 represents the vehicle status characteristic, which is defined as the current load rate, and the calculation formula is: f3=W c / W m , where W c is the current load, W m is the maximum load of the vehicle, f4,f5,…,f n Indicates other extended features, which are defined according to actual delivery requirements; The distribution environment state is constructed based on the extracted feature vectors. The expression formula of the environment state is: S e =g(f1,f2,…,f n ), Among them, S e is the global state of the distribution environment, g is the feature mapping function, which is normalized and defined as: Among them, min(f i ) and max(f i ) are the features f i The minimum and maximum values ​​of The method of marking the urgency of an order is: For each order, the urgency weight is calculated based on the order type. The urgency weight formula is: u j =h(d j ,t j ), Among them, u j represents the urgency weight of order j, d j represents the delivery distance of order j, t j represents the deadline of order j, h(d j ,t j ) is the urgency calculation function, defined as: Among them, τ represents the current time, and k represents the weight coefficient of the delivery distance; The feature vector and the urgency mark are combined to form a delivery status vector. The combination formula is: S c =[S e ,u1,u2,…,u m ], Among them, S c is the comprehensive state vector, u1,u2,…,u m Represents the set of urgency weights for all orders.

5. The artificial intelligence-based logistics rapid delivery analysis system according to claim 4, characterized in that: The steps of generating a local solution based on the received distribution environment state using the MADDPG algorithm strategy, including the path planning and vehicle allocation solution for each agent, are as follows: The global state S of the distribution environment e Divided into multiple local states, the local state formula is: o i,t =f i (S e ), Among them, i,t is the local observation of agent i at time t, f i Extract the local state function, which is segmented according to the agent's geographical location or task assignment. Each agent generates actions based on local state input through the policy network. The policy function is defined as: a i,t =π i (the i,t ;θ i ), Among them, a i,t is the action performed by agent i at time t, including path planning and vehicle allocation scheme, π i is the strategy function of agent i, with parameter θ i ,θ i is the trainable parameter of the policy network; The local delivery plan is formed by the action set of all agents, which is expressed as: A t =[a 1,t ,a 2,t ,…,a m,t ], Among them, A t is the action set of all agents at time t, m is the total number of agents, Define the reward function r i,t , the formula is: Among them, T i is the time for agent i to complete the current task, u i is the urgency weight of the order currently processed by the agent, α, β, γ are the weight coefficients of the reward function, R g is the global efficiency reward, defined as: Using the centralized training and distributed execution mechanism of the MADDPG algorithm, the strategy parameters are updated by gradient descent. The update formula is: in, J(θ i ) is the objective function of agent i, defined as: J(θ i )E[r i,t ·log(π i (you are i,t θ i ))], Among them, η is the learning rate.

6. The artificial intelligence-based logistics rapid delivery analysis system according to claim 5, characterized in that: The collaborative optimization method is: Utilize the interaction mechanism in multi-agent reinforcement learning to adjust their respective delivery paths.

7. The artificial intelligence-based logistics rapid delivery analysis system according to claim 6, characterized in that: The agents share local solutions through the communication module and perform collaborative optimization in combination with the environment state. After the agent's action is executed, the global state of the delivery environment is updated, and the update formula is: S e,t+1 =f(S e,t ,A t ), Among them, S e,t+1 is the global state at time t+1, S e,t is the global state at time t, A t is the set of actions performed by all agents at time t, f is the state transition function, and the state is dynamically updated based on traffic flow and order completion. Each agent shares part of the global reward and exchanges local states through a communication mechanism. The collaboration formula is: in, is the collaboration reward of agent i, r j,t is the individual reward of agent j, γ is the collaboration coefficient, which adjusts the degree of reward sharing; The objective function is designed by integrating individual and collaborative rewards. The objective function formula is: By optimizing the policy parameters through gradient descent, the update formula is: in, η is the learning rate, is the gradient of the objective function, Combining local states and collaborative rewards, the agent generates optimized path planning and vehicle allocation solutions. The path planning formula is: Among them, P i 新 is the new path of agent i, C(P i ,S e ) is the path P i Cost function, combining time and urgency; The vehicle allocation formula is: Among them, V i 新 is the vehicle allocation plan of agent i, U(V i ,S e ) is the benefit function of vehicle allocation, combining task balance and transportation capacity, Define the overall benefit function to comprehensively evaluate the effect of agent collaborative optimization. The overall benefit function is: Among them, E g is the global benefit, α, β are weight coefficients.

8. The artificial intelligence-based logistics rapid delivery analysis system according to claim 7, characterized in that: The adjusted execution results include: The agent’s adjusted path nodes and estimated arrival time each time, Priority changes based on order urgency, As well as the number of delivery tasks and mileage delivered for each vehicle.

9. The artificial intelligence-based logistics rapid delivery analysis system according to claim 8, characterized in that: The steps of dynamically updating the delivery path and delivery priority are: The agent calculates the priority of each order based on the remaining delivery time of the order and the urgency of the current task. Each agent replans its path based on the current traffic conditions and order priorities, using dynamic programming and heuristic algorithms for planning. After completing the route adjustment, each agent generates an updated delivery action based on the new priority information and route plan, and sends it to the delivery vehicle in real time.

10. The artificial intelligence-based logistics rapid delivery analysis system according to claim 9, characterized in that: The step of re-evaluating the MADDPG algorithm strategy in step S2 using the multi-objective optimization NSGA-II algorithm based on the execution result of the overall distribution is: Collect the execution data of all agents after completing the delivery task, including actual delivery time, path cost and task allocation. Based on the execution data, define the objective function, the objectives include: Minimize the total delivery time, Minimize delivery costs, and maximize the balance of task allocation, The NSGA-II algorithm is used to generate the initial population based on the objective function, and the Pareto frontier solution set is selected through crossover, mutation and non-dominated sorting. During the optimization process, each solution balances multiple objectives, and a solution is selected from the frontier solution set to readjust the MADDPG strategy.

Images