Goods transportation integrated logistics management system

Abstract

The invention provides an integrated logistics management system for cargo transportation. The integrated logistics management system comprises a multi-mode collaborative perception module, a semantic reasoning module, a scene adaptation decision module, an intelligent execution module and a logistics trust chain module. The multi-mode collaborative sensing module collects logistics full-scene state data through a sensor network; the semantic reasoning module is used for constructing a knowledge graph based on a graph attention network and realizing unified modeling of multi-source information; the scene adaptation decision module is combined with a pulse coupled neural network (PCNN) and a swarm intelligence collaboration mechanism to complete routing optimization; the intelligent execution module adopts a hierarchical task management and dynamic adjustment mechanism, and adjusts a task scheme according to execution abnormity and credit evaluation; the logistics trust chain module constructs a logistics trust system through a hierarchical consensus mechanism, a credit scoring model and a trust propagation and early warning mechanism, and realizes risk prevention and control of the whole process. The system solves the technical problems of a traditional logistics system in the aspects of data integration, dynamic supervision, decision support, collaborative scheduling, safety traceability and the like.

Description

Technical Field

[0001] The present invention relates to the field of logistics information technology, and in particular to a cargo transportation integrated logistics management system. Background Art

[0002] With the deepening development of the global economy and the rapid rise of new retail models, modern logistics and transportation are facing unprecedented challenges. The current logistics management system mainly has the following problems:

[0003] Insufficient system integration. Functional modules such as order management, warehouse scheduling, and transportation tracking are independent of each other, with serious data barriers; it is difficult for systems of different participants to achieve seamless connection; in multimodal transport scenarios, information exchange between various modes of transport is blocked; customs clearance, commodity inspection and other links in cross-border logistics are not well connected with the transportation system.

[0004] Dynamic monitoring capabilities are limited. Single GPS positioning and RFID identification cannot achieve all-round status monitoring; there are insufficient specialized monitoring methods for special goods; the real-time and accuracy of monitoring data is difficult to guarantee; and monitoring blind spots are common.

[0005] Insufficient decision support. Historical data analysis remains superficial, and deep value mining is insufficient; there is a lack of intelligent decision-making capabilities under multiple scenarios and constraints; and decision recommendations lack explainability.

[0006] The efficiency of collaborative dispatch is low. Fleet management lacks overall planning and is mainly based on single-vehicle dispatch; cross-regional collaborative capabilities are weak, and resources between regions are difficult to share; and there is a lack of a unified command and dispatch mechanism for multi-vehicle collaborative operations.

[0007] Safety tracing is difficult. There is no mechanism for tracking and verifying the integrity of goods; there is insufficient monitoring of physical conditions during transportation; and safety hazards caused by human factors are difficult to detect in a timely manner.

[0008] To sum up, the integration, intelligence and coordination of traditional logistics management systems are difficult to meet the needs of the complex environment of modern logistics, and they urgently need to be optimized and improved through innovative technologies. Summary of the invention

[0009] The present invention provides a cargo transportation integrated logistics management system, which realizes intelligent management of the entire cargo transportation process through the systematic integration of multiple functional modules, and is used to solve the technical problems of the existing logistics system in data integration, dynamic supervision, decision support, collaborative scheduling and safety traceability. The system builds a complete technical framework from cargo transportation perception, analysis to execution, and each module cooperates to support the integrated management of the entire cargo transportation process. Specifically, it includes the following technical modules:

[0010] 1. Multimodal collaborative perception module:

[0011] As the data foundation layer of the system, the temperature, humidity, vibration and inclination sensor networks set up in vehicles, warehouses and loading and unloading areas realize the status monitoring of the entire logistics scene; a deep belief network with a specific network structure is adopted, in which each type of sensor data is configured with an independent feature extraction layer to ensure the feature expression of different types of data; the edge computing node realizes local data processing through the preset federated learning model, transmits feature information to the central node, and reduces the data transmission load; this module provides real-time multi-dimensional perception data support for the semantic reasoning module and the scenario adaptation decision module.

[0012] 2. Semantic Reasoning Module:

[0013] As the knowledge processing layer of the system, the knowledge graph constructed based on the graph attention network will unify the modeling of the data obtained from the multimodal collaborative perception module with the order information, warehousing status, transportation route, vehicle information and cargo attributes; assign attention scores to different types of information through the node weight matrix to highlight key information; use the set knowledge graph completion algorithm to infer node relationships and fill in information gaps; use the ontology mapping method to achieve semantic connections between road, rail and air transportation; this module provides structured knowledge support for the scenario adaptation decision module, and at the same time provides the processed semantic information to the logistics trust chain module for credibility assessment.

[0014] 3.Scene Adaptation Decision Module:

[0015] As the intelligent decision-making layer of the system, it includes a pulse coupled neural network (PCNN) layer and a swarm intelligence collaboration layer. The PCNN layer constructs the logistics network into a neuron matrix structure, where each neuron corresponds to a logistics node. The connection weights between neurons are determined by normalized weighted calculation of four indicators: path distance, travel time, transportation cost, and path reliability. The maximum processing capacity of each node is the core basis for the activation threshold, and it is dynamically adjusted in combination with processing time and cache capacity. The swarm intelligence collaboration layer sets up multiple decision-making agents, each of which is responsible for an area within a specified radius. Its status information includes road condition index, vehicle availability, storage capacity, and emergency order ratio. Agents in adjacent areas complete status information exchange within a fixed period through a preset hierarchical communication protocol. The initial routing solution output by the PCNN layer serves as a soft constraint for the agent's decision. The agent optimizes the route based on the regional status information it has mastered, and the optimization result dynamically updates the connection weights of the PCNN layer according to the preset learning rate parameter. Finally, the module transmits the decision result to the intelligent execution module and provides structured decision data support for the logistics trust chain module.

[0016] 4. Intelligent execution module:

[0017] As the task execution layer of the system, a hierarchical task management mechanism is adopted to decompose the transportation task output by the scenario adaptation decision module into three sub-tasks: trunk line transportation, branch line scheduling and terminal delivery. The constraint model based on time window, path traffic status and node processing capacity is used to ensure the rationality and feasibility of the execution plan. The task execution plan is dynamically adjusted through the anomaly detection results of the multimodal collaborative perception module and the credit evaluation of the logistics trust chain module. At the same time, the real-time status of task execution is fed back to the multimodal collaborative perception module to form a closed-loop control, thereby enhancing the adaptability and stability of the system.

[0018] 5. Logistics trust chain module:

[0019] As the trust guarantee layer of the system, a logistics trust system is built through a hierarchical consensus mechanism, a credit scoring model, a trust propagation and early warning mechanism; based on the role attributes of carriers, warehouses, consignees and consignors, consensus weights are allocated in combination with cumulative credit scores and transaction data; the on-time rate, transaction volume share, praise rate, and complaint rate indicators are evaluated through a credit scoring model, and the initial trust assessment results of the intelligent execution module are inherited; the trust propagation model is used to calculate the trust relationship between nodes, and a hierarchical early warning mechanism is implemented, and data collection strategies, path planning decisions, resource allocation methods, and transaction consensus requirements are adjusted according to the trust and risk assessment results; at the same time, the module continuously receives real-time data from the multimodal collaborative perception module and task feedback from the intelligent execution module, dynamically updates the trust assessment results, and forms an efficient closed-loop trust management mechanism.

[0020] Beneficial effects of the present invention:

[0021] 1. Global perception monitoring: Use multimodal sensor networks and federated learning models to improve the coverage and data processing efficiency of logistics scenario status monitoring and reduce system monitoring blind spots.

[0022] 2. Semantic information fusion: Build a knowledge graph through the graph attention network to achieve the interconnection of multimodal transport information, reduce information islands, and improve information utilization efficiency.

[0023] 3. Intelligent decision-making optimization: Combining PCNN with group intelligence collaboration mechanism, route planning is completed under multiple constraints to improve the accuracy and timeliness of system decisions.

[0024] 4. Execution task control: Based on hierarchical task management and multi-dimensional constraint models, coordinated dispatching of trunk and branch lines is achieved to ensure the feasibility and efficiency of the execution plan.

[0025] 5. Full-process trust verification: Through the layered consensus mechanism and smart contracts, combined with dynamic credit scoring, a credibility assessment system for the logistics process is established to enhance security traceability capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The system architecture diagram of the present invention is shown;

[0027] Figure 2 The system working flow chart of the present invention is shown. DETAILED DESCRIPTION

[0028] Exemplary embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0029] Combination Figure 1 The present invention provides a cargo transportation integrated logistics management system, which is composed of the following:

[0030] 1. Multimodal collaborative perception module:

[0031] As the data foundation layer of the system, efficient status monitoring of the entire logistics scenario is achieved through scenario-optimized sensor network deployment, deep belief network feature extraction, and edge computing combined with the distributed processing mechanism of federated learning. The specific implementation includes:

[0032] A1). Sensor network deployment

[0033] Deploy temperature, humidity, vibration, and tilt sensors in transport vehicles, storage facilities, and loading and unloading areas, and use a scene-adaptive layered coverage mechanism for optimized deployment:

[0034] a). Dynamic calculation of coverage

[0035] The effective coverage of the sensor is dynamically adjusted by the perception quality function Q, as follows:

[0036]

[0037] Where: Q 0 is the initial perception quality; α is the distance attenuation coefficient; e is a natural constant; d(s,p): the distance from sensor s to monitoring point p; f env (p): Environmental impact factors.

[0038] b). Layered deployment

[0039] According to the characteristics of logistics scenarios, sensors are deployed in two layers:

[0040] Core monitoring layer: deployed at key nodes (loading and unloading ports, storage entrances), using high-precision sensors to ensure the accuracy of data in key areas.

[0041] Collaborative perception layer: deployed in regular areas, using low-cost sensors to provide wide coverage support.

[0042] A2). Deep Belief Network Structure (DBN)

[0043] A deep belief network with a specific network structure is used to extract features from multimodal data. A constrained hierarchical feature extraction mechanism is designed for temperature, humidity, vibration, and inclination data in logistics scenarios:

[0044] a). DBN energy function

[0045] The energy function of DBN feature extraction is:

[0046]

[0047] in:

[0048] v is the state vector of the neurons in the visible layer; v i is the state vector of the i-th neuron in the visible layer;

[0049] h is the hidden layer neuron state vector; h j is the jth neuron state vector in the hidden layer;

[0050] m and n are the number of neurons in the hidden and visible layers, respectively;

[0051] w ij is the connection weight between the visible layer and the hidden layer; b i ,c j is the bias term;

[0052] R(v,h) is the logistics scenario constraint, including temperature constraint, vibration amplitude constraint, and inclination range constraint, and is defined as:

[0053] R(v,h)=λ 1 R temp (v)+λ 2 R vib (v)+λ 3 R ang (v)

[0054] Where:

[0055] λ 1 ,λ 2 ,λ 3 is the weight coefficient; i ∈[0,1], satisfying Used to balance various constraints;

[0056] R temp (v) is the temperature change rate constraint:

[0057]

[0058] Limit the continuity of temperature data and suppress abnormal fluctuations. In the formula, α is the smoothing coefficient;

[0059] R vib (v) is the vibration amplitude constraint:

[0060] R vib (v)=βmax(0,‖v‖ 2 -V max )

[0061] Limit the vibration amplitude to exceed the set threshold V max , where β is the vibration suppression coefficient; ‖v‖ 2 is the second norm of the vibration amplitude;

[0062] R ang (v) is the tilt range constraint:

[0063]

[0064] Limit the inclination angle to exceed the safety range θ max , where γ is the inclination constraint coefficient; It is an indicator function that returns 1 if the condition is met, otherwise it returns 0.

[0065] b). Feature extraction layer design

[0066] According to the characteristics of multimodal data, the following independent feature extraction layers are designed respectively:

[0067] Temperature and humidity data: input dimension 1, the number of hidden layer nodes is [64, 32, 16];

[0068] Vibration data: input dimension 3, the number of hidden layer nodes is [128, 64, 32];

[0069] Inclination data: input dimension is 2, and the number of hidden layer nodes is [64, 32, 16].

[0070] A3). Edge computing and federated learning

[0071] Combining edge computing with federated learning enables distributed data processing and model training, effectively reducing the data transmission load and improving computing efficiency:

[0072] a). Local training on edge nodes

[0073] Each edge node is based on the local dataset D k Perform model training and update parameters

[0074]

[0075] in: They represent the model parameters of the edge nodes at time steps t+1 and t respectively; η is the learning rate, and its initial value is 0.01; is the gradient operator, used to differentiate the function; is the loss function based on the node dataset, defined as:

[0076]

[0077] in: is the model parameter of edge node k in the tth iteration; D k The local dataset used by edge node k; |D k | is the data set D k The size of the dataset is the number of samples in the dataset; x is the dataset D k An input sample in ; is the predicted output reconstructed from the input data x by the model; μ is the regularization coefficient; is the model parameter The square of the second norm of ; For the input data x and its reconstructed data The second norm squared error between .

[0078] b). Model aggregation and optimization

[0079] The central node aggregates the edge node model according to the following weighted average formula:

[0080]

[0081] γ k is the reliability coefficient of the node;

[0082] Q k (t) is the data quality score of the node, defined as:

[0083] Q k (t) = ω 1 C k (t)+ω 2 F k (t)+ω 3 T k (t)

[0084] Where: 1 ,ω 2 ,ω 3 : Weight coefficient, used to balance the impact of different data quality indicators, satisfying:

[0085] C k (t) is the completeness score. The completeness score measures the proportion of missing values ​​in the data. The higher the proportion, the more complete the data is. It is defined as:

[0086] Where: Nmiss is the number of missing data points; N total is the total number of data points in the dataset.

[0087] F k (t) is the freshness score, which measures the timeliness of the data. The higher the score, the fresher the data is, ensuring that the model uses data with strong timeliness. Data with low freshness may affect the real-time and accuracy of the model. It is defined as:

[0088] Where: t now is the current timestamp; t last is the timestamp of the most recent data update; β is the time decay coefficient, which controls the speed of time decay.

[0089] T k (t) is the accuracy score, which measures the accuracy of the data based on outlier detection. The higher the score, the more accurate the data is, and is defined as:

[0090] in: is the mean of historical data; σ is the standard deviation of historical data; I(·) is the indicator function. If the deviation of data x from the historical mean is greater than 3 times the standard deviation, it is judged as an outlier; |D k | is the data set D k The number of samples in .

[0091] This scoring mechanism takes into account the integrity, timeliness and accuracy of the data, and is used to weight the data quality of different nodes during the model aggregation process. High-quality data can be given a higher weight, prompting the model to rely more on the data of these nodes during aggregation, ultimately improving the overall model performance.

[0092] A4). Multi-dimensional perception data support

[0093] Through the joint processing of sensor networks and deep belief networks, high-quality data support is provided for semantic reasoning and decision-making modules:

[0094] Real-time data on temperature, humidity, vibration and inclination are converted into structured perception information after feature extraction; the real-time and integrity of the data are ensured through the efficient collaboration of edge computing and federated learning.

[0095] 2. Semantic Reasoning Module

[0096] As the knowledge processing layer of the system, a knowledge graph is constructed based on the graph attention network (GAT), and the multi-dimensional data collected by the multimodal collaborative perception module is uniformly modeled with order information, storage status, transportation routes, vehicle information and cargo attributes. The semantic association weights between nodes are assigned through the dynamic attention mechanism to highlight key information; the knowledge graph completion algorithm is used to predict and improve node relationships and fill in information gaps; and the ontology mapping method is combined to achieve semantic connections in multimodal transport scenarios. The specific implementation includes:

[0097] B1). Knowledge graph construction

[0098] The knowledge graph G = (V, E, R) is constructed, which includes the node set V, the edge set E and the relationship set R, and is used to uniformly model the heterogeneous information in the logistics system.

[0099] a). Node set definition

[0100] The node set V includes the following types:

[0101] Order Node V o : Contains order ID, cargo type, quantity and timeliness requirements, which are used to describe the core information of logistics requirements;

[0102] Storage Node V w : Including warehouse ID, storage location, capacity and inventory status, reflecting the storage information of goods;

[0103] Path Node V p : Record the road section ID, starting and ending points, distance and traffic status to describe the transportation route;

[0104] Vehicle Node V v : Including vehicle ID, type, load and current location, reflecting the real-time status of logistics transportation tools;

[0105] Cargo Node V g : Contains cargo ID, attributes, status and location information, used to monitor cargo flow.

[0106] b). Edge set definition

[0107] The edge set E is defined as a specific relationship instance between entity nodes, including:

[0108] E={(v i ,r k ,v j ,p)|v i ,v j ∈V,r k ∈R,p∈P}

[0109] Where: v i is the starting node; rk is the relationship type; v j As the target node, define the same i ; p is the edge attribute; V is the node set; R is the relationship type set; P is the edge attribute set, including the following categories:

[0110] Time attributes: relationship establishment time, relationship termination time (if in progress, estimated termination time), and last status update time;

[0111] State attributes: current state identification, previous state record, expected state transition;

[0112] Weight attributes: basic relationship weight, timeliness weight, service quality weight, cost-related weight

[0113] Business attributes: capacity / load related parameters, efficiency related parameters, resource occupancy parameters, and constraint condition parameters.

[0114] c). Relationship set definition

[0115] The relation set R includes:

[0116] R = {stored in, transported in, belonging to, connected to, carried by, including}

[0117] in:

[0118] Stored in: The relationship between goods and storage, indicating that the goods are stored in a warehouse;

[0119] Transported by: The relationship between goods and vehicles, indicating that the goods are transported by a certain vehicle;

[0120] Belongs to: The relationship between a vehicle and a company or the relationship between a warehouse and a logistics center, indicating that a vehicle belongs to a certain company or a certain warehouse belongs to a certain logistics center;

[0121] Connection: The relationship between path nodes, indicating the connection relationship of the transportation path;

[0122] Transport: The relationship between a vehicle and an order, indicating that the vehicle transports a certain order;

[0123] Contains: The relationship between an order and goods, or the relationship between a warehouse and a storage location, indicating that an order contains multiple goods, or a warehouse contains multiple storage locations.

[0124] B2). Graph Attention Network Implementation

[0125] This module uses graph attention network to model the knowledge graph and complete the dynamic weight allocation and information aggregation of node features.

[0126] a).Node feature encoding

[0127] Encode node features according to node type:

[0128]

[0129] Where: h i represents the feature vector of node i; φ(v i ) is the node type; x i is the original feature of the node; is a specific encoder for the node type.

[0130] The encoder parameters are configured as follows:

[0131] Order node: input dimension 12, output dimension 64; warehouse node: input dimension 8, output dimension 64; path node: input dimension 10, output dimension 64; vehicle node: input dimension 6, output dimension 64; cargo node: input dimension 8, output dimension 64.

[0132] b) Attention Mechanism

[0133] The semantic association weights between nodes are assigned through the graph attention network, and the calculation formula is as follows:

[0134]

[0135] in:

[0136] α ij represents the attention weight between node i and its neighbor node j; h i ,h j ,h k is the feature vector of nodes i, j and k; W is the feature transformation weight matrix; a is the attention weight vector; is the neighbor set of node i; exp represents the natural exponential function; LeakyReLU is the linear rectification activation function with leakage correction; || represents the vector concatenation operation.

[0137] c). Information aggregation

[0138] Update node features through multi-head attention mechanism:

[0139] in:

[0140] K is the number of attention heads (set to 8), which enhances the diversity of information aggregation between nodes; σ is the activation function used to introduce nonlinear features; || represents the concatenation operation of the outputs of K attention heads.

[0141] B3). Knowledge graph completion

[0142] In order to fill the information gaps in the knowledge graph, the TransE model is used for relationship prediction and completion:

[0143] a). Relationship prediction

[0144] Predict missing relations based on the embedding vectors of head entity, tail entity, and relation:

[0145] ‖h+rt‖ 2

[0146] Among them: h, r, t are the vectors of the head entity, relation and tail entity respectively; the embedding dimension is set to 100.

[0147] b). Confidence assessment

[0148] Confidence assessment of predicted relationships:

[0149] conf(h,r,t)=σ(-‖h+rt‖ 2 )

[0150] Where: σ is the sigmoid function; the confidence threshold is set to 0.8, which is used to screen reliable relationships.

[0151] B4). Multimodal transport semantic mapping

[0152] In order to achieve semantic connection between road, rail and air transport, ontology mapping technology is used for standardization:

[0153] a). Ontology construction

[0154] Constructing Highway Transport Ontology r 、Railway transport t and air transport entity O a , a unified description of the core concepts of each mode of transportation.

[0155] b. Ontology mapping

[0156] Compute semantic similarity between concepts through information content measurement:

[0157]

[0158] Where: c 1 ,c 2 are two concepts to be mapped; MICA(c 1 ,c 2 ) is the most recent common ancestor; IC is the information content, reflecting the importance of the concept.

[0159] c. Rule Generation

[0160] Generate cross-transportation mode rules through mapping relationships:

[0161] Rule i→j :P i (x 1 ,...,x n )∧Map i,j →P j (y 1 ,...,y m )

[0162] Among them: Rule i→j represents the rule from state i to j; P i (x 1 ,...,x n ) represents a certain attribute or condition in state i; ∧ is the logical AND operator, indicating that two conditions must be met at the same time; Map i,j represents the mapping relationship from state i to j; P j (y 1 ,...,y m ) represents a certain attribute or condition under state j.

[0163] The processed semantic information is passed to the decision-making module for path planning and scheduling optimization; the node relationship and confidence are given to the trust chain module for dynamically adjusting the credibility of the logistics network.

[0164] 3. Scene Adaptation Decision Module

[0165] As the intelligent decision-making layer of the system, this module constructs an adaptive routing optimization framework under multiple constraints through the innovative combination of pulse coupled neural network (PCNN) and swarm intelligence coordination mechanism. This framework fully considers the node processing capacity, path characteristics and regional status information in the logistics scenario, and can achieve efficient coordination between global planning and local optimization, providing reliable support for intelligent decision-making in complex logistics networks. Specific implementation includes:

[0166] 3.1 PCNN network construction and optimization

[0167] C1). Logistics network modeling

[0168] Map each logistics node in the logistics network to a neuron of PCNN. The activation threshold of the neuron is θ i Reflects the maximum processing capacity of the node. The calculation formula is:

[0169] θ i =k 1 C i +k 2 T i -k 3 B i

[0170] Where: Ci is the maximum processing capacity of node i (pieces / hour), obtained from historical data statistics; T i is the average processing time (hours) of node i, indicating the processing efficiency; B i is the cache capacity of node i (pieces), which is used to reflect the temporary storage capacity; k 1 ,k 2 ,k 3 is the weight coefficient, which is obtained through regression optimization of historical operation data.

[0171] C2). Connection weight optimization

[0172] The path connection weight W between nodes ijkl Taking into account the path distance, travel time, transportation cost and path reliability, the calculation formula is as follows:

[0173]

[0174] Where: d ijkl is the path distance, reflecting the physical distance between nodes; t ijkl is the travel time, which indicates the impact of traffic conditions on path efficiency; c ijkl is the transportation cost, indicating the path economy; R ijkl is the path reliability index, calculated by the historical on-time rate of the path; ω 1 ,ω 2 ,ω 3 ,ω 4 : Weight coefficient, satisfying ∑ω i =1, obtained through experimental optimization.

[0175] Neuron Dynamic Equations The neuron dynamic behavior of PCNN is described by the following equations:

[0176]

[0177] U ij [n] = F ij [n](1+βL ij [n]),

[0178]

[0179] in:

[0180] F ij [n], L ij [n]: indicates feed input and link input respectively; U ij [n] is the internal activity of the neuron; Y ij [n] is the output state, 1 means the node is activated, 0 means it is not activated; α F ,αL is the input attenuation coefficient; V F ,V L is the input amplitude coefficient; β is the link strength.

[0181] C3). Group intelligence collaborative optimization

[0182] Each decision agent a i Maintain a state vector to describe the logistics status of the area it is responsible for:

[0183] s i (t) = [ρ i (t),v i (t),w i (t),u i (t)]

[0184] in:

[0185] ρ i (t) is the road condition index, ranging from [0,1], reflecting the traffic conditions, calculated by the ratio of the current vehicle speed to the ideal free flow speed;

[0186] v i (t) is the vehicle availability, ranging from [0,1], indicating the proportion of dispatchable vehicles in the current area; w i (t) is the storage capacity, ranging from [0,1], reflecting the storage utilization rate, which is calculated by the ratio of available space to total space; u i (t) is the emergency order ratio, which indicates the proportion of emergency orders in the region. When the remaining time is less than 20% of the standard time, it is considered an emergency order.

[0187] The agents use a hierarchical gossip protocol to exchange state information. The state update formula is as follows:

[0188]

[0189] in:

[0190] Respectively represent the direct neighboring area and the second-order neighboring area of ​​agent i; α 1 ,α 2 are the direct neighbor and distant neighbor influence factors respectively; ∈ is the information fusion coefficient.

[0191] C4). Two-tier collaborative decision-making mechanism

[0192] The activation state output by PCNN is used as the soft constraint of the agent, and the formula is as follows:

[0193] C(a i )=β 1 D(a i,Y)+β 2 V(a i )

[0194] in:

[0195] β 1 ,β 2 is the weight parameter;

[0196] D(a i ,Y): measures the deviation between the agent’s decision and the PCNN routing solution, which is calculated by the ratio of the deviation between the actual path and the planned path and the planned path;

[0197] V(a i ): Local feasibility evaluation, used to limit infeasible paths, the calculation formula is:

[0198] V(a i )=min{V 1 ,V 2 ,V 3}

[0199] in:

[0200] V 1 Time feasibility, indicating whether there is sufficient time; V 2 is resource feasibility, indicating whether resource supply and demand match. V 3 Cost feasibility, indicating whether the budget is sufficient; Represents the minimum value function;

[0201] Combining timeliness, economy, path reliability and load balancing, the reward function is defined as follows:

[0202]

[0203] in:

[0204] λ 1 , 2 , 3 , 4 is the weight parameter;

[0205] t actual ,t limit They represent the actual time required to complete the task and the time limit to complete the task respectively;

[0206] c actual 、c budget Represent the actual cost and budgeted cost respectively;

[0207] R path is the path reliability score, and the calculation formula is:

[0208]

[0209] in:

[0210] R T is time reliability, which indicates the proportion of tasks completed on time. R C Cost reliability reflects the deviation between actual cost and planned cost. R S It is the safety reliability, which is the reverse indicator of the accident rate.

[0211] C5). Decision-making process and convergent judgment

[0212] Initialize the PCNN network and agent state, set learning parameters and path weights; generate an initial routing plan through PCNN, then optimize the path plan through the agent, update the PCNN weights, and check the convergence conditions. The convergence condition is that the objective function change rate is less than the threshold value.

[0213]

[0214] Where: ∈ = 0.01, f t 、f t-1 Represent the objective function values ​​at time t and time t-1 respectively.

[0215] 4. Intelligent execution module

[0216] As the task execution layer of the system, it realizes accurate execution and real-time adjustment of transportation tasks through hierarchical task management mechanism, multi-dimensional constraints, dynamic feedback and abnormal response mechanism and resource optimization strategy, ensuring efficient and reliable operation of the system in complex logistics scenarios. Specific implementation includes:

[0217] D1). Task decomposition mechanism

[0218] The task decomposition adopts a three-layer structure, namely:

[0219] T 1 : Trunk transport task set; T 2 : branch line scheduling task set; T 3 : A collection of terminal delivery tasks.

[0220] The total task T is defined as:

[0221] T=T 1 ∪T 2 ∪T 3

[0222] The set of tasks at each level meets the following conditions:

[0223]

[0224] in:

[0225] t ij represents the jth subtask in the i-th layer task; k is the level; n i is the number of tasks at the i-th layer.

[0226] Task decomposition is achieved through the optimization model, and the optimization goal is to minimize the execution cost and interaction cost of the task:

[0227] when

[0228] in:

[0229] t ij : The jth subtask of the i-th layer; C(t ij ): execution cost of a single task; η(t ij ,t ik ): task interaction cost; λ: weight factor of task interaction cost.

[0230] D2). Multidimensional Constraint Model

[0231] a). Time window constraints

[0232] The tasks must be completed within the specified time frame:

[0233] ET ij ≤AT ij ≤LT ij ,DT ij -AT ij ≤PT ij ,

[0234] Among them: ET ij ,LT ij AT is the earliest and latest arrival time of the task; ij is the actual arrival time; DT ij Departure time; PT ij is the maximum processing time.

[0235] b). Path traffic status constraints

[0236] Path feasibility is assessed by the following model:

[0237]

[0238] Among them: U ij[n] is the internal activity of PCNN neurons, which dynamically reflects the real-time changes of path status;

[0239] ρ·θ i The activation threshold θ i and the proportionality coefficient ρ, which represents the dynamic lower limit of path reliability; R ijkl is the path reliability index, which is dynamically calculated by the network training process of PCNN.

[0240] Cong(p ij ) is the congestion degree, and the calculation formula is:

[0241]

[0242] Where: t actual ,c actual is the measured travel time and transportation cost; t ijkl ,c ijkl is the preset travel time and cost; μ 1 ,μ 2 is the weight coefficient, satisfying μ 1 +μ 2 =1.

[0243] c). Node processing capacity constraints

[0244] The processing power of a node is defined by the following model:

[0245]

[0246] in:

[0247] x ijk For task t ij The variable v that uses resource k at node j; k is the unit demand of resource k; Q j is the maximum processing capacity of node j.

[0248] D3). Dynamic adjustment mechanism

[0249] a). Abnormal event assessment

[0250] Abnormal events are evaluated through the following model to assess the risk value of abnormal events:

[0251] Risk(e i )=P(e i )·I(e i ),

[0252] in:

[0253] P(e i ) is the probability of abnormal events, and the calculation method is:

[0254] P(e i )=[1-Q k (t)]·exp(-ψ·R ijkl )

[0255] Where: Q k (t) is the data quality score; ψ is the influence coefficient, which controls the attenuation effect of path reliability on probability, and takes ψ = 0.5; R ijkl It is the path reliability index, which dynamically reflects the reliability of the path.

[0256] The higher the data quality score, the lower the probability of abnormal events; the higher the path reliability index, the probability of abnormal events decreases exponentially.

[0257] I(e i ) is the impact intensity, and the calculation formula is:

[0258]

[0259] Where: t delay is the delay time, which means the difference between the actual travel time and the preset travel time; t ijkl is the preset travel time; c extra is the additional cost, including the increased cost caused by the current road conditions or other emergencies; c ijkl is the preset travel cost; 1 ,η 2 is the weight coefficient and η 1 +η 2 =1.

[0260] b). Dynamic adjustment of tasks

[0261] Among all available options S, choose the one with the utility U(s,t ij ) and initial trust Trust init The solution s with the largest weighted product value of (s) * :

[0262]

[0263] in:

[0264] U(s,t ij ) is the utility function, defined as:

[0265]

[0266] Among them: U ij [n] is the internal activity of neurons in PCNN; θ i is the neuron activation threshold in PCNN, which represents the evaluation baseline of path performance; t actualis the actual travel time; t ijkl is the preset travel time; 1 ,φ 2 is the weight coefficient, satisfying φ 1 +φ 2 =1.

[0267] Trust init (s) is the initial trust, defined as:

[0268] Trust init (s) = exp(-κ·E s )

[0269] Where: E s is the cumulative impact of execution deviation, E s =∑(d k / d max ) 2 , where d k is the deviation of the kth execution, calculated by the actual execution result and the preset value, d max is the maximum allowed deviation, which is used for normalization; κ is the attenuation coefficient, which controls the impact of deviation on trust.

[0270] D4). Real-time feedback mechanism

[0271] a).Task status monitoring

[0272] Task state vector definition:

[0273] State(t ij )={Location,Progress,Resource,Exception}

[0274] Among them: Location is the task execution location, based on the node index of PCNN; Progress is the task completion progress, standardized to [0,1]; Resource is the resource usage; Exception is the exception mark, which takes 1 for exception and 0 for normal.

[0275] Overall status score: K

[0276]

[0277] h k (State k ) is the scoring function:

[0278] Where: State ref is the planning reference value from the scene adaptation decision module; δk is the state deviation coefficient, which is set according to the importance of the indicator; I() is the indicator function, which is used for discrete state judgment.

[0279]

[0280] T base : basic feedback cycle; τ 1 =0.8 is the high state threshold. τ 2 =0.6 is the low state threshold.

[0281] D5). Execute optimization strategy

[0282] a). Resource scheduling optimization

[0283] Dynamic programming optimization goal: V t (s t )=max[R(s t ,a t )+γ·V t+1 (s t+1 )]

[0284] Where: R(s t ,a t ) represents the immediate reward function, R(s t ,a t )=ξ 1 ·EP(a t )+ξ 2 ·RC(a t ), where ξ 1 ,ξ 2 is the weight coefficient and ξ 1 +ξ 2 =1;EP(a t ) is the execution performance evaluation, and H(t ij ) standardization value; RC(a t ) Resource consumption assessment, RC(a t )=1-∑c actual / ∑c ijkl , where c actual is the actual resource consumption cost; c ijkl The preset resource cost.

[0285] b). Dynamic adjustment of task priority

[0286] Priority(t ij )=α 1 ·U(t ij )+α 2 ·D(t ij )+α 3 ·C(t ij )

[0287] in:

[0288] Weight coefficient α 1 , α 2 , α 3 , satisfying α 1 +α 2 +α 3 =1;

[0289] Urgency U(t ij ):

[0290]

[0291] Where, t ddl is the task deadline; t cur is the current time; t total is the total allowed execution time; ν is the time urgency coefficient.

[0292] Dependence D(t ij ):

[0293] Where N dep N is the number of subsequent tasks that depend on this task; total is the total number of tasks in the current system.

[0294] Completion degree C(t ij ):

[0295]

[0296] 5. Logistics Trust Chain Module

[0297] As the trust guarantee layer of the system, an efficient and reliable logistics trust system is built through layered consensus mechanism, smart contracts and dynamic credit evaluation. The specific implementation includes:

[0298] E1). Layered consensus mechanism

[0299] The hierarchical consensus mechanism supports fair participation and decision-making of multiple parties in the logistics system through role attributes, credit scoring, dynamic weight distribution and flexible consensus rules.

[0300] a). Role attribute definition

[0301] Basic role: Carrier R car , responsible for cargo transportation; warehousing party R war , responsible for the storage and allocation of goods; consignor and consignee R con , responsible for the sending and receiving of goods.

[0302] b).Cumulative credit score

[0303] The credit score is calculated based on a set of quantitative indicators defined by the business key points of the role type and calculated comprehensively by weight.

[0304] Scoring formula:

[0305]

[0306] in:

[0307] Carrier rating indicators (type = car):

[0308] Punctuality rate 1 , the proportion of tasks completed on time; path matching degree Q 2 , the degree of consistency between the actual and planned routes; the cargo integrity rate Q 3 , the complement of the proportion of damaged goods; abnormal response rate Q 4 , the proportion of abnormal situations handled in a timely manner; weight φ k The values ​​are 0.3, 0.2, 0.3, and 0.2 respectively.

[0309] Warehouse scoring indicators (type = war):

[0310] Storage accuracy Q 5 , the proportion of storage records consistent with actual goods; loading and unloading efficiency Q 6 , the amount of loading and unloading tasks completed per unit time (unit: pieces / hour) is standardized; inventory accuracy Q 7 , the proportion of inventory records consistent with actual goods; error rate complement Q 8 , the proportional complement of storage or operation errors; the weights are 0.3, 0.3, 0.2, 0.2 respectively;

[0311] Scoring indicators for consignors and receivers (type=con):

[0312] Timely settlement Q 9 , on-time payment ratio; information accuracy Q 10 , the ratio of information transmitted to actual consistency; degree of coordination Q 11 , measured by successful completion of tasks, and record the success rate; document validity Q 12 , the completeness of the documents transmitted and confirmed; the weights are 0.2, 0.3, 0.2, and 0.3 respectively.

[0313] c) Consensus weight distribution

[0314] The consensus weight is dynamically calculated based on role attributes and historical behavior.

[0315] Weight formula:

[0316]

[0317] Among them: Role type weight (π type ): Carrier weight π car =0.4, warehouse weight π war =0.3,

[0318] Weight of consignor and consignee π con =0.3.

[0319] Credit score (C i ): C i =δ·Trust init (i)+(1-δ)·Credit i (type)

[0320] Where: trust inheritance coefficient δ = 0.4;

[0321] Transaction frequency T i =N i (type) / max(N 1 ,...,N n ), N i (type) is the number of transactions for a specific role;

[0322] Transaction amount V i =M i (type)max(M 1 ,...,M n ), M i (type): the total transaction amount of a specific role;

[0323] The weight coefficients α=0.5, β=0.3, and γ=0.2.

[0324] d) Consensus reaching mechanism

[0325] If ∑(W i ·V i )≥θ(type), then the consensus variable Agreement=1 (reaching consensus);

[0326] If ∑(W i ·V i )<θ(type), then the consensus variable Agreement=0 (no consensus reached).

[0327] Among them, ∑(W i ·V i ) is the weight of each node W i And the corresponding value V i The sum of the products of , θ(type) is the consensus threshold determined according to the decision type, and the calculation method is:

[0328]

[0329] Where f is the maximum number of fault-tolerant nodes, f = (n-1) / 3 and rounded down; n is the total number of nodes; ∈ type is the adjustment coefficient, which is determined by the decision type. For carrier-related decisions, ∈ car = 1.1, for warehousing-related decisions, ∈ war =1.0, for the relevant decisions of the consignor and consignee,∈ con =0.9.

[0330] In addition, the participation requirements for different roles are different: carriers require that at least 40% of carrier nodes must participate in the decision-making vote. Only when this participation requirement is met, the calculation results of the consensus formula are meaningful; warehouses require that at least 30% of warehouse nodes must participate in the decision-making vote; shippers and consignees require both parties to clearly confirm before passing, and are not subject to unilateral constraints on the calculation results of the consensus formula.

[0331] E2). Smart Contract

[0332] a). Status verification and conversion

[0333] Verify the current operation A t Whether all rules are met k If it passes, the state transition is allowed. The state transition is based on the current state S t 、Operation A t and condition set C t .

[0334] b). Positioning verification

[0335] Check whether the deviation between the current position and the planned position is within the allowable range.

[0336] c). Timeliness verification

[0337] Check whether the deviation of the operation completion time from the planned time is as expected.

[0338] E3). Credit scoring model

[0339] a). Rating calculation model

[0340] Credit score is calculated by weighting multiple indicators:

[0341]

[0342] Where: w k is the indicator weight, w k =w k,base ·exp(μ·ΔI k ), where μ is the weight update coefficient, which takes a value of 0.3; ΔI kis the index change range; w k,base is the benchmark indicator weight; I k It is the index value, including: punctuality rate I 1 , normal transaction volume accounts for I 2 , favorable rate I 3 , complaint rate complement I 4 , the initial trust I obtained by the intelligent execution module 5 .

[0343] b). Dynamic update mechanism

[0344] The credit score is dynamically updated using the following formula:

[0345] Score new =(1-α)·Score old +α·[β·Trust init +(1-β)·ΔScore]

[0346] Among them: α = 0.2 is the update coefficient, which controls the fusion ratio of new and old scores; β = 0.3 is the inheritance coefficient, which balances the impact of initial trust and newly added scores; ΔScore: the score change of the newly added indicator.

[0347] E4). Dynamic trust assessment and risk warning mechanism

[0348] a). Trust Propagation Model

[0349] The formula for trust propagation among nodes is:

[0350]

[0351] Where: T ij : the trust of node i in node j; T ik ,T kj : A chain of trust propagated through intermediate nodes.

[0352] b). System monitoring and early warning

[0353] The warning level is determined by the comprehensive risk value:

[0354]

[0355] Among them: Risk(e i ) is the abnormal event risk value obtained by the intelligent execution module; θ h ,θ l High and low risk thresholds.

[0356] E5). Early warning response and coordinated regulation

[0357] Based on the aforementioned risk warning mechanism, the system takes corresponding coordinated control measures for different warning levels:

[0358] a). High risk status

[0359] Transport control: restrict carriers from adding new waybills; increase the frequency of GPS positioning and sensor data collection to once every 5 minutes; require vehicles to travel along the main route planned by PCNN, prohibit alternative routes; and enforce cargo status checks at each handover point.

[0360] Warehouse control: Raise the processing capacity threshold of warehouse nodes to 120%; increase the frequency of inventory counts to once a day; require warehouses to provide additional video surveillance records.

[0361] Information control: All transaction information requires additional confirmation nodes to participate in consensus; increase the node association threshold of the knowledge graph; shorten the execution timeout of smart contracts to 50% of the standard value.

[0362] b) Medium risk status

[0363] Transport control: New waybills are allowed, but additional node confirmation is required; the sensor data collection frequency is adjusted to once every 15 minutes; alternative routes planned by PCNN are allowed, but route changes must be reported in real time; cargo status checks are performed at key handover points.

[0364] Warehouse control: Maintain standard processing capacity thresholds for warehouse nodes; maintain regular inventory counting frequency, but require detailed records; strengthen monitoring of key areas.

[0365] Information control: transaction information requires a benchmark number of nodes to participate in consensus; maintain standard association thresholds for knowledge graphs; and execute standard timeout mechanisms for smart contracts.

[0366] c) Low risk status

[0367] Transport control: New waybills are added normally; sensor data collection frequency is maintained at 30-minute intervals; all optional routes planned by PCNN are allowed to be flexibly used; and cargo status is checked according to standard procedures.

[0368] Warehouse control: Allow warehouse nodes to flexibly adjust processing capacity; implement regular inventory management mechanisms; maintain standard monitoring frequency.

[0369] Information control: Transaction information is processed according to the standard consensus mechanism; dynamic association threshold adjustment of knowledge graphs is allowed; smart contracts are executed according to standard procedures.

[0370] Combination Figure 2 As shown, the cargo transportation integrated logistics management system of the present invention works in the following five stages:

[0371] Phase 1: Multimodal Data Perception and Edge Processing

[0372] The system uses a multimodal collaborative sensing module to deploy a sensor network in transport vehicles, storage facilities, and loading and unloading areas to collect logistics status data such as temperature, humidity, vibration, and inclination. The sensor network adopts the following hierarchical deployment method:

[0373] Core monitoring layer: High-precision sensors are placed at key nodes such as loading and unloading ports and storage entrances to ensure the accuracy of status monitoring in key areas.

[0374] Collaborative sensing layer: Low-cost sensors are deployed in regular areas to expand monitoring coverage.

[0375] The data is transmitted to the edge computing node, and the deep belief network (DBN) is used to perform hierarchical feature extraction on the collected multimodal data to generate hidden feature representation.

[0376] Federated learning model processing: The edge computing node completes model training based on the local data set and uploads the model parameters to the central node. The central node integrates the parameters of each edge node through a weighted aggregation algorithm to generate compressed global feature data, providing real-time, multi-dimensional data support for the semantic reasoning module.

[0377] Phase 2: Semantic Reasoning and Knowledge Graph Construction

[0378] The system receives the feature data transmitted by the multimodal collaborative perception module through the semantic reasoning module, and combines the order information, storage status, transportation route, vehicle information and cargo attributes to construct a knowledge graph G = (V, E, R):

[0379] Node set V: including order node V o 、Storage Node V w 、Path Node V p 、Vehicle node V v and cargo node V g .

[0380] Edge set E and relationship set R: define the semantic relationship and association type between nodes respectively.

[0381] Use Graph Attention Network (GAT) to dynamically assign semantic weights between nodes and aggregate key information.

[0382] In order to address the information gaps in the knowledge graph, the TransE model is used to predict and complete relationships, and reliable completion results are screened through confidence evaluation.

[0383] To support multimodal transport scenarios, the semantic reasoning module uses ontology mapping technology to build a unified semantic model among road, rail and air transport, providing structured knowledge support for the scenario adaptation decision module.

[0384] Phase 3: PCNN and swarm intelligence routing decision

[0385] Pulse Coupled Neural Network (PCNN) modeling: The scenario adaptation decision module maps the logistics network into the neuron matrix structure of PCNN and dynamically calculates the activation threshold of neurons according to the following rules: node processing capacity and path characteristics (distance, travel time, transportation cost and reliability) are used as key decision-making indicators; the connection weights between neurons are generated by normalized weighted calculation based on the above indicators.

[0386] Swarm intelligence collaborative optimization mechanism: The scenario adaptation decision module sets up multiple decision-making agents. Each agent maintains the state vector of the area it is responsible for, including road condition index, vehicle availability, storage capacity and emergency order ratio; neighboring agents periodically exchange state data through a hierarchical communication protocol; the initial routing solution output by PCNN serves as a soft constraint condition for the agent's decision. The agent optimizes the route based on the state information of the area, and the optimization result dynamically updates the connection weight of PCNN.

[0387] The collaboratively optimized routing solution is passed to the intelligent execution module, while providing structured decision data for the logistics trust chain module.

[0388] Phase 4: Hierarchical logistics task execution

[0389] Task decomposition: The intelligent execution module receives the routing plan output by the scenario adaptation decision module and decomposes it into three levels of tasks: trunk transportation, branch line scheduling and terminal delivery. Each task must meet the constraints of time window, path traffic status and node processing capacity.

[0390] Dynamic adjustment mechanism: When the perception module detects an abnormal event (such as path interruption, vehicle failure, etc.), the intelligent execution module dynamically adjusts the task execution plan based on the abnormal detection results and the credit evaluation provided by the logistics trust chain module to ensure the smooth completion of the task.

[0391] Real-time feedback mechanism: The task execution status is transmitted to the multimodal collaborative perception module in real time through the feedback module, forming a closed-loop control to support subsequent task optimization.

[0392] Phase 5: Logistics trust chain construction and risk prevention and control

[0393] Hierarchical consensus mechanism: The logistics trust chain module is based on the role attribute definitions of carriers, warehouses, and consignees and consignees. It calculates the cumulative credit score based on role-specific indicators such as on-time fulfillment rate, storage accuracy, and settlement timeliness. It also allocates consensus weights based on factors such as transaction frequency and transaction amount to achieve trust evaluation of all parties in the logistics network.

[0394] Credit scoring and propagation: The module uses a scoring model to evaluate indicators such as punctuality rate, transaction volume ratio, praise rate, complaint rate, etc., and updates the credit score in combination with a time decay mechanism; it uses a trust propagation model to calculate the trust relationship between nodes and build a trust network.

[0395] Risk warning and coordinated regulation: Based on the trust level and risk assessment results, graded warning is implemented, and differentiated management and control measures are taken for different risk levels, including adjusting data collection frequency, restricting path selection, changing resource allocation methods and adjusting consensus mechanism parameters, and coordinating with other modules to complete risk prevention and control.

[0396] Through the collaborative work of the above five stages, the cargo transportation integrated logistics management system of the present invention realizes closed-loop operation from data collection and processing, knowledge reasoning, intelligent decision-making to task execution and risk prevention and control. The collaborative cooperation between the modules of the system ensures the controllability of the logistics process.

[0397] In summary, the present invention provides a cargo transportation integrated logistics management system, which constructs a complete technical framework from logistics scene perception, knowledge processing to task execution through the hierarchical perception network and edge computing processing of the multimodal collaborative perception module, the knowledge graph construction and ontology mapping of the semantic reasoning module, the PCNN and group intelligence collaboration of the scene adaptation decision module, the hierarchical task management of the intelligent execution module and the hierarchical consensus mechanism of the logistics trust chain module. The system realizes the intelligent management of the entire logistics process through the data interaction and feedback mechanism between the modules, and effectively solves the technical problems of the existing logistics system in data integration, dynamic supervision, decision support, collaborative scheduling and safety traceability.

[0398] Although this document shows the specific application of the present invention through specific embodiments, these embodiments are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims, and any appropriate modification, equivalent substitution or improvement based on the principle of the present invention should be deemed to fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be given the broadest interpretation to cover all such modifications, equivalent structures and functions.

Claims

1. A cargo transportation integrated logistics management system, characterized in that: include: The multimodal collaborative perception module is used to collect multi-dimensional status data of logistics scenarios through the deployed temperature, humidity, vibration and tilt sensor network, and realize local data processing and feature information aggregation of central nodes based on edge computing nodes and federated learning models; The semantic reasoning module is used to build a knowledge graph based on the graph attention network, unify the modeling of multimodal data with order information, storage status, transportation routes, vehicle information and cargo attributes, and realize the semantic connection of multimodal transport scenarios through ontology mapping method; The scenario-adaptive decision-making module is used to perform adaptive routing optimization of logistics networks under multiple constraints of path distance, transportation cost, and node processing capacity based on pulse coupled neural network (PCNN) and swarm intelligence coordination mechanism; Intelligent execution module, which is used to decompose the transportation tasks output by the decision module into trunk transportation, branch line scheduling and terminal delivery tasks, and adjust the task execution plan based on the real-time feedback mechanism; The logistics trust chain module is used to evaluate the trust relationship between carriers, warehouses, and consignees and consignors based on a hierarchical consensus mechanism and a dynamic credit scoring model, and to generate graded warnings and dynamic control measures based on risk assessment.

2. The cargo transportation integrated logistics management system according to claim 1, wherein: The multimodal collaborative perception module uses a deep belief network (DBN) to extract features from temperature, humidity, vibration and inclination data, and optimizes data coverage through the core monitoring layer and collaborative perception layer of the hierarchical perception network.

3. The cargo transportation integrated logistics management system according to claim 1, wherein: The multimodal collaborative perception module optimizes the aggregation model of the central node through the distributed training mechanism of federated learning, and the weight distribution is based on the data integrity, freshness and accuracy scores.

4. The cargo transportation integrated logistics management system according to claim 1, wherein: The semantic reasoning module predicts missing relationships between nodes through the knowledge graph completion algorithm and screens reliable relationships through confidence evaluation.

5. The cargo transportation integrated logistics management system according to claim 1, wherein: The semantic reasoning module uses the ontology mapping method to realize the semantic connection between road, rail and air transportation, and combines the graph attention network to assign semantic weights between nodes to highlight key information.

6. The cargo transportation integrated logistics management system according to claim 1, wherein: The scenario adaptation decision module models logistics nodes through pulse coupled neural network (PCNN), and the activation threshold of the node is dynamically calculated by the maximum processing capacity, average processing time and cache capacity.

7. The cargo transportation integrated logistics management system according to claim 1, wherein: The scenario-adaptive decision-making module optimizes the routing plan through a group intelligence collaboration mechanism, and the decision-making agent adjusts the route based on the road condition index, vehicle availability, storage capacity and proportion of emergency orders in the area.

8. The cargo transportation integrated logistics management system according to claim 1, wherein: The intelligent execution module dynamically decomposes and adjusts trunk line transportation, branch line scheduling and terminal delivery tasks based on the constraints of time windows, path traffic status and node processing capabilities.

9. The cargo transportation integrated logistics management system according to claim 1, wherein: The logistics trust chain module calculates the trust relationship between nodes through the trust propagation model, and dynamically adjusts the data collection frequency, resource allocation method and path planning strategy based on the risk value of abnormal events.

Images