Intelligent cold chain integrated management method and system based on AI big data

Abstract

The present invention relates to the field of AI big data technology, and discloses an intelligent cold chain integrated management method and system based on AI big data, wherein the method comprises: collecting feature data of nodes in a cold chain system, obtaining original data of the nodes and constructing an initial adjacency matrix of a cold chain network; inputting a temperature control parameter identification model to perform graph convolution operation, and obtaining a temperature control parameter identification result; performing task decomposition and resource allocation calculation to obtain a multi-region collaborative task allocation scheme; performing dynamic load calculation and inter-regional task allocation based on the multi-region collaborative task allocation scheme to obtain real-time task execution status data; performing hierarchical processing on the real-time task execution status data and collecting health assessment data of the cold chain system; performing strategy optimization to obtain control instructions of the cold chain system, wherein the method realizes early warning of temperature control anomalies and collaborative control at the equipment level and system level, and forms a complete set of intelligent cold chain integrated management solutions.

Description

Technical Field

[0001] The present invention relates to the field of AI big data technology, and in particular to an intelligent cold chain integrated management method and system based on AI big data. Background Art

[0002] With the rapid development of the cold chain logistics industry, the scale and complexity of the cold chain system are constantly increasing. Traditional cold chain management methods can no longer meet the refined management needs of modern cold chain logistics. In particular, the lack of systematic solutions in multi-regional collaborative operations, real-time status monitoring, and dynamic adjustment of temperature control parameters has led to low utilization of cold chain resources and insufficient accuracy in temperature control parameter identification.

[0003] In active cold chain systems, frequent changes in network topology hinder the accurate identification of temperature control parameters. At the same time, the temperature and humidity parameters are sensitive to natural disaster factors, making real-time monitoring and early warning of the system more difficult. In addition, due to the wide spatial distribution of many temperature control sensor devices, scalability and performance issues are brought about in the process of IoT integration. Existing data collection solutions often lead to network congestion, reduced throughput and increased latency. In edge computing environments, the uneven geographical distribution of tasks may lead to unbalanced loads on cold storage and transportation equipment, and some large-scale cold chain transportation tasks are difficult to complete entirely on a single cold chain node, and independent cold storage cannot fully utilize cold chain network resources. At the same time, the existing health monitoring system lacks an effective anomaly detection and early warning mechanism, and cannot promptly detect and handle temperature control anomalies and equipment failures. Summary of the invention

[0004] The present invention provides an intelligent cold chain integrated management method and system based on AI big data. The present invention realizes early warning of temperature control abnormalities and coordinated control at the equipment level and system level, forming a complete intelligent cold chain integrated management solution.

[0005] In a first aspect, the present invention provides a smart cold chain integrated management method based on AI big data, and the smart cold chain integrated management method based on AI big data includes:

[0006] Collect characteristic data of nodes in the cold chain system, obtain original node data and construct the initial adjacency matrix of the cold chain network;

[0007] Inputting the original node data and the initial adjacency matrix of the cold chain network into the temperature control parameter identification model for graph convolution operation to obtain a temperature control parameter identification result;

[0008] Performing task decomposition and resource allocation calculation on the cold chain transportation task data and the temperature control parameter identification results to obtain a multi-region collaborative task allocation plan;

[0009] Based on the multi-region collaborative task allocation scheme, dynamic load calculation and inter-region task allocation are performed to obtain real-time task execution status data;

[0010] Performing hierarchical processing on the real-time task execution status data and collecting cold chain system health assessment data;

[0011] Through the digital twin model, strategy optimization is performed based on the cold chain system health assessment data to obtain cold chain system control instructions.

[0012] In a second aspect, the present invention provides an intelligent cold chain integrated management system based on AI big data, and the intelligent cold chain integrated management system based on AI big data includes:

[0013] The acquisition module is used to collect characteristic data of nodes in the cold chain system, obtain the original data of the nodes and construct the initial adjacency matrix of the cold chain network;

[0014] An identification module, used for inputting the original data of the node and the initial adjacency matrix of the cold chain network into a temperature control parameter identification model to perform a graph convolution operation to obtain a temperature control parameter identification result;

[0015] A calculation module, used to perform task decomposition and resource allocation calculation on the cold chain transportation task data and the temperature control parameter identification results to obtain a multi-region collaborative task allocation plan;

[0016] A dispatching module, used for performing dynamic load calculation and inter-region task dispatching based on the multi-region collaborative task dispatching scheme, and obtaining real-time task execution status data;

[0017] A processing module, used for performing hierarchical processing on the real-time task execution status data and collecting cold chain system health assessment data;

[0018] The optimization module is used to perform strategy optimization based on the cold chain system health assessment data through a digital twin model to obtain cold chain system control instructions.

[0019] The third aspect of the present invention provides an intelligent cold chain integrated management device based on AI big data, comprising: a memory and at least one processor, the memory storing instructions; the at least one processor calls the instructions in the memory to enable the intelligent cold chain integrated management device based on AI big data to execute the above-mentioned intelligent cold chain integrated management method based on AI big data.

[0020] The fourth aspect of the present invention provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the above-mentioned intelligent cold chain integrated management method based on AI big data.

[0021] In the technical solution provided by the present invention, real-time update of the cold chain network topology and accurate identification of temperature control parameters are realized through a dynamic graph fusion network. The teaching optimization algorithm is used to solve the problems of multi-region task allocation and resource load balancing, thereby improving the overall utilization efficiency of cold chain resources. Through a dual-mode data acquisition strategy and a multi-level sampling mechanism, network congestion is effectively reduced, and the system throughput is improved compared to the strategy based on historical records. Combined with the cold chain health monitoring system and the digital twin model, a complete multi-dimensional health assessment and emergency response mechanism is established, which realizes early warning of temperature control anomalies and coordinated control at the equipment and system levels, forming a complete set of intelligent cold chain integrated management solutions.

[0022] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a schematic diagram of an embodiment of an intelligent cold chain integrated management method based on AI big data in an embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram of an embodiment of an intelligent cold chain integrated management system based on AI big data in an embodiment of the present invention;

[0026] Figure 3 This is a schematic diagram of an embodiment of an intelligent cold chain integrated management device based on AI big data in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0029] To facilitate understanding of this embodiment, firstly, a smart cold chain integrated management method based on AI big data disclosed in an embodiment of the present invention is introduced in detail. Figure 1 As shown, the method comprises the following steps:

[0030] 101. Collect characteristic data of nodes in the cold chain system, obtain original data of nodes and construct an initial adjacency matrix of the cold chain network;

[0031] It is understandable that the execution subject of the present invention can be an intelligent cold chain integrated management system based on AI big data, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking the server as the execution subject as an example.

[0032] Specifically, the location coordinate information of all nodes in the cold chain system is collected. The nodes of the cold chain system include cold storage, transport vehicles and distribution sites, which form the basis of the entire cold chain network. The node locations are accurately collected and classified according to their characteristics. The location coordinate information is divided into cold storage node data, transport vehicle node data and distribution site node data to obtain node location classification data. The temperature parameters, humidity parameters and equipment operation status parameters of each node in the cold chain system are collected in real time. During the operation of each node in the cold chain system, the control of its temperature and humidity is crucial to the product quality of the entire cold chain, so these parameters need to be monitored and recorded uninterruptedly. At the same time, the operation status parameters of the equipment are collected. The operation status parameters are directly related to whether the equipment is in normal working condition and whether there are potential faults. The temperature, humidity and equipment operation status data are associated with the node location classification data to obtain the characteristic data describing the status of each node. The node distance between the node location classification data is calculated. By calculating the geographical distance between the nodes and based on a preset logistics connection distance threshold, it is determined which nodes have a direct logistics connection relationship. The determination of the logistics connection relationship provides the basis for the construction of the cold chain network: if the distance between two nodes is less than the preset threshold, then a connection relationship is established between them; otherwise, no connection is established. The logistics connection relationship data is transformed into a matrix. When there is a direct logistics connection between two nodes, the corresponding position is marked as 1 in the matrix, and when there is no direct connection, the corresponding position is marked as 0, thereby obtaining the initial connection matrix of the cold chain system. Deep feature extraction is performed on the node state feature data. The temperature change feature vector, humidity change feature vector and equipment state feature vector are extracted respectively to describe the dynamic state and change trend of the node, and a node feature vector group is obtained. The feature vector is fused with the node location classification data to obtain the feature representation data of the node. Based on the node feature representation data, the feature similarity between the nodes is calculated. The feature similarity reflects the similarity between the state features of the two nodes. By setting a preset similarity threshold, the connection relationship between the nodes is updated. If the feature similarity of the two nodes is higher than the threshold, it is considered that there is some potential connection between them, so a connection is added to the connection relationship. Through this step, the dynamic connection matrix of the cold chain system is obtained. The dynamic connection matrix is ​​combined with the initial connection matrix to obtain the initial adjacency matrix of the cold chain network. The adjacency matrix combines the information of geographical location and state characteristics to describe the connection relationship between each node in the cold chain system.

[0033] 102. Inputting the original node data and the initial adjacency matrix of the cold chain network into the temperature control parameter identification model to perform graph convolution operation to obtain the temperature control parameter identification result;

[0034] Specifically, the original data of the node is separated according to temperature data, humidity data and equipment status data, so as to facilitate the subsequent targeted feature extraction. The separated temperature, humidity and equipment status data are respectively input into three parallel feature extraction networks for processing. Each feature extraction network consists of three convolutional layers, which can extract the deep features of the data layer by layer. The first convolutional layer of the feature extraction network uses 64 3×3 convolution kernels and adopts the LeakyReLU activation function, with the purpose of extracting more basic features in the earlier convolution process; the number of convolution kernels in the second convolutional layer is increased to 128, the size of the convolution kernel remains unchanged, and the LeakyReLU activation function is used to increase the capacity of the network to extract deeper features; the third convolutional layer uses 256 3×3 convolution kernels and the LeakyReLU activation function to fully extract the features of the input data. Each convolutional layer is followed by a batch normalization layer to standardize the feature map, reduce the problem of feature distribution changes during training, make the network easier to train, and improve the generalization ability of the model. After processing by three layers of convolutional layers, three sets of feature maps are obtained, including deep features of temperature, humidity and equipment status. The three sets of feature maps are fused to express the information of each data source in a unified manner. Feature mapping is performed through three layers of fully connected layers to achieve data fusion and transformation in the feature space. The output dimension of the first fully connected layer is 512, and the features are expressed in high dimensions to retain as much information as possible; the output dimension of the second fully connected layer is reduced to 256, and redundant information is removed by reducing the dimension of the features, while highlighting the important features in the data; the output dimension of the third fully connected layer is reduced to 128, forming a concise and effective node feature vector. After each layer of fully connected layers, the ReLU activation function is used to introduce nonlinear factors so that the model can better fit complex feature relationships. The node feature vector and the initial adjacency matrix of the cold chain network are input into the graph convolution layer of the temperature control parameter recognition model. The graph convolution layer includes four graph convolution units connected in series, each of which propagates and updates the features of the nodes to better capture the relationships and dependencies between nodes. The input of the graph convolution unit includes the central node features and the adjacent node features, and the features are propagated through two layers of graph convolution networks. In the first layer of graph convolution, the output feature dimension is 128. The main function of this layer is to map the node features from the original space to a higher-dimensional feature space so as to better capture the complex interaction relationship between adjacent nodes; the second layer of graph convolution network reduces the feature dimension to 64. This process highlights key information through feature compression while reducing the complexity of calculation. Each layer of graph convolution network uses the ReLU activation function to introduce nonlinearity and make the network more expressive. At the same time, the GraphNorm normalization method is used to standardize the node features to reduce the distribution differences between features and improve the stability and training effect of the model.Through the feature propagation of these two layers of graph convolutional networks, the updated features of the nodes are obtained, and the updated features of the nodes can better reflect the characteristics of the nodes themselves and their neighboring nodes. The updated features of the nodes are globally pooled. The feature information of all nodes is fused by weighted average of attention. The introduction of the attention mechanism can assign different weights according to the importance of each node, so as to highlight the features of key nodes in the global feature representation, obtain the feature representation at the graph level, and reflect the comprehensive state of the entire cold chain network. The graph-level feature representation is input into the dual-branch prediction network for the prediction of temperature and humidity respectively. The temperature prediction branch and the humidity prediction branch each contain two fully connected layers, the output dimension of the first layer is 32, and the output dimension of the second layer is 1. Finally, the Sigmoid activation function is used for nonlinear mapping to ensure that the output prediction value is within a reasonable range. Through this structure, the predicted values ​​of temperature parameters and humidity parameters are obtained respectively. The predicted values ​​of temperature parameters and humidity parameters are combined to obtain the temperature control parameter identification results of the entire system.

[0035] 103. Perform task decomposition and resource allocation calculation on the cold chain transportation task data and temperature control parameter identification results to obtain a multi-region collaborative task allocation plan;

[0036] Specifically, the cold chain transportation task data is statistically divided into regions. According to the geographical distribution of tasks and the distribution of cold storage nodes, the density matrix of tasks is calculated to reflect the task load in each region, reflect the task concentration of each region, and obtain the regional task load data. The temperature control parameter identification results are matched and analyzed with the regional task load data. A connection is established between the geographical distribution of tasks and the temperature control parameter identification results to ensure the rational use of resources. In the matching analysis process, based on the capacity of cold storage and the capacity of transportation equipment, the resource utilization rate of each region is calculated to obtain the regional resource load data. Through these data, the current resource usage of each region is clarified, and regions with high or low resource utilization are identified. Threshold analysis is performed on the regional resource load data. For regions with high task density, their load usually exceeds the preset threshold, and the tasks in this region are decomposed. By decomposing high-density tasks into multiple sub-task units, a fine-grained task decomposition sequence is obtained to improve the flexibility and efficiency of task allocation. A teaching optimization model is constructed based on the fine-grained task decomposition sequence. The global optimal solution is used as the teacher, and the remaining solutions are used as learners to construct the initial task allocation strategy. The teaching optimization model makes full use of the guiding role of the optimal solution, so that the learner solution can gradually approach the optimal solution, accelerate the optimization process of task allocation, and ensure the effectiveness of the initial allocation strategy. The initial task allocation strategy is horizontally coordinated and optimized. There is often a complementary relationship between resources between regions in the cold chain system. Some regions have surplus cold storage capacity or transportation equipment, while other regions face resource shortages. Based on the complementary relationship between adjacent regions, a collaborative scheduling matrix is ​​established to optimize resource utilization between regions. Through this step, a regional collaborative scheme is obtained, which can ensure that resources between regions are mutually scheduled and supported, and realize resource optimization of the overall cold chain system. In order to improve the efficiency and rationality of task allocation, the regional collaborative scheme is input into the multi-objective optimizer to construct the optimization objective function. The optimization objectives include minimizing the delivery time and balancing the regional load, ensuring the timeliness of the cold chain logistics process, and avoiding the situation where some regions are overloaded. Through the calculation of the multi-objective optimizer, the optimization result of task scheduling is obtained, and the reasonable allocation and efficient execution of tasks are realized globally. The user-cold storage matching degree is calculated for the task scheduling optimization result. Based on the location relationship of nodes and the rating of service capabilities, a matching matrix is ​​established to measure the matching between different users and different cold storages, and the optimal matching solution is obtained. The optimal matching solution is integrated with the regional coordination solution to generate a multi-regional collaborative task allocation solution. This solution combines the decomposition of tasks, the allocation of resources, the coordination between regions, and the matching of user cold storages. It can optimally allocate and schedule the tasks of the cold chain system on a global scale, and realize efficient management of the cold chain system.

[0037] 104. Perform dynamic load calculation and inter-regional task allocation based on the multi-region collaborative task allocation scheme to obtain real-time task execution status data;

[0038] Specifically, the multi-region collaborative task allocation scheme is input into the distributed scheduling node in the edge computing environment, and the advantages of edge computing are used to improve the real-time and efficiency of data processing. After the scheme is input, each scheduling node is assigned cold storage and transportation equipment in its management area to generate regional management allocation data. Resource statistical analysis is performed on the regional management allocation data. The storage capacity of the cold storage, the vehicle transportation capacity, and the status of the temperature control equipment are calculated to generate regional resource capacity data to reflect the resource status of each region, including the utilization efficiency and remaining capacity of the resources. Based on the establishment of regional resource capacity data, an inter-regional communication mechanism is formed, and resource status information and task requirement information are transmitted between scheduling nodes through a distributed communication protocol to generate resource sharing data. The resource sharing method can improve the efficiency of inter-regional collaboration and provide timely feedback on the resource usage of each region. After obtaining the resource sharing data, load analysis is performed. By monitoring the resource usage of each region, when it is detected that the resource load of a certain region exceeds the preset threshold, the task is automatically assigned to the adjacent region with lower load, and a task redistribution scheme is obtained. Through the dynamic adjustment mechanism, service interruption caused by resource overload in a specific region can be effectively avoided, and resources can be reasonably allocated to achieve balanced operation of the overall cold chain system. The task redistribution plan is input into the task execution monitoring module to track the task execution process in real time. The monitoring module focuses on the real-time acquisition and analysis of storage temperature, transportation temperature and location information to ensure that all temperature control parameters meet the standards during the execution process and maintain the quality and safety of cold chain products. By continuously monitoring and recording the execution status of the task, potential problems can be discovered in a timely manner to provide data support for subsequent exception handling. On the basis of execution monitoring, anomaly detection is performed on the task monitoring data. By setting pre-defined temperature control parameter thresholds and time limits, the execution status of the task is evaluated and task anomaly marking data is generated in a timely manner. The task anomaly marking data is input into the task redistribution mechanism, and the task is re-optimized based on the current system status to form a task adjustment plan. In terms of execution status recording, the key parameter changes during the task execution process are collected through multi-dimensional indicators, which can track and evaluate the execution effect of the task in real time, and finally obtain comprehensive real-time task execution status data.

[0039] 105. Perform hierarchical processing on real-time task execution status data and collect health assessment data of cold chain system;

[0040] Specifically, the IoT sensor device data is collected for the real-time task execution status data, the temperature and humidity data are collected by the first sensor unit, the operation status and location information of the equipment are collected by the second sensor unit, and the environmental parameter data are collected by the third sensor unit. Through this step, a comprehensive node monitoring data set is formed. Two-level abnormality analysis is performed on the nodes in the node monitoring data set. In the first-level analysis, the abnormality level coefficient is obtained by calculating the frequency of temperature control abnormalities and the duration of abnormalities. The high or low level of this coefficient can directly reflect the reliability and stability of the node in temperature control. In the second-level analysis, the weight coefficient of the task is calculated by evaluating the timeliness of the task and the value of the goods to reflect the importance of the task. Through these two levels of analysis, the node priority classification data is obtained. Based on the results of the node priority classification data, the three-level sampling frequency is set. The sampling interval of 120 seconds is set for the first-level node, the sampling interval of 300 seconds is set for the second-level node, and the sampling interval of 600 seconds is set for the third-level node. Through the multi-level sampling strategy, the conventional monitoring status data can be effectively obtained to ensure that the monitoring frequency of nodes with different priorities matches their importance, so as to detect potential abnormal situations in time. The conventional monitoring status data is monitored with three thresholds. The monitoring process involves the judgment of temperature deviation, humidity deviation and equipment status deviation. The three key indicators are monitored by setting the first, second and third thresholds respectively. When any indicator exceeds the set threshold, the system will immediately switch to the alarm state and generate alarm status data. The generated alarm status data is input into the two-dimensional historical analysis module to analyze the development trend of the abnormal situation. In the first dimension, the change trend of the abnormality is calculated based on the time series analysis method, while in the second dimension, the risk of abnormal diffusion is evaluated through spatial correlation analysis. The two-dimensional analysis method can fully understand the evolution process of the abnormal situation, provide guidance for subsequent response measures, and obtain alarm processing data. The alarm processing data and the conventional monitoring status data are resolved in three layers. The first layer focuses on time conflicts to ensure the timeliness of monitoring and alarm information; the second layer handles spatial conflicts to ensure accurate data transmission between nodes; the third layer is the processing of priority conflicts. Through hierarchical processing strategies, it is ensured that resources can be reasonably allocated when multiple situations occur at the same time to obtain collaborative collection data. The collaboratively collected data is input into the four-dimensional health assessment model, and the health status of the cold chain system is understood through a comprehensive assessment of multiple dimensions. In the first dimension, the stability of temperature control is evaluated; the second dimension evaluates the reliability of equipment; the third dimension evaluates the quality of network communication; and the fourth dimension evaluates the efficiency of task execution. Through this systematic evaluation method, detailed health assessment indicators are obtained. A two-level trend analysis is performed on the health assessment indicators. In the first level, a 15-minute sliding window is used to calculate the short-term health trend, and in the second level, a 4-hour sliding window is used to analyze the long-term health trend.Hierarchical trend analysis can monitor the system's operating status in real time and make predictions about future operating conditions, thus obtaining complete cold chain system health assessment data.

[0041] 106. Through the digital twin model, strategy optimization is carried out based on the health assessment data of the cold chain system to obtain the control instructions of the cold chain system.

[0042] Specifically, the health assessment data of the cold chain system is divided into three layers of states. In the first layer, the temperature control state is discretized to generate a temperature control state space to describe the performance of the system under different temperature control conditions; in the second layer, the device state is similarly discretized to obtain the device state space, which reflects the state changes of the device under various operating conditions; and in the third layer, the system throughput is discretized to generate the system state space, which reflects the overall performance of the system. Through these three layers of state division, a multidimensional state vector is formed. Based on the multidimensional state vector, the first Markov chain model is constructed. In this model, the transition probability between different states is calculated by the maximum likelihood estimation method to obtain the first state transfer matrix. The state transfer matrix is ​​expanded by the second order, and the second Markov chain model is constructed by calculating the conditional probability of the state sequence to generate the second state transfer matrix. This process can capture the dynamic characteristics of state changes and provide a basis for future state prediction. The second state transfer matrix is ​​input into the digital twin modeling unit, and the digital twin of the temperature control system is constructed through the first virtual mapping module, while the digital twin of the device operation is constructed through the second virtual mapping module to form a complete digital twin model of the system. The digital twin model can reflect the actual operation status of the cold chain system in real time. The system digital twin model is simulated for three scenarios. These three scenarios simulate the normal operation status, equipment failure status and network congestion status respectively. By simulating these three scenarios, scenario simulation data is obtained to reflect the performance of the system under different conditions. Through the simulation data, potential system bottlenecks and risks can be identified, providing direction for subsequent optimization. Multi-objective optimization calculation is performed based on scenario simulation data. The throughput of the system is optimized through the first sub-goal to ensure that the system can still operate efficiently under high load; the transmission delay is optimized through the second sub-goal to reduce the overall time for task completion and improve the response speed; the energy consumption is optimized through the third sub-goal to ensure that the operation of the cold chain system is within the economically feasible range. By comprehensively considering these three sub-goals, an optimization strategy set is obtained. After the optimization strategy set is input into the emergency response processing module, it is graded based on the temperature control abnormality level and the equipment alarm level to determine the intelligent control scheme required under different conditions to ensure that the system can respond quickly under abnormal conditions. The generated intelligent control scheme is decomposed into execution instructions, and the optimization strategy is converted into device-level control instructions and system-level scheduling instructions through the control instruction generator, and the cold chain system control instructions are output.

[0043] In the embodiment of the present invention, real-time update of the cold chain network topology and accurate identification of temperature control parameters are achieved through a dynamic graph fusion network, and a teaching optimization algorithm is used to solve the problems of multi-region task allocation and resource load balancing, thereby improving the overall utilization efficiency of cold chain resources; through a dual-mode data acquisition strategy and a multi-level sampling mechanism, network congestion is effectively reduced, and the system throughput is improved compared to the strategy based on historical records; combined with the cold chain health monitoring system and the digital twin model, a complete multi-dimensional health assessment and emergency response mechanism is established, which realizes early warning of temperature control anomalies and coordinated control at the equipment level and system level, forming a complete set of intelligent cold chain integrated management solutions.

[0044] In a specific embodiment, the process of executing step 101 may specifically include the following steps:

[0045] (1) Collect the location coordinate information of all nodes in the cold chain system, and classify the location coordinate information into cold storage node data, transportation vehicle node data, and distribution site node data to obtain node location classification data;

[0046] (2) Collect the temperature parameters, humidity parameters and equipment operation status parameters of all nodes in the cold chain system in real time, and associate the collected data with the node location classification data to obtain node status characteristic data;

[0047] (3) Calculate the distance between nodes for the node position classification data, determine the connection relationship between nodes based on the preset logistics connection distance threshold, obtain logistics connection relationship data, and perform matrix conversion on the logistics connection relationship data. When there is a direct logistics connection between nodes, the corresponding position is marked as 1, and when there is no direct connection, the corresponding position is marked as 0, and the initial connection matrix is ​​obtained;

[0048] (4) Perform deep feature extraction on the node status feature data, extract the temperature change feature vector, humidity change feature vector and device status feature vector respectively, and obtain a node feature vector group;

[0049] (5) The node feature vector group is fused with the node position classification data to obtain node feature representation data, and based on the node feature representation data, the feature similarity between nodes is calculated. The connection relationship between nodes is updated according to the preset similarity threshold to obtain a dynamic connection matrix.

[0050] (6) Combine the dynamic connection matrix and the initial connection matrix to obtain the initial adjacency matrix of the cold chain network.

[0051] Specifically, the specific location coordinate information of each node is obtained by using the Global Positioning System (GPS) and other positioning technologies. These nodes include cold storage, transport vehicles, and distribution sites. In this process, the collected location coordinate information is classified according to the node type to form cold storage node data, transport vehicle node data, and distribution site node data. The temperature parameters, humidity parameters, and equipment operating status parameters of all nodes in the cold chain system are collected in real time. These data are monitored in real time by sensors and IoT devices, and are associated with previous node location classification data to generate node status feature data. The feature data contains environmental monitoring information and reflects the equipment operating status of each node. The node distance between the node location classification data is calculated. Based on the node location classification data, the Euclidean distance formula is used to calculate the distance between each pair of nodes. The Euclidean distance formula is as follows:

[0052] ;

[0053] in, Representation Node and nodes The distance between and Respectively represent nodes The horizontal and vertical coordinates of and The node Based on the preset logistics connection distance threshold (for example, set to 100 kilometers), the connection relationship between nodes is determined by comparing the distances between them. If it is less than or equal to the threshold, there is a direct logistics connection between the two nodes, which is marked as 1; otherwise, it is marked as 0. The obtained logistics connection relationship data is converted into an initial connection matrix to form the following matrix form:

[0054] ;

[0055] Among them, the matrix The elements in represent the connection relationship between nodes, 1 indicates that there is a connection, and 0 indicates that there is no connection. Perform deep feature extraction on the node status feature data. By using a deep learning model, the temperature change feature vector, humidity change feature vector, and device status feature vector are extracted respectively. The extraction process of these feature vectors is completed through models such as convolutional neural networks to generate a node feature vector group, which brings together the comprehensive feature information of each node in environmental parameters and device status. The node feature vector group is fused and calculated with the node position classification data. Using the weighted sum model, the specific formula is as follows:

[0056] ;

[0057] in, Is a node Characterization data of Is a node The temperature eigenvector of is the humidity eigenvector, is the device state feature vector, and , , are the weights of these feature vectors respectively. After fusion calculation, comprehensive node feature representation data is generated. By calculating the feature similarity between these nodes, the connection relationship between the nodes is updated. According to the preset similarity threshold, the connection relationship between the nodes is dynamically updated to obtain a dynamic connection matrix. For example, if the feature similarity between node 1 and node 2 is higher than the set threshold, their connection status is updated to 1 in the dynamic connection matrix. The dynamic connection matrix and the initial connection matrix are combined to form the initial adjacency matrix of the cold chain network. For example, in a cold chain logistics system, it is assumed that there are three nodes: cold storage (A), transport vehicle (B) and distribution site (C). The location coordinates of the cold storage are (10,20), the location coordinates of the transport vehicle are (15,25), and the location coordinates of the distribution site are (30,35). By calculating the distance between the nodes, the distance between A and B is 7.07, the distance between B and C is 21.21, and the distance between A and C is 22.36. If the connection threshold is set to 15, a connection is established between A and B, marked as 1, while no connection is established between A and C, or between B and C, marked as 0. A clear cold chain network structure is formed.

[0058] In a specific embodiment, the process of executing step 102 may specifically include the following steps:

[0059] (1) The node raw data is separated into temperature data, humidity data and equipment status data, and the separated data are input into three parallel feature extraction networks respectively; each feature extraction network includes three convolutional layers. The first convolutional layer uses 64 3×3 convolutional kernels and LeakyReLU activation function, the second convolutional layer uses 128 3×3 convolutional kernels and LeakyReLU activation function, and the third convolutional layer uses 256 3×3 convolutional kernels and LeakyReLU activation function. Each layer is followed by a batch normalization layer to obtain three sets of feature maps;

[0060] (2) Feature fusion is performed on the three sets of feature maps, and feature mapping is performed through three fully connected layers. The output dimension of the first fully connected layer is 512, the output dimension of the second fully connected layer is 256, and the output dimension of the third fully connected layer is 128. The ReLU activation function is used in each layer to obtain the node feature vector;

[0061] (3) The node feature vector and the initial adjacency matrix of the cold chain network are input into the graph convolution layer of the temperature control parameter identification model. The graph convolution layer includes four graph convolution units connected in series. The input of the graph convolution unit includes the central node feature and the adjacent node feature. The feature propagation is performed through a two-layer graph convolution network. The first layer outputs 128-dimensional features and the second layer outputs 64-dimensional features. Each layer uses the ReLU activation function and GraphNorm normalization to obtain the node update feature.

[0062] (4) Perform a global pooling operation on the node update features, fuse the feature information of all nodes by weighted average of attention, and obtain graph-level feature representation;

[0063] (5) The graph-level feature representation is input into the dual-branch prediction network. The temperature prediction branch and the humidity prediction branch each contain two fully connected layers. The output dimension of the first layer is 32, and the output dimension of the second layer is 1. The Sigmoid activation function is used to obtain the temperature parameter prediction value and the humidity parameter prediction value. The temperature parameter prediction value and the humidity parameter prediction value are combined to obtain the temperature control parameter recognition result.

[0064] Specifically, the node raw data is separated according to temperature data, humidity data and device status data. By writing a data processing script, the raw data set is screened and classified to extract the corresponding feature information. The temperature data will be extracted into a dedicated array, and the humidity data and device status data are processed in the same way. The separated data are input into three parallel feature extraction networks respectively. The three feature extraction networks run independently and extract features for specific input data. Each feature extraction network includes three convolutional layers. In the first convolutional layer, 64 3×3 convolution kernels are used, and the LeakyReLU activation function is applied to increase the nonlinear ability of the network. The output of the convolutional layer will be standardized through the batch normalization layer to improve the convergence speed and stability of the model. In the second convolutional layer, the number of convolution kernels is increased to 128, and the size of 3×3 and the LeakyReLU activation function are still used to extract more complex features. The third convolutional layer uses 256 3×3 convolution kernels, and the LeakyReLU activation function and batch normalization layer are also applied. Through this series of convolution operations, three sets of feature maps are obtained, each of which represents the deep feature representation of temperature, humidity and device status. The three sets of feature maps are fused. The feature fusion process is completed through three fully connected layers. The output dimension of the first fully connected layer is set to 512, the second layer is 256, and the third layer is 128. Each layer uses the ReLU activation function. Feature fusion is expressed by the following formula:

[0065] ;

[0066] in, represents the output features of the fully connected layer, is the weight matrix, is the input feature vector, is the bias term. Through calculation, the model can effectively combine different features to obtain a node feature vector, which integrates the characteristic information of temperature, humidity and equipment status. The node feature vector and the initial adjacency matrix of the cold chain network are input into the graph convolution layer of the temperature control parameter recognition model. The graph convolution layer consists of four graph convolution units connected in series, where the input of each graph convolution unit includes the central node feature and the adjacent node feature. Through this design, graph convolution can effectively capture the relationship and structural information between nodes. The feature propagation of graph convolution is expressed as:

[0067] ;

[0068] in, It is The feature matrix of the layer, is the adjacency matrix of the graph, is the diagonal matrix of node degrees, It is The weight matrix of the layer, is the activation function. The output of the first layer will be 128-dimensional features, while the second layer will output 64-dimensional features. Each layer applies the ReLU activation function and GraphNorm normalization to ensure the stability of feature propagation. After the feature propagation is completed, the node update features will be globally pooled, and the feature information of all nodes will be fused by attention weighted averaging to obtain the graph-level feature representation. The attention mechanism can effectively emphasize important features, so that the model pays more attention to key nodes when making decisions. The graph-level features are expressed by the following formula:

[0069] ;

[0070] in, is a graph-level feature representation, It is The feature vector of each node is is the attention weight, which indicates the importance of the node in the entire graph. The graph-level feature representation is input into the dual-branch prediction network for temperature and humidity prediction respectively. The temperature prediction branch and the humidity prediction branch each contain two fully connected layers, the output dimension of the first layer is 32, and the output dimension of the second layer is 1, both using the Sigmoid activation function. The temperature parameter prediction value and the humidity parameter prediction value are obtained by combining the following formula:

[0071] ;

[0072] in, is the final temperature control parameter identification result, is the predicted temperature value, is the predicted humidity value.

[0073] In a specific embodiment, the process of executing step 103 may specifically include the following steps:

[0074] (1) Perform regional statistics on cold chain transportation task data, calculate the task density matrix based on the task geographical location distribution and cold storage node distribution, and obtain regional task load data;

[0075] (2) Match and analyze the temperature control parameter identification results with the regional task load data, calculate the resource utilization rate of each region based on the cold storage capacity and transportation equipment capacity, and obtain the regional resource load data;

[0076] (3) Perform threshold analysis on regional resource load data, decompose regional tasks with high task density into multiple subtask units, and obtain a fine-grained task decomposition sequence;

[0077] (4) Construct a teaching optimization model based on the fine-grained task decomposition sequence, taking the global optimal solution as the teacher and the remaining solutions as learners to obtain the initial task allocation strategy;

[0078] (5) Perform horizontal collaborative optimization on the initial task allocation strategy, establish a collaborative scheduling matrix based on the resource complementarity relationship between adjacent regions, obtain a regional collaborative plan, and input the regional collaborative plan into the multi-objective optimizer. By minimizing the delivery time and balancing the regional load, the optimization objective function is constructed to obtain the task scheduling optimization result;

[0079] (6) The user-cold storage matching degree is calculated for the task scheduling optimization results, and a matching degree matrix is ​​established based on the node location relationship and service capability score to obtain the optimal matching solution. The optimal matching solution is then integrated with the regional coordination solution to generate a multi-regional collaborative task allocation solution.

[0080] Specifically, the cold chain transportation task data is statistically divided into regions. According to the geographical distribution of the tasks and the distribution of cold storage nodes, the task density matrix is ​​calculated to obtain the regional task load data. The task density matrix is ​​expressed by the following formula:

[0081] ;

[0082] in, Indicates area Neidi The density of tasks, is in the area The first The number of tasks, It is a region In this way, it is possible to effectively identify which areas have a high task density and form regional task load data. The temperature control parameter identification results are matched and analyzed with the regional task load data to better understand the resource utilization of each area. On this basis, combined with the capacity of the cold storage and the capacity of the transportation equipment, the resource utilization rate of each area is calculated to obtain the regional resource load data. The utilization rate is expressed by the following formula:

[0083] ;

[0084] in, It is a region Resource utilization, Indicates the actual resource usage of the region. is the total resource capacity of the area. This analysis helps to identify areas with tight resources. Threshold analysis is performed on the regional resource load data, and fine-grained task decomposition is performed on areas with high task density to improve the manageability of tasks and effectively improve the flexibility of cold chain transportation. Fine-grained task decomposition identifies high-density areas, sorts the regional resource load data, and determines areas where the task density is higher than a certain threshold. The tasks in the identified areas are split to generate multiple subtask units. These subtask units are processed in different time periods or using different resources to obtain a fine-grained task decomposition sequence. Based on the fine-grained task decomposition sequence, a teaching optimization model is constructed. In this model, the global optimal solution is regarded as a "teacher" and other solutions are regarded as "learners". In this way, the task allocation strategy is effectively learned to ensure the optimization of the overall efficiency of the task, which is expressed by the following formula:

[0085] ;

[0086] in, Indicates The cost of a task, is the time required to complete the task. By optimizing the objective function, the initial task allocation strategy is obtained. The initial task allocation strategy is horizontally coordinated and optimized, and a coordinated scheduling matrix is ​​established based on the resource complementarity relationship between adjacent regions to achieve efficient utilization and sharing of resources, and finally form a regional coordination plan. The optimization goal of coordinated scheduling is expressed by the following formula:

[0087] ;

[0088] in, Indicates area Delivery time, It is a region The load imbalance degree. This optimization will be processed by the multi-objective optimizer to obtain the task scheduling optimization result. After completing the task scheduling optimization, the matching degree between the user and the cold storage is calculated. Based on the node position relationship and service capability score, a matching degree matrix is ​​established, which can reflect the service matching between each user and the cold storage. The matching degree is expressed as:

[0089] ;

[0090] in, Is a user With cold storage The matching degree, Is a user service needs, It is a cold storage service capabilities, and is the distance between the user and the cold storage. Through this calculation, the optimal matching solution is obtained. The optimal matching solution is integrated with the regional coordination solution to generate the final multi-regional collaborative task allocation solution. This solution helps to improve the efficiency of cold chain transportation, ensure that the goods can reach the destination in the best condition, and maximize the use of existing resources.

[0091] In a specific embodiment, the process of executing step 104 may specifically include the following steps:

[0092] (1) Input the multi-region collaborative task allocation scheme into the distributed scheduling nodes in the edge computing environment, assign cold storage and transportation equipment within the management area to each scheduling node, and obtain regional management allocation data;

[0093] (2) Perform resource statistical analysis on regional management allocation data, and obtain regional resource capacity data by calculating cold storage capacity, vehicle transportation capacity, and temperature control equipment status;

[0094] (3) Establish an inter-regional communication mechanism based on regional resource capacity data, transmit resource status information and task requirement information between scheduling nodes through a distributed communication protocol, and obtain resource sharing data;

[0095] (4) Perform load analysis on resource sharing data. When it is detected that the resource load of a region exceeds a preset threshold, the task is assigned to an adjacent region with a lower load, and a task reallocation plan is obtained.

[0096] (5) Input the task redistribution plan into the task execution monitoring module, track the storage temperature, transportation temperature and location information during the task execution in real time, and obtain task monitoring data;

[0097] (6) Perform anomaly detection on the task monitoring data, evaluate the task execution status based on the preset temperature control parameter threshold and time limit, and obtain task anomaly marking data;

[0098] (7) Input the task anomaly mark data into the task reallocation mechanism, re-optimize the task allocation based on the current system status, obtain the task adjustment plan, and record the execution status of the task adjustment plan. Through multi-dimensional indicators, the changes in key parameters during the task execution process are collected to obtain real-time task execution status data.

[0099] Specifically, the multi-region collaborative task allocation scheme is input into the distributed scheduling nodes in the edge computing environment. Each scheduling node manages the cold storage and transportation equipment in a specific area and generates regional management allocation data. The scheduling nodes in this process need to clearly define the cold storage and transportation equipment they manage, and integrate the resource information in the area to form an efficient management strategy. Perform resource statistical analysis on the regional management allocation data, calculate the storage capacity of the cold storage, the vehicle transportation capacity, and the status of the temperature control equipment, and obtain the regional resource capacity data. Use the following formula to represent the total resource capacity of the cold storage and transportation equipment:

[0100] ;

[0101] in, Represents the total resource capacity of the region, is the storage capacity of the cold storage, is the capacity of the transport equipment, is the capacity of the temperature control equipment. Based on the regional resource capacity data, an inter-regional communication mechanism is established. The resource status information and task requirement information are transmitted between the scheduling nodes through a distributed communication protocol to obtain resource sharing data. This communication mechanism enables each scheduling node to receive real-time changes in resource status and task requirements in other regions. For example, node A sends the available capacity of its cold storage and the current transportation task to node B through the network to achieve optimal resource allocation. Load analysis is performed on the resource sharing data. If it is detected that the resource load of a certain area exceeds the preset threshold, the task is assigned to the adjacent area with lower load. This is achieved through the following formula:

[0102] ;

[0103] in, It is a region For Region The load ratio, Representative area of resource usage, and It is a region The total resource capacity. When it exceeds 1, it indicates the area The load is too high, so some tasks are reallocated to the area with lower load. , the task redistribution scheme obtained at this time can effectively alleviate the regional The task reallocation plan is generated to input the task execution monitoring module to track the storage temperature, transportation temperature and location information in real time during the task execution and obtain the task monitoring data. Sensor technology is used to monitor the temperature and location data of each node to ensure that all environmental parameters are always within the controllable range during the entire cold chain transportation process. Anomaly detection is performed on the task monitoring data, and the task execution status is evaluated based on the preset temperature control parameter thresholds and time limits to obtain the task abnormality marking data. For example, when the temperature of a node exceeds the preset safety threshold, the node will be marked as abnormal, and the system will automatically start the alarm program. This process is expressed by the following formula:

[0104] ;

[0105] in, Is a node Mission The abnormal mark, is the actual temperature during task execution, is the preset safety temperature threshold. The task abnormality mark data is input into the task reallocation mechanism, and the task is optimally allocated based on the current system status to generate a task adjustment plan. In this process, machine learning and data analysis techniques are used to evaluate the system status in real time and make optimization suggestions. The final task adjustment plan can flexibly respond to various emergencies. When implementing this plan, the key parameter changes during the task execution process, such as temperature fluctuations, position offsets, and resource utilization, are collected through multi-dimensional indicators to obtain real-time task execution status data. For example, suppose that in a certain cold chain transportation, the temperature monitoring data of the cold storage in area A shows that its temperature exceeds the set threshold, resulting in an abnormal mark. At this time, the system will start the reallocation mechanism to transfer the task to the nearby area B. In this process, area B needs to have sufficient resources to receive the transferred tasks, and ensure that its storage and transportation temperatures are within a safe range through real-time monitoring, so as to achieve efficient operation of the entire cold chain system.

[0106] In a specific embodiment, the process of executing step 105 may specifically include the following steps:

[0107] (1) Collecting IoT sensor device data for real-time task execution status data: collecting temperature data and humidity data through the first sensor unit, collecting equipment operation status data and location data through the second sensor unit, and collecting environmental parameter data through the third sensor unit to obtain a node monitoring data set;

[0108] (2) Perform two-level anomaly analysis on the nodes in the node monitoring data set. The first level calculates the anomaly level coefficient based on the temperature control anomaly frequency and anomaly duration. The second level calculates the task weight coefficient based on the task timeliness and cargo value to obtain the node priority classification data.

[0109] (3) A three-level sampling frequency is set based on the node priority classification data. A sampling interval of 120 seconds is set for the first-level nodes, a sampling interval of 300 seconds is set for the second-level nodes, and a sampling interval of 600 seconds is set for the third-level nodes. The conventional monitoring status data is obtained through a multi-level sampling strategy;

[0110] (4) Perform three-threshold monitoring on the conventional monitoring status data, determine the temperature deviation through the first threshold, determine the humidity deviation through the second threshold, and determine the equipment status deviation through the third threshold. When any indicator exceeds the threshold, switch to the alarm state and obtain the alarm state data;

[0111] (5) Input the alarm status data into the two-dimensional historical analysis module. The first dimension calculates the abnormal development trend based on time series analysis, and the second dimension calculates the abnormal diffusion risk based on spatial correlation analysis to obtain the alarm processing data.

[0112] (6) Perform three-layer conflict resolution on alarm processing data and routine monitoring status data. The first layer processes time conflicts, the second layer processes space conflicts, and the third layer processes priority conflicts. Collaboratively collected data is obtained through a hierarchical processing strategy.

[0113] (7) Input the collaboratively collected data into the four-dimensional health assessment model, evaluate the temperature control stability through the first dimension, evaluate the equipment reliability through the second dimension, evaluate the network communication quality through the third dimension, and evaluate the task execution efficiency through the fourth dimension to obtain the health assessment index;

[0114] (8) A two-level trend analysis is performed on the health assessment indicators. The first level calculates the short-term health trend through a 15-minute sliding window, and the second level calculates the long-term health trend through a 4-hour sliding window to obtain the health assessment data of the cold chain system.

[0115] Specifically, data is collected through IoT sensor devices. The first sensor unit is used to collect temperature and humidity data, the second sensor unit collects the operating status and location information of the equipment, and the third sensor unit collects environmental parameters. Through these different sensor units, a comprehensive node monitoring data set is constructed, which contains multi-dimensional data such as temperature, humidity, device status and environmental information. A two-level anomaly analysis is performed on the nodes in the node monitoring data set. The first level of analysis evaluates the node's anomaly level coefficient by calculating the frequency and duration of temperature control anomalies. It is expressed as the following formula:

[0116] ;

[0117] in, Is a node The abnormal level coefficient, Is a node The frequency of temperature control anomalies during the observation period, and is the total monitoring time of the node in the same cycle. This analysis result helps to identify which nodes have greater risks in temperature control. The second level of analysis focuses on the timeliness of the task and the value of the goods, and calculates the weight coefficient of the task, which is expressed by the following formula:

[0118] ;

[0119] in, It's a task The weight coefficient of Representative tasks The timeliness of is the value of the goods, and It is the resource consumption associated with the task. The calculation of the weight coefficient will help the system prioritize important and urgent tasks when resources are limited. Based on the node priority classification data, set the three-level sampling frequency to optimize the efficiency of data collection and the use of resources. For the first-level nodes, set a sampling interval of 120 seconds; for the second-level nodes, set a sampling interval of 300 seconds; and for the third-level nodes, set a sampling interval of 600 seconds. Through the multi-level sampling strategy, obtain the regular monitoring status data to ensure that the data collection frequency of key nodes is high enough to better reflect the operating status of the system. Perform three-threshold monitoring on the regular monitoring status data. The temperature deviation is judged by the first threshold, the humidity deviation is judged by the second threshold, and the equipment status deviation is judged by the third threshold. When any indicator exceeds the preset threshold, the system will automatically switch to the alarm state, and the alarm status data generated at this time will trigger subsequent processing. The implementation of the alarm mechanism is expressed by the following formula:

[0120] ;

[0121] in, Is a node On Task The alarm status, , and are the actual values ​​of temperature, humidity and device status, respectively, and , and is the corresponding threshold. Through this mechanism, we can respond as soon as the problem occurs. The alarm status data is input into the two-dimensional historical analysis module for processing. In this module, the first dimension uses time series analysis to calculate the abnormal development trend, while the second dimension calculates the abnormal diffusion risk based on spatial correlation analysis to obtain the alarm processing data. On this basis, the alarm processing data and the conventional monitoring status data are resolved in three layers, firstly, the time conflict is processed, then the spatial conflict is processed, and finally the priority conflict is processed. The hierarchical processing strategy aims to ensure the consistency and availability of the data and obtain the collaborative collection data. Through this step, the data from different sources are integrated to improve the overall monitoring efficiency. The collaborative collection data is input into the four-dimensional health assessment model. The assessment dimensions include temperature control stability, equipment reliability, network communication quality and task execution efficiency. The health assessment indicators formed by the assessment results will provide guidance for further optimization of the system. On this basis, the health assessment indicators are analyzed in two levels. The first level calculates the short-term health trend through a 15-minute sliding window, and the second level calculates the long-term health trend through a 4-hour sliding window. These analysis results constitute the health assessment data of the cold chain system, which can help managers grasp the system status in time.

[0122] In a specific embodiment, the process of executing step 106 may specifically include the following steps:

[0123] (1) The health assessment data of the cold chain system is divided into three layers. The first layer discretizes the temperature control state to obtain the temperature control state space, the second layer discretizes the equipment state to obtain the equipment state space, and the third layer discretizes the system throughput to obtain the system state space, thus obtaining a multi-dimensional state vector.

[0124] (2) Based on the multidimensional state vector, a first Markov chain model is constructed, and the transition probability between states is calculated by the maximum likelihood estimation method to obtain the first state transfer matrix. The first state transfer matrix is ​​expanded to the second order, and a second Markov chain model is constructed by calculating the conditional probability of the state sequence to obtain the second state transfer matrix.

[0125] (3) Inputting the second state transfer matrix into the digital twin modeling unit, constructing the digital twin of the temperature control system through the first virtual mapping module, and constructing the digital twin of the equipment operation through the second virtual mapping module, thereby obtaining a digital twin model of the system;

[0126] (4) Perform three-scenario simulation calculations on the system digital twin model, simulating the normal operation state through the first scenario, simulating the equipment failure state through the second scenario, and simulating the network congestion state through the third scenario to obtain scenario simulation data;

[0127] (5) Perform multi-objective optimization calculations based on scenario simulation data, optimize system throughput through the first sub-objective, optimize transmission delay through the second sub-objective, and optimize energy consumption through the third sub-objective, and obtain an optimization strategy set;

[0128] (6) The optimization strategy set is input into the emergency response processing module, and the module performs hierarchical processing based on the temperature control abnormality level and the equipment alarm level to obtain an intelligent control solution. The intelligent control solution is then decomposed into execution instructions. The optimization strategy is converted into device-level control instructions and system-level scheduling instructions through the control instruction generator, and the cold chain system control instructions are output.

[0129] Specifically, the health assessment data of the cold chain system is divided into three levels of states to effectively manage different aspects of the system. The first level focuses on the temperature control state, and the temperature control state space is obtained through discretization. This processing uses the interval division method, such as dividing the temperature range into three categories: "normal", "low" and "high", so as to clearly identify each state. The second level focuses on the discretization of equipment status, such as dividing the operating status of the equipment into "normal", "fault" and "pending maintenance" to form an equipment state space. The third level is the discretization of the system throughput, such as defining different levels of throughput, such as "high", "medium" and "low", to obtain a multi-dimensional state vector that can comprehensively reflect the health of the system. The first Markov chain model is constructed based on the multi-dimensional state vector to better understand the transition between states. The maximum likelihood estimation method is used to calculate the transition probability between states, which is expressed by the following formula:

[0130] ;

[0131] in, From the state Transfer to state The probability of is from the state during the observation period To status The number of transfers, Yes Status The total number of visits in the same cycle. Through these probabilities, the first state transfer matrix is ​​obtained, which is used to describe the transition characteristics of the system between different states. Based on the first state transfer matrix, a second-order expansion is performed, and the second Markov chain model is constructed by calculating the conditional probability of the state sequence to obtain the second state transfer matrix. The second state transfer matrix is ​​input into the digital twin modeling unit, which is the basis for virtual simulation and analysis of the system. Through the first virtual mapping module, the digital twin of the temperature control system is constructed to replicate the operating state of the temperature control system in a virtual environment. At the same time, through the second virtual mapping module, the digital twin of the equipment operation is constructed, and the state and operating characteristics of the equipment are presented in digital form to obtain a complete system digital twin model. The digital twin model is applied to the simulation calculation of three scenarios. The first scenario simulates the normal operation state, and verifies the performance of the system in the best state by real-time tracking and recording of various parameters. The second scenario simulates the equipment failure state to help evaluate the system's response capability and recovery strategy in the event of equipment failure. The third scenario simulates the network congestion state to verify the stability and countermeasures of the system when resource competition is fierce. Through the simulation calculation of these scenarios, the corresponding scenario simulation data is obtained. Multi-objective optimization calculation is performed based on the scenario simulation data. The throughput of the optimized system is expressed by the following formula:

[0132] ;

[0133] in, is the optimized system throughput, and Represents all possible throughput data. Secondly, optimize the transmission delay and energy consumption, which are expressed as:

[0134] ;

[0135] in, and are the optimized transmission delay and energy consumption, and For all corresponding data sets. Through multiple optimization objectives, an optimization strategy set is generated to help the system achieve the best operating efficiency. The optimization strategy set is input into the emergency response processing module, and the temperature control abnormality level and equipment alarm level are graded to obtain an intelligent control solution. By decomposing the execution instructions of the intelligent control solution, the control instruction generator is used to convert the optimization strategy into device-level control instructions and system-level scheduling instructions, and finally the cold chain system control instructions are output.

[0136] The above describes the intelligent cold chain integrated management method based on AI big data in the embodiment of the present invention. The following describes the intelligent cold chain integrated management system based on AI big data in the embodiment of the present invention. Figure 2In one embodiment of the present invention, an intelligent cold chain integrated management system based on AI big data includes:

[0137] The collection module 201 is used to collect characteristic data of nodes in the cold chain system, obtain the original data of the nodes and construct the initial adjacency matrix of the cold chain network;

[0138] The identification module 202 is used to input the original node data and the initial adjacency matrix of the cold chain network into the temperature control parameter identification model to perform graph convolution operation to obtain the temperature control parameter identification result;

[0139] The calculation module 203 is used to perform task decomposition and resource allocation calculation on the cold chain transportation task data and the temperature control parameter identification results to obtain a multi-region collaborative task allocation plan;

[0140] The deployment module 204 is used to perform dynamic load calculation and inter-region task deployment based on the multi-region collaborative task allocation scheme to obtain real-time task execution status data;

[0141] Processing module 205, for performing hierarchical processing on real-time task execution status data and collecting cold chain system health assessment data;

[0142] The optimization module 206 is used to perform strategy optimization based on the cold chain system health assessment data through the digital twin model to obtain cold chain system control instructions.

[0143] Through the collaborative cooperation of the above-mentioned components, real-time updating of the cold chain network topology and accurate identification of temperature control parameters are achieved through the dynamic graph fusion network. The teaching optimization algorithm is used to solve the problems of multi-region task allocation and resource load balancing, thereby improving the overall utilization efficiency of cold chain resources. The dual-mode data acquisition strategy and multi-level sampling mechanism are used to effectively reduce network congestion and improve the system throughput performance compared with the historical record-based strategy. Combined with the cold chain health monitoring system and the digital twin model, a complete multi-dimensional health assessment and emergency response mechanism is established, which realizes early warning of temperature control anomalies and coordinated control at the equipment and system levels, forming a complete set of intelligent cold chain integrated management solutions.

[0144] above Figure 2 The intelligent cold chain integrated management system based on AI big data in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The intelligent cold chain integrated management device based on AI big data in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0145] Figure 3: This is a structural schematic diagram of an intelligent cold chain integrated management device based on AI big data provided by an embodiment of the present invention. The intelligent cold chain integrated management device 300 based on AI big data may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the intelligent cold chain integrated management device 300 based on AI big data. Furthermore, the processor 310 can be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the intelligent cold chain integrated management device 300 based on AI big data to implement the steps of the above-mentioned intelligent cold chain integrated management method based on AI big data.

[0146] The intelligent cold chain integrated management device 300 based on AI big data may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that Figure 3 The structure of the intelligent cold chain integrated management equipment based on AI big data shown does not constitute a limitation of the intelligent cold chain integrated management equipment based on AI big data provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0147] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the intelligent cold chain integrated management method based on AI big data.

[0148] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0149] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0150] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart cold chain integrated management method based on AI big data, characterized in that: The method comprises: Collect characteristic data of nodes in the cold chain system, obtain original node data and construct the initial adjacency matrix of the cold chain network; Inputting the original node data and the initial adjacency matrix of the cold chain network into the temperature control parameter identification model for graph convolution operation to obtain a temperature control parameter identification result; The cold chain transportation task data and the temperature control parameter identification results are subjected to task decomposition and resource allocation calculation to obtain a multi-region collaborative task allocation plan; specifically, the plan includes: performing regional statistics on the cold chain transportation task data, calculating the task density matrix according to the geographical location distribution of the tasks and the cold storage node distribution, and obtaining regional task load data; matching and analyzing the temperature control parameter identification results with the regional task load data, calculating the resource utilization rate of each region based on the cold storage capacity and the transportation equipment capacity, and obtaining regional resource load data; performing threshold analysis on the regional resource load data, decomposing the regional tasks with high task density into multiple subtask units, and obtaining a fine-grained task decomposition sequence; constructing a task decomposition sequence based on the fine-grained task decomposition sequence. The teaching optimization model uses the global optimal solution as the teacher and the remaining solutions as learners to obtain the initial task allocation strategy; the initial task allocation strategy is horizontally collaboratively optimized, a collaborative scheduling matrix is ​​established based on the resource complementarity relationship between adjacent regions, a regional collaborative plan is obtained, and the regional collaborative plan is input into the multi-objective optimizer, and the optimization objective function is constructed by minimizing the delivery time and balancing the regional load to obtain the task scheduling optimization result; the user-cold storage matching degree is calculated for the task scheduling optimization result, and a matching degree matrix is ​​established based on the node location relationship and service capability score to obtain the optimal matching plan, and the optimal matching plan is integrated with the regional collaborative plan to generate a multi-region collaborative task allocation plan; Based on the multi-region collaborative task allocation scheme, dynamic load calculation and inter-region task allocation are performed to obtain real-time task execution status data; Performing hierarchical processing on the real-time task execution status data and collecting cold chain system health assessment data; Through the digital twin model, strategy optimization is performed based on the cold chain system health assessment data to obtain cold chain system control instructions.

2. The intelligent cold chain integrated management method based on AI big data according to claim 1 is characterized in that: The method of collecting characteristic data of nodes in the cold chain system, obtaining original node data and constructing an initial adjacency matrix of the cold chain network includes: The location coordinate information of all nodes in the cold chain system is collected, and the location coordinate information is classified into cold storage node data, transport vehicle node data and distribution site node data to obtain node location classification data; The temperature parameters, humidity parameters and equipment operation status parameters of all nodes in the cold chain system are collected in real time, and the collected data are associated with the node location classification data to obtain node status characteristic data; Calculate the distance between nodes for the node position classification data, determine the connection relationship between nodes based on a preset logistics connection distance threshold, obtain logistics connection relationship data, and perform matrix conversion on the logistics connection relationship data. When there is a direct logistics connection between nodes, the corresponding position is marked as 1, and when there is no direct connection, the corresponding position is marked as 0, to obtain an initial connection matrix; Performing deep feature extraction on the node state feature data, respectively extracting a temperature change feature vector, a humidity change feature vector, and a device state feature vector to obtain a node feature vector group; The node feature vector group is fused with the node position classification data to obtain node feature representation data, and based on the node feature representation data, feature similarity between nodes is calculated, and connection relationships between nodes are updated according to a preset similarity threshold to obtain a dynamic connection matrix; The dynamic connection matrix and the initial connection matrix are combined to obtain an initial adjacency matrix of the cold chain network.

3. The intelligent cold chain integrated management method based on AI big data according to claim 2 is characterized in that: The step of inputting the original node data and the initial adjacency matrix of the cold chain network into the temperature control parameter identification model for graph convolution operation to obtain the temperature control parameter identification result includes: The node raw data is separated according to temperature data, humidity data and equipment status data, and the separated data are respectively input into three parallel feature extraction networks; each feature extraction network includes 3 convolution layers, the first convolution layer uses 64 3×3 convolution kernels and LeakyReLU activation function, the second convolution layer uses 128 3×3 convolution kernels and LeakyReLU activation function, and the third convolution layer uses 256 3×3 convolution kernels and LeakyReLU activation function, and each layer is followed by a batch normalization layer to obtain three sets of feature maps; The three sets of feature maps are feature fused, and feature mapping is performed through three fully connected layers, where the output dimension of the first fully connected layer is 512, the output dimension of the second fully connected layer is 256, and the output dimension of the third fully connected layer is 128. Each layer uses a ReLU activation function to obtain a node feature vector; The node feature vector and the initial adjacency matrix of the cold chain network are input into the graph convolution layer of the temperature control parameter recognition model, and the graph convolution layer includes 4 graph convolution units connected in series; the input of the graph convolution unit includes the central node feature and the adjacent node feature, and the feature propagation is performed through a two-layer graph convolution network, the first layer outputs 128-dimensional features, and the second layer outputs 64-dimensional features. Each layer uses the ReLU activation function and GraphNorm normalization to obtain the node update feature; Performing a global pooling operation on the node update features, fusing the feature information of all nodes by means of attention weighted averaging, and obtaining a graph-level feature representation; The graph-level feature representation is input into a dual-branch prediction network. The temperature prediction branch and the humidity prediction branch each contain 2 fully connected layers. The output dimension of the first layer is 32, and the output dimension of the second layer is 1. The Sigmoid activation function is used to obtain the temperature parameter prediction value and the humidity parameter prediction value. The temperature parameter prediction value and the humidity parameter prediction value are combined to obtain the temperature control parameter recognition result.

4. The intelligent cold chain integrated management method based on AI big data according to claim 1 is characterized in that: The dynamic load calculation and inter-region task allocation based on the multi-region collaborative task allocation scheme are performed to obtain real-time task execution status data, including: The multi-region collaborative task allocation scheme is input into the distributed scheduling node in the edge computing environment, and each scheduling node is assigned to manage the cold storage and transportation equipment in the area to obtain the regional management allocation data; Perform resource statistical analysis on the regional management allocation data, and obtain regional resource capacity data by calculating cold storage capacity, vehicle transportation capacity, and temperature control equipment status; An inter-regional communication mechanism is established based on the regional resource capacity data, and resource status information and task requirement information are transmitted between scheduling nodes through a distributed communication protocol to obtain resource sharing data; Performing load analysis on the resource sharing data, and when it is detected that the resource load of a region exceeds a preset threshold, allocating tasks to adjacent regions with lower loads, thereby obtaining a task reallocation plan; The task reallocation plan is input into the task execution monitoring module, and the storage temperature, transportation temperature and location information during the task execution are tracked in real time to obtain task monitoring data; Performing anomaly detection on the task monitoring data, evaluating the task execution status based on preset temperature control parameter thresholds and time limits, and obtaining task anomaly marking data; The task exception marking data is input into the task reallocation mechanism, and the task is re-optimized and allocated based on the current system status to obtain a task adjustment plan. The execution status of the task adjustment plan is recorded, and the changes in key parameters during the task execution process are collected through multi-dimensional indicators to obtain real-time task execution status data.

5. The intelligent cold chain integrated management method based on AI big data according to claim 4 is characterized in that: The step of performing hierarchical processing on the real-time task execution status data and collecting cold chain system health assessment data includes: The IoT sensor device data is collected for the real-time task execution status data, temperature data and humidity data are collected through the first sensor unit, equipment operation status data and location data are collected through the second sensor unit, and environmental parameter data are collected through the third sensor unit to obtain a node monitoring data set; A two-level abnormality analysis is performed on the nodes in the node monitoring data set. The first level calculates the abnormality level coefficient by the temperature control abnormality frequency and the abnormality duration. The second level calculates the task weight coefficient by the task timeliness and the cargo value to obtain the node priority classification data. A three-level sampling frequency is set based on the node priority classification data, a 120-second sampling interval is set for the first-level node, a 300-second sampling interval is set for the second-level node, and a 600-second sampling interval is set for the third-level node, and conventional monitoring status data is obtained through a multi-level sampling strategy; Perform three-threshold monitoring on the conventional monitoring status data, judge the temperature deviation through the first threshold, judge the humidity deviation through the second threshold, and judge the equipment status deviation through the third threshold, and switch to the alarm state when any indicator exceeds the threshold to obtain the alarm state data; The alarm status data is input into a two-dimensional history analysis module, the first dimension calculates the abnormal development trend based on time series analysis, and the second dimension calculates the abnormal diffusion risk based on spatial correlation analysis to obtain alarm processing data; Perform three-layer conflict resolution on the alarm processing data and the conventional monitoring status data, wherein the first layer processes time conflicts, the second layer processes space conflicts, and the third layer processes priority conflicts, and obtains collaboratively collected data through a hierarchical processing strategy; The collaboratively collected data is input into a four-dimensional health assessment model, and the temperature control stability is evaluated through the first dimension, the equipment reliability is evaluated through the second dimension, the network communication quality is evaluated through the third dimension, and the task execution efficiency is evaluated through the fourth dimension to obtain a health assessment index; A two-level trend analysis is performed on the health assessment indicators. The first level calculates the short-term health trend through a 15-minute sliding window, and the second level calculates the long-term health trend through a 4-hour sliding window to obtain the health assessment data of the cold chain system.

6. The intelligent cold chain integrated management method based on AI big data according to claim 5 is characterized in that: The digital twin model is used to optimize the strategy based on the cold chain system health assessment data to obtain the cold chain system control instructions, including: The cold chain system health assessment data is divided into three layers of states. The first layer discretizes the temperature control state to obtain the temperature control state space, the second layer discretizes the equipment state to obtain the equipment state space, and the third layer discretizes the system throughput to obtain the system state space, thereby obtaining a multi-dimensional state vector. A first Markov chain model is constructed based on the multidimensional state vector, the transition probability between states is calculated by a maximum likelihood estimation method to obtain a first state transfer matrix, and the first state transfer matrix is ​​expanded by a second order, and a second Markov chain model is constructed by calculating the conditional probability of the state sequence to obtain a second state transfer matrix; Inputting the second state transfer matrix into the digital twin modeling unit, constructing a digital twin of the temperature control system through the first virtual mapping module, and constructing a digital twin of the equipment operation through the second virtual mapping module to obtain a digital twin model of the system; Perform three-scenario simulation calculations on the digital twin model of the system, simulate a normal operating state through a first scenario, simulate a device failure state through a second scenario, and simulate a network congestion state through a third scenario to obtain scenario simulation data; Perform multi-objective optimization calculation based on the scenario simulation data, optimize system throughput through a first sub-objective, optimize transmission delay through a second sub-objective, and optimize energy consumption through a third sub-objective, to obtain an optimization strategy set; The optimization strategy set is input into the emergency response processing module, and graded processing is performed based on the temperature control abnormality level and the equipment alarm level to obtain an intelligent control solution, and the execution instructions of the intelligent control solution are decomposed. The optimization strategy is converted into device-level control instructions and system-level scheduling instructions through the control instruction generator, and the cold chain system control instructions are output.

7. An intelligent cold chain integrated management system based on AI big data, characterized in that: Used to execute the intelligent cold chain integrated management method based on AI big data as described in any one of claims 1 to 6, the system includes: The acquisition module is used to collect characteristic data of nodes in the cold chain system, obtain the original data of the nodes and construct the initial adjacency matrix of the cold chain network; An identification module, used for inputting the original data of the node and the initial adjacency matrix of the cold chain network into a temperature control parameter identification model to perform a graph convolution operation to obtain a temperature control parameter identification result; A calculation module, used to perform task decomposition and resource allocation calculation on the cold chain transportation task data and the temperature control parameter identification results to obtain a multi-region collaborative task allocation plan; A dispatching module, used for performing dynamic load calculation and inter-region task dispatching based on the multi-region collaborative task dispatching scheme, and obtaining real-time task execution status data; A processing module, used for performing hierarchical processing on the real-time task execution status data and collecting cold chain system health assessment data; The optimization module is used to perform strategy optimization based on the cold chain system health assessment data through a digital twin model to obtain cold chain system control instructions.

8. A smart cold chain integrated management device based on AI big data, characterized in that: The intelligent cold chain integrated management device based on AI big data includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory so that the intelligent cold chain integrated management device based on AI big data executes the intelligent cold chain integrated management method based on AI big data as described in any one of claims 1-6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by the processor, the intelligent cold chain integrated management method based on AI big data as described in any one of claims 1-6 is implemented.

Images