Multi-mode AIGC cold-chain logistics abnormal event early warning method and system

Abstract

The invention relates to the technical field of cold chain management, and discloses a multi-modal AIGC cold chain logistics abnormal event early warning method and system, and the method comprises the steps: collecting the multi-modal data of a cold chain node through an Internet of Things sensing system, carrying out the data mapping through a cross-modal alignment algorithm, generating a multi-modal monitoring feature sequence, and carrying out the early warning of the abnormal event of the cold chain logistics. According to the method, cold chain network evaluation feature distribution is obtained through high-dimensional feature analysis, local and global abnormal features are extracted by using a space-time diagram convolutional network, an abnormal probability information set is generated, the information set is subjected to multi-strategy deduction and diffusion simulation, prediction data are corrected, and finally accurate abnormal event early warning information is generated. According to the method, the abnormal early warning precision and response speed of cold-chain logistics can be improved, the risk and loss are reduced, the spatial topology and time evolution of a cold-chain network are captured, the method is superior to traditional isolated node detection, and the problem that in the prior art, single monitoring data prediction has the limitation of recognition capacity is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of cold chain management, and in particular, to a method and system for warning of abnormal events in cold chain logistics based on multi-modal AIGC. Background Art

[0002] In modern cold chain logistics, it is crucial to ensure temperature control and environmental monitoring of goods during transportation and storage. In cold chain logistics systems, especially when dealing with high-value and high-time-sensitivity goods such as food and medicine, existing cold chain logistics abnormal warning systems usually rely on single monitoring data or traditional prediction methods, which limits the ability to identify various potential abnormal events. At the same time, the transportation of cold chain goods passes through multiple cold chain nodes, and the information between each cold chain node is not interconnected, making it impossible to warn of the safety risks of transported cold chain goods from an overall perspective. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for warning of abnormal events in cold chain logistics based on multi-modal AIGC, aiming to solve the problem of limited identification ability in single monitoring data prediction in the prior art.

[0004] The present invention is implemented as follows. In the first aspect, the present invention provides a method for warning of abnormal events in cold chain logistics based on multi-modal AIGC, including: Collecting multi-modal data from each cold chain node of the cold chain network through a pre-deployed Internet of Things sensing system to obtain the original monitoring data of each cold chain node; Performing multi-modal semantic space mapping of the original monitoring data of the cold chain node in its own form and interaction form through a pre-trained cross-modal alignment algorithm to obtain the multi-modal monitoring feature sequence of the cold chain node; Performing high-dimensional feature analysis of the multi-modal monitoring feature sequence of each cold chain node for cold chain warehouse goods management and cold chain logistics transportation path to obtain the cold chain network evaluation feature distribution; Extracting features of local and global anomalies from the cold chain network evaluation feature distribution according to the spatio-temporal graph convolutional network to generate an abnormal probability information set of the cold chain network; Feeding back the abnormal probability information set to the cold chain network evaluation feature distribution to perform multi-strategy effect deduction on the abnormal probability information set and generate abnormal event prediction data; Performing diffusion simulation of the impact of the abnormal event on the abnormal event prediction data based on the cold chain network evaluation feature distribution, and correcting the abnormal event prediction data according to the diffusion simulation result to obtain abnormal event warning information.

[0005] In a second aspect, the present invention provides a cold chain logistics abnormal event warning system for multi-modal AIGC, which is used to implement the method for warning abnormal events in cold chain logistics of multi-modal AIGC described in any one of the first aspects, including: A data monitoring module, which is used to collect multi-modal data of each cold chain node in the cold chain network through a pre-deployed Internet of Things sensing system to obtain the original monitoring data of each cold chain node; A data analysis module, which is used to perform multi-modal semantic space mapping of the original monitoring data of the cold chain node in its own form and interaction form through a pre-trained cross-modal alignment algorithm to obtain the multi-modal monitoring feature sequence of the cold chain node; An overall evaluation module, which is used to perform high-dimensional feature analysis of the cold chain storage goods management and the cold chain logistics transportation path on the multi-modal monitoring feature sequences of each cold chain node to obtain the cold chain network evaluation feature distribution; A probability analysis module, which is used to extract local and global abnormal features from the cold chain network evaluation feature distribution according to the spatio-temporal graph convolutional network to generate an abnormal probability information set of the cold chain network; A strategy deduction module, which is used to feedback the abnormal probability information set to the cold chain network evaluation feature distribution to perform multi-strategy effect deduction on the abnormal probability information set and generate abnormal event prediction data; An event warning module, which is used to perform diffusion simulation of the impact of the abnormal event on the abnormal event prediction data based on the cold chain network evaluation feature distribution, and correct the abnormal event prediction data according to the diffusion simulation result to obtain abnormal event warning information.

[0006] The present invention provides a method for warning abnormal events in cold chain logistics of multi-modal AIGC, which has the following beneficial effects: The present invention collects multi-modal data of cold chain nodes through the Internet of Things sensing system, performs data mapping through the cross-modal alignment algorithm to generate a multi-modal monitoring feature sequence, obtains the cold chain network evaluation feature distribution through high-dimensional feature analysis, and uses the spatio-temporal graph convolutional network to extract local and global abnormal features to generate an abnormal probability information set. This information set undergoes multi-strategy deduction and diffusion simulation to correct the prediction data, and finally generates accurate abnormal event warning information. This method can improve the accuracy and response speed of abnormal warning in cold chain logistics, reduce risks and losses, capture the spatial topology and time evolution of the cold chain network, is superior to traditional isolated node detection, and solves the problem of limited recognition ability in single monitoring data prediction in the prior art. Description of the Drawings

[0007] Figure 1 is a step schematic diagram of a method for warning abnormal events in cold chain logistics of multi-modal AIGC provided by an embodiment of the present invention; Figure 2 It is a schematic structural diagram of a cold chain logistics abnormal event warning system for multi-modal AIGC provided by an embodiment of the present invention. Specific implementation manners

[0008] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0009] The implementation of the present invention will be described in detail below with reference to specific embodiments.

[0010] Refer to Figure 1 、 Figure 2 As shown, it is a preferred embodiment provided by the present invention.

[0011] In a first aspect, the present invention provides a method for warning abnormal events in cold chain logistics of multi-modal AIGC, including: S1: Collect multi-modal data from each cold chain node of the cold chain network through a pre-deployed Internet of Things sensing system to obtain the original monitoring data of each cold chain node; S2: Perform multi-modal semantic space mapping of the original monitoring data of the cold chain node in its own form and interaction form through a pre-trained cross-modal alignment algorithm to obtain the multi-modal monitoring feature sequence of the cold chain node; S3: Perform high-dimensional feature analysis on the multi-modal monitoring feature sequences of each of the cold chain nodes for cold chain storage goods management and cold chain logistics transportation paths to obtain the cold chain network evaluation feature distribution; S4: Extract features of local anomalies and global anomalies from the cold chain network evaluation feature distribution according to the spatio-temporal graph convolutional network to generate an abnormal probability information set of the cold chain network; S5: Feed back the abnormal probability information set to the cold chain network evaluation feature distribution to perform multi-strategy effect deduction on the abnormal probability information set to generate abnormal event prediction data; S6: Perform diffusion simulation of the impact of the abnormal event on the abnormal event prediction data based on the cold chain network evaluation feature distribution, and correct the abnormal event prediction data according to the diffusion simulation result to obtain abnormal event warning information.

[0012] Specifically, in step S1 of the embodiment provided by the present invention, according to the actual needs of cold chain logistics, key cold chain nodes are selected for sensor deployment, and these nodes include fixed cold chain nodes (such as cold chain warehouses, cold storages, distribution centers, etc.) and mobile cold chain nodes (such as cold chain transport vehicles, refrigerated trucks, containers, etc.).

[0013] More specifically, multiple sensors are deployed at each cold chain node to monitor different types of data. Common sensor types include: Visual sensors: Monitor the placement location of cold chain goods. Label recognition sensors: Register the labels of cold chain goods entering the cold chain node. Temperature sensors: Used to monitor the temperature changes in the cold chain node to ensure that the goods are kept within a safe temperature range. Humidity sensors: Used to detect the humidity level in the cold chain environment, which is particularly crucial for products that require humidity control (such as fresh food, pharmaceuticals, etc.). Air pressure sensors: Monitor air pressure changes, which are particularly important for certain goods that require a special climate environment (such as pharmaceuticals, perishable goods). GPS positioning sensors: Used to track the location and driving trajectory of cold chain transportation nodes (such as transport vehicles, containers, etc.). Vibration sensors: Monitor the vibration conditions during transportation to ensure that the goods are not damaged during transportation. CO2 sensors (optional): Monitor the carbon dioxide concentration in the cold chain environment, especially in environments where food or pharmaceuticals are stored.

[0014] More specifically, sensors continuously collect environmental data and transmit it to the data center. This data includes parameters such as temperature, humidity, location, vibration, etc. The sensor data of each node is collected at set time intervals, such as every minute, every hour, etc. The collected data is transmitted to the central server or cloud platform through wireless communication technologies (such as Wi-Fi, NB-IoT, LoRa, 5G, etc.). Wi-Fi and 5G are suitable for environments with stable network connections such as warehouses. NB-IoT and LoRa are suitable for remote or low-power areas, such as cold chain transportation in remote areas. To ensure the accuracy of real-time data collection, the Internet of Things system needs to adopt low-latency, high-bandwidth communication technologies to ensure the timely transmission of data.

[0015] More specifically, after the data is transmitted to the central server, data cleaning and preprocessing are first performed to ensure data quality. This includes removing noise data, handling missing values, and data format conversion, etc. The data from different sensors needs to be fused. For example, temperature and humidity data and location data need to be combined to better understand the state of cold chain logistics. Data fusion technology can help combine the data of multiple sensors to form a unified, multi-modal dataset that can be used for analysis.

[0016] More specifically, the central server or cloud platform will monitor the status of each cold chain node in real time and visualize the real-time data. Through charts, dashboards, etc., the current status of each node can be quickly understood. The system detects anomalies in real time through preset thresholds. For example, if the temperature of a certain node exceeds the set safe range, the system will immediately issue an alarm, and the alarm information will be transmitted to relevant personnel through methods such as text messages, emails, and APP notifications, so that timely measures can be taken.

[0017] More specifically, all the collected data will be stored in the cloud or a local database to form a historical data archive. Data such as the temperature and humidity records of cold-chain goods, transportation routes, and vehicle status can be stored long-term. The storage of data enables the historical data to be traced back in case of anomalies or quality issues, and the status of the entire cold-chain logistics process can be checked, which helps to identify potential loopholes or anomalies and ensures the transparency and traceability of cold-chain logistics.

[0018] It can be understood that the Internet of Things sensing system enables each node in the cold-chain network to be monitored in real time, ensuring that environmental conditions (such as temperature, humidity, etc.) are always within a safe range. This helps to improve the transparency of cold-chain logistics and provides real-time decision-making support for management personnel. The system can issue an alarm in a timely manner when an anomaly occurs at a cold-chain node. For example, when the temperature rises abnormally or the humidity is too low, the system can notify relevant personnel through the set alarm mechanism for a quick response and corrective measures.

[0019] More specifically, based on the collected multi-modal data, the cold-chain system can be continuously optimized through data analysis. For example, by analyzing the temperature and humidity data of different cold-chain nodes, the conditions during cold-chain transportation can be optimized to further improve the efficiency of cold-chain logistics. By storing and tracing the monitoring data of each node, it is ensured that there are complete data records for the entire process of each batch of goods from departure to transportation and finally to the destination. This enhances the traceability of cold-chain logistics, especially in the food and pharmaceutical fields, where it can guarantee their quality and safety. Real-time data collection and monitoring can detect potential risks early, helping to avoid damage or deterioration of goods in the cold chain, thus reducing the losses and risks of enterprises.

[0020] Specifically, in step S2 of the embodiment provided by the present invention, multiple sensors are deployed at cold-chain nodes to collect different types of original monitoring data, such as temperature, humidity, GPS location, vibration, air pressure, etc. The original data generated by each sensor has different time scales and data forms. For example, temperature and humidity data are usually continuous numerical data, while location and vibration data may be discrete time-series data, providing comprehensive monitoring data for cold-chain nodes and laying a foundation for subsequent data fusion and cross-modal alignment. The multi-modal data includes environmental data, physical states, and location information, laying a foundation for more comprehensive monitoring and analysis.

[0021] More specifically, preprocess the raw data collected by each sensor, such as data denoising, missing value filling, standardization, normalization, etc., and extract key features from the raw data. For example, extract features such as temperature peaks and fluctuation ranges from temperature data, and extract location change trends from GPS location data. Through data preprocessing and feature extraction, convert the raw data into a usable format for cross-modal alignment, providing clear and standardized input data for the multi-modal alignment algorithm, and improving the efficiency and accuracy of the subsequent mapping process.

[0022] More specifically, use deep learning (such as convolutional neural networks, recurrent neural networks) or traditional machine learning methods (such as PCA, canonical correlation analysis) to train a cross-modal alignment model. The goal is to learn the mapping relationship from different modalities (such as temperature and humidity data, location data, vibration data, etc.) to a unified semantic space. During the training process, by constructing a joint loss function, the data of each modality are made to maintain mutual correlation and consistency in the mapped space. This usually includes aligning their semantic features and time series features. Using the trained model, map the data of each modality of the cold chain node to a shared semantic space, achieving the alignment of different modality data, enabling data from different sensors to be compared and analyzed in the same semantic space. Through cross-modal alignment, the system can obtain comprehensive cold chain monitoring features, rather than relying solely on a single data source.

[0023] More specifically, combine the aligned multi-modal data sequences (such as temperature and humidity sequences, location change sequences, vibration sequences, etc.) in the semantic space to generate a comprehensive multi-modal monitoring feature sequence. The monitoring feature sequence of each cold chain node includes the multi-modal monitoring features of the node at different time points, such as environmental status, transportation status, location change, etc. Through time series modeling (such as LSTM, GRU, etc.), fuse these feature sequences into a time-series multi-modal feature sequence, providing data support for subsequent anomaly detection, trend prediction, and optimization decision-making. A time-series multi-modal monitoring feature sequence is generated, enabling the state of the cold chain node to be dynamically tracked in the time dimension. The multi-modal feature sequence provides richer and more comprehensive information, which can provide strong support for the next step of prediction, pattern recognition, and anomaly detection.

[0024] More specifically, by inputting the generated multi-modal monitoring feature sequences into subsequent analysis algorithms, such as anomaly detection, risk assessment, trend prediction, etc., to achieve the intelligent management of the cold chain network. Machine learning algorithms (such as support vector machines, random forests, deep neural networks) can be used to train the feature sequences to identify potential risks in the cold chain or predict future trends. The multi-modal feature sequences provide deep data support for the real-time monitoring of the cold chain network, can identify potential problems in the cold chain (such as abnormal temperature, excessive vibration, etc.), realize the intelligent decision-making support for cold chain logistics, and improve the management efficiency and response speed.

[0025] It can be understood that through the cross-modal alignment algorithm, heterogeneous data from different sensors is mapped to a unified semantic space, achieving the goal of information integration. The features of different data sources can be effectively compared and analyzed in the same space, eliminating the differences between different data formats. The multi-modal monitoring feature sequences generated after cross-modal alignment can provide comprehensive cold chain node status information, including environmental factors (such as temperature and humidity), equipment status (such as vibration, position change), etc. These feature sequences can be used for dynamic monitoring and analysis of the overall health status of cold chain logistics.

[0026] More specifically, through the multi-modal feature sequences, abnormal events in the cold chain network (such as temperature fluctuations, abnormal transportation vibrations, etc.) can be quickly identified. The prediction model can predict the future cold chain status based on historical data, providing decision-making support for managers. Using a unified semantic space to represent all modal data makes the multi-modal data fusion and analysis more efficient, reduces the complexity of data processing, enables the system to identify potential risks faster and respond, improving the safety and efficiency of cold chain logistics. By combining the application of cross-modal alignment and multi-modal feature sequences, the cold chain network can achieve more intelligent and automated management. The system can not only monitor cold chain nodes in real time, but also predict potential risks in the cold chain through data analysis and provide optimized decisions.

[0027] Specifically, in step S3 of the embodiment provided by the present invention, the multi-modal monitoring feature sequences (such as temperature, humidity, location, vibration, etc.) collected from each cold chain node (such as warehousing, transportation, distribution, etc.) are integrated into high-dimensional feature vectors. These feature vectors contain the environmental information, status information, etc. of each node in the cold chain network during a specific period. A model is built for cold chain warehousing goods management and cold chain logistics transportation routes to generate high-dimensional features. These features include not only temperature and humidity data, but also information such as transportation routes, warehousing conditions, transportation time, and vibration of transportation vehicles. For example, by considering the storage time of cold chain goods, transportation timeliness, and goods handling processes, multi-dimensional features regarding the logistics route and goods are generated. Multiple modal data (such as temperature and humidity, location information, etc.) are integrated into high-dimensional vectors to generate a feature space with rich information, which is convenient for subsequent analysis. The high-dimensional feature space provides detailed data support for the evaluation of the cold chain network, helping to identify and understand the comprehensive performance of each node in the cold chain.

[0028] More specifically, since the multi-modal feature space may have a high dimension, dimensionality reduction algorithms (such as PCA principal component analysis, t-SNE, UMAP, etc.) are used to perform dimensionality reduction on the high-dimensional feature vectors, reducing the data dimension while retaining the main information of the data. The dimensionality-reduced data is used for clustering analysis, such as K-means, DBSCAN, or hierarchical clustering. According to the characteristics of different cold chain nodes, similar node groups are identified to further analyze the performance and optimization solutions of cold chain nodes. Dimensionality reduction techniques help reduce the data complexity and remove redundant information, making subsequent analysis more efficient. Clustering analysis can reveal node groups with similar behaviors in the cold chain network (such as efficient nodes, inefficient nodes, etc.), providing a basis for network optimization.

[0029] More specifically, based on the multi-modal monitoring feature sequences of cold chain nodes, key factors in the cold chain warehousing process are analyzed, such as temperature and humidity fluctuations, storage time, and goods shelf life. Using the features in the high-dimensional feature space, such as storage environment, temperature and humidity control accuracy, and timeliness of goods storage, the effectiveness of warehousing management is evaluated. Through high-dimensional analysis of cold chain warehousing management data, key factors affecting warehousing efficiency are identified, such as the frequency of temperature and humidity control failures and the goods loss rate. High-dimensional feature analysis can accurately identify weak links in cold chain warehousing management, such as areas with unstable temperature and humidity control and links with poor warehousing timeliness, providing guidance for optimizing warehousing management and data support for improving the warehousing environment and management strategies.

[0030] More specifically, analyze the transportation path characteristics of cold chain logistics, evaluate the path selection, transportation efficiency, and potential risks during transportation. Through the multi-modal monitoring data of cold chain nodes, analyze information such as temperature and humidity changes, transportation duration, and vibration of transportation vehicles on the transportation path, evaluate the rationality and efficiency of the path, and use characteristics such as transportation time, temperature and humidity control, and vibration level in the high-dimensional feature space for path optimization analysis, evaluate the quality and efficiency of cold chain logistics paths. High-dimensional feature analysis provides data support for the path optimization of cold chain logistics, can identify the optimal path and potential transportation bottlenecks, ensure the quality and timeliness during the transportation of cold chain goods. In path evaluation, it can identify which paths are more efficient and which paths have potential risks (such as unstable temperature and humidity, transportation delays, etc.), providing key insights for the optimization of cold chain logistics networks.

[0031] More specifically, based on the high-dimensional analysis of cold chain warehouse goods management and logistics transportation paths, further conduct distribution analysis on the evaluation characteristics of cold chain networks. This can display the distribution of the high-dimensional feature space through visualization tools (such as t-SNE, PCA), helping to identify abnormal patterns or efficient nodes in the cold chain network. By analyzing the feature distributions of each cold chain node, comprehensively evaluate the overall performance of the cold chain network. The evaluation content includes indicators such as temperature and humidity control accuracy, goods management efficiency, and transportation timeliness of each node in the network. Feature distribution analysis provides a clear perspective for the global evaluation of cold chain networks, can clearly identify the advantages and weak links existing in the network. Through high-dimensional feature distributions, the performance of cold chain networks can be quantified, helping enterprises to overall control the cold chain operation status and optimize it targeted.

[0032] It can be understood that by constructing and analyzing high-dimensional features from the multi-modal monitoring features of cold chain nodes, the complex relationships and potential operation patterns in cold chain logistics can be captured, providing a comprehensive feature perspective for cold chain management. Dimensionality reduction processing reduces the complexity of data, and at the same time, clustering analysis can identify different types of node groups, helping cold chain managers to optimize various nodes and management links targeted. Based on high-dimensional feature analysis, the efficiency of warehouse management and logistics transportation paths can be accurately evaluated, problems can be discovered, and data support can be provided for optimization strategies.

[0033] More specifically, through the distribution analysis of cold chain node characteristics, the overall health status of the cold chain network is evaluated. This comprehensive evaluation can help managers fully understand the operation efficiency of the cold chain network, discover and solve problems in a timely manner, and improve the management level of cold chain logistics. High-dimensional feature analysis and its results support more intelligent cold chain management decisions, helping enterprises to achieve automated optimization and refined management of cold chain logistics, ensuring the quality of goods, reducing losses, and improving transportation efficiency.

[0034] Specifically, in step S4 of the embodiment provided by the present invention, a spatio-temporal graph of the cold chain network needs to be constructed. Each node represents a cold chain node (such as a warehouse, a transport vehicle, a distribution center, etc.), and the edges between the nodes represent the physical or logical relationships between them (such as transport routes, goods circulation routes, etc.). The spatio-temporal characteristics (such as temperature, humidity, transport status, etc.) in the cold chain network will change over time. The cold chain network is represented as a graph structure, where the nodes of the graph have time series characteristics, and the characteristics of each node include real-time data of the cold chain node, such as temperature, humidity, transport timeliness, etc. Through graph convolution operations, the spatial relationships between cold chain nodes (such as the interactions between different cold chain nodes) and the dynamic changes of the time series can be captured.

[0035] More specifically, in ST-GCN, spatio-temporal convolution can effectively fuse spatial information (such as the connection relationships between cold chain nodes) and time information (such as the sequential changes of nodes). Graph convolution is used to capture spatial dependencies, and temporal convolution is used to capture dynamic changes over time. This way, the characteristics of the cold chain system at different times and in different spaces can be captured. The spatio-temporal graph convolution network can process the spatio-temporal data in the cold chain network, capture the spatial dependencies and the time variation rules between the nodes in the cold chain network, and provide strong data support for subsequent anomaly detection.

[0036] More specifically, through the spatio-temporal graph convolution network, local anomaly detection is first performed, that is, the spatio-temporal feature sequence of a single cold chain node is analyzed. For example, the abnormal fluctuations of temperature and humidity in a certain cold chain warehouse, or the delay in a certain transport link, etc. The ST-GCN model is used to analyze the spatio-temporal characteristics of the node, calculate the anomaly probability of each node, and determine whether there is a local anomaly. On the basis of local anomalies, overall anomaly detection is further performed, that is, through the global cold chain network structure and spatio-temporal characteristics, the anomaly patterns of the entire cold chain network are analyzed. For example, the abnormal movements of multiple nodes in the entire cold chain network are analyzed to find potential global problems in the cold chain system (such as the simultaneous temperature and humidity fluctuations of multiple nodes, or the simultaneous delays of multiple logistics paths).

[0037] More specifically, the ST-GCN model will output the anomaly probabilities of each node and the entire network. These probability values indicate whether there is an anomaly in a certain cold chain node or the entire network under given spatio-temporal conditions. By training the network, the model learns the normal mode of the cold chain system and detects the nodes and paths that deviate from the normal mode. Local and overall anomaly detection can help identify potential problems in the cold chain network, whether it is the anomaly of a single node or the anomaly in the overall operation of the cold chain system. Through the analysis of the spatio-temporal graph convolution network, the anomaly patterns of cold chain nodes can be captured more accurately, providing timely warnings.

[0038] More specifically, the anomaly probability values generated by the spatio-temporal graph convolutional network can be aggregated into a probability information set. This information set will contain the anomaly probability values of each cold chain node, as well as the overall anomaly probability of the entire cold chain network. This information can be used for the dynamic monitoring, anomaly warning, and further decision-making support of the cold chain logistics system. As time goes by, the state of the cold chain network will constantly change. The spatio-temporal graph convolutional network can continuously process and update spatio-temporal data, dynamically generating the anomaly probability information set, which is crucial for the real-time monitoring of the cold chain system, enabling the timely discovery of potential problems in the system and taking corresponding intervention measures. The generated anomaly probability information set provides real-time and accurate anomaly detection results for cold chain managers, helping to quickly discover potential problems in the cold chain network and react in a timely manner. This information set can be used in the decision-making support system to optimize the operation and management of cold chain logistics and improve the reliability and safety of the cold chain system.

[0039] More specifically, by analyzing the anomaly probability information set of the cold chain network, potential anomaly patterns in the cold chain system can be identified. For example, frequent anomalies in certain nodes may indicate management loopholes, or transportation delays on certain routes may require route optimization or maintenance of transport vehicles. Based on the anomaly probability information set, the cold chain system can establish a real-time monitoring and warning mechanism. When the anomaly probability exceeds a certain threshold, the system will automatically trigger an alarm to notify the management staff for inspection and handling. By analyzing the anomaly probability, it can help the management staff optimize the operation strategy of the cold chain network, adjust storage conditions, transportation routes, cargo scheduling, etc., to improve the overall efficiency and reliability of the cold chain. Through the analysis of the anomaly probability, targeted optimization strategies can be provided for the cold chain logistics system to ensure the stability of system operation. The real-time monitoring and warning mechanism can effectively reduce the risks in the cold chain logistics process and prevent cargo losses and cold chain interruptions caused by anomalies.

[0040] It can be understood that the spatio-temporal graph convolutional network can effectively process the complex spatio-temporal data in the cold chain network, combine the dual information of space and time, and perform efficient feature extraction and analysis. The joint detection method for local and global anomalies can accurately capture potential problems in the cold chain system. Whether it is local anomalies in individual nodes or global anomalies in the overall system, they can be discovered in a timely manner. The anomaly probability information set generated by the spatio-temporal graph convolutional network provides real-time anomaly monitoring data for cold chain managers, helping decision-makers make timely and effective responses in cold chain logistics operations. Through real-time anomaly detection and dynamic update, the operation and management of cold chain logistics will become more intelligent and precise, reducing human intervention and improving the automation and decision-making efficiency of the system. Anomaly detection and the probability information set can help identify bottlenecks and potential problems in the cold chain system, providing valuable data support for cold chain path planning, warehouse management, transportation scheduling, etc., and optimizing the overall operation efficiency of the cold chain network.

[0041] Specifically, in step S5 of the embodiment provided by the present invention, the generated abnormal probability information set is fed back into the evaluation feature distribution of the cold chain network. This feedback mechanism can dynamically adjust the feature distribution of the existing network through the learning ability of the system. The cold chain network evaluation feature distribution includes the spatio-temporal features of each node in the cold chain network, such as temperature, humidity, transportation timeliness, etc. The fed-back abnormal probability information can be regarded as the dynamic state changes of the nodes and paths in the network. The fed-back abnormal probability information set adds an "abnormal weight" to each node and path in the cold chain network, indicating the probability of abnormal occurrence of the node or path within a certain time window. These weights will affect the evaluation feature distribution of each node in the cold chain network, so that the original network evaluation features (such as temperature, humidity) incorporate new "abnormal behavior" factors during the deduction process. By integrating the abnormal probability information set with the existing cold chain feature data (such as transportation status, environmental factors, etc.), a richer and more dynamic feature set is formed. This process enhances the perception ability of the real-time state of the cold chain network. By feeding back the abnormal probability information set to the evaluation feature distribution, the cold chain system can continuously update its state information, making the system's response to abnormal patterns more flexible and dynamic. The system's abnormal perception ability is enhanced, and it can capture abnormal factors that may affect the stability of the cold chain network in spatio-temporal changes, thereby improving the cold chain management strategy.

[0042] More specifically, based on the cold chain network evaluation feature distribution, a multi-strategy deduction model is constructed. This model can simulate different intervention strategies to deal with different abnormal events. For example, for temperature anomalies, a strategy of increasing the operating intensity of the warehouse air conditioning system can be adopted; for transportation delays, strategies such as adjusting the transportation route or introducing backup transportation tools may be required. Machine learning algorithms (such as reinforcement learning, decision trees, etc.) are used to deduce the effects of different strategies. The deduction model learns the effects of different strategies when implemented in the cold chain network through the feedback mechanism, and then evaluates the advantages and disadvantages of each strategy. For example, a certain strategy may be effective in the case of temperature anomalies, while another strategy is more effective in the case of humidity anomalies. Through deduction, the model can generate the effectiveness evaluation of each strategy under specific abnormal patterns.

[0043] More specifically, based on the results of the deduction, the system can obtain the expected effects of different strategies in response to different abnormal events. According to the deduced effects, the system can recommend the best intervention strategies to ensure that the cold chain network can effectively respond to various abnormal events. Through the deduction of the effects of multiple strategies, the system can simulate different emergency measures and intervention strategies and identify the most effective response measures. This deduction ability can provide a scientific basis for cold chain managers, thereby reducing the impact of abnormal events on the stability of the cold chain network. The application of methods such as reinforcement learning enables the system to autonomously discover the optimal strategy and adjust response measures according to the changing cold chain environment, improving the intelligent level of cold chain management.

[0044] More specifically, based on the evaluated feature distribution of the cold chain network after feedback and the results of multi-strategy deduction, a prediction model for abnormal events is established. This model can predict the trends of abnormal events based on techniques such as regression analysis and time series prediction (such as LSTM, ARIMA), and output the occurrence probability, time, and possible impact range of each abnormal event. Through the prediction model, the system can generate prediction data for future abnormal events. For example, within a certain period in the future, the system predicts that a temperature abnormality may occur at a certain cold chain node, or a delay may occur on a certain transportation route. The system predicts the occurrence frequency, duration, etc. of abnormal events through the time series model.

[0045] More specifically, finally, the system outputs the prediction data of abnormal events, including abnormal types, occurrence probabilities, occurrence times, potentially affected nodes and paths, etc. These prediction data provide a scientific basis for the early warning system of the cold chain network, helping managers take preventive measures in advance. Through the generated prediction data of abnormal events, the cold chain system can identify potential risks in advance and provide early warnings. Such early warnings can reduce losses caused by abnormal events and provide time for decision-makers to prepare intervention measures. Through accurate abnormal predictions, the resource scheduling and emergency response of the cold chain network can be more forward-looking, enhancing the risk resistance ability and operation efficiency of the cold chain system.

[0046] More specifically, based on the abnormal event prediction data, the cold chain system can provide intelligent decision support. By analyzing the prediction data, the system can propose specific intervention measures. For example, when it is predicted that the temperature in a certain warehouse may be abnormal, the system can automatically adjust the operation strategy of the temperature control equipment; when it is predicted that there will be a transportation delay, the system can dispatch backup transport vehicles in advance or adjust the transportation route. With the real-time feedback of abnormal events, the system can make dynamic adjustments according to the new data. For example, if the initial prediction data is not completely accurate, the system can correct the prediction results through feedback and optimize the emergency intervention measures. Real-time abnormal prediction and intervention decision-making can greatly reduce the potential risks in cold chain logistics, ensure the safety and quality of cold chain products, and the intelligent decision support of the system can improve the response speed and flexibility of the cold chain network and enhance the overall operation efficiency.

[0047] It can be understood that by feeding back the abnormal probability information set to the cold chain network evaluation feature distribution, the cold chain system can perceive abnormalities in real time and dynamically adjust its own features, thereby improving the system's ability to respond to abnormalities. Through multi-strategy deduction, the system can predict the effects of different intervention measures, so as to provide the optimal strategy for cold chain managers and reduce the impact of cold chain abnormal events. Through the abnormal event prediction model, the system can give early warnings of the occurrence of abnormal events, provide time for the cold chain system to intervene and adjust, and effectively reduce risks. The combination of abnormal event prediction and intervention decision-making makes the management of the cold chain network more intelligent and precise, providing strong support for the safe, stable and efficient operation of cold chain logistics.

[0048] Specifically, in step S6 of the embodiment provided by the present invention, a diffusion model reflecting the mutual influence between each node and path in the cold chain network is constructed. This model can be based on a graph theory model, where each node in the cold chain network (such as a transport vehicle, a warehouse, a distribution center, etc.) represents an entity in the cold chain system, and the path represents the connection between entities. This diffusion model can use models similar to epidemic propagation models and propagation models in complex networks (such as the SIR model, diffusion model, etc.) to simulate the propagation process of abnormal events.

[0049] More specifically, in a cold chain network, the spread of abnormal events (such as temperature fluctuations, transportation delays, etc.) is not a simple linear relationship, but is affected by factors of mutual dependence and influence among nodes. By defining spread rules, such as the propagation probability and speed between nodes, the impact scope and time process of abnormal events are simulated. For example, a temperature anomaly may spread to other nodes through certain nodes in the supply chain, resulting in more areas being affected. Using the above model to simulate the spread of abnormal events, by inputting initial abnormal event data (such as a temperature anomaly in a certain warehouse, a delay in a certain transportation route, etc.), the model will calculate how the abnormal event spreads through each node of the cold chain network according to the propagation rules, identify potential affected areas and nodes. Through the spread simulation, the system can identify the potential impact scope and time of the abnormal event, predict the areas where the event may spread, and help managers understand the multiple chain reactions that may occur after the abnormal event. This spread simulation provides a systematic way to evaluate the overall impact of abnormal events and helps to more accurately understand the potential risks in the cold chain network.

[0050] More specifically, according to the results of the spread simulation, the impact scope and severity are fed back to correct the prediction data of the abnormal event. Through the analysis of the spread results, the system can judge the inaccuracies in the prediction data, especially the parts where the impact scope is underestimated or the impact time is overestimated. For example, if the initial predicted data of the temperature anomaly only considers the anomaly in a certain area, but the spread simulation shows that other areas may also be affected, the prediction data needs to be corrected accordingly. The results of the spread simulation can provide a higher confidence level for the original prediction data of the abnormal event, especially in a network environment with multiple nodes and multiple paths. By combining the simulation results, the system can adjust the abnormal prediction values of each node and improve the overall prediction accuracy of the system.

[0051] More specifically, the time range and event intensity of the abnormal event prediction are corrected. For example, if the spread simulation shows that the abnormal event may affect a longer time or spread to more areas, the system will appropriately extend the warning time of the abnormal event and increase the intensity to ensure the sufficiency of the warning information. By correcting the prediction data of the abnormal event through the spread simulation, the system can self-correct under real-time conditions and avoid prediction deviations. The corrected prediction data of the abnormal event can more accurately reflect the occurrence and spread of the actual abnormal event and provide more scientific warning information for cold chain managers.

[0052] More specifically, by integrating diffusion simulation and corrected abnormal event prediction data, the system generates final abnormal event warning information, which includes possible abnormal events, the occurrence probability of the events, the affected range, the time span, and the potential impact on the cold chain network. The system generates warning information based on the corrected data and notifies relevant personnel through visual interfaces, reports, text messages, or emails. The warning information should not only include the type and time of the abnormal event but also detail the possible affected nodes, paths, and possible countermeasures. As the state of the cold chain network changes, the warning information needs to be updated dynamically. For example, when a new abnormal event occurs, the system should correct it in real time and provide new warning data. By continuously tracking the results of diffusion simulation, the system can continuously optimize the warning information to ensure its high consistency with the actual situation. Based on the corrected abnormal event prediction data, the system can issue warning information in a timely and accurate manner, ensuring that relevant personnel in the cold chain network can take effective measures before the abnormal event occurs. Through detailed warning information, managers can obtain a comprehensive picture of the abnormal event, understand the possible risks, and take effective emergency measures in a timely manner to avoid or reduce losses.

[0053] It can be understood that through the cold chain network diffusion model, the propagation path and influence range of abnormal events in the network can be simulated. The system can not only identify local anomalies but also analyze their global impacts, thus providing a more comprehensive risk assessment. The results of diffusion simulation can continuously correct the abnormal event prediction data, improve the accuracy of abnormal event prediction, and ensure that the cold chain system can keep abreast of the actual development of events in real time. Based on the corrected prediction data, the system generates more accurate and timely warning information, ensuring that managers can quickly take effective countermeasures according to the warning information and avoid greater losses caused by cold chain abnormal events. Diffusion simulation can help the system identify the global impacts of abnormal events, optimize the management decision-making of the cold chain network, and ensure the stability and safety of cold chain transportation. The corrected abnormal event warning information provides a scientific basis for cold chain management, helps decision-makers make more accurate decisions in a dynamically changing environment, and improves the system's emergency response ability and overall efficiency.

[0054] The present invention provides a method for warning abnormal events in cold chain logistics based on multi-modal AIGC, which has the following beneficial effects: The present invention collects multi-modal data of cold chain nodes through an Internet of Things sensing system, performs data mapping through a cross-modal alignment algorithm to generate a multi-modal monitoring feature sequence, obtains the cold chain network evaluation feature distribution through high-dimensional feature analysis, and uses a spatio-temporal graph convolutional network to extract local and global abnormal features to generate an abnormal probability information set. This information set is corrected through multi-strategy deduction and diffusion simulation to generate accurate abnormal event warning information. This method can improve the abnormal warning accuracy and response speed of cold chain logistics, reduce risks and losses, and capture the spatial topology and time evolution of the cold chain network, which is superior to traditional isolated node detection and solves the problem of the limitation of the recognition ability in the prediction of single monitoring data in the prior art.

[0055] Preferably, the step of collecting multi-modal data from each cold chain node of the cold chain network through a pre-deployed Internet of Things sensing system to obtain the original monitoring data of each cold chain node includes: S11: Perform clock synchronization processing on the sensor modules pre-deployed at each cold chain node of the cold chain network to generate a data recording axis for the sensor modules of the Internet of Things sensing system deployed at each cold chain node; S12: Collect data from multiple monitoring channels of the cold chain node through several sensor modules with different functions deployed at the cold chain node, and generate and mark time stamps for the sensing data collected by the sensor modules of each function based on the data recording axis, so as to obtain the original monitoring data in which the sensing data with time stamp marks are arranged in sequence according to the time stamp marks; S13: When the fixed cold chain node representing the cold chain warehouse conducts goods handover with the mobile cold chain node representing the cold chain vehicle, perform monitoring data exchange and data recording axis synchronization of the original monitoring data of the fixed cold chain node and the mobile cold chain node participating in the goods handover, so as to update the original monitoring data of each cold chain node.

[0056] Specifically, different sensor modules are deployed at each cold chain node, and these sensors need to work together to accurately record and match the monitoring data. If the time of each sensor is not synchronized, it may lead to incorrect time stamps of the data, which in turn affects data analysis and decision-making. Standard clock synchronization technologies such as the Network Time Protocol can be used to ensure the clock synchronization of each sensor module, and the control system adjusts the clocks of all sensors to make their acquisition times unified, providing an accurate time basis for subsequent data recording and analysis.

[0057] More specifically, once the clock synchronization of each sensor module is completed, a data recording axis can be created for each sensor module, that is, a time series data recording framework. All the collected data will be sorted and marked according to this unified time axis, ensuring that the data of all sensors are based on a unified time reference, thus guaranteeing the time consistency between data. Clock synchronization improves the accuracy of data collection, enabling the monitoring data of each node in the cold chain network to accurately reflect the chronological order of events.

[0058] More specifically, the sensor modules deployed at cold chain nodes have multiple functions. Commonly included are temperature sensors, humidity sensors, GPS positioning sensors, access control sensors, CO2 sensors, etc. There will be multiple sensors on each cold chain node responsible for different monitoring tasks, collecting multi-modal data. Each sensor module real-time collects environmental data such as temperature, humidity, location and other information, and sends the collected data to the central control system. The collected data may have different formats and precisions according to different sensor types. The data collected by each sensor is timestamped through the data recording axis. The generation rule of the timestamp is based on the unified clock of the sensor module, ensuring that each data record can be arranged in the order of collection time. According to the timestamp, the system will automatically arrange the data from different sensors in chronological order to form a time series of the original monitoring data. Multi-modal data integration can simultaneously obtain and integrate multi-dimensional data of cold chain nodes, comprehensively reflecting the operating state of the cold chain. Through timestamp marking and unified clock management, it is ensured that the data is accurately arranged in chronological order, which helps with later analysis and prediction.

[0059] More specifically, when cold chain goods are handed over between a warehouse and a transport vehicle, it involves data exchange between a fixed cold chain node (warehouse) and a mobile cold chain node (transport vehicle). During the handover, the monitoring data of the two nodes needs to be synchronized to ensure that all the original monitoring data can be kept consistent. During the process of goods handover, the fixed cold chain node and the mobile cold chain node conduct data exchange through wireless network, Bluetooth, or other communication protocols. The fixed cold chain node and the mobile cold chain node will transmit their respective monitoring data to each other and synchronize the data recording axis to ensure that the data of both is seamlessly connected at the handover point. To maintain the integrity and continuity of the data, the cold chain node will update the data recording axis according to the content of the data exchange, which includes adding timestamps to new data, updating the status of the original data, and transmitting the synchronized data to the cloud platform or the centralized management system.

[0060] More specifically, by synchronizing with the recording axis through monitoring data exchange, it is ensured that the data of each node in the cold chain can be transmitted continuously and completely, thereby maintaining the data consistency of the entire cold chain system, and thus ensuring the identification of the transportation trajectory of cold chain goods during the cold chain transportation process, and identifying which cold chain goods each cold chain node in the cold chain grid has. Through synchronization and exchange, it can be ensured that at the node of goods handover, the data of each cold chain node is not omitted or repeated, improving the reliability of the data.

[0061] More specifically, after the fixed cold chain node and the mobile cold chain node complete the data exchange, it is necessary to update and verify the original monitoring data at both ends. Any new monitoring data needs to be compared with the existing data to check its integrity and accuracy. The updated original monitoring data will be transmitted to the centralized management system or the cloud platform. The system can perform real-time monitoring, analysis, and prediction based on these data, providing data support for the status monitoring of the cold chain, ensuring the data consistency of each cold chain node, and preventing data loss or errors caused by handover. The updated original monitoring data can provide an accurate basis for subsequent data analysis, abnormal event detection, and early warning systems, helping decision-makers to understand the cold chain transportation status in real time.

[0062] It can be understood that by using multi-modal sensors and clock synchronization technology, the system can collect key monitoring data of each cold chain node in real time and with high precision, ensuring that the environmental data of each link in the cold chain transportation is accurately recorded. The data collected by multiple sensors are integrated through timestamp marking and data recording axis to ensure the time consistency between data, providing a reliable basis for later analysis and early warning.

[0063] More specifically, during the goods handover, the cold chain nodes can smoothly exchange data and maintain the integrity and consistency of the data in the cold chain network through recording axis synchronization. The data update and verification mechanism ensures the integrity and accuracy of the monitoring data, improving the real-time analysis and decision support capabilities of the cold chain management system for data. Through precise collection and update of the original monitoring data, the cold chain transportation status can be monitored in real time, improving the transparency and reliability of goods transportation, and avoiding goods losses caused by factors such as temperature and humidity.

[0064] Preferably, the steps of obtaining the multi-modal monitoring feature sequence of the cold chain node by performing multi-modal semantic space mapping of the original monitoring data of the cold chain node in its own form and interaction form through a pre-trained cross-modal alignment algorithm include: S21: Judge the monitoring form of the original monitoring data of each cold chain node, and at the same time, verify and supervise each other according to the judgment results of the mutually related cold chain nodes, so as to divide the original monitoring data of each cold chain node into its own form monitoring part and interactive form monitoring part; among them, the self-form monitoring part is used to describe the time period when the fixed cold chain node representing the cold chain warehouse and the mobile cold chain node representing the cold chain vehicle do not conduct goods handover, and the interactive form monitoring part is used to describe the time period when the fixed cold chain node representing the cold chain warehouse and the mobile cold chain node representing the cold chain vehicle conduct goods handover; S22: Perform multi-modal semantic space mapping on the self-form monitoring part of the cold chain node through a pre-trained cross-modal alignment algorithm to obtain the multi-modal monitoring features of the self-form monitoring part of the cold chain node; S23: Perform multi-modal semantic space mapping on the interactive form monitoring part of the cold chain node through a pre-trained cross-modal alignment algorithm to obtain the multi-modal monitoring features of the interactive form monitoring part of the cold chain node; S24: Perform temporal arrangement processing on the multi-modal monitoring features of the self-form monitoring part and the multi-modal monitoring features of the interactive form monitoring part to obtain the multi-modal monitoring feature sequence of the cold chain node.

[0065] Specifically, analyze the original monitoring data of each cold chain node to judge whether the data reflects the self-form monitoring part or the interactive form monitoring part. The self-form monitoring part describes the status monitoring data of the cold chain warehouse (fixed cold chain node) and the cold chain vehicle (mobile cold chain node) during the time period when no goods handover occurs. These data usually include information such as temperature and humidity in the warehouse, inventory status, etc., or environmental data during vehicle transportation. The interactive form monitoring part describes the monitoring data during the time period when the cold chain warehouse and the cold chain vehicle conduct goods handover. These data include information such as temperature and humidity changes, the goods handling process, and the temperature control exchange between the warehouse and the vehicle. By clearly distinguishing the two monitoring forms of the original monitoring data of the cold chain node, it is possible to better understand the source and content of the data, lay a foundation for subsequent multi-modal semantic mapping, and ensure that each type of data can be analyzed in the correct time period and context through clear judgment of the monitoring form, avoiding confusion of different types of data and improving the accuracy of data processing.

[0066] More specifically, according to the mutual correlation between cold chain nodes (such as the relationship between cold chain warehouses and cold chain vehicles), the respective monitoring data are verified. If data exchange or handover operations occur between a cold chain warehouse and a cold chain vehicle at the same time point, it is necessary to ensure the consistency of the monitoring data of the two nodes. Through the supervision and feedback in the data verification process, it is ensured that the data of each cold chain node can accurately reflect its state, and at the same time, its accuracy and consistency can be further improved through data exchange with each other. Through verification and supervision, it can be ensured that the data between different cold chain nodes can be correctly synchronized during handover, avoiding data mismatches or omissions. Through the supervision mechanism, invalid or incorrect data can be excluded, thus ensuring the quality of the monitoring data of the entire cold chain system.

[0067] More specifically, according to the judgment result of the previous step, the original monitoring data of the cold chain node is divided into two parts. The self-form monitoring part includes the data of the cold chain node in the non-handover state, such as the temperature and humidity data inside the warehouse, the temperature record of the vehicle, the record of the stored goods, etc. The interactive-form monitoring part includes the data of the cold chain node when handing over goods, such as the temperature and humidity changes at the handover point, the monitoring data during handling operations, the transfer record of the goods, etc. By dividing the data, the monitoring data of the cold chain node can be structured, which is convenient for subsequent processing and analysis, ensuring that each part of the monitoring data can be independently analyzed in different situations, so as to extract more refined monitoring features.

[0068] More specifically, a pre-trained cross-modal alignment algorithm (such as an alignment network or a mapping network based on deep learning) is used to perform multi-modal semantic space mapping on the self-form monitoring part and the interactive-form monitoring part of the cold chain node. The cross-modal alignment algorithm maps different types of data (such as temperature, humidity, location, etc.) collected by the cold chain node into a unified semantic space, so that data from different sensors can be compared and fused in the same space. Through mapping, multi-modal monitoring features of the cold chain node are extracted. These features can reflect the state of the cold chain node and provide multi-dimensional data support for subsequent analysis. Cross-modal data fusion can effectively fuse different types of sensor data into the same semantic space, extract the key features of the cold chain node. Through multi-modal semantic space mapping, the deviation of single-modal data can be avoided, thus improving the comprehensiveness and accuracy of data analysis.

[0069] More specifically, the multimodal monitoring features of the self-form monitoring part of the cold chain node and the multimodal monitoring features of the interactive form monitoring part are arranged in time series to ensure that the feature data can reflect the state changes of the cold chain node in different time periods. By arranging the time series features, a complete multimodal monitoring feature sequence of the cold chain node is generated. This sequence will cover various monitoring states of the cold chain nodes and reflect the entire process of the cold chain transportation process. It can form a complete multimodal feature sequence through time series arrangement, which provides a basis for subsequent time series analysis (such as anomaly detection, trend prediction, etc.). The generated multimodal monitoring feature sequence comprehensively displays the state changes of the cold chain nodes, which is convenient for further analysis of the quality and efficiency of cold chain transportation.

[0070] It can be understood that through the cross-modal alignment algorithm, the different sensor data of the cold chain nodes are mapped into a unified semantic space, realizing the deep fusion and feature extraction of multimodal data, and the clear division of the self-form monitoring part and the interactive form monitoring part, so that the monitoring data under different states can be processed in a targeted manner, ensuring that the state of the cold chain nodes can be accurately described. Through the generation of time series arrangement and feature sequences, it can provide a rich data foundation for subsequent time series analysis, which is helpful for real-time monitoring and prediction and early warning of cold chain status. After integrating multimodal information, the cold chain monitoring system can provide more comprehensive and accurate monitoring results, improve the monitoring accuracy and reliability of the entire cold chain network, and thus reduce the risks and losses in cold chain transportation.

[0071] Preferably, the step of performing multimodal semantic space mapping on the self-form monitoring part of the cold chain node by a pre-trained cross-modal alignment algorithm to obtain the multimodal monitoring features of the self-form monitoring part of the cold chain node comprises: S221: Acquire working environment information of the cold chain node as the analysis object and monitoring mode information of the configured sensor module, and adaptively deploy algorithm parameters of the pre-trained cross-modal alignment algorithm according to the working environment information and the monitoring mode information, so that the cross-modal alignment algorithm is adapted to the cold chain node as the analysis object; S222: performing homomodal feature analysis on the self-modal monitoring part of the cold chain node according to the cross-modal alignment algorithm, and mapping the feedback information of each modality to a specified cross-modal semantic space; wherein the feedback information includes several possible information features and corresponding feature confidence factors; S223: performing cross-modal sample alignment on feedback information of each modality mapped to the cross-modal semantic space, so that the feedback information of each modality is in an aligned state marked by a timestamp; S224: Based on the cross-modal alignment algorithm, perform multi-modal collaborative verification based on feature confidence factors for the information features of several possibilities of each modality, and at the same time perform semantic reinforcement constraints on the multi-modal collaborative verification based on the pre-constructed cold chain transportation knowledge graph to obtain the multi-modal monitoring features of the self-form monitoring part.

[0072] Specifically, the working environment information of the cold chain node includes background information such as the temperature, humidity, and logistics status of the environment where the cold chain node is located. These information help to understand the working conditions of the cold chain node and affect the data collection method and effectiveness of the sensor. The monitoring mode information of the sensor module The configuration and monitoring mode of the sensor module include the types of sensors used by the cold chain node (such as temperature and humidity sensors, GPS positioning, acceleration sensors, etc.), as well as the monitoring frequency and data transmission method of these sensors.

[0073] More specifically, based on the working environment information and monitoring mode information of the cold chain node, perform adaptive algorithm parameter configuration on the pre-trained cross-modal alignment algorithm to ensure that the cross-modal alignment algorithm can accurately adapt to the specific characteristics and monitoring requirements of the cold chain node. By performing adaptive parameter deployment according to the working environment and monitoring mode of the cold chain node, it is ensured that the cross-modal alignment algorithm can fully consider the impact of the environment on data collection, improve the adaptability and accuracy of the algorithm. Since the environment and monitoring methods of each cold chain node are different, being able to adjust the algorithm parameters according to different conditions can ensure the effective processing of data from different cold chain nodes.

[0074] More specifically, according to the pre-trained cross-modal alignment algorithm, perform feature analysis on the self-form monitoring part of the cold chain node (such as single-type sensor data such as temperature, humidity, and location). This process includes the separate analysis of each modality data, extracting its effective information, and mapping the feedback information of each modality (such as possible features and their corresponding confidence factors) to a predefined cross-modal semantic space. This semantic space converts the data of multiple modalities into vector forms in the same space, facilitating subsequent comparison and fusion. By mapping the same-modal data to the cross-modal semantic space, a shared semantic framework can be provided for the data of different modalities, enabling these data to be more effectively fused and analyzed. Through the same-modal feature analysis, the key information related to the state of the cold chain node can be effectively extracted, providing a reliable basis for subsequent multi-modal collaborative verification.

[0075] More specifically, through the cross-modal alignment algorithm, the feedback information of each modality mapped into the cross-modal semantic space is aligned to ensure that this information is matched and synchronized according to the timestamp markings. For example, information such as temperature, humidity, and location may come from different sensors, and the cross-modal alignment algorithm will ensure their temporal consistency. On the basis of ensuring temporal synchronization, the cross-modal alignment algorithm merges data of different modalities and provides them with a common time dimension for facilitating the collaborative analysis of multi-modal data. By aligning the timestamps of samples, it ensures that data of different modalities are consistent in the time dimension, which helps to improve the accuracy of analysis, guarantees the temporal consistency between cross-modal data, and contributes to generating a traceable multi-modal monitoring feature sequence.

[0076] More specifically, the cross-modal alignment algorithm performs multi-modal collaborative verification based on the information characteristics and corresponding confidence factors of each modality. Each piece of feedback information of a modality has a confidence factor used to represent the credibility of the modality information. Data of multiple modalities will be verified and fused under the weighting of the confidence factors to obtain a more accurate multi-modal monitoring feature. When performing multi-modal collaborative verification, the pre-constructed cold chain transportation knowledge graph is used to semantically reinforce and constrain the results of multi-modal collaborative verification. The knowledge graph provides domain knowledge about cold chain logistics, such as cargo temperature control requirements, transportation processes, etc., which can provide semantic constraint conditions for the verification of cross-modal data and further improve the accuracy of data fusion. Through multi-modal collaborative verification based on confidence factors, data of different modalities can corroborate each other, improving the credibility of monitoring features. Through the semantic reinforcement of the knowledge graph, it can effectively constrain the verification process of cross-modal data, making the final multi-modal monitoring features more in line with the actual needs of cold chain transportation and avoiding the interference of irrelevant data.

[0077] More specifically, after multi-modal collaborative verification and semantic reinforcement and constraint, the multi-modal monitoring features of the self-form monitoring part of the cold chain node are extracted. These features describe the state of the cold chain node at different time periods and can effectively reflect important information such as the conditions and environmental changes of cold chain transportation. Through the whole process, the finally extracted multi-modal monitoring features can comprehensively and accurately describe the working state of the cold chain node. These multi-modal monitoring features provide profound insights for the cold chain management system and contribute to the real-time monitoring, anomaly detection, and predictive analysis of the cold chain process.

[0078] It is understandable that by performing cross-modal analysis, mapping, alignment, collaborative verification, and semantic enhancement on data of different modalities, multi-dimensional data from different sensors can be efficiently fused, comprehensive and accurate multi-modal monitoring features can be extracted, and algorithm parameters can be adaptively adjusted according to the working environments and monitoring modes of different cold chain nodes, enabling personalized processing and analysis of the data of each cold chain node, improving the adaptability of the overall monitoring system. Based on multi-modal collaborative verification with feature confidence factors and the introduction of semantic enhancement constraints, the accuracy and credibility of the data have been significantly improved, noise data and irrelevant factors can be effectively excluded, and the state of the cold chain node can be comprehensively reflected through the finally extracted multi-modal monitoring features, thereby improving the real-time monitoring and anomaly warning capabilities of the cold chain management system and optimizing the cold chain transportation efficiency and safety.

[0079] Preferably, the steps of performing multi-modal semantic space mapping on the interactive form monitoring part of the cold chain node through a pre-trained cross-modal alignment algorithm to obtain the multi-modal monitoring features of the interactive form monitoring part of the cold chain node include: S231: Mark the interactive form monitoring part of the cold chain node as the reference part to be analyzed, and mark the interactive form monitoring part of another cold chain node that conducts goods handover with the reference part as the interactive part; S232: Process the reference part and the interactive part respectively according to the multi-modal semantic space mapping method corresponding to their own form monitoring parts to obtain the first monitoring feature corresponding to the reference part and the second monitoring feature corresponding to the interactive part; S233: Perform timestamp alignment processing on the first monitoring feature and the second monitoring feature, and conduct information change analysis and consistency verification on the monitoring feedback content of the first monitoring feature and the second monitoring feature to obtain the cold chain goods transfer process information corresponding to the reference part; S234: Obtain the working environment information of the cold chain node corresponding to the reference part and the cold chain node corresponding to the interactive part, and perform feature encoding on the working environment information for the spatio-temporal structure of the cold chain node. Based on the spatio-temporal structure feature encoding of the two cold chain nodes participating in the interaction, perform digital simulation on the key steps of the execution process of the cold chain goods transfer process information, and extract key features from the cold chain goods transfer process information according to the digital simulation results to obtain the key features of cold chain goods transfer; S235: Combine the key features of cold chain goods transfer with the first monitoring feature for monitoring feedback content to obtain the multi-modal monitoring features of the interactive form monitoring part of the cold chain node.

[0080] Specifically, the interaction form monitoring part of the cold chain node to be analyzed is marked as the "reference part". The reference part refers to the location of one of the cold chain nodes in the cold chain logistics process, which serves as the starting point for data analysis. The interaction form monitoring part of another cold chain node that conducts goods handover with the reference part is marked as the "interaction part". The interaction part represents the location of another cold chain node that hands over goods with the reference part. By marking the reference part and the interaction part, the interaction process between the two is clarified, and a foundation for subsequent data comparison and analysis is laid. By marking the reference part and the interaction part, the interaction relationship between key nodes in the cold chain logistics can be clearly identified, helping to understand the flow process of goods.

[0081] More specifically, based on the multi-modal semantic space mapping method of the self-form monitoring part, the monitoring data of the reference part and the interaction part are processed respectively, that is, feature extraction and mapping are performed on the data of these two parts respectively to obtain the "first monitoring feature" of the reference part and the "second monitoring feature" of the interaction part. The monitoring data is mapped to the cross-modal semantic space to achieve unified representation of different modal data (such as temperature, humidity, location, timestamp, etc.). Through the cross-modal semantic space mapping method, corresponding monitoring features can be extracted from the monitoring data of the reference part and the interaction part, providing a high-quality data basis for subsequent data analysis and comparison. Unifying different types of monitoring data into a semantic space ensures that data between different modalities can be compared and fused within the same framework.

[0082] More specifically, timestamp alignment is performed on the first monitoring feature of the reference part and the second monitoring feature of the interaction part to ensure that the data of the two is consistent in time, facilitating subsequent analysis. Information change analysis of the monitoring feedback content is performed on the monitoring features after timestamp alignment, comparing the changes that occur in the reference part and the interaction part during the interaction process, and performing consistency verification to check for inconsistencies or anomalies. Through timestamp alignment and consistency verification, the reliability and accuracy of the data are ensured. Timestamp alignment ensures the consistency in time of the monitoring features of the reference part and the interaction part, avoiding errors caused by time asynchronization. Through consistency verification, anomalies or deviations in the interaction process can be effectively identified, improving the credibility of the monitoring results.

[0083] More specifically, obtain the working environment information of the cold chain nodes corresponding to the reference part and the cold chain nodes corresponding to the interaction part, such as temperature and humidity, equipment operation status, location, etc. Perform feature encoding on the working environment information of the cold chain nodes to construct the spatio-temporal structure features of the cold chain nodes. This feature encoding can represent the distribution, movement, and status of the cold chain nodes in space and time. Through feature encoding, the working environment information of the cold chain nodes can be transformed into data available for analysis, helping to model the spatio-temporal distribution and status of the cold chain nodes. The spatio-temporal structure features of the cold chain nodes provide deeper background information and can play an important role in subsequent simulations and verifications.

[0084] More specifically, based on the spatio-temporal structure feature encoding of two cold chain nodes, digitally simulate the cold chain cargo transfer process. Through the simulation, the state changes and influencing factors of the cold chain cargo during the handover process can be predicted. According to the digital simulation results, extract the key steps in the cold chain cargo transfer process and identify the features crucial to the cold chain cargo transfer process. Digital simulation can help predict and optimize the cold chain cargo transfer process, discover potential risk points or efficiency bottlenecks in advance. Through the simulation process, the most critical features for the cold chain cargo transfer can be extracted, which helps to improve the response speed and accuracy of the monitoring system.

[0085] More specifically, analyze the key features of the cold chain cargo transfer extracted from the digital simulation results and combine these features with the first monitoring features of the reference part. Combine the key features of the cold chain cargo transfer with the feedback content in the monitoring data to obtain the multi-modal monitoring features of the cold chain node interaction process. These features not only include the changes in the monitoring data itself but also take into account all the key links in the cold chain cargo transfer process. The combined monitoring features provide a multi-dimensional monitoring view for the cold chain node interaction process, making the monitoring more comprehensive and accurate. By combining the key features of the cold chain cargo transfer with the monitoring feedback content, the entire process of the cargo flow can be tracked, thereby improving the accuracy and efficiency of cold chain management.

[0086] More specifically, according to the outputs of all the above steps, finally obtain the multi-modal monitoring features of the cold chain node interaction form monitoring part. These features comprehensively reflect the status and dynamic changes of the reference part and the interaction part during the entire cargo handover process. By fusing the multi-modal features from the reference part and the interaction part, a more comprehensive and accurate monitoring view is provided. These multi-modal monitoring features can help cold chain managers control and schedule the cargo handover process more precisely, improving the efficiency and reliability of cold chain transportation.

[0087] It can be understood that by aligning timestamps, analyzing information changes, and performing consistency verification, the data of the benchmark part and the interaction part can be accurately compared to ensure the effective fusion and comparison of multimodal data, thereby improving the monitoring accuracy. Through digital simulation, the process of cold chain cargo transfer can be optimized and predicted, so as to identify potential problems in advance and make adjustments, improving the reliability and efficiency of cold chain transportation. Combining the key features of cold chain cargo transfer with the monitoring feedback data can achieve comprehensive monitoring and tracking of the interaction process of cold chain nodes, providing stronger decision-making support for cold chain management.

[0088] Preferably, the steps of performing high-dimensional feature analysis on the multimodal monitoring feature sequences of each of the cold chain nodes for cold chain warehouse cargo management and cold chain logistics transportation routes to obtain the cold chain network evaluation feature distribution include: S31: Identify the functional uses of the cold chain nodes to classify the cold chain nodes into fixed cold chain nodes representing cold chain warehouses and mobile cold chain nodes representing cold chain vehicles; S32: Construct a basic cold chain network based on the relative positional relationships between the fixed cold chain nodes, and at the same time deploy the monitoring features of each time period for the basic cold chain network according to the multimodal monitoring sequences of the fixed cold chain nodes to obtain the inventory actual cold chain network corresponding to each time period; S33: Dynamically construct the node links of the basic cold chain network according to the multimodal monitoring feature sequences of the mobile cold chain nodes to obtain the transportation actual cold chain network corresponding to each time period; S34: Combine the information of the inventory actual cold chain network and the transportation actual cold chain network of each segment to obtain a number of cold chain comprehensive actual networks arranged in sequence according to time development; S35: Perform traceability processing on the cold chain goods of each fixed cold chain node based on the cold chain comprehensive actual network to obtain the cold chain goods traceability characteristics of each fixed cold chain node; S36: Perform tracking processing on the movement trajectories of each mobile cold chain node based on the cold chain comprehensive actual network to obtain the cold chain transportation trajectory characteristics of each mobile cold chain node; S37: Assign the cold chain goods traceability characteristics and the cold chain transportation trajectory characteristics to the corresponding cold chain nodes on the cold chain comprehensive actual network, and the cold chain comprehensive actual networks arranged in sequence according to time together form the cold chain network evaluation feature distribution.

[0089] Specifically, identify the functional uses of cold chain nodes and classify them into two categories: Fixed cold chain nodes usually represent cold chain warehouses, which are responsible for storing goods and maintaining stable temperature and humidity conditions. Mobile cold chain nodes represent cold chain vehicles, which are responsible for transporting goods between different locations. Define the functional attributes of different cold chain nodes to facilitate subsequent targeted processing of data for fixed and mobile cold chain nodes. By clarifying the functional attributes of cold chain nodes, it is possible to ensure that subsequent feature analysis and data processing have clear goals and scopes, facilitating subsequent individual optimization of each type of node.

[0090] More specifically, based on the relative position relationships between individual fixed cold chain nodes, construct a basic cold chain network that reflects the layout and interconnection relationships of cold chain storage nodes in space. According to the multi-modal monitoring feature sequences of individual fixed cold chain nodes, deploy the monitoring features for different time periods on the basic cold chain network. That is, at each time node of the basic network, embed the cold chain monitoring data for different time periods respectively to construct the corresponding cold chain network of actual inventory status. By deploying the monitoring data of the cold chain network, it is possible to accurately present the actual inventory status of the cold chain network in each time period on the time axis, and the real-time deployment of multi-modal monitoring data helps cold chain managers monitor the inventory status in different time periods and ensure the accuracy and timeliness of inventory data.

[0091] More specifically, according to the multi-modal monitoring feature sequences of individual mobile cold chain nodes, construct dynamic node links for the basic cold chain network to form a "cold chain network of transportation status" corresponding to each time period. This process aims to track in real time and dynamically construct the connections between mobile cold chain nodes (such as cold chain vehicles), thereby reflecting the cargo transfer paths in the cold chain logistics process, enabling dynamic tracking of the cold chain logistics transportation paths, providing real-time logistics status and path optimization information for cold chain managers, and real-time monitoring and dynamic construction of the transportation network can promptly respond to and adjust cold chain transportation plans and optimize distribution routes.

[0092] More specifically, combine the information of the "cold chain network of actual inventory status" and the "cold chain network of transportation status" for each time period. The combined cold chain comprehensive actual situation network is arranged in sequence according to the development of time, reflecting the overall actual situation of the cold chain network from storage to transportation. By combining inventory information and logistics path information, a comprehensive cold chain network actual situation is formed for subsequent analysis and evaluation. By integrating storage and transportation information, it is possible to provide a comprehensive cold chain logistics status map, enhancing the overall visualization and transparency of cold chain management, helping cold chain managers conduct comprehensive evaluations of the entire network in different time periods and make real-time decisions.

[0093] More specifically, based on the comprehensive real-time network of the cold chain, the goods in the fixed cold chain nodes (cold chain warehouses) are traced to obtain the traceability characteristics of the cold chain goods. These characteristics include the temperature and humidity history, storage time, location and other information of the goods. Based on the comprehensive real-time network of the cold chain, the mobile trajectory of the mobile cold chain nodes (such as transport vehicles) is tracked, and the transportation trajectory characteristics of the cold chain vehicles (such as transportation path, residence time, temperature control status, etc.) are extracted. Through tracing and trajectory tracking, the entire process of cold chain goods from warehousing to transportation can be understood to ensure the quality and safety of the goods in the entire link. By tracing the cold chain goods and the cold chain transportation trajectory, managers can realize full monitoring and backtracking of the cold chain goods, ensure the traceability and transparency of the cold chain transportation, and monitor the key factors such as the temperature and humidity of the goods in real time during the transportation process to ensure the quality and safety of the cold chain goods.

[0094] More specifically, the cold chain cargo traceability characteristics and cold chain transportation trajectory characteristics are respectively assigned to the corresponding cold chain nodes, the cold chain cargo traceability characteristics are assigned to fixed cold chain nodes, and the cold chain transportation trajectory characteristics are assigned to mobile cold chain nodes. The various cold chain comprehensive real-time networks are arranged in chronological order to ensure that the information in the cold chain network is continuously updated and improved over time. Through assignment and arrangement, the cold chain comprehensive real-time network can not only display the real-time status of each node, but also reflect its dynamic changes in the entire cold chain network. By assigning traceability characteristics and trajectory characteristics to the comprehensive real-time network, cold chain managers can conduct time-based dynamic cold chain evaluation, timely grasp the cold chain transportation status, and form a complete cold chain network evaluation feature distribution, which can help cold chain managers to carry out refined management and optimization at multiple time points and different links.

[0095] It is understandable that the combination of inventory management, cold chain logistics transportation routes and cargo quality traceability has achieved comprehensive real-time monitoring of the cold chain network, ensuring that every link in the cold chain transportation process can be accurately tracked and evaluated. Through high-dimensional feature analysis, a comprehensive evaluation of the status of cold chain nodes in each time period is conducted, which is conducive to real-time optimization and adjustment of cold chain logistics, improving logistics efficiency and resource utilization. Through the traceability of cold chain goods and the tracking of cold chain transportation trajectories, the quality and safety of goods are ensured, and at the same time, a more accurate feature distribution is provided for the evaluation of the cold chain network, further enhancing the transparency and traceability of cold chain management. The comprehensive evaluation feature distribution of the cold chain network provides intelligent decision-making support for cold chain managers, helping to optimize the cold chain network, risk warning and decision-making guidance, thereby improving the efficiency and reliability of cold chain logistics.

[0096] Preferably, the step of extracting local anomalies and overall anomalies from the cold chain network evaluation feature distribution according to the spatiotemporal graph convolutional network to generate an anomaly probability information set of the cold chain network includes: S41: Taking the cold-chain goods participating in the cold-chain network transportation as the benchmark object, perform path construction processing on the fixed cold-chain nodes and mobile cold-chain nodes for the cold-chain network evaluation feature distribution to obtain the cold-chain storage and transportation path of the cold-chain goods; S42: Retrieve the features related to the benchmark object for the cold-chain storage and transportation path according to the cold-chain network evaluation feature distribution and assign them to the cold-chain storage and transportation path. Analyze the correlation tightness of each cold-chain node in the cold-chain network evaluation feature distribution based on all the cold-chain storage and transportation paths, so as to set the detection target based on the correlation tightness for the cold-chain network evaluation feature distribution, and divide the cold-chain network evaluation feature distribution into several local networks; S43: Perform temporal convolution and anomaly recognition on the storage environment data and transportation path data of each local network according to the pre-trained spatio-temporal graph convolutional network to obtain the anomaly probability information composed of the suspected anomaly features and corresponding anomaly probability values of each local network on the storage environment data and transportation path data; S44: Analyze the overall distribution relationship of each local network according to the cold-chain network evaluation feature distribution to perform high-level information clustering processing on the anomaly probability information of each local network. Verify and adjust the anomaly probability information according to the information clustering result, and repeatedly perform high-level information clustering processing until the anomaly probability information of each cold-chain node in the global network is obtained as the anomaly probability information set of the cold-chain network.

[0097] Specifically, selecting the cold-chain goods participating in the cold-chain network transportation as the benchmark object means that the path, storage environment and its changes of the goods in the cold-chain will be the core of the analysis. Based on the cold-chain network evaluation feature distribution, construct the cold-chain storage and transportation path of the cold-chain goods. This path includes the path relationship between the fixed cold-chain nodes (warehouses) and mobile cold-chain nodes (transport vehicles), reflecting the flow of cold-chain goods from the warehouse to the transport vehicle and then to the target location. By clarifying the transportation path of the cold-chain goods, the flow process of the goods in the cold-chain network can be clearly understood, providing a specific path basis for subsequent anomaly detection. Selecting the cold-chain goods as the benchmark object can achieve data-driven path analysis and facilitate the precise positioning of potential problems in the cold-chain network.

[0098] More specifically, according to the characteristic distribution of the cold chain network assessment, characteristic information related to the cold storage and transportation path of cold chain goods is retrieved, including temperature and humidity, location, transportation time, etc. These characteristics help to comprehensively understand the state of cold chain goods during the entire transportation process. Based on the extracted characteristic information, the correlation tightness of each cold chain node in the cold chain network is analyzed, that is, the interaction and dependence relationship between each node, especially the connection between the storage node and the transportation node. This analysis helps to identify the structure and potential anomalies of each local network in the cold chain network. By analyzing the correlation between nodes, the interaction mechanism between each node in the cold chain network can be better understood, providing a basis for subsequent anomaly detection. Through correlation analysis, it is possible to identify whether there are anomalies or the possibility of anomalies in each node of the cold chain network, improving the accuracy of anomaly detection.

[0099] More specifically, according to the correlation tightness of cold chain nodes, detection targets are set. These detection targets represent the key nodes and important links in the cold chain network, focusing on the areas where anomalies may occur. The cold chain network assessment characteristic distribution is divided into several local networks according to the set detection targets. Each local network represents a subgraph composed of nodes and paths, which may contain cold chain logistics information in certain specific areas. Dividing the entire network into multiple local networks helps to focus on the anomalies in local areas, improving the accuracy of anomaly detection. The local network division can significantly reduce the computational amount, making the subsequent anomaly detection more efficient.

[0100] More specifically, a pre-trained spatio-temporal graph convolutional network is used to analyze each local network, especially performing temporal convolution processing on the storage environment data and transportation path data. Through spatio-temporal convolution, the network can consider the combined influence of time series data and spatial location data, thereby more accurately identifying anomalies. During the convolution process, the model will identify the abnormal characteristics in the data, especially abnormal situations such as temperature and humidity changes in the storage environment and delays in the transportation path. The spatio-temporal convolutional network generates suspected abnormal characteristics of each local network on the storage environment data and transportation path data, and generates an anomaly probability value based on these characteristics. The spatio-temporal graph convolutional network can process data synchronously in time and space, considering the temporal and spatial dependencies of the cold chain network, thereby effectively identifying the abnormal patterns in the cold chain network. Combining the storage environment data and transportation path data, the model can comprehensively identify the anomalies in the cold chain network, especially multi-dimensional factors such as temperature and humidity and path anomalies.

[0101] More specifically, based on the evaluation of the feature distribution of the cold chain network, the overall distribution relationship of each local network is analyzed. This step helps to understand the interactions and associations between different local networks, and performs high-level information clustering on the abnormal probability information of each local network. Clustering analysis helps to discover the common features of abnormal patterns and trends, and generates multiple clustering results. According to the information clustering results, the abnormal probability information is interactively verified and adjusted to optimize the clustering model and the abnormal recognition process, ensuring the accuracy of abnormal detection. The high-level information clustering process is repeated until stable abnormal probability information is obtained. Through the clustering process, potential abnormal patterns in different local networks can be identified, and combined with their overall distribution, the abnormal probability information in the global cold chain network can be obtained. Interactive verification and adjustment can continuously optimize the abnormal probability information, improve the adaptive ability and accuracy of the model. Through high-level analysis and clustering, the precision of cold chain management can be effectively improved, helping managers to identify problems in the cold chain network in a timely manner and make corresponding adjustments.

[0102] More specifically, finally, through the interactive verification of the high-level information clustering results and abnormal probabilities, an abnormal probability information set for each cold chain node in the cold chain network is generated. These information sets contain the abnormal states of each node in the cold chain network and provide a basis for subsequent cold chain management and optimization. The abnormal probability information set provides a global assessment for the abnormal detection of the cold chain network, helping managers to comprehensively understand the operating conditions of the cold chain system. Through the abnormal probability information set, potential risks in the cold chain network can be identified in advance, and effective preventive measures can be taken.

[0103] It can be understood that through the joint analysis of time and space, the spatio-temporal graph convolutional network can achieve precise detection of cold chain network anomalies, especially in the identification of anomalies in warehouse environments and transportation path data. Through local network division and high-level information clustering, the computational complexity can be effectively reduced. At the same time, through interactive verification and adjustment, the abnormal detection model is optimized, and the accuracy of abnormal detection is improved. The generated abnormal probability information set provides the abnormal state information of each node in the cold chain network, providing a complete abnormal detection report for cold chain managers, which helps to monitor the cold chain in real time and adjust the cold chain transportation plan in a timely manner. Through accurate abnormal detection and real-time early warning, the transparency of cold chain logistics management can be improved, and the manager's ability to respond to potential problems and decision-making efficiency can be enhanced.

[0104] Preferably, the step of feeding back the abnormal probability information set to the cold chain network evaluation feature distribution to deduce the effects of multiple strategies on the abnormal probability information set and generate abnormal event prediction data includes: S51: Analyze the abnormal probability information set to feed each abnormal probability information back to the corresponding cold chain node in the cold chain network evaluation feature distribution; S52: Successively use each cold chain node in the cold chain network evaluation feature distribution as an event deduction object according to the abnormal probability information, and retrieve each abnormal probability information and the corresponding abnormal event record of the event deduction object in the past time period to evaluate the safety bearing capacity of the event deduction object, so as to obtain the safety bearing capacity information of the event deduction object; S53: Configure several risk control strategies for the event deduction object to predict potential risk events based on the abnormal probability information for the event deduction object according to various risk control strategies, and at the same time evaluate the severity of the events for the prediction of potential risk events based on the safety bearing capacity information, so as to obtain the abnormal event prediction data of the event deduction object corresponding to various risk control strategies.

[0105] Specifically, parse the abnormal probability information set generated by the spatio-temporal graph convolutional network. This set contains the abnormal probabilities of each node in the cold chain network, representing the potential abnormal probabilities of each node in the cold chain network. Feed back each abnormal probability information to the cold chain nodes in the corresponding cold chain network evaluation feature distribution. Through feedback, abnormal information can be embedded into the original data structure of the cold chain network, providing support for subsequent event deduction and risk assessment. By combining the abnormal probability information with the nodes in the cold chain network evaluation feature distribution, the abnormal detection results can be better integrated into the network evaluation model, improving the relevance and consistency of the data. The feedback of the abnormal probability information helps to provide real-time data for the cold chain management system, enabling managers to quickly identify potential abnormal areas in the network and providing accurate input data for subsequent deduction.

[0106] More specifically, according to the abnormal probability information set, successively select each cold chain node in the cold chain network evaluation feature distribution as an event deduction object. These nodes represent the key links in the cold chain network. For each event deduction object, retrieve its abnormal probability information and the corresponding abnormal event record in the past time period. These historical data provide the background information of the abnormal events that occurred at this node in the past period, helping to predict future potential risks. By retrieving the historical abnormal event records, it is possible to more accurately infer the possible abnormalities of the event deduction object in the future, especially the abnormal trends under specific historical patterns. Combining the historical abnormal data with the real-time feedback abnormal probability information can effectively improve the accuracy and foresight of event prediction.

[0107] More specifically, based on the retrieved historical data, the safety bearing capacity of each event deduction object is evaluated. The safety bearing capacity refers to the response measures that a cold chain node can take, as well as its ability and resources to respond to anomalies when faced with abnormal events. This evaluation can consider factors such as the handling ability, response speed, and facility capacity of historical abnormal events, so as to obtain the safety bearing capacity information of each cold chain node. The evaluation of the safety bearing capacity enables each node in the cold chain network to have specific emergency response ability data, helping managers understand which nodes can handle events quickly and effectively and which nodes may face greater pressure in the event of an anomaly. The evaluation of the safety bearing capacity provides real-time decision-making support for managers, helping them allocate resources and prioritize the handling of emergency risk events to avoid system collapse or losses.

[0108] More specifically, several risk control strategies are configured for each event deduction object. Different cold chain nodes may face different types of risks, so the configured risk control strategies should be customized according to the specific situation of the nodes. These strategies include, but are not limited to, temperature and humidity monitoring, real-time monitoring during transportation, accident response measures, etc. Based on the abnormal probability information of each event deduction object and in combination with the configured risk control strategies, potential risk event prediction is carried out. This step will provide managers with early warnings of future potential risks by analyzing possible abnormal events and predicting their occurrence probabilities. At the same time, based on the safety bearing capacity information, the severity of potential risk events is evaluated. This step will judge the severity of the predicted events according to the emergency handling ability of the nodes, and then decide whether emergency measures need to be taken. By combining the abnormal probability information with the risk control strategies, accurate prediction of potential risk events can be achieved, identifying key problems that may occur in the cold chain network in advance, helping managers take preventive measures in advance. The application of multiple risk control strategies can effectively manage potential risks in the cold chain network, improve the response ability of cold chain logistics. At the same time, based on the evaluation of the event severity, resources can be reasonably allocated to avoid over-response or resource waste.

[0109] More specifically, based on the prediction results of each event deduction object under different risk control strategies, final abnormal event prediction data is generated. These data include the types of abnormal events that may occur at each node in the cold chain network under specific strategies and their occurrence probabilities, for managers' reference. By generating abnormal event prediction data, managers can understand the potential risks in the cold chain network in real time and make scientific prevention and response decisions. By continuously deducing the risk predictions under different strategies, managers can flexibly adjust the cold chain transportation and storage plans to avoid losses caused by unexpected events.

[0110] It is understandable that by combining abnormal probability information and historical data, through safety bearing capacity assessment and risk control strategies, abnormal events that may occur in the cold chain network can be predicted in advance, and their severity can be evaluated. Different risk control strategies can be configured for different nodes, and appropriate countermeasures can be selected according to the characteristics of the nodes and the risks they face, improving the flexibility and adaptability of cold chain management. Based on the generation of abnormal event prediction data, managers of the cold chain network can obtain clear early warnings and decision-making suggestions, helping them to conduct resource scheduling and emergency response more efficiently. The combination of safety bearing capacity and abnormal events improves the accuracy of risk management, reduces the possibility of over-management or neglect of risks, and thus enhances the overall efficiency and safety of the cold chain network.

[0111] Preferably, the steps of performing diffusion simulation of the impact of the abnormal event on the abnormal event prediction data based on the cold chain network evaluation feature distribution and correcting the abnormal event prediction data according to the diffusion simulation result to obtain abnormal event early warning information include: S61: Perform correlation analysis of the remaining cold chain nodes on the event deduction object according to the cold chain network evaluation feature distribution to obtain an impact diffusion map of the event deduction object relative to the cold chain network evaluation feature distribution; S62: Perform impact diffusion simulation of the control strategies on the various risk control strategies configured for the event deduction object according to the impact diffusion map to obtain impact feedback tendency information of the various risk control strategies of the remaining cold chain nodes in the cold chain network evaluation feature distribution corresponding to the event deduction object; S63: Cycle through each cold chain node in the cold chain network evaluation feature distribution as the event deduction object to generate impact feedback information of the various risk control strategies of each cold chain node corresponding to each event deduction object, and use the impact feedback information of the various risk control strategies of each cold chain node corresponding to each event deduction object as the overall impact information of the cold chain node; S64: Perform positive value analysis of the strategy synergy and negative value analysis of the strategy conflict of the risk control strategies of each event deduction object, and perform impact simulation of the optimal implementation of the risk control strategy on the overall impact information of each cold chain node based on the results of the two-way value analysis to obtain abnormal event early warning information of each cold chain node.

[0112] Specifically, based on the evaluation of the feature distribution of the cold chain network, the relevance between each event deduction object and other cold chain nodes is analyzed. This analysis aims to find out the relationship between the event deduction object and the surrounding nodes, especially the potential impact of abnormal events on other nodes. Through the relevance analysis, an influence diffusion map of the event deduction object is generated. This map shows the propagation path and degree of the abnormal event from the event deduction object to other cold chain nodes. This map helps to identify key nodes and potential patterns of abnormal event diffusion. Through the relevance analysis and the generation of the influence diffusion map, it is possible to clearly identify which nodes in the cold chain network may be directly affected by abnormal events and the possible chains of abnormal propagation. The topological structure of the cold chain network and the interaction relationship between nodes are clarified, providing accurate input data for subsequent diffusion simulation.

[0113] More specifically, based on the influence diffusion map, for each event deduction object, a diffusion simulation of risk control strategies is carried out separately. This step mainly simulates how abnormal events spread to other cold chain nodes under different control strategies and evaluates the impact of the strategies. The feedback information obtained through the diffusion simulation can reflect the potential impact of each control strategy on other nodes and analyze its positive or negative feedback. This feedback information helps to further understand the abnormal event response ability of each cold chain node. Through the diffusion simulation, different control strategies will affect the propagation of abnormal events in the network, thereby helping decision-makers identify which strategies can more effectively control risks and slow down or contain the spread of abnormal events. Through the feedback tendency information, the overall impact of each strategy on the cold chain network during actual implementation can be evaluated, helping to optimize control measures.

[0114] More specifically, a cyclic deduction is performed on each cold chain node in the evaluation feature distribution of the cold chain network. Each deduction takes the cold chain node as the event deduction object. According to the previous diffusion simulation results, feedback information on the impact of control strategies corresponding to each node is generated. The feedback information of each node is integrated to form the overall impact information of each node. These information show the possible impact of each node in the cold chain network on the entire network under various risk control strategies. Through the cyclic deduction, it is possible to comprehensively evaluate the response of each node in the cold chain network to abnormal events and its impact on the entire network, ensuring that the performance and impact of each node are accurately recorded. By integrating the impact information of each node, it helps to optimize the risk control strategies in the cold chain network from a global perspective, avoid the deficiencies of local optimization, and ensure the stability of the entire system.

[0115] More specifically, for the positive value analysis of strategy synergy, conduct a positive value analysis on the risk control strategies of each cold chain node, evaluate the synergy between different strategies, analyze how each strategy cooperates with each other to produce positive effects, and for the reverse value analysis of strategy conflict, at the same time, conduct a reverse value analysis of strategy conflict, evaluate whether there are conflicts between different control strategies and their possible negative impacts, and avoid poor effects caused by mutual interference between strategies. Through the analysis of strategy synergy, the combination of risk control strategies can be optimized, and the comprehensive benefits of the strategies can be improved. On the contrary, through reverse analysis, possible strategy conflicts can be identified and avoided in a timely manner, ensuring the smooth execution of strategies. The analysis of synergy and conflict provides a scientific basis for strategy selection for cold chain managers and helps to implement the best control measures.

[0116] More specifically, based on the results of the analysis of strategy synergy and conflict, simulate the optimal implementation of risk control strategies for the overall impact information of each cold chain node. Through this simulation, the most effective control plan can be identified. Finally, based on the optimal simulation results, generate early warning information for abnormal events at each cold chain node. These early warning information include possible abnormal events, the probability of event occurrence, and recommended control strategies. The early warning information for abnormal events generated through the optimal implementation simulation can accurately identify potential abnormal events and their impacts, thus providing timely warnings for managers. The generated early warning information will help managers take appropriate measures to reduce the occurrence and spread of abnormal events, thereby improving the stability and reliability of the cold chain network.

[0117] It can be understood that by analyzing the relevance between the event deduction object and other cold chain nodes, the propagation path of abnormal events in the cold chain network can be accurately identified, providing a basis for the subsequent implementation of risk control strategies. By simulating the diffusion effects of different risk control strategies, the impact of various strategies on network stability can be evaluated in advance, providing scientific guidance for strategy optimization. Through the analysis of strategy synergy and conflict, the strategy combination can be optimized, strategy conflicts can be avoided, and various control strategies in the cold chain network can be ensured to cooperate with each other to achieve the best effect. Based on the overall impact information and the optimal simulation results, the generated early warning information for abnormal events can help managers take countermeasures in a timely manner, effectively preventing potential threats of abnormal events to the cold chain network. The final early warning information for abnormal events helps to strengthen the risk management ability of the cold chain network, reduce the occurrence frequency of abnormal events, and ensure that the cold chain system can maintain efficient and stable operation under different risk situations.

[0118] Refer to Figure 2 As shown, in the second aspect, the present invention provides a cold chain logistics abnormal event early warning system for multi-modal AIGC, which is used to implement the method for early warning of cold chain logistics abnormal events for multi-modal AIGC described in any one of the first aspects, including: A data monitoring module, which is used to collect multimodal data from each cold chain node of the cold chain network through a pre-deployed Internet of Things sensing system, so as to obtain the original monitoring data of each cold chain node; A data analysis module, which is used to perform multimodal semantic space mapping of its own form and interaction form on the original monitoring data of the cold chain node through a pre-trained cross-modal alignment algorithm, so as to obtain the multimodal monitoring feature sequence of the cold chain node; An overall evaluation module, which is used to perform high-dimensional feature analysis of cold chain storage goods management and cold chain logistics transportation paths on the multimodal monitoring feature sequences of each of the cold chain nodes, so as to obtain the cold chain network evaluation feature distribution; A probability analysis module, which is used to extract local and global abnormal features from the cold chain network evaluation feature distribution according to a spatio-temporal graph convolutional network, so as to generate an abnormal probability information set of the cold chain network; A strategy deduction module, which is used to feedback the abnormal probability information set to the cold chain network evaluation feature distribution, so as to deduce the effects of multiple strategies on the abnormal probability information set and generate abnormal event prediction data; An event warning module, which is used to perform diffusion simulation of the impact of the abnormal event on the abnormal event prediction data based on the cold chain network evaluation feature distribution, and correct the abnormal event prediction data according to the diffusion simulation result to obtain abnormal event warning information.

[0119] In this embodiment, for the specific implementation of each module in the above system embodiment, please refer to that described in the above method embodiment, and details are not described herein again.

[0120] The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A warning method for abnormal events in cold chain logistics of multi-modal AIGC, characterized in that, Including: Performing multi-modal data collection on each cold chain node of the cold chain network through a pre-deployed Internet of Things sensing system to obtain the original monitoring data of each cold chain node; Performing multi-modal semantic space mapping of the original monitoring data of the cold chain node in its own form and interaction form through a pre-trained cross-modal alignment algorithm to obtain the multi-modal monitoring feature sequence of the cold chain node; Performing high-dimensional feature analysis on the multi-modal monitoring feature sequences of each of the cold chain nodes for cold chain storage goods management and cold chain logistics transportation path to obtain the cold chain network evaluation feature distribution; Performing feature extraction of local anomalies and global anomalies on the cold chain network evaluation feature distribution according to the spatio-temporal graph convolutional network to generate an abnormal probability information set of the cold chain network; Feeding back the abnormal probability information set to the cold chain network evaluation feature distribution to perform multi-strategy effect deduction on the abnormal probability information set and generate abnormal event prediction data; Performing diffusion simulation of the impact of the abnormal event on the abnormal event prediction data based on the cold chain network evaluation feature distribution, and correcting the abnormal event prediction data according to the diffusion simulation result to obtain abnormal event warning information.

2. The cold chain logistics abnormal event warning method for multi-modal AIGC according to claim 1, wherein, The steps of performing multi-modal data collection on each cold chain node of the cold chain network through a pre-deployed Internet of Things sensing system to obtain the original monitoring data of each cold chain node include: Performing clock synchronization processing on the sensor modules of each cold chain node pre-deployed in the cold chain network to generate a data recording axis for the sensor modules of the Internet of Things sensing system deployed on each cold chain node; Performing data collection on the cold chain node through several sensor modules with different functions deployed on the cold chain node, and generating and marking time stamps for the sensing data collected by the sensor modules of each function based on the data recording axis to obtain the original monitoring data in which the sensing data with time stamp marks are arranged in sequence according to the time stamp marks; When the fixed cold chain node representing the cold chain warehouse and the mobile cold chain node representing the cold chain vehicle perform goods handover, performing monitoring data exchange and data recording axis synchronization of the original monitoring data on the fixed cold chain node and the mobile cold chain node participating in the goods handover to update the data of the original monitoring data of each cold chain node.

3. The cold chain logistics abnormal event warning method for multi-modal AIGC according to claim 1, characterized in that, The steps of performing multi-modal semantic space mapping of the original monitoring data of the cold chain node in its own form and interaction form through a pre-trained cross-modal alignment algorithm to obtain the multi-modal monitoring feature sequence of the cold chain node include: Judging the monitoring form of the original monitoring data of each cold chain node, and at the same time performing mutual verification and supervision according to the judgment results of the mutually related cold chain nodes to divide the original monitoring data of each cold chain node into its own form monitoring part and interaction form monitoring part; wherein, the own form monitoring part is used to describe the time period when the fixed cold chain node representing the cold chain warehouse and the mobile cold chain node representing the cold chain vehicle do not perform goods handover, and the interaction form monitoring part is used to describe the time period when the fixed cold chain node representing the cold chain warehouse and the mobile cold chain node representing the cold chain vehicle perform goods handover. Perform multimodal semantic space mapping on the self-form monitoring part of the cold chain node through a pre-trained cross-modal alignment algorithm to obtain the multimodal monitoring features of the self-form monitoring part of the cold chain node; Perform multimodal semantic space mapping on the interaction-form monitoring part of the cold chain node through a pre-trained cross-modal alignment algorithm to obtain the multimodal monitoring features of the interaction-form monitoring part of the cold chain node; Perform temporal arrangement processing on the multimodal monitoring features of the self-form monitoring part and the multimodal monitoring features of the interaction-form monitoring part to obtain the multimodal monitoring feature sequence of the cold chain node.

4. The cold chain logistics abnormal event warning method for multi-modal AIGC according to claim 3, wherein, The steps of performing multimodal semantic space mapping on the self-form monitoring part of the cold chain node through a pre-trained cross-modal alignment algorithm to obtain the multimodal monitoring features of the self-form monitoring part of the cold chain node include: Obtain the working environment information of the cold chain node as the analysis object and the monitoring mode information of the configured sensor module, and perform adaptive algorithm parameter deployment on the pre-trained cross-modal alignment algorithm according to the working environment information and the monitoring mode information, so that the cross-modal alignment algorithm adapts to the cold chain node as the analysis object; Perform unimodal feature analysis on the self-form monitoring part of the cold chain node according to the cross-modal alignment algorithm, and map the feedback information of each modality to a specified cross-modal semantic space; wherein, the feedback information includes several possible information features and corresponding feature confidence factors; Perform cross-modal sample alignment on the feedback information of each modality mapped to the cross-modal semantic space, so that the feedback information of each modality is in an aligned state marked by timestamps; Perform multimodal collaborative verification based on feature confidence factors on several possible information features of each modality according to the cross-modal alignment algorithm, and at the same time perform semantic reinforcement constraints on the multimodal collaborative verification based on the pre-constructed cold chain transportation knowledge graph to obtain the multimodal monitoring features of the self-form monitoring part.

5. The cold chain logistics abnormal event warning method for multi-modal AIGC according to claim 3, characterized in that, The steps of performing multimodal semantic space mapping on the interaction-form monitoring part of the cold chain node through a pre-trained cross-modal alignment algorithm to obtain the multimodal monitoring features of the interaction-form monitoring part of the cold chain node include: Mark the interaction-form monitoring part of the cold chain node as the reference part, and mark the interaction-form monitoring part of another cold chain node that conducts goods handover with the reference part as the interaction part; Process the reference part and the interaction part respectively according to the multimodal semantic space mapping method corresponding to the self-form monitoring part to obtain the first monitoring feature corresponding to the reference part and the second monitoring feature corresponding to the interaction part; Perform timestamp alignment processing on the first monitoring feature and the second monitoring feature, and perform information change analysis and consistency verification on the monitoring feedback content of the first monitoring feature and the second monitoring feature to obtain the cold chain goods transfer process information corresponding to the reference part. Obtain the working environment information of the cold chain nodes corresponding to the reference part and the cold chain nodes corresponding to the interaction part, and perform feature encoding of the spatio-temporal structure of the cold chain nodes on the working environment information. Based on the spatio-temporal structure feature encoding of the two cold chain nodes participating in the interaction, perform digital simulation of the key steps in the execution process of the cold chain cargo transfer process information, so as to extract key features of the cold chain cargo transfer process information according to the digital simulation results, and obtain the key features of cold chain cargo transfer; Combine the key features of the cold chain cargo transfer with the first monitoring features to obtain the multi-modal monitoring features of the interaction form monitoring part of the cold chain nodes.

6. The cold chain logistics abnormal event warning method for multi-modal AIGC according to claim 1, characterized in that, The steps of performing high-dimensional feature analysis on the multi-modal monitoring feature sequences of each cold chain node for cold chain storage cargo management and cold chain logistics transportation paths to obtain the cold chain network evaluation feature distribution include: Identify the functional uses of the cold chain nodes to classify the cold chain nodes into fixed cold chain nodes representing cold chain warehouses and mobile cold chain nodes representing cold chain vehicles; Construct a basic cold chain network according to the relative position relationship between each fixed cold chain node, and at the same time deploy monitoring features for each time period on the basic cold chain network according to the multi-modal monitoring sequences of each fixed cold chain node to obtain the inventory actual cold chain network corresponding to each time period; Dynamically construct node links of the basic cold chain network according to the multi-modal monitoring feature sequences of each mobile cold chain node to obtain the transportation actual cold chain network corresponding to each time period; Combine the information of the inventory actual cold chain network and the transportation actual cold chain network of each segment to obtain several cold chain comprehensive actual networks arranged in sequence according to time development; Based on the cold chain comprehensive actual network, perform traceability processing on the cold chain goods of each fixed cold chain node to obtain the cold chain goods traceability features of each fixed cold chain node; Based on the cold chain comprehensive actual network, perform tracking processing on the movement trajectories of each mobile cold chain node to obtain the cold chain transportation trajectory features of each mobile cold chain node; Assign the cold chain goods traceability features and the cold chain transportation trajectory features to the corresponding cold chain nodes on the cold chain comprehensive actual network, and the cold chain comprehensive actual networks arranged in sequence according to time order together constitute the cold chain network evaluation feature distribution.

7. The cold chain logistics abnormal event warning method for multi-modal AIGC according to claim 6, wherein, The steps of extracting local and global anomaly features from the cold chain network evaluation feature distribution according to the spatio-temporal graph convolutional network to generate an anomaly probability information set of the cold chain network include: Taking the cold chain goods participating in the cold chain network transportation as the reference object, perform path construction processing on the cold chain network evaluation feature distribution for fixed cold chain nodes and mobile cold chain nodes to obtain the cold chain storage transportation path of the cold chain goods; Retrieve the relevant features of the benchmark object for the cold chain storage and transportation path according to the cold chain network evaluation feature distribution, and assign them to the cold chain storage and transportation path. Analyze the association tightness of each cold chain node in the cold chain network evaluation feature distribution based on all the cold chain storage and transportation paths, so as to set the detection target based on the association tightness for the cold chain network evaluation feature distribution, and divide the cold chain network evaluation feature distribution into several local networks; Perform temporal convolution and anomaly recognition on the storage environment data and transportation path data of each local network according to the pre-trained spatio-temporal graph convolutional network, so as to obtain the anomaly probability information composed of the suspected anomaly features and the corresponding anomaly probability values of each local network on the storage environment data and transportation path data; Analyze the overall distribution relationship of each local network according to the cold chain network evaluation feature distribution, so as to perform information clustering processing at a high level on the anomaly probability information of each local network. Verify and adjust the anomaly probability information according to the information clustering result, and repeatedly perform information clustering processing at a high level until the anomaly probability information of each cold chain node in the global network is obtained, so as to serve as the anomaly probability information set of the cold chain network.

8. The cold chain logistics abnormal event warning method for multi-modal AIGC according to claim 1, characterized in that, The steps of feeding back the anomaly probability information set to the cold chain network evaluation feature distribution to perform multi-strategy effect deduction on the anomaly probability information set and generating anomaly event prediction data include: Analyze the anomaly probability information set to feed back each anomaly probability information to the corresponding cold chain node in the cold chain network evaluation feature distribution; Take each cold chain node in the cold chain network evaluation feature distribution as an event deduction object in turn according to the anomaly probability information, and retrieve each anomaly probability information and the corresponding anomaly event record of the event deduction object in the past time period, so as to evaluate the safety bearing capacity of the event deduction object, so as to obtain the safety bearing capacity information of the event deduction object; Configure several risk control strategies for the event deduction object, so as to predict potential risk events based on the anomaly probability information for the event deduction object according to various risk control strategies, and at the same time evaluate the severity of the potential risk events based on the safety bearing capacity information, so as to obtain the anomaly event prediction data of the event deduction object corresponding to various risk control strategies.

9. The cold chain logistics abnormal event warning method for multi-modal AIGC according to claim 8, characterized in that, The steps of performing diffusion simulation on the impact of the anomaly event prediction data based on the cold chain network evaluation feature distribution and correcting the anomaly event prediction data according to the diffusion simulation result to obtain the anomaly event warning information include: Perform relevance analysis of the remaining cold chain nodes on the event deduction object according to the cold chain network evaluation feature distribution, so as to obtain the impact diffusion map of the event deduction object relative to the cold chain network evaluation feature distribution; For each of the various risk control strategies configured for the event deduction object according to the influence diffusion map, perform an influence diffusion simulation of the control strategy to obtain the influence feedback tendency information of each of the remaining cold chain nodes in the cold chain network evaluation feature distribution corresponding to the various risk control strategies of the event deduction object; Loop through each cold chain node in the cold chain network evaluation feature distribution as an event deduction object to generate the influence feedback information of each cold chain node corresponding to the various risk control strategies of each event deduction object, and use the influence feedback information of each cold chain node corresponding to the various risk control strategies of each event deduction object as the overall influence information of the cold chain node; Conduct a positive value analysis of the strategy synergy and a negative value analysis of the strategy conflict for the risk control strategies of each event deduction object, and based on the results of the two-way value analysis, perform an influence simulation of the optimal implementation of the risk control strategy on the overall influence information of each cold chain node to obtain the abnormal event warning information of each cold chain node.

10. A cold chain logistics abnormal event warning system for multi-modal AIGC, characterized in that, A cold chain logistics abnormal event warning method for implementing the multi-modal AIGC according to any one of claims 1-9, comprising: A data monitoring module, configured to collect multi-modal data of each cold chain node in the cold chain network through a pre-deployed Internet of Things sensing system to obtain the original monitoring data of each cold chain node; A data analysis module, configured to perform a multi-modal semantic space mapping of the original monitoring data of the cold chain node in its own form and interaction form through a pre-trained cross-modal alignment algorithm to obtain the multi-modal monitoring feature sequence of the cold chain node; An overall evaluation module, configured to perform a high-dimensional feature analysis of the cold chain storage goods management and the cold chain logistics transportation path on the multi-modal monitoring feature sequence of each cold chain node to obtain the cold chain network evaluation feature distribution; A probability analysis module, configured to extract local and global abnormal features from the cold chain network evaluation feature distribution according to the spatio-temporal graph convolutional network to generate an abnormal probability information set of the cold chain network; A strategy deduction module, configured to feedback the abnormal probability information set to the cold chain network evaluation feature distribution to perform a multi-strategy effect deduction on the abnormal probability information set to generate abnormal event prediction data; An event warning module, configured to perform a diffusion simulation of the influence of the abnormal event on the abnormal event prediction data based on the cold chain network evaluation feature distribution, and correct the abnormal event prediction data according to the diffusion simulation result to obtain the abnormal event warning information.

Images