Data analysis method and system based on cold chain transportation

The data analysis method integrates logistics and environmental data through a risk prediction network, improving risk identification accuracy and response speed in pharmaceutical cold chain transportation.

CN118761698BActive Publication Date: 2024-12-06CHENGDU YISU LOGISTICS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411239049.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-12-06
Estimated Expiration
2044-09-05

AI Technical Summary

Technical Problem

Existing pharmaceutical cold chain transportation management relies heavily on manual monitoring and empirical judgment, lacking real-time data analysis and failing to integrate logistics and environmental data effectively, leading to inaccurate risk predictions.

Method used

A data analysis method and system that utilizes a cold chain transportation risk prediction network with feature extraction and prediction subnetworks to integrate logistics and environmental data, generating vector sets for correlation analysis and risk knowledge point identification.

Benefits of technology

Enhances the accuracy and speed of risk identification in pharmaceutical cold chain transportation by efficiently extracting and correlating logistics and environmental data, providing comprehensive risk assessment and prompt decision-making support.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118761698B_ABST
    Figure CN118761698B_ABST
Patent Text Reader

Abstract

The embodiment of the present application provides a data analysis method and system based on cold chain transportation. By constructing a dual network architecture, a target cold chain transportation risk prediction network including a feature extraction subnetwork and a feature prediction subnetwork is implemented to realize efficient feature extraction and integration of the entire chain data of cold chain transportation, including logistics data and its corresponding transportation environment change data, thereby generating a target vector set that characterizes logistics status and environmental changes. By using this target vector set and risk knowledge vector set for correlation analysis, a correlation weight set between each risk knowledge point and cold chain transportation logistics data can be accurately generated, thereby marking the target risk knowledge points for the cold chain transportation logistics data based on these correlation weights, significantly improving the accuracy and response speed of risk identification during cold chain transportation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of pharmaceutical cold chain services, and more specifically, to a data analysis method and system based on cold chain transportation. Background Art

[0002] In the pharmaceutical cold chain field, it is crucial to ensure the safety and quality of high-value and highly sensitive pharmaceutical products throughout the transportation process. These products are often extremely sensitive to environmental factors such as temperature and humidity. Any slight environmental change may cause them to fail or deteriorate, thus causing serious impact on the life and health of patients. Therefore, pharmaceutical cold chain transportation requires extremely high accuracy and reliability.

[0003] Traditional pharmaceutical cold chain transportation management mainly relies on manual monitoring and empirical judgment, but this method has many limitations. First, manual monitoring is difficult to grasp the environmental changes during transportation in real time and comprehensively, and it is easy to miss inspections or misjudge. Second, empirical judgments are often based on historical data and personal experience, lacking scientificity and objectivity, and are difficult to cope with complex and changing transportation environments.

[0004] With the rapid development of Internet of Things technology, logistics data and environmental change data in the pharmaceutical cold chain transportation process can be collected and recorded in real time, providing a rich basis for data analysis. However, how to effectively use these massive data, extract valuable information, and make accurate risk predictions is still a major challenge facing the pharmaceutical cold chain field.

[0005] Most existing data analysis methods focus on processing a single type of data, such as mining logistics data or analyzing environmental data, ignoring the intrinsic connection between logistics data and environmental data. In addition, these methods often lack customized designs for the characteristics of the pharmaceutical cold chain, which limits the accuracy and reliability of risk prediction. Summary of the invention

[0006] In view of this, the purpose of this application is to provide a data analysis method and system based on cold chain transportation.

[0007] According to a first aspect of the present application, a data analysis method based on cold chain transportation is provided, which is applied to a cold chain monitoring system, and the method comprises:

[0008] Loading the cold chain transportation logistics data and the transportation environment change data corresponding to the cold chain transportation logistics data into the target cold chain transportation risk prediction network; the target cold chain transportation risk prediction network includes a feature extraction subnetwork and a feature prediction subnetwork;

[0009] Utilizing the feature extraction subnetwork in the target cold chain transportation risk prediction network to extract features from the cold chain transportation logistics data and the transportation environment change data, generating a set of logistics knowledge logic vectors corresponding to the cold chain transportation logistics data and a transportation environment change vector corresponding to the transportation environment change data, integrating the logistics knowledge logic vector set and the transportation environment change vector to generate a target vector set;

[0010] A risk knowledge vector set corresponding to X risk knowledge points is obtained, and a correlation analysis is performed on the target vector set and the risk knowledge vector set using the feature prediction subnetwork to generate a correlation weight set corresponding to the risk knowledge vector set; each correlation weight parameter in the correlation weight set represents the correlation between the cold chain transport logistics data and a risk knowledge point; X is a positive integer;

[0011] Based on the relevance weight set, matching scores corresponding to the X risk knowledge points are generated respectively, and based on the matching scores, target risk knowledge points among the X risk knowledge points are marked for the cold chain transportation logistics data.

[0012] According to the second aspect of the present application, a cold chain monitoring system is provided, which includes a processor and a readable storage medium, wherein the readable storage medium stores a program, and when the program is executed by the processor, the aforementioned data analysis method based on cold chain transportation is implemented.

[0013] According to the third aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are monitored to be executed, the aforementioned data analysis method based on cold chain transportation is implemented.

[0014] According to any of the above aspects, the embodiment of the present application realizes efficient feature extraction and integration of cold chain transportation logistics data and its corresponding transportation environment change data by constructing a target cold chain transportation risk prediction network including a feature extraction subnetwork and a feature prediction subnetwork, thereby generating a target vector set that characterizes logistics status and environmental changes. By using this target vector set and the risk knowledge vector set for correlation analysis, a set of correlation weights between each risk knowledge point and the cold chain transportation logistics data can be accurately generated, thereby marking the target risk knowledge points for the cold chain transportation logistics data based on these correlation weights, significantly improving the accuracy and response speed of risk identification during cold chain transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A schematic diagram of a process flow of a data analysis method based on cold chain transportation provided in an embodiment of the present application;

[0017] Figure 2 A schematic diagram of the component structure of a cold chain monitoring system provided in an embodiment of the present application for implementing the above-mentioned data analysis method based on cold chain transportation is shown.

[0018] Figure numbers: 100 - cold chain monitoring system; 111 - memory; 112 - storage controller; 113 - processor; 114 - peripheral interface; 115 - input and output unit; 116 - audio unit; 117 - display unit; 118 - radio frequency unit; 210 - application. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application.

[0020] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0021] Figure 1The flowchart of the data analysis method based on cold chain transportation provided by the embodiment of the present application is shown. It should be understood that in other embodiments, the order of some steps in the data analysis method based on cold chain transportation of this embodiment can be interchanged according to actual needs, or some steps can be omitted or deleted. The detailed steps of the data analysis method based on cold chain transportation are introduced as follows.

[0022] Step S110, loading the cold chain transportation logistics data and the transportation environment change data corresponding to the cold chain transportation logistics data into the target cold chain transportation risk prediction network. The target cold chain transportation risk prediction network includes a feature extraction subnetwork and a feature prediction subnetwork.

[0023] In this embodiment, in the field of pharmaceutical cold chain, the server receives a batch of cold chain transportation logistics data and corresponding transportation environment change data. It is assumed that this batch of data comes from a logistics company specializing in the transportation of various pharmaceutical and biological products (such as blood products, cell preparations, etc.).

[0024] Cold chain transport logistics data may include: detailed information on transport vehicles (such as vehicle model, number, maintenance records), departure and destination (specific to the name of the hospital department or medical research institution), transport route (accurate to the specific road and city passed through), transport time (accurate to the minute), detailed information on each batch of pharmaceutical and biological products (such as product type, specification, batch number, expiration date), real-time temperature monitoring records during transportation (including temperature data at different locations in the carriage), humidity monitoring records, vibration monitoring data, light monitoring data, etc.

[0025] The data on changes in the transportation environment may include: detailed weather conditions in different areas during transportation (such as temperature, rainfall, wind speed, wind direction, haze index), road conditions (such as road flatness, slope, curve angle, degree of traffic congestion, specific road construction conditions), power supply stability (including normal power supply duration, power outage duration, voltage fluctuations in the passing areas), network signal stability (signal strength changes, presence or absence of signal blind spots and duration), etc.

[0026] The server loads these rich and detailed data into the trained target cold chain transportation risk prediction network. The feature extraction subnetwork and feature prediction subnetwork in the target cold chain transportation risk prediction network are ready to process and analyze these data.

[0027] Step S120, utilize the feature extraction subnetwork in the target cold chain transportation risk prediction network to perform feature extraction on the cold chain transportation logistics data and the transportation environment change data, generate a logistics knowledge logic vector set corresponding to the cold chain transportation logistics data, and a transportation environment change vector corresponding to the transportation environment change data, integrate the logistics knowledge logic vector set and the transportation environment change vector, and generate a target vector set.

[0028] In this embodiment, the server uses the feature extraction subnetwork in the target cold chain transportation risk prediction network to perform feature extraction on the above-mentioned cold chain transportation logistics data and transportation environment change data.

[0029] For cold chain transportation logistics data, the server generates multiple cold chain transportation knowledge graphs based on multiple transportation records in the transportation logistics data. For example, for a transportation task from a blood product production base to a blood transfusion department of a large hospital, the server generates a detailed cold chain transportation knowledge graph based on the temperature, humidity, vibration and other multi-dimensional data of the vehicle at different time periods and locations. This knowledge graph can clearly display all-round key information in the transportation process.

[0030] The feature extraction subnetwork is used to perform deep feature extraction on these knowledge graphs to generate graph logic data corresponding to each knowledge graph. For example, complex features such as the temperature being below the threshold in a specific time period and lasting longer than the preset value during transportation, the specific pattern of abnormal humidity fluctuations in a certain interval, and the frequency of vibration amplitude exceeding the safe range are extracted. Then, rich logistics node information is added to these graph logic data, such as which specific highway intersection in which city the vehicle briefly stopped at, which warehouse in which logistics park performed the transfer operation and the length of stay, etc., to generate graph logic update data. Further fine feature extraction is performed on these graph logic update data to generate multiple highly accurate graph representation vectors, and a set of logistics knowledge logic vectors corresponding to the cold chain transportation logistics data is generated based on these graph representation vectors.

[0031] For the transportation environment change data, the server uses the feature extraction subnetwork to break it down into multiple fine environment change path clusters. For example, the weather changes during transportation are divided according to different sections and more precise time periods (such as one interval every 10 minutes) to form multiple path clusters. These path clusters are subjected to deep feature extraction to generate the environmental change logical features corresponding to each path cluster, such as the impact of the specific changes in wind speed and wind direction on the cabin temperature in a certain section within a specific 10 minutes. Similarly, detailed logistics node information is added to these logical features, such as the precise location and speed of the vehicle when encountering strong winds on a certain section of the road, to generate environmental change logical update features. These updated features are then subjected to in-depth feature extraction to generate highly accurate second graph representation vectors corresponding to multiple environmental change path clusters, and transportation environment change vectors corresponding to the transportation environment change data are generated based on these second graph representation vectors.

[0032] Finally, the server integrates the logistics knowledge logic vector set and the transportation environment change vector to generate a comprehensive and accurate target vector set.

[0033] Step S130, obtaining a risk knowledge vector set corresponding to X risk knowledge points, using the feature prediction subnetwork to perform a correlation analysis on the target vector set and the risk knowledge vector set, and generating a correlation weight set corresponding to the risk knowledge vector set. Each correlation weight parameter in the correlation weight set represents the correlation between the cold chain transportation logistics data and a risk knowledge point. X is a positive integer.

[0034] In this embodiment, the server obtains a set of risk knowledge vectors corresponding to multiple (X) complex and specific risk knowledge points related to the pharmaceutical cold chain. These risk knowledge points are, for example:

[0035] 1. If the cumulative time during transportation of certain pharmaceutical and biological products (such as a certain type of cell preparation) when the temperature is below the lower threshold and the humidity is simultaneously above the upper threshold exceeds the preset critical value, the activity of the product may be impaired.

[0036] 2. The transport vehicle stays in a remote area due to a sudden mechanical failure, and the time during which the temperature in the car remains in the critical range is close to the tolerable limit of the product, increasing the risk of product deterioration.

[0037] 3. Continuous heavy rainfall during transportation caused serious waterlogging on the roads, significantly reducing the vehicle speed. At the same time, local leakage occurred in the sealing of the carriage, which greatly increased the probability of the product getting damp.

[0038] 4. The transportation route planning is unreasonable. The products pass through many high-altitude and extremely low-temperature areas. In addition, the performance of the vehicle's thermal insulation equipment has declined, causing the products to be in a low-temperature stress state for a long time, affecting their quality and efficacy.

[0039] For each risk knowledge point, the server has a corresponding vector representation. Then, the server uses the feature prediction sub-network to perform correlation analysis on the previously generated target vector set and this risk knowledge vector set.

[0040] For example, for the risk knowledge point "the cumulative time during the transportation of a specific pharmaceutical biological product (such as a certain type of cell preparation) when the temperature is lower than the lower threshold and the humidity is higher than the upper threshold exceeds the preset critical value, which may cause the activity of the product to be damaged", the feature prediction subnetwork will analyze the change trend and cumulative time of the temperature and humidity monitoring data in the target vector set in detail, and conduct a deep comparison with the conditions and thresholds set for this risk knowledge point. If a situation that meets the description of this risk knowledge point is found in the target vector set, the correlation between the two will be high.

[0041] The calculation generates a set of correlation weights corresponding to the risk knowledge vector set. Each weight parameter in this set of correlation weights accurately represents the degree of correlation between the cold chain transportation logistics data and a specific complex risk knowledge point.

[0042] Step S140: generating matching scores corresponding to the X risk knowledge points respectively based on the correlation weight set, and marking target risk knowledge points among the X risk knowledge points for the cold chain transportation logistics data based on the matching scores.

[0043] In this embodiment, based on the above generated relevance weight set, the server generates an accurate matching score for each complex risk knowledge point. For example, for the first risk knowledge point mentioned above, if the cumulative duration of temperature and humidity anomalies found in the target vector set is very close to or even exceeds the preset key value, then its relevance weight will be very high, and the generated matching score will be high; conversely, if the cumulative duration of the relevant abnormal situation is short and does not reach the key threshold, the relevance weight is low, and the matching score is low.

[0044] Then, the server determines which risk knowledge points are target risk knowledge points based on the set strict threshold value. Assuming the threshold value is set to 85 points, and the matching score of a risk knowledge point (such as the second risk knowledge point mentioned above) is 92 points, the server will mark this risk knowledge point as a target risk knowledge point and make a prominent mark in the relevant cold chain transportation logistics data, and promptly remind relevant personnel to attach great importance to it and take targeted measures quickly, such as immediately checking the quality status of this batch of products, conducting a comprehensive review of the transportation process to find the root cause of the problem, optimizing subsequent transportation plans, and strengthening vehicle maintenance.

[0045] As a result, the complex risks in pharmaceutical cold chain transportation can be predicted and labeled in depth, comprehensively and accurately, providing strong technical support and decision-making basis for ensuring the safety, reliability and effectiveness of pharmaceutical cold chain transportation.

[0046] Based on the above steps, the embodiment of the present application realizes efficient feature extraction and integration of cold chain transportation logistics data and its corresponding transportation environment change data by constructing a target cold chain transportation risk prediction network including a feature extraction subnetwork and a feature prediction subnetwork, thereby generating a target vector set that characterizes logistics status and environmental changes. By using this target vector set and the risk knowledge vector set for correlation analysis, a set of correlation weights between each risk knowledge point and the cold chain transportation logistics data can be accurately generated, thereby marking the target risk knowledge points for the cold chain transportation logistics data based on these correlation weights, significantly improving the accuracy and efficiency of risk identification during cold chain transportation.

[0047] In a possible implementation, step S130 includes:

[0048] Step A110: In the feature prediction subnetwork, correlation analysis is performed on the target vector set and the risk knowledge vector set to generate a first basic correlation vector.

[0049] Step A120: performing regularization conversion on the first basic correlation vector to generate a first regularized correlation vector.

[0050] Step A130, perform self-attention processing on the first regularized correlation vector to generate a first self-attention vector, aggregate the first regularized correlation vector and the first self-attention vector to generate a correlation aggregation vector, perform regularization transformation on the correlation aggregation vector, and generate a correlation weight set corresponding to the risk knowledge vector set.

[0051] In this embodiment, in the scenario of cold chain transportation of medicines, the server receives rich and detailed cold chain transportation logistics data and transportation environment change data, and generates a target vector set through the feature extraction sub-network. At the same time, the server also obtains a risk knowledge vector set corresponding to multiple risk knowledge points, and prepares to use the feature prediction sub-network for correlation analysis.

[0052] First, in the feature prediction subnetwork, the correlation analysis of the target vector set and the risk knowledge vector set is performed to generate the first basic correlation vector. Assume that the server processes the cold chain transportation data of a batch of blood products. The target vector set contains detailed information about the transport vehicle, such as the vehicle model A123, number B456, and good maintenance record; the departure point is a blood product production base, and the destination is the hematology department of a large hospital in a certain city; the transportation route passes through multiple cities, including highways and urban roads; the transportation time starts at 8 am and lasts for 5 hours, accurate to the minute; detailed information about each batch of blood products, such as the product type is plasma of a specific blood type, the specification is 500 ml / bag, the batch number is C789, and the validity period is until the next month; the real-time temperature monitoring record during the transportation process shows that the temperature in the car is slightly lower than the set optimal temperature range during the 2 hours to 2 hours and 30 minutes on the way; the humidity monitoring record is normal; the vibration monitoring data has a short and large amplitude vibration when passing through a bumpy section; the light monitoring data is normal. The data on changes in the transportation environment include weather conditions in different areas during transportation. For example, there may be light rain in the area passed through during 1 to 2 hours, the temperature may drop by 2 degrees, and the wind speed may be relatively low. The road conditions may be slightly congested in a certain section of urban roads. The power supply may be stable, with no power outages or voltage fluctuations. The network signal stability may be normal.

[0053] One of the risk knowledge points in the risk knowledge vector set is "If the temperature of blood products is lower than the lower threshold during transportation and lasts for more than 30 minutes, the quality of the products may be reduced." The server compares and analyzes the temperature data in the target vector set with this risk knowledge point. It is found that the temperature during transportation is lower than the lower threshold for 40 minutes, which indicates that there is a certain correlation between the two. Through complex mathematical calculations and pattern matching, the server generates a first basic correlation vector that represents the degree of this correlation.

[0054] Next, the first basic correlation vector is subjected to regularization conversion to generate a first regularized correlation vector. The purpose of regularization conversion is to map the value of the first basic correlation vector to a specific range for subsequent processing and comparison. Assuming that the value of the first basic correlation vector is originally between -10 and 10, the server uses a specific regularization function to convert it to a value between 0 and 1. After regularization conversion, the first basic correlation vector originally represented as 5 may be converted to 0.5, thereby obtaining the first regularized correlation vector.

[0055] Then, the first regularized correlation vector is processed by self-attention to generate the first self-attention vector. The self-attention mechanism allows the server to pay more attention to the important parts of the first regularized correlation vector. Assuming that the part of the first regularized correlation vector related to the duration of the temperature being below the lower threshold is considered important, the weight of this part will be increased through the self-attention mechanism. After processing, the first self-attention vector is generated.

[0056] Next, the first regularized relevance vector and the first self-attention vector are aggregated to generate a relevance aggregation vector. The server may use a weighted summation method for aggregation. For example, the weight of the first regularized relevance vector is 0.6, the weight of the first self-attention vector is 0.4, and then the two are weighted summed to obtain a relevance aggregation vector.

[0057] Finally, the correlation aggregation vector is transformed into a regularized form to generate a set of correlation weights corresponding to the risk knowledge vector set. The regularized transformation in this step is similar to the previous one, which also maps the value of the correlation aggregation vector to a specific range to more clearly represent the strength of the correlation. Assuming that the value of the correlation aggregation vector is between 0 and 10, after the regularized transformation, it is mapped to between 0 and 1 to obtain the final set of correlation weights. Each value in this set of correlation weights accurately represents the degree of correlation between the cold chain transportation logistics data and a risk knowledge point.

[0058] For another risk knowledge point, such as "transport vehicles encounter severe road congestion during transportation and the duration exceeds 1 hour, increasing the risk of blood product deterioration", the server also processes it according to the above steps. First, the road condition data in the target vector set is analyzed, and it is found that during the transportation process, a serious congestion was encountered in a certain urban road section, which lasted for 1.5 hours, and the corresponding first basic correlation vector is generated. After regularization conversion, self-attention processing, aggregation, and re-regularization conversion, the correlation weight corresponding to this risk knowledge point is obtained.

[0059] By repeating the above process for multiple risk knowledge points, the server finally generates a complete set of correlation weights, which provides a key basis for the subsequent accurate assessment and labeling of risks in cold chain transportation. In this way, relevant personnel can quickly identify potential high-risk situations based on these correlation weights and take corresponding measures to ensure the quality and safety of pharmaceutical cold chain transportation.

[0060] For example, there is another risk knowledge point: "The vibration amplitude of the transport vehicle exceeds the safety range frequently, which may affect the stability of blood products." The server searches for vibration monitoring data in the target vector set and finds that when the vehicle passes through a rugged mountain road, the vibration amplitude exceeds the safety range a certain number of times. After a series of processing steps, the correlation weight corresponding to this risk knowledge point is generated.

[0061] For the risk knowledge point of "unreasonable transportation route planning, passing through multiple high-temperature areas and long residence time, affecting the quality of blood products", the server analyzes data such as transportation route and residence time, determines its relevance to the risk knowledge point, and generates relevance weights through the same processing flow.

[0062] Through detailed analysis and processing of these different risk knowledge points, the server can comprehensively and accurately assess various potential risks in the cold chain transportation process, providing strong support for ensuring the reliability and effectiveness of pharmaceutical cold chain transportation.

[0063] In a possible implementation, the feature prediction subnetwork includes Y prediction modules, the Y prediction modules are sequentially linked, Y is a positive integer, the Y prediction modules include a prediction module Za, and a is a positive integer not greater than Y. Step S130 includes:

[0064] Step B110, taking the progressive analysis vector corresponding to the prediction module Za and the target vector set as the progressive loading vector of the prediction module Za, performing correlation analysis on the progressive loading vector of the prediction module Za in the prediction module Za, and generating the progressive prediction vector of the prediction module Za.

[0065] Step B120: If the prediction module Za is the first prediction module among the Y prediction modules, the progressive analysis vector corresponding to the prediction module Za is the risk knowledge vector set.

[0066] Step B130, using the progressive prediction vector of the prediction module Za as the progressive analysis vector corresponding to the prediction module Z(a+1), using the progressive analysis vector corresponding to the prediction module Z(a+1) and the target vector set as the progressive loading vector of the prediction module Z(a+1), performing correlation analysis on the progressive loading vector of the prediction module Z(a+1) in the prediction module Z(a+1), generating the progressive prediction vector of the prediction module Z(a+1), until the progressive prediction vector of the last prediction module in the Y prediction modules is generated, and outputting the progressive prediction vector of the last prediction module as the correlation weight set corresponding to the risk knowledge vector set. The prediction module Z(a+1) is the backward connected prediction module of the prediction module Za.

[0067] Assume that in the pharmaceutical cold chain transportation scenario, the feature prediction subnetwork includes 5 prediction modules (Y=5), namely prediction modules Z1, Z2, Z3, Z4, and Z5.

[0068] First, for prediction module Z1. Since it is the first prediction module among the five prediction modules, its corresponding progressive analysis vector is a risk knowledge vector set. Assume that this risk knowledge vector set describes that "a certain pharmaceutical biological product may have reduced activity if its temperature is higher than the upper threshold for a long time during transportation."

[0069] The server uses this risk knowledge vector set and target vector set as the progressive loading vector of prediction module Z1. The target vector set contains detailed information about the transportation, such as the real-time temperature monitoring record of the transportation vehicle, which shows that the temperature in the compartment is higher than the set upper threshold during the 3-4 hour period of transportation. Prediction module Z1 performs a correlation analysis on this progressive loading vector, and generates a progressive prediction vector of prediction module Z1 through complex calculations and comparisons.

[0070] Next, for prediction module Z2. At this point, the progressive prediction vector of prediction module Z1 becomes the progressive analysis vector corresponding to prediction module Z2. The server uses this progressive analysis vector and the target vector set again as the progressive loading vector of prediction module Z2. Assume that prediction module Z2 is concerned with the relationship between humidity changes and risks during transportation. The humidity monitoring records in the target vector set show that the humidity fluctuates abnormally over a certain period of time. Prediction module Z2 performs a correlation analysis on this progressive loading vector to generate a progressive prediction vector for prediction module Z2.

[0071] Then comes prediction module Z3. The progressive prediction vector of prediction module Z2 becomes the progressive analysis vector corresponding to prediction module Z3. The server uses it and the target vector set (including information such as road conditions in the transportation route) as the progressive loading vector of prediction module Z3. Assume that prediction module Z3 analyzes the impact of road bumps on transportation risks. After correlation analysis, the progressive prediction vector of prediction module Z3 is generated.

[0072] Continue to prediction module Z4. The progressive prediction vector of prediction module Z3 becomes the progressive analysis vector corresponding to prediction module Z4, and together with the target vector set (including information such as power supply stability) is used as a progressive loading vector. If prediction module Z4 studies the impact of unstable power supply on transportation equipment and the resulting risks, after processing, the progressive prediction vector of prediction module Z4 is generated.

[0073] Finally, there is the prediction module Z5. The progressive prediction vector of the prediction module Z4 becomes the progressive analysis vector corresponding to the prediction module Z5, and forms a progressive loading vector with the target vector set (including information such as network signal stability). Assume that the prediction module Z5 is concerned with the risk brought by the impact of unstable network signals on transportation monitoring. After correlation analysis, the progressive prediction vector of the prediction module Z5 is generated.

[0074] Since the prediction module Z5 is the last prediction module among the Y prediction modules, the progressive prediction vector generated by it is output as the correlation weight set corresponding to the risk knowledge vector set.

[0075] Through this series of sequential processing and analysis, the server can comprehensively and deeply evaluate the complex correlations between various potential risks in cold chain transportation and the target vector set, thereby providing strong support for accurate prediction and labeling of risks.

[0076] In a possible implementation, step B110 includes:

[0077] Step B111: In the prediction module Za, correlation analysis is performed on the progressive analysis vector and the target vector set to generate a second basic correlation vector.

[0078] Step B112, performing regularization conversion on the second basic correlation vector to generate a second regularized correlation vector.

[0079] Step B113, perform self-attention processing on the second regularized correlation vector to generate a second self-attention vector, aggregate the second regularized correlation vector and the second self-attention vector to generate a target aggregate vector, perform regularization transformation on the target aggregate vector, and generate a progressive prediction vector of the prediction module Za.

[0080] In this embodiment, assuming that the pharmaceutical cold chain transportation scenario is used, the prediction module Z3 in the feature prediction subnetwork is still taken as an example for explanation.

[0081] In the prediction module Z3, the correlation analysis is first performed on the progressive analysis vector and the target vector set to generate a second basic correlation vector. Assuming that the progressive analysis vector is based on the output of the previous prediction module Z2, it describes a certain preliminary relationship between abnormal humidity fluctuations and risks during transportation. The target vector set contains road condition information in the transportation route, such as a section of highway with a large slope and many bends. The server uses complex algorithms and models to perform in-depth comparison and analysis of the two, and finds that the slope and bend conditions of the road are related to abnormal humidity fluctuations to a certain extent, thereby generating a second basic correlation vector.

[0082] Next, the second basic correlation vector is subjected to regularization conversion to generate a second regularized correlation vector. Assuming that the value range of the second basic correlation vector is between -5 and 10, the server uses a specific regularization function to convert it to a value between 0 and 1. After the conversion, the second basic correlation vector with an original value of 6 may be converted to 0.6 to obtain the second regularized correlation vector.

[0083] Then, the second regularized correlation vector is processed by self-attention to generate the second self-attention vector. In this process, the server will focus on the part of the second regularized correlation vector that is more critical to the road conditions and humidity fluctuations. For example, if it is found that the humidity fluctuations at the bends of the road have a more significant impact, then this part will get a higher weight after self-attention processing, thereby generating the second self-attention vector.

[0084] After that, the second regularized relevance vector and the second self-attention vector are aggregated to generate a target aggregate vector. Assuming that the weight assigned to the second regularized relevance vector is 0.4 and the weight assigned to the second self-attention vector is 0.6, the target aggregate vector is obtained by weighted summation.

[0085] Finally, the target aggregation vector is transformed by regularization to generate the progressive prediction vector of the prediction module Z3. Assuming that the value range of the target aggregation vector is between 0 and 8, it is mapped to between 0 and 1 through regularization transformation to obtain the final progressive prediction vector, which provides input for the subsequent prediction module Z4.

[0086] For other prediction modules, such as Z1, Z2, Z4 and Z5, similar steps will be followed, except that the risk factors and target vector sets that each module focuses on will be different, but the processing logic and process are consistent. Through such gradual in-depth and detailed processing, the server can more accurately and comprehensively assess the risks in cold chain transportation and generate the final set of correlation weights, providing strong support for ensuring the safety and quality of pharmaceutical cold chain transportation.

[0087] In a possible implementation, step B111 includes:

[0088] Step B1111, in the prediction module Za, the progressive analysis vector is used as the search vector in the correlation mapping space, the target vector set is used as the reference vector in the correlation mapping space, and the target vector set is used as the result vector in the correlation mapping space.

[0089] Step B1112: Use the correlation mapping space to output the fusion result of the search vector and the reference vector as a first aggregated vector.

[0090] Step B1113, obtaining the knowledge element vector corresponding to the target vector set, and outputting the fusion result of the first aggregation vector and the inverse element of the knowledge element vector as a second aggregation vector.

[0091] Step B1114: convert the second aggregated vector into a first nonlinear transformation vector according to the nonlinear transformation unit in the correlation mapping space.

[0092] Step B1115: output the fusion result of the first nonlinear transformation vector and the result vector as a third aggregate vector, and output the fusion calculation result of the third aggregate vector and the search vector as a second basic correlation vector.

[0093] In this embodiment, the prediction module Z3 in the pharmaceutical cold chain transportation scenario is still taken as an example.

[0094] In the prediction module Z3, the progressive analysis vector is first used as the search vector in the correlation mapping space. It is assumed that the progressive analysis vector describes a certain correlation feature between abnormal humidity fluctuations and unstable power supply during transportation. The target vector set is used as the reference vector and the result vector in the correlation mapping space, where the reference vector contains detailed information about the road conditions of the transportation route, such as the flatness of the road, the number of bends, etc., and the result vector contains information about the type and specifications of specific pharmaceutical and biological products.

[0095] The server uses the correlation mapping space to fuse the search vector with the reference vector. For example, the unstable power supply feature in the search vector interacts with the poor road smoothness feature in the reference vector, and the output is the first aggregated vector.

[0096] Then, obtain the knowledge element vector corresponding to the target vector set. Assume that the knowledge element vector describes information such as the sensitivity of different pharmaceutical and biological products to environmental conditions. Fuse the first aggregate vector and the inverse element of the knowledge element vector. For example, the knowledge element vector indicates that a certain cell preparation is extremely sensitive to humidity changes. Its inverse element is fused with the information about road conditions in the first aggregate vector and output as the second aggregate vector.

[0097] Next, the second aggregated vector is transformed according to the nonlinear transformation unit in the correlation mapping space. For example, the value of the second aggregated vector is amplified or reduced by a certain nonlinear function to be transformed into the first nonlinear transformation vector.

[0098] Afterwards, the first nonlinear transformation vector and the result vector are fused. The information such as the type and specification of the pharmaceutical biological product in the result vector interacts with the comprehensive information about the environment and knowledge elements in the first nonlinear transformation vector, and is output as a third aggregate vector.

[0099] Finally, the third aggregate vector and the search vector are fused and calculated. For example, the humidity and power supply characteristics in the search vector interact with the comprehensive information of the road, knowledge elements and result vector contained in the third aggregate vector again, and the output is the second basic relevance vector.

[0100] For other prediction modules, such as Z1, Z2, Z4 and Z5, similar detailed steps and logic will be followed for processing, except that the specific features and information processed by each module will vary according to its position and function in the feature prediction subnetwork, ultimately providing the basis for generating accurate progressive prediction vectors and relevance weight sets.

[0101] In a possible implementation, step B112 includes:

[0102] Step B1121, obtain the correlation value in the second basic correlation vector, determine the average value and the dispersion of the correlation value, and output the deviation value between the second basic correlation vector and the average value as a loss vector.

[0103] Step B1122, obtain the regularization coefficient, perform a square root operation on the fusion calculation result of the discrete degree and the regularization coefficient to generate an adjustment coefficient, and output the fusion result between the loss vector and the inverse element of the adjustment coefficient as a second regularization correlation vector.

[0104] Let’s take the prediction module Z3 in the pharmaceutical cold chain transportation scenario as an example.

[0105] Obtain the correlation value in the second basic correlation vector. Assume that the second basic correlation vector contains multiple values ​​about the relationship between different factors in the transportation process, for example, the correlation values ​​between humidity fluctuation and road conditions, power supply and other factors are 0.8, -0.5, 0.3, etc. respectively.

[0106] Determine the average of these correlation values. By calculation, the average of the above values ​​is assumed to be 0.2.

[0107] At the same time, the dispersion of the correlation values ​​is determined. The dispersion is calculated by statistical methods and is assumed to be 0.4.

[0108] The deviation between the second basic correlation vector and the average value is output as a loss vector. For example, for a value of 0.8, its deviation from the average value of 0.2 is 0.6; for a value of -0.5, the deviation is -0.7, etc. These deviations constitute a loss vector.

[0109] Get the regularization coefficient, assuming the regularization coefficient is 0.5.

[0110] Take the square root of the fusion calculation result of the discrete degree 0.4 and the regularization coefficient 0.5. First calculate the fusion result as 0.2, and take the square root of it to get 0.447 as the adjustment coefficient.

[0111] The fusion result between the loss vector and the inverse element of the adjustment coefficient is output as the second regularized correlation vector. Assuming that the inverse element of the adjustment coefficient is 2.222, each deviation value in the loss vector is multiplied by the inverse element respectively, and the result obtained constitutes the second regularized correlation vector.

[0112] For other prediction modules, such as Z1, Z2, Z4 and Z5, similar detailed steps will be followed to achieve regularized transformation of the second basic correlation vector and provide standardized data for subsequent processing.

[0113] In a possible implementation, the target cold chain transportation risk prediction network further includes a linear mapping unit, and step S140 includes:

[0114] Step S141 , linear mapping is performed on the correlation weight set according to the conversion function of the linear mapping unit to generate matching scores corresponding to the X risk knowledge points respectively.

[0115] Step S142: output the matching score that is not less than the set threshold value among the X matching scores as the target matching score, and mark the correlation between the risk knowledge point corresponding to the target matching score and the cold chain transportation logistics data.

[0116] In the actual scenario of pharmaceutical cold chain transportation, it is assumed that a set of correlation weights for the transportation of a batch of vaccines has been obtained.

[0117] First, the correlation weight set is linearly mapped according to the conversion function of the linear mapping unit. For example, the weight values ​​in the correlation weight set are 0.6, 0.7, 0.8, etc. The conversion function of the linear mapping unit may be to multiply the weight value by 100 and add 50. Then, after linear mapping, the corresponding matching scores are generated as 110, 120, 130, etc.

[0118] Next, the threshold is set to 120. Among the X matching scores, the matching scores that are not less than the set threshold are output as the target matching scores. In the above example, 120 and 130 are the target matching scores.

[0119] Then, the risk knowledge points corresponding to the target matching score are labeled with the cold chain transportation logistics data. Assuming that the risk knowledge point corresponding to 120 is "the cumulative time that the temperature of a specific pharmaceutical biological product (such as a certain type of cell preparation) is lower than the lower threshold and the humidity is higher than the upper threshold during transportation exceeds the preset critical value, which may cause the activity of the product to be damaged", and the risk knowledge point corresponding to 130 is "the transport vehicle stops in a remote area due to a sudden mechanical failure, and the time during which the temperature in the compartment is maintained in the critical range is close to the limit that the product can withstand, increasing the risk of product deterioration", the server will clearly mark these two risk knowledge points with the cold chain transportation logistics data of this vaccine, such as marking "high risk" next to the relevant records in the database, or highlighting these risk points in the transportation report, so that relevant personnel can quickly and clearly understand the more significant risks in this transportation, so as to take corresponding measures in time, such as checking the quality of vaccines, strengthening the maintenance of transportation equipment, etc., to ensure the safety and quality of subsequent cold chain transportation.

[0120] In a possible implementation manner, the step of obtaining a set of risk knowledge vectors corresponding to the X risk knowledge points includes:

[0121] Step S131, obtain X risk knowledge points, deconstruct the X risk knowledge points respectively, and generate risk content unit sets corresponding to the X risk knowledge points respectively. A risk content unit in the risk content unit set represents a basic element structure generated after a risk knowledge point is deconstructed.

[0122] Step S132, using the feature extraction subnetwork in the target cold chain transportation risk prediction network, feature representation is performed on X risk content unit sets to generate knowledge point content representation vector sets corresponding to the X risk content unit sets.

[0123] Step S133 , performing self-attention processing on the X knowledge point content representation vector sets respectively, to generate risk knowledge point vectors corresponding to the X risk knowledge points respectively.

[0124] Step S134: fuse the X risk knowledge point vectors to generate a risk knowledge vector set corresponding to the X risk knowledge points.

[0125] In this embodiment, it is assumed that a risk knowledge vector set corresponding to four risk knowledge points is to be obtained.

[0126] First, get these 4 risk knowledge points:

[0127] 1. If the cumulative time during transportation of certain pharmaceutical and biological products (such as a certain type of cell preparation) when the temperature is below the lower threshold and the humidity is simultaneously above the upper threshold exceeds the preset critical value, the activity of the product may be impaired.

[0128] 2. The transport vehicle stays in a remote area due to a sudden mechanical failure, and the time during which the temperature in the car remains in the critical range is close to the tolerable limit of the product, increasing the risk of product deterioration.

[0129] 3. Continuous heavy rainfall during transportation caused serious waterlogging on the roads, significantly reducing the vehicle speed. At the same time, local leakage occurred in the sealing of the carriage, which greatly increased the probability of the product getting damp.

[0130] 4. The transportation route planning is unreasonable. The products pass through many high-altitude and extremely low-temperature areas. In addition, the performance of the vehicle's thermal insulation equipment has declined, causing the products to be in a low-temperature stress state for a long time, affecting their quality and efficacy.

[0131] Then, these four risk knowledge points are deconstructed separately to generate their corresponding risk content unit sets.

[0132] For the first risk knowledge point, the deconstructed risk content unit set may include basic elements such as "a certain type of cell preparation", "transportation process", "temperature below the lower threshold", "humidity above the upper threshold", "cumulative time exceeds the preset critical value", and "product activity is impaired".

[0133] Next, the feature extraction subnetwork in the target cold chain transportation risk prediction network is used to represent the features of these four risk content unit sets. For example, for the first risk content unit set, the feature extraction subnetwork may associate "a certain type of cell preparation" with a specific category code, "transportation process" with a specific stage identifier, "temperature below the lower threshold" and "humidity above the upper threshold" with the corresponding monitoring data feature values, "cumulative duration exceeds the preset critical value" with the time feature, and "product activity impairment" with the risk consequence feature, thereby generating a corresponding knowledge point content representation vector set.

[0134] Afterwards, the four knowledge point content representation vector sets are processed by self-attention. For example, in the first knowledge point content representation vector set, the self-attention mechanism may pay more attention to the feature of "the cumulative duration exceeds the preset critical value" and assign a higher weight to it, thereby generating the risk knowledge point vector corresponding to the first risk knowledge point.

[0135] Finally, the four risk knowledge point vectors are fused, possibly by weighted summation or other fusion methods, such as assigning different weights to each risk knowledge point according to its importance, and integrating them into a unified risk knowledge vector set for subsequent correlation analysis and risk assessment with cold chain transportation logistics data.

[0136] In a possible implementation, the cold chain transport logistics data includes Q cold chain transport records, where Q is a positive integer. Step S120 includes:

[0137] Step S121, based on the Q cold chain transport records in the cold chain transport logistics data, generate W cold chain transport knowledge graphs, use the feature extraction subnetwork in the target cold chain transport risk prediction network to perform feature extraction on the W cold chain transport knowledge graphs, generate graph logic data corresponding to the W cold chain transport knowledge graphs, add logistics node information to the W graph logic data, and generate W graph logic update data. The logistics node information refers to the information of the logistics node of a cold chain transport knowledge graph in the cold chain transport logistics data. W is a positive integer. The cold chain transport knowledge graph is composed of at least one cold chain transport record among the Q cold chain transport records.

[0138] Step S122: perform feature extraction on the W graph logic update data to generate first graph representation vectors corresponding to the W cold chain transportation knowledge graphs respectively; and based on the first graph representation vectors corresponding to the W cold chain transportation knowledge graphs respectively, generate a logistics knowledge logic vector set corresponding to the cold chain transportation logistics data.

[0139] Step S123, using the feature extraction subnetwork in the target cold chain transportation risk prediction network, the transportation environment change data is decomposed into T environment change path clusters, where T is a positive integer.

[0140] Step S124, extract features from the T environment change path clusters, generate environment change logic features corresponding to the T environment change path clusters, add logistics node information to the T environment change logic features, and generate T environment change logic update features. The logistics node information refers to the information of the logistics node of an environment change path cluster in the transportation environment change data.

[0141] Step S125, extracting features of the T environment change logic update features, generating second graph representation vectors corresponding to the T environment change path clusters respectively, and generating a transportation environment change vector corresponding to the transportation environment change data based on the second graph representation vectors corresponding to the T environment change path clusters respectively.

[0142] In this embodiment, it is assumed that a pharmaceutical cold chain transportation platform has accumulated 100 cold chain transportation records (Q=100) within one month.

[0143] First, based on these 100 cold chain transport records, 20 cold chain transport knowledge graphs (W=20) are generated. For example, the first cold chain transport knowledge graph consists of 5 transport records, all of which transport a batch of specific cell preparations from the same biopharmaceutical factory to different large hospitals' oncology departments. For these 5 transports, the server collected detailed information about the transport vehicles, such as the vehicle model "refrigerated truck A01", the number "12345", and the most recent maintenance record was good; the departure point was a specific warehouse of the pharmaceutical factory, and the destinations were the oncology departments of Hospital A, Hospital B, and Hospital C; the transport routes covered highways, urban trunk roads, etc., and were specific to every road and city passed through; the transport time was accurate to the minute, such as the first transport started at 8:15 am and ended at 10:30 am. At the same time, there are detailed information about each batch of pharmaceutical and biological products, such as the product type is a specific type of cell preparation, the specification is 10 ml per bottle, the batch number is "BC001", and the validity period is until the next month; the real-time temperature monitoring record during transportation shows that the temperature in the middle of the vehicle is slightly lower than the set optimal temperature limit between 9:00 and 9:15, and the temperature in other locations is normal; the humidity monitoring record remains stable during the entire transportation process; the vibration monitoring data shows that there is a slight vibration when passing through a road repair section; the light monitoring data is normal. The server generates a comprehensive cold chain transportation knowledge graph based on these detailed data, clearly showing the key information of these five transportations.

[0144] The feature extraction subnetwork in the target cold chain transportation risk prediction network is used to extract features from these 20 cold chain transportation knowledge graphs, generating 20 corresponding graph logic data. Taking the first cold chain transportation knowledge graph as an example, the feature extraction subnetwork may extract the following features: the temperature in the middle of the vehicle is lower than the lower threshold for 15 minutes in a specific time period (such as 9:00 to 9:15) during transportation; the humidity remains in the normal range during the entire transportation process without abnormal fluctuations; the vibration amplitude is within a safe range, but the vibration frequency is slightly higher than the average when passing through the road repair section. These features constitute the graph logic data corresponding to the first cold chain transportation knowledge graph.

[0145] Then, logistics node information is added to each of the 20 graph logic data. Taking the first graph logic data as an example, the added logistics node information may include: the vehicle made a short stop of 10 minutes at a service area on the highway to check the condition of the goods; the vehicle made a transfer operation at Warehouse No. 3 in a logistics park, and the stop time was 20 minutes. These detailed logistics node information are combined with the previously extracted graph logic data to generate graph logic update data.

[0146] Next, we perform more in-depth feature extraction on these 20 graph logic update data to generate the first graph representation vectors corresponding to the 20 cold chain transportation knowledge graphs. For example, for the first graph logic update data, we further extract the potential correlation features between the time period when the temperature is below the lower limit threshold and the vehicle stop and transfer operations through the deep learning algorithm to generate a highly accurate first graph representation vector. Based on these 20 first graph representation vectors, we comprehensively consider the weights and interrelationships of various features to generate a set of logistics knowledge logic vectors corresponding to the cold chain transportation logistics data.

[0147] For the transportation environment change data, the server uses the feature extraction subnetwork to disassemble it into 15 environment change path clusters (T=15). For example, the first environment change path cluster is the change in weather and road conditions during transportation through mountainous sections, including between 9 and 10 a.m., the temperature dropped from 20 degrees Celsius to 15 degrees Celsius, the rainfall gradually increased, the wind speed increased from 5 meters per second to 8 meters per second, the wind direction changed from southeast wind to northwest wind, and the haze index rose from 50 to 80; in terms of road conditions, the road flatness decreased, the slope increased from 5% to 8%, the curve angle increased, the traffic congestion level changed from mild congestion to moderate congestion, and there was no road construction.

[0148] The 15 environmental change path clusters are feature extracted to generate 15 corresponding environmental change logic features. Taking the first environmental change path cluster as an example, the extracted environmental change logic feature may be: in the 10 minutes from 9:30 to 9:40, the change in wind speed and wind direction caused the cabin temperature to drop by 2 degrees Celsius.

[0149] Next, add logistics node information to each of the 15 environmental change logic features. For the first environmental change logic feature, the added logistics node information may be: the specific location of the vehicle on the mountain road is 50 kilometers from the starting point, the speed is 40 kilometers per hour, and the cabin is well sealed. These logistics node information are combined with the environmental change logic features to generate the environmental change logic update features.

[0150] Finally, the 15 environmental change logic update features are feature extracted to generate the second graph representation vectors corresponding to the 15 environmental change path clusters. For example, for the first environmental change logic update feature, the deep relationship between vehicle speed, cabin tightness, and the impact of environmental changes on cabin temperature is extracted to generate the corresponding second graph representation vector. Based on these 15 second graph representation vectors, various factors are comprehensively considered to generate the transportation environment change vector corresponding to the transportation environment change data.

[0151] The server integrates the generated logistics knowledge logic vector set and the transportation environment change vector to form a comprehensive and accurate target vector set. This target vector set contains various key information and features in the cold chain transportation process, providing a solid data foundation for subsequent risk analysis and prediction.

[0152] In actual pharmaceutical cold chain transportation scenarios, such detailed analysis and feature extraction can help the server more accurately assess potential risks during transportation and promptly identify factors that may affect the quality and safety of pharmaceutical and biological products, thereby ensuring the reliability and effectiveness of cold chain transportation. For example, if a risk knowledge point is found to have a high correlation with the target vector set in subsequent risk analysis, the server can issue an alarm in time to remind relevant personnel to take emergency measures, such as adjusting the transportation plan, checking product quality, etc., to reduce the losses caused by the risk.

[0153] In a possible implementation, the cold chain transport logistics data includes Q cold chain transport records, the number of target risk knowledge points is U, Q and U are both positive integers, and the target risk knowledge points include risk knowledge points Yb, where b is a positive integer not greater than U. The method further includes:

[0154] Step C110, based on the Q cold chain transport records in the cold chain transport logistics data, generate W cold chain transport knowledge graphs, obtain the knowledge graph feature set corresponding to the W cold chain transport knowledge graphs and the target risk knowledge vector corresponding to the risk knowledge point Yb. W is a positive integer. The cold chain transport knowledge graph is composed of the transport entity of at least one cold chain transport record in the Q cold chain transport records.

[0155] Step C120, using the feature prediction subnetwork to perform correlation analysis on the knowledge graph feature set and the target risk knowledge vector respectively, to generate a knowledge graph correlation weight set corresponding to the risk knowledge point Yb. Each correlation weight parameter in the knowledge graph correlation weight set represents the correlation between the risk knowledge point Yb and a cold chain transportation knowledge graph.

[0156] Step C130, based on the knowledge graph relevance weight set, generates cold chain transportation knowledge graph matching scores corresponding to the W cold chain transportation knowledge graphs respectively, and based on the cold chain transportation knowledge graph matching scores, obtains the target cold chain transportation knowledge graph corresponding to the risk knowledge point Yb in the W cold chain transportation knowledge graphs, and marks the risk knowledge point Yb for the target cold chain transportation knowledge graph.

[0157] Continuing to assume that a pharmaceutical cold chain transportation platform has accumulated 500 cold chain transportation records (Q=500) within a quarter, and has identified three target risk knowledge points (U=3), namely risk knowledge points Y1, Y2 and Y3.

[0158] First, based on these 500 cold chain transportation records, 80 cold chain transportation knowledge graphs (W=80) are generated. For example, the first cold chain transportation knowledge graph consists of 10 transportation records, and all of these 10 transportations are of the same vaccine from the vaccine production plant to the CDC in different cities. For these 10 transports, the server obtained detailed transport information, including the specific model of the transport vehicle, which was "refrigerated truck X001", number "67890", and good maintenance records; the departure points were all specific warehouses of the vaccine production plant, and the destinations were the CDCs of City A, City B, and City C; the transport routes were accurate to the specific highway numbers and names of the cities along the way; the transport time was accurate to the minute, such as the first transport started at 3 a.m. and ended at 7 a.m.; detailed information on each batch of vaccines, such as the vaccine type was COVID-19 vaccine, the specification was 0.5 ml per dose, the batch number was "VAC001", and the validity period was six months later; the real-time temperature monitoring records during the transport process showed that the temperature was close to the lower threshold between 5 a.m. and 5:10 a.m. at the rear of the vehicle; the humidity monitoring records were normal; the vibration monitoring data had slight vibrations when passing a bridge; and the light monitoring data was normal. The server generates a comprehensive cold chain transport knowledge graph based on these detailed data.

[0159] Get the knowledge graph feature set corresponding to these 80 cold chain transportation knowledge graphs, and get the target risk knowledge vector corresponding to the risk knowledge point Y1. Assume that the risk knowledge point Y1 is "the temperature during transportation is close to the lower limit threshold and lasts for more than 30 minutes, which may affect the efficacy of the vaccine."

[0160] The feature prediction subnetwork is used to perform correlation analysis on the knowledge graph feature set and the target risk knowledge vector respectively, and a knowledge graph correlation weight set corresponding to the risk knowledge point Y1 is generated. For example, for the first cold chain transportation knowledge graph, the feature prediction subnetwork analysis found that the duration of the temperature approaching the lower limit threshold was 25 minutes, which is close to the 30 minutes set for the risk knowledge point Y1, thereby generating a corresponding correlation weight parameter. This analysis was performed on all 80 cold chain transportation knowledge graphs to obtain a knowledge graph correlation weight set.

[0161] Based on the knowledge graph relevance weight set, the cold chain transportation knowledge graph matching scores corresponding to 80 cold chain transportation knowledge graphs are generated. Assuming that the relevance weight of the first cold chain transportation knowledge graph is higher, the calculated matching score is 85 points.

[0162] Based on the cold chain transportation knowledge graph matching score, the target cold chain transportation knowledge graph corresponding to the risk knowledge point Y1 is obtained from the 80 cold chain transportation knowledge graphs. Assuming that the set matching score threshold is 80 points, the cold chain transportation knowledge graphs with a score higher than 80 points are determined as the target cold chain transportation knowledge graphs. For example, there are 20 cold chain transportation knowledge graphs with matching scores higher than 80 points, and they are determined as the target cold chain transportation knowledge graphs related to the risk knowledge point Y1.

[0163] Finally, mark the risk knowledge point Y1 for these target cold chain transportation knowledge diagrams so that relevant personnel can clearly understand the situations related to this risk knowledge point in these transportation processes and take corresponding measures, such as quality inspection of relevant batches of vaccines and strengthening temperature monitoring during transportation.

[0164] For risk knowledge points Y2 and Y3, similar steps are also followed for processing and analysis, comprehensively evaluating the relationship between various risks in the cold chain transportation process and different transportation knowledge graphs, and accurately marking and processing them.

[0165] In a possible implementation, the method further includes:

[0166] Step S101, loading sample cold chain transportation logistics data and sample transportation environment change data corresponding to the sample cold chain transportation logistics data into a basic cold chain transportation risk prediction network. The basic cold chain transportation risk prediction network includes a basic feature extraction subnetwork and a basic feature prediction subnetwork.

[0167] Step S102, using the basic feature extraction subnetwork in the basic cold chain transportation risk prediction network to perform feature extraction on the sample cold chain transportation logistics data and the sample transportation environment change data, generate a sample logistics knowledge logic vector set corresponding to the sample cold chain transportation logistics data, and a sample transportation environment change vector corresponding to the sample transportation environment change data, integrate the sample logistics knowledge logic vector set and the sample transportation environment change vector to generate a sample integrated vector set.

[0168] Step S103, obtain a set of sample risk knowledge vectors corresponding to K sample risk knowledge points, use the basic feature prediction subnetwork in the basic cold chain transportation risk prediction network to perform correlation analysis on the sample integrated vector set and the sample risk knowledge vector set, and generate a correlation weight sample set corresponding to the sample risk knowledge vector set. Each correlation weight parameter in the correlation weight sample set represents the correlation between the sample cold chain transportation logistics data and a sample risk knowledge point. K is a positive integer.

[0169] Step S104, training the basic feature extraction subnetwork and the basic feature prediction subnetwork in the basic cold chain transportation risk prediction network based on the correlation weight sample set to generate a target cold chain transportation risk prediction network, and the target cold chain transportation risk prediction network is used to annotate risk knowledge points of cold chain transportation logistics data.

[0170] Suppose there is a data set containing 100 sets of sample data, including sample cold chain transportation logistics data and corresponding sample transportation environment change data.

[0171] First, the server loads these 100 sets of sample cold chain transportation logistics data and the corresponding sample transportation environment change data into the basic cold chain transportation risk prediction network. These sample data come from different pharmaceutical cold chain transportation tasks, such as:

[0172] Example 1: A batch of insulin is being transported by a refrigerated truck of a specific model. The departure point is a pharmaceutical factory and the destination is a hospital. The transportation route passes through multiple cities and the transportation time is within a specific time period. The monitoring data of temperature, humidity, vibration, etc. during the transportation process are recorded in detail. The transportation environment change data includes the weather, roads, electricity, and network signals in the areas passed along the way.

[0173] Example 2: A batch of vaccines is being transported, with different transport vehicle information, departure and destination, and also detailed transport and environmental change data records.

[0174] Then, the basic feature extraction subnetwork in the basic cold chain transportation risk prediction network is used to extract features from these sample data. For the cold chain transportation logistics data in Sample 1, the corresponding sample logistics knowledge logic vector is generated, such as extracting features such as the temperature briefly exceeding the appropriate range in a specific city during transportation and the humidity slightly rising for a period of time; for the transportation environment change data of Sample 1, a sample transportation environment change vector is generated, such as extracting the impact of strong wind weather on the temperature of the carriage on a certain section of road. This processing is performed on all 100 groups of sample data, and then the sample logistics knowledge logic vector set and the sample transportation environment change vector are integrated to generate a sample integrated vector set.

[0175] Next, obtain the sample risk knowledge vector set corresponding to the 50 sample risk knowledge points. For example:

[0176] Sample risk knowledge point 1: If the temperature exceeds the upper threshold during transportation and persists for a certain period of time, the drug may become ineffective.

[0177] Sample risk knowledge point 2: When a transport vehicle stays for a long time in bad weather, changes in the environment inside the vehicle affect the quality of the drugs.

[0178] The basic feature prediction subnetwork in the basic cold chain transportation risk prediction network is used to perform correlation analysis on the sample integration vector set and the sample risk knowledge vector set to generate a correlation weight sample set corresponding to each sample risk knowledge vector. For example, for the integration vector of sample 1, its correlation with sample risk knowledge point 1 is analyzed and the corresponding weight parameter is calculated.

[0179] Finally, the basic feature extraction subnetwork and basic feature prediction subnetwork in the basic cold chain transportation risk prediction network are trained based on the obtained correlation weight sample set. The server adjusts the parameters in the network, such as the connection weights and biases of neurons, according to these weight sample sets. Through multiple iterative optimizations, the network can more accurately predict and analyze the risks in cold chain transportation, and finally generate a target cold chain transportation risk prediction network that can be used for risk knowledge point annotation of actual cold chain transportation logistics data.

[0180] In a possible implementation, the basic feature extraction subnetwork includes a first graph autoencoder and a second graph autoencoder. Step S104 includes:

[0181] Step S1041, obtaining the prior logistics knowledge logic vector corresponding to the sample cold chain transportation logistics data, and training the first graph autoencoder based on the prior logistics knowledge logic vector and the sample logistics knowledge logic vector set.

[0182] Step S1042: Obtain a priori transportation environment change vector corresponding to the sample transportation environment change data, and train the second graph autoencoder based on the priori transportation environment change vector and the sample transportation environment change vector.

[0183] Step S1043, obtaining the sample annotated risk knowledge points corresponding to the sample cold chain transportation logistics data, and generating sample matching scores corresponding to the K sample risk knowledge points respectively based on the correlation weighted sample set.

[0184] Step S1044, performing exponential conversion on the K sample matching scores respectively to generate K score conversion values.

[0185] Step S1045: Determine the label values ​​corresponding to the K sample risk knowledge points respectively according to the sample annotated risk knowledge points and the sample matching scores. The label values ​​include a first label result or a second label result, wherein the first label result indicates that there are risk knowledge points in the sample annotated risk knowledge points that are consistent with the sample risk knowledge points corresponding to the first label result, and the second label result indicates that there are no risk knowledge points in the sample annotated risk knowledge points that are consistent with the sample risk knowledge points corresponding to the second label result.

[0186] Step S1046: Generate a training error parameter based on the label values ​​corresponding to the K sample risk knowledge points and the K score conversion values.

[0187] Step S1047: training the basic feature prediction subnetwork in the basic cold chain transportation risk prediction network based on the training error parameter to generate a target cold chain transportation risk prediction network.

[0188] In this embodiment, it is further assumed that there is a sample data set containing 50 samples for training the basic cold chain transportation risk prediction network.

[0189] First, obtain the prior logistics knowledge logic vector corresponding to the sample cold chain transportation logistics data. For example, for a sample of transporting blood samples, the prior logistics knowledge logic vector may contain features such as expected temperature and humidity ranges based on past experience or standard settings. Based on this prior logistics knowledge logic vector and the set of sample logistics knowledge logic vectors actually extracted, the first graph autoencoder is trained. For example, the sample logistics knowledge logic vector actually extracted shows that the temperature exceeds the a priori set range at a certain time during the transportation process. The first graph autoencoder will adjust its own parameters based on this deviation to better learn and extract features from the cold chain transportation logistics data.

[0190] Next, obtain the prior transportation environment change vector corresponding to the sample transportation environment change data. Assume that for a sample of transporting medicines, the prior transportation environment change vector sets the expected degree of impact of weather conditions on transportation. Based on this prior transportation environment change vector and the sample transportation environment change vector actually extracted, the second image autoencoder is trained. For example, if the actual sample transportation environment change vector shows that the impact on medicines in a certain period of bad weather exceeds the prior expectations, the second image autoencoder will adjust the parameters accordingly to improve the ability to extract the characteristics of transportation environment changes.

[0191] Then, obtain the sample annotated risk knowledge points corresponding to the sample cold chain transportation logistics data. For example, for a sample of transporting vaccines, the annotated risk knowledge point may be "the temperature during transportation is lower than the lower limit threshold for more than 1 hour". Generate sample matching scores corresponding to 50 sample risk knowledge points based on the correlation weighted sample set. For example, a sample risk knowledge point has a high correlation with the actual data, and its matching score is 80 points.

[0192] Perform exponential conversion on 50 sample matching scores respectively to generate 50 score conversion values. Assuming that a sample matching score is 70 points, a corresponding conversion value is obtained after exponential conversion.

[0193] Based on the sample annotated risk knowledge points and the sample matching scores, determine the label values ​​corresponding to the 50 sample risk knowledge points. If there is a risk knowledge point in the sample annotated risk knowledge points that is consistent with a certain sample risk knowledge point, its label value is the first label result; if not, it is the second label result. For example, for the above-mentioned sample of transporting vaccines, if the actual situation is that the temperature is indeed below the lower limit threshold for more than 1 hour, the corresponding label value is the first label result; otherwise, it is the second label result.

[0194] Based on the label values ​​corresponding to the 50 sample risk knowledge points and the 50 score conversion values, a training error parameter is generated. For example, by calculating the difference between the label value and the score conversion value, a total training error parameter is obtained by combining multiple samples.

[0195] Finally, the basic feature prediction subnetwork in the basic cold chain transportation risk prediction network is trained based on the training error parameters. By adjusting the parameters in the subnetwork, such as the weights and biases of neurons, the prediction ability of the network is continuously optimized until the error parameters meet the preset requirements or convergence is achieved, thereby generating a target cold chain transportation risk prediction network that can accurately predict risks.

[0196] Further, Figure 2 FIG. 1 is a schematic diagram showing a hardware structure of a cold chain monitoring system 100 for implementing the method provided in an embodiment of the present application. Figure 2 As shown, the cold chain monitoring system 100 includes a memory 111, a storage controller 112, a processor 113, a peripheral interface 114, an input and output unit 115, an audio unit 116, a display unit 117 and a radio frequency unit 118.

[0197] The memory 111, the storage controller 112, the processor 113, the peripheral interface 114, the input / output unit 115, the audio unit 116, the display unit 117, and the radio frequency unit 118 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines.

[0198] The memory 111 may be, but not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electric erasable programmable read-only memory (EEPROM), etc. The memory 111 is used to store the application 210, and the processor 113 executes the application 210 after receiving the execution instruction. The processor 113 and other possible components may access the memory 111 under the control of the storage controller 112.

[0199] The processor 113 may be an integrated circuit chip having the ability to process signals. The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present invention may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0200] The peripheral interface 114 couples various input / output devices to the processor 113 and the memory 111. In some embodiments, the peripheral interface 114, the processor 113 and the memory controller 112 can be implemented in a single chip. In other embodiments, they can be implemented by separate chips.

[0201] The input-output unit 115 is used to provide input data to the user to implement the interaction between the user and the cold chain monitoring system 100. The input-output unit 115 can be, but is not limited to, a mouse and a keyboard.

[0202] The audio unit 116 provides an audio interface to the user and may include one or more microphones, one or more speakers, and an audio circuit.

[0203] The display unit 117 provides an interactive interface (such as a user operation interface) between the cold chain monitoring system 100 and the user or is used to display process control data. In this embodiment, the display unit 117 can be a liquid crystal display or a touch display. If it is a touch display, it can be a capacitive touch screen or a resistive touch screen that supports single-point and multi-point touch operations. Supporting single-point and multi-point touch operations means that the touch display can sense touch operations generated from one or more positions on the touch display, and the sensed touch operations are handed over to the processor for calculation and processing.

[0204] The radio frequency unit 118 is used to receive and send radio wave signals (such as electromagnetic waves) to achieve mutual conversion between radio waves and electrical signals, thereby realizing communication between the cold chain monitoring system 100 and the network or other communication devices.

[0205] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of the embodiments of the present application are described. Other embodiments are within the scope of the attached claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or can be advantageous.

[0206] Each embodiment in the embodiments of the present application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the above different embodiments, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0207] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program. The above program may be stored in a computer-readable storage medium, and the above storage medium may be a read-only memory, a disk or an optical disk, etc.

Claims

1. A data analysis method based on cold chain transportation, characterized in that: The method comprises: Loading the cold chain transportation logistics data and the transportation environment change data corresponding to the cold chain transportation logistics data into the target cold chain transportation risk prediction network; the target cold chain transportation risk prediction network includes a feature extraction subnetwork and a feature prediction subnetwork; Utilizing the feature extraction subnetwork in the target cold chain transportation risk prediction network to extract features from the cold chain transportation logistics data and the transportation environment change data, generating a set of logistics knowledge logic vectors corresponding to the cold chain transportation logistics data and a transportation environment change vector corresponding to the transportation environment change data, integrating the logistics knowledge logic vector set and the transportation environment change vector to generate a target vector set; A risk knowledge vector set corresponding to X risk knowledge points is obtained, and a correlation analysis is performed on the target vector set and the risk knowledge vector set using the feature prediction subnetwork to generate a correlation weight set corresponding to the risk knowledge vector set; each correlation weight parameter in the correlation weight set represents the correlation between the cold chain transport logistics data and a risk knowledge point; X is a positive integer; Generate matching scores corresponding to the X risk knowledge points based on the correlation weight set, and mark the target risk knowledge points among the X risk knowledge points for the cold chain transportation logistics data based on the matching scores; The cold chain transport logistics data includes Q cold chain transport records, the number of the target risk knowledge points is U, Q and U are both positive integers, the target risk knowledge points include risk knowledge points Yb, b is a positive integer not greater than U; the method further includes: According to the Q cold chain transport records in the cold chain transport logistics data, W cold chain transport knowledge graphs are generated, and the knowledge graph feature sets corresponding to the W cold chain transport knowledge graphs and the target risk knowledge vector corresponding to the risk knowledge point Yb are obtained; W is a positive integer; the cold chain transport knowledge graph is composed of a transport entity of at least one cold chain transport record in the Q cold chain transport records; The feature prediction subnetwork is used to perform correlation analysis on the knowledge graph feature set and the target risk knowledge vector respectively, and a knowledge graph correlation weight set corresponding to the risk knowledge point Yb is generated; each correlation weight parameter in the knowledge graph correlation weight set represents the correlation between the risk knowledge point Yb and a cold chain transportation knowledge graph; Based on the knowledge graph relevance weight set, cold chain transportation knowledge graph matching scores corresponding to the W cold chain transportation knowledge graphs are generated; based on the cold chain transportation knowledge graph matching scores, a target cold chain transportation knowledge graph corresponding to the risk knowledge point Yb is obtained from the W cold chain transportation knowledge graphs, and the risk knowledge point Yb is marked on the target cold chain transportation knowledge graph.

2. The data analysis method based on cold chain transportation according to claim 1 is characterized in that: The using the feature prediction subnetwork to perform correlation analysis on the target vector set and the risk knowledge vector set to generate a correlation weight set corresponding to the risk knowledge vector set includes: In the feature prediction subnetwork, correlation analysis is performed on the target vector set and the risk knowledge vector set to generate a first basic correlation vector; Performing regularization conversion on the first basic correlation vector to generate a first regularized correlation vector; Perform self-attention processing on the first regularized correlation vector to generate a first self-attention vector, aggregate the first regularized correlation vector and the first self-attention vector to generate a correlation aggregation vector, perform regularization transformation on the correlation aggregation vector, and generate a correlation weight set corresponding to the risk knowledge vector set.

3. The data analysis method based on cold chain transportation according to claim 1 is characterized in that: The feature prediction subnetwork includes Y prediction modules, the Y prediction modules are sequentially linked, Y is a positive integer, the Y prediction modules include a prediction module Za, and a is a positive integer not greater than Y; the feature prediction subnetwork is used to perform a correlation analysis on the target vector set and the risk knowledge vector set to generate a correlation weight set corresponding to the risk knowledge vector set, including: The progressive analysis vector corresponding to the prediction module Za and the target vector set are used as the progressive loading vector of the prediction module Za, and a correlation analysis is performed on the progressive loading vector of the prediction module Za in the prediction module Za to generate the progressive prediction vector of the prediction module Za; If the prediction module Za is the first prediction module among the Y prediction modules, the progressive analysis vector corresponding to the prediction module Za is the risk knowledge vector set; The progressive prediction vector of the prediction module Za is used as the progressive analysis vector corresponding to the prediction module Z(a+1), the progressive analysis vector corresponding to the prediction module Z(a+1) and the target vector set are used as the progressive loading vector of the prediction module Z(a+1), and the progressive loading vector of the prediction module Z(a+1) is subjected to correlation analysis in the prediction module Z(a+1) to generate the progressive prediction vector of the prediction module Z(a+1), until the progressive prediction vector of the last prediction module among the Y prediction modules is generated, and the progressive prediction vector of the last prediction module is output as the correlation weight set corresponding to the risk knowledge vector set; the prediction module Z(a+1) is a backward-connected prediction module of the prediction module Za.

4. The data analysis method based on cold chain transportation according to claim 3 is characterized in that: The step of using the progressive analysis vector corresponding to the prediction module Za and the target vector set as the progressive loading vector of the prediction module Za, performing correlation analysis on the progressive loading vector of the prediction module Za in the prediction module Za, and generating the progressive prediction vector of the prediction module Za includes: In the prediction module Za, correlation analysis is performed on the progressive analysis vector and the target vector set to generate a second basic correlation vector; Performing regularization transformation on the second basic correlation vector to generate a second regularized correlation vector; Performing self-attention processing on the second regularized correlation vector to generate a second self-attention vector, aggregating the second regularized correlation vector and the second self-attention vector to generate a target aggregated vector, and performing regularization conversion on the target aggregated vector to generate a progressive prediction vector of the prediction module Za; Wherein, in the prediction module Za, the correlation analysis is performed on the progressive analysis vector and the target vector set to generate a second basic correlation vector, including: In the prediction module Za, the progressive analysis vector is used as a search vector in a correlation mapping space, the target vector set is used as a reference vector in the correlation mapping space, and the target vector set is used as a result vector in the correlation mapping space; Outputting a fusion result of the search vector and the reference vector as a first aggregated vector by using the correlation mapping space; Acquire the knowledge element vector corresponding to the target vector set, and output the fusion result of the first aggregation vector and the inverse element of the knowledge element vector as a second aggregation vector; According to the nonlinear transformation unit in the correlation mapping space, the second aggregated vector is transformed into a first nonlinear transformation vector; Outputting a fusion result of the first nonlinear transformation vector and the result vector as a third aggregate vector, and outputting a fusion calculation result of the third aggregate vector and the search vector as a second basic correlation vector; The step of performing regularization conversion on the second basic correlation vector to generate a second regularized correlation vector includes: Obtaining correlation values ​​in the second basic correlation vector, determining an average value and a dispersion degree of the correlation values, and outputting a deviation value between the second basic correlation vector and the average value as a loss vector; Obtain a regularization coefficient, perform a square root operation on a fusion calculation result of the discrete degree and the regularization coefficient to generate an adjustment coefficient, and output a fusion result between the loss vector and the inverse element of the adjustment coefficient as a second regularization correlation vector.

5. The data analysis method based on cold chain transportation according to claim 1 is characterized in that: The target cold chain transportation risk prediction network also includes a linear mapping unit, which generates matching scores corresponding to the X risk knowledge points based on the correlation weight set, and labels the target risk knowledge points among the X risk knowledge points for the cold chain transportation logistics data based on the matching scores, including: Performing linear mapping on the correlation weight set according to the conversion function of the linear mapping unit to generate matching scores corresponding to the X risk knowledge points respectively; The matching score that is not less than the set threshold value among the X matching scores is output as the target matching score, and the risk knowledge point corresponding to the target matching score is marked with the correlation with the cold chain transportation logistics data.

6. The data analysis method based on cold chain transportation according to claim 1 is characterized in that: The step of obtaining a risk knowledge vector set corresponding to the X risk knowledge points includes: Obtain X risk knowledge points, deconstruct the X risk knowledge points respectively, and generate risk content unit sets corresponding to the X risk knowledge points respectively; a risk content unit in the risk content unit set represents a basic element composition generated after a risk knowledge point is deconstructed; Using the feature extraction subnetwork in the target cold chain transportation risk prediction network, feature representation is performed on X risk content unit sets to generate knowledge point content representation vector sets corresponding to the X risk content unit sets respectively; Perform self-attention processing on the X knowledge point content representation vector sets respectively to generate risk knowledge point vectors corresponding to the X risk knowledge points respectively; The X risk knowledge point vectors are fused to generate a risk knowledge vector set corresponding to the X risk knowledge points.

7. The data analysis method based on cold chain transportation according to claim 1 is characterized in that: The cold chain transportation logistics data includes Q cold chain transportation records, where Q is a positive integer; the feature extraction subnetwork in the target cold chain transportation risk prediction network is used to extract features from the cold chain transportation logistics data and the transportation environment change data, and a set of logistics knowledge logic vectors corresponding to the cold chain transportation logistics data and a transportation environment change vector corresponding to the transportation environment change data are generated, including: Based on the Q cold chain transport records in the cold chain transport logistics data, W cold chain transport knowledge graphs are generated, and the feature extraction subnetwork in the target cold chain transport risk prediction network is used to extract features from the W cold chain transport knowledge graphs to generate graph logic data corresponding to the W cold chain transport knowledge graphs, and logistics node information is added to the W graph logic data to generate W graph logic update data; the logistics node information refers to the information of a logistics node of a cold chain transport knowledge graph in the cold chain transport logistics data; W is a positive integer; the cold chain transport knowledge graph is composed of at least one cold chain transport record among the Q cold chain transport records; Perform feature extraction on the W graph logic update data to generate first graph representation vectors corresponding to the W cold chain transportation knowledge graphs respectively, and generate a set of logistics knowledge logic vectors corresponding to the cold chain transportation logistics data based on the first graph representation vectors corresponding to the W cold chain transportation knowledge graphs respectively; Using the feature extraction subnetwork in the target cold chain transportation risk prediction network, the transportation environment change data is decomposed into T environment change path clusters; T is a positive integer; Extract features from the T environmental change path clusters, generate environmental change logic features corresponding to the T environmental change path clusters, add logistics node information to the T environmental change logic features, and generate T environmental change logic update features; the logistics node information refers to the information of the logistics node of an environmental change path cluster in the transport environment change data; Feature extraction is performed on the T environment change logic update features to generate second graph representation vectors corresponding to the T environment change path clusters respectively, and based on the second graph representation vectors corresponding to the T environment change path clusters respectively, a transportation environment change vector corresponding to the transportation environment change data is generated.

8. The data analysis method based on cold chain transportation according to claim 1 is characterized in that: The method further comprises: Loading sample cold chain transportation logistics data and sample transportation environment change data corresponding to the sample cold chain transportation logistics data into a basic cold chain transportation risk prediction network; the basic cold chain transportation risk prediction network includes a basic feature extraction subnetwork and a basic feature prediction subnetwork; Utilizing the basic feature extraction subnetwork in the basic cold chain transportation risk prediction network to perform feature extraction on the sample cold chain transportation logistics data and the sample transportation environment change data, generating a sample logistics knowledge logic vector set corresponding to the sample cold chain transportation logistics data, and a sample transportation environment change vector corresponding to the sample transportation environment change data, integrating the sample logistics knowledge logic vector set and the sample transportation environment change vector to generate a sample integrated vector set; Obtain a set of sample risk knowledge vectors corresponding to K sample risk knowledge points, use the basic feature prediction subnetwork in the basic cold chain transportation risk prediction network to perform correlation analysis on the sample integrated vector set and the sample risk knowledge vector set, and generate a correlation weight sample set corresponding to the sample risk knowledge vector set; each correlation weight parameter in the correlation weight sample set represents the correlation between the sample cold chain transportation logistics data and a sample risk knowledge point; K is a positive integer; Based on the correlation weight sample set, the basic feature extraction subnetwork and the basic feature prediction subnetwork in the basic cold chain transportation risk prediction network are trained to generate a target cold chain transportation risk prediction network, and the target cold chain transportation risk prediction network is used to annotate risk knowledge points of cold chain transportation logistics data; The basic feature extraction subnetwork includes a first graph autoencoder and a second graph autoencoder; the basic feature extraction subnetwork and the basic feature prediction subnetwork in the basic cold chain transportation risk prediction network are trained based on the correlation weight sample set to generate a target cold chain transportation risk prediction network, including: Obtaining a priori logistics knowledge logic vector corresponding to the sample cold chain transportation logistics data, and training the first graph autoencoder based on the priori logistics knowledge logic vector and the set of sample logistics knowledge logic vectors; Acquire a priori transportation environment change vector corresponding to the sample transportation environment change data, and train the second graph autoencoder based on the priori transportation environment change vector and the sample transportation environment change vector; Obtaining the sample annotated risk knowledge points corresponding to the sample cold chain transport logistics data, and generating sample matching scores corresponding to the K sample risk knowledge points respectively based on the correlation weighted sample set; Perform exponential conversion on the K sample matching scores respectively to generate K score conversion values; Determine the label values ​​corresponding to the K sample risk knowledge points respectively according to the sample annotated risk knowledge points and the sample matching scores; the label values ​​include a first label result or a second label result, the first label result indicates that there are risk knowledge points consistent with the sample risk knowledge points corresponding to the first label result in the sample annotated risk knowledge points, and the second label result indicates that there are no risk knowledge points consistent with the sample risk knowledge points corresponding to the second label result in the sample annotated risk knowledge points; Generate a training error parameter based on the label values ​​corresponding to the K sample risk knowledge points and the K score conversion values; The basic feature prediction subnetwork in the basic cold chain transportation risk prediction network is trained based on the training error parameter to generate a target cold chain transportation risk prediction network.

9. A cold chain monitoring system, characterized in that: The cold chain monitoring system includes a processor and a readable storage medium, wherein the readable storage medium stores a program, and when the program is executed by the processor, the data analysis method based on cold chain transportation according to any one of claims 1 to 8 is implemented.