Multidimensional analysis method and system of pharmaceutical cold chain logistics data based on AI robot

Abstract

The present invention provides a method and system for multidimensional analysis of pharmaceutical cold chain logistics data based on an AI robot, which relates to the field of artificial intelligence technology. By obtaining a sample pharmaceutical cold chain logistics data sequence and determining a corresponding cold chain transportation element sequence, an initialized AI robot network is used to perform a multidimensional knowledge space conversion on the cold chain transportation elements to generate sample multidimensional knowledge space features. By identifying the first cold chain transportation element and the second cold chain transportation element in a target element dependency state, and obtaining the corresponding target dependency state features, the element knowledge features of the second cold chain transportation element can be accurately determined. Based on the deviation between the element knowledge features and the second sample multidimensional knowledge space features, the target training error of the AI ​​robot network is calculated, and the network weight information is updated accordingly to generate an updated AI robot network. This method effectively improves the accuracy and efficiency of pharmaceutical cold chain logistics data analysis and optimizes the application performance in the field of pharmaceutical cold chain logistics.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a multidimensional analysis method and system for pharmaceutical cold chain logistics data based on an AI robot. Background Art

[0002] In the pharmaceutical field, cold chain logistics plays a vital role. Pharmaceutical products, especially biological products and vaccines, are extremely sensitive to environmental conditions such as temperature and humidity. Therefore, strict temperature control and environmental monitoring must be maintained during transportation. With the rapid development of the Internet of Things, big data and artificial intelligence technologies, the digital and intelligent management of pharmaceutical cold chain logistics has become possible.

[0003] Traditional pharmaceutical cold chain logistics data analysis methods often rely on manual experience or simple statistical analysis, which are incapable of processing massive, multi-dimensional cold chain logistics data. On the one hand, manual analysis is difficult to capture the complex relationships and potential laws in the data; on the other hand, simple statistical analysis methods often cannot fully mine the valuable information in the data and are difficult to meet the refined management needs of pharmaceutical cold chain logistics. Summary of the invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a multidimensional analysis method of pharmaceutical cold chain logistics data based on an AI robot, the method comprising:

[0005] Acquire a sample pharmaceutical cold chain logistics data sequence, and determine a cold chain transportation element sequence corresponding to the sample pharmaceutical cold chain logistics data sequence;

[0006] Based on the sample pharmaceutical cold chain logistics data in the sample pharmaceutical cold chain logistics data sequence and the initialized AI robot network, each cold chain transportation element in the cold chain transportation element sequence is converted into a multidimensional knowledge space to generate a sample multidimensional knowledge space feature;

[0007] Determine a first cold chain transport element and a second cold chain transport element in the cold chain transport element sequence that have a target element dependency state, and obtain a target dependency state feature corresponding to the target element dependency state;

[0008] Based on the first sample multidimensional knowledge space feature corresponding to the first cold chain transportation element and the target dependent state feature, determine the element knowledge feature corresponding to the second cold chain transportation element, and determine the target training error corresponding to the AI ​​robot network based on the second sample multidimensional knowledge space feature corresponding to the second cold chain transportation element and the deviation of the element knowledge feature;

[0009] The network weight information of the AI ​​robot network is updated based on the target training error to generate an updated AI robot network.

[0010] On the other hand, an embodiment of the present invention also provides a multidimensional analysis system for pharmaceutical cold chain logistics data based on an AI robot, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0011] Based on the above aspects, the embodiment of the present application obtains a sample pharmaceutical cold chain logistics data sequence and determines the corresponding cold chain transport element sequence, and uses the initialized AI robot network to perform multi-dimensional knowledge space conversion on the cold chain transport elements to generate sample multi-dimensional knowledge space features. Furthermore, by identifying the first cold chain transport element and the second cold chain transport element that exist in a target element dependency state, and obtaining the corresponding target dependency state characteristics, the element knowledge characteristics of the second cold chain transport element can be accurately determined. Based on the deviation between the element knowledge characteristics and the second sample multi-dimensional knowledge space characteristics, the target training error of the AI ​​robot network is calculated, and the network weight information is updated accordingly to generate an updated AI robot network. This method effectively improves the accuracy and efficiency of pharmaceutical cold chain logistics data analysis, optimizes the application performance of the AI ​​robot network in the field of pharmaceutical cold chain logistics, and thus contributes to the intelligent management and decision-making of pharmaceutical cold chain logistics. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of the execution flow of the multidimensional analysis method of pharmaceutical cold chain logistics data based on AI robots provided in an embodiment of the present invention.

[0013] Figure 2 It is a schematic diagram of the hardware architecture of the multidimensional analysis system for pharmaceutical cold chain logistics data based on AI robots provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a multidimensional analysis method for pharmaceutical cold chain logistics data based on an AI robot provided by an embodiment of the present invention. The multidimensional analysis method for pharmaceutical cold chain logistics data based on an AI robot is introduced in detail below.

[0015] Step S110, obtaining a sample pharmaceutical cold chain logistics data sequence, and determining a cold chain transportation element sequence corresponding to the sample pharmaceutical cold chain logistics data sequence.

[0016] In this embodiment, for a certain pharmaceutical cold chain logistics platform, the sample pharmaceutical cold chain logistics data sequence can be obtained from the logistics management system of the pharmaceutical cold chain logistics platform. The logistics management system records various types of data during each vaccine transportation process, such as the start time and end time of the transportation, the temperature change record during the transportation process (the temperature value is recorded once every certain interval), the model of the transportation vehicle, the driver’s qualification information, the transportation route, the loading and unloading location information, and the type and batch of the vaccine, etc., thereby forming a sample pharmaceutical cold chain logistics data sequence.

[0017] For the cold chain transportation element sequence, taking the transportation of vaccines as an example, the cold chain transportation elements can include transportation tools (such as refrigerated trucks), temperature control equipment (such as refrigeration units), packaging materials (such as insulated boxes), transportation personnel (drivers, etc.), transportation routes (including cities passed through, road conditions, etc.), and goods (vaccines). Each sample pharmaceutical cold chain logistics data is associated with these cold chain transportation elements. For example, the transportation start time and end time are related to the transportation route and transportation personnel, the temperature change record is related to the temperature control equipment and packaging materials, the vehicle model is related to the transportation tool, the driver qualification information is related to the transportation personnel, and the type and batch of vaccines are related to the goods. Through this association relationship, the cold chain transportation element sequence corresponding to the sample pharmaceutical cold chain logistics data sequence can be determined.

[0018] Step S120, based on the sample medical cold chain logistics data in the sample medical cold chain logistics data sequence and the initialized AI robot network, each cold chain transportation element in the cold chain transportation element sequence is converted into a multi-dimensional knowledge space to generate a sample multi-dimensional knowledge space feature.

[0019] Continuing with the example of vaccine transportation, after the AI ​​robot network is initialized, the cold chain transportation elements are processed. Exemplarily, the initialized AI robot network is a multi-layer neural network structure. For the cold chain transportation element of transportation tools, relevant data of refrigerated trucks are extracted from the sample medical cold chain logistics data, such as the purchase time, maintenance records, refrigeration capacity, etc. of the vehicle. Then, the above relevant data are input into the AI ​​robot network, and the neurons in the AI ​​robot network process these relevant data according to pre-set weights and activation functions. In the process of multi-dimensional knowledge space conversion, the purchase time of the vehicle can be converted into the concept of vehicle service life, the maintenance records can be converted into vehicle reliability indicators, the refrigeration capacity can be converted into refrigeration efficiency indicators, etc. These converted indicators together constitute some of the characteristics of the transportation tools in the multi-dimensional knowledge space.

[0020] Looking at the temperature control equipment factor, the temperature change record is input as part of the sample pharmaceutical cold chain logistics data. The AI ​​robot network may analyze the temperature fluctuation range, fluctuation frequency, and the degree of deviation from the preset standard temperature. These analysis results are converted into temperature stability indicators, temperature adjustment timeliness indicators, etc. These indicators become the characteristic components of temperature control equipment in the multidimensional knowledge space.

[0021] For packaging materials, data such as the material and thickness of the insulation box are input. The network will calculate indicators such as insulation effect and thermal insulation performance based on these data, thereby constructing the characteristics of packaging materials in the multidimensional knowledge space.

[0022] For transport personnel, data such as the driver’s driving experience (estimated based on qualification information and previous transport records) and fatigue driving risk (judged based on data such as transport duration) are processed and converted into multi-dimensional knowledge space features such as personnel professional quality indicators and safety risk indicators.

[0023] In terms of transportation routes, data such as traffic congestion in the cities along the route (obtained through external traffic data) and road slope (geographic information data) are converted into route smoothness indicators, transportation difficulty indicators, etc.

[0024] In terms of goods (vaccines), data such as the type of vaccine (for example, some vaccines are more sensitive to temperature) and batch quality requirements are converted into goods sensitivity indicators, quality assurance demand indicators, etc. All these indicators converted for each cold chain transportation element together constitute the multi-dimensional knowledge space characteristics of the sample.

[0025] Step S130, determining the first cold chain transportation element and the second cold chain transportation element in the cold chain transportation element sequence that have a target element dependency state, and obtaining target dependency state features corresponding to the target element dependency state.

[0026] In the pharmaceutical cold chain logistics, the transport tool (refrigerated truck) is considered as the first cold chain transport element, and the temperature control equipment (refrigeration unit) is considered as the second cold chain transport element. There is a target element dependency state between them. This is because the refrigeration effect of the refrigerated truck depends on the normal operation of the refrigeration unit.

[0027] In order to obtain the target dependency state characteristics corresponding to the target element dependency state, the target dependency state category corresponding to the target element dependency state is first determined. In this example, since the refrigeration unit is the core component of the refrigeration function of the refrigerated truck, this dependency relationship can be determined as an inclusion dependency state category. For the inclusion dependency state category, the corresponding dependency state characteristics may include the contribution rate of the refrigeration unit to the overall refrigeration function of the refrigerated truck, the degree of influence of the refrigeration unit failure on the refrigeration effect of the refrigerated truck, etc.

[0028] Specifically, from the equipment maintenance records and technical specifications of logistics companies, we can obtain the proportion of the refrigeration power of the refrigeration unit to the total refrigeration capacity of the refrigerated truck, which is a manifestation of the contribution rate. At the same time, by simulating refrigeration unit failure experiments or actual failure case analysis, we can determine the speed at which the temperature inside the refrigerated truck rises when the refrigeration unit fails, the time to reach a dangerous temperature (a temperature that may affect the quality of vaccines), etc. These data reflect the degree of impact of the failure on the refrigeration effect, thereby determining the target dependent state characteristics.

[0029] For another example, the transportation route is taken as the first cold chain transportation element, and the packaging materials are taken as the second cold chain transportation element. The geographical environment of the transportation route (such as many mountainous areas and changeable climate) may affect the requirements for packaging materials, which is also a factor dependency state. This dependency state may belong to the category of environmental impact dependency state. The corresponding dependency state characteristics may include the additional protection capabilities that packaging materials need to have in different geographical environments (such as better shock resistance in mountainous areas and better moisture resistance in areas with changeable climates). By analyzing historical transportation data and comparing data such as the damage rate and moisture rate of the same vaccine packaging on different routes, these target dependency state characteristics can be determined.

[0030] Step S140, based on the first sample multidimensional knowledge space characteristics corresponding to the first cold chain transport element and the target dependent state characteristics, determine the element knowledge characteristics corresponding to the second cold chain transport element, and determine the target training error corresponding to the AI ​​robot network based on the second sample multidimensional knowledge space characteristics corresponding to the second cold chain transport element and the deviation of the element knowledge characteristics.

[0031] Take the transport tool (refrigerated truck) as the first cold chain transport element and the temperature control equipment (refrigeration unit) as the second cold chain transport element as an example. The first sample multidimensional knowledge space characteristics corresponding to the refrigerated truck include indicators such as vehicle age and refrigeration efficiency. The target dependent state characteristics corresponding to the refrigeration unit include the contribution rate of the refrigeration unit in the overall refrigeration function of the refrigerated truck, the impact of the refrigeration unit failure on the refrigeration effect of the refrigerated truck, etc.

[0032] The preset deep neural network model is used to perform preliminary feature extraction on the multi-dimensional knowledge space features of the first sample of refrigerated trucks. Assume that this deep neural network model has multiple hidden layers, and neurons are connected by nonlinear functions. For data such as vehicle age and refrigeration efficiency, in the first hidden layer of the network, neurons perform weighted summation based on the input value and their own weights, and then process them through activation functions (such as ReLU functions) to obtain preliminary low-level feature representations. For example, the vehicle age and refrigeration efficiency may be comprehensively converted into a feature that represents the trend of the vehicle's overall refrigeration performance over time.

[0033] This low-level feature representation is input into the feature encoding layer, which uses an autoencoder structure. The encoder part of the autoencoder compresses and encodes this low-level feature, converting it into a more compact encoded feature, and then the decoder part decodes and reconstructs this encoded feature. In this way, some redundant information can be removed and a more representative compressed encoded feature can be obtained.

[0034] The compressed coded features are semantically parsed by the semantic parsing module. For example, the coded features are mapped into a semantic space representing the performance of refrigeration equipment to generate a first parsed feature vector with corresponding semantic meaning, which may contain basic attribute information of the refrigeration performance of the refrigerated truck (such as the refrigeration capacity level) and its semantic association in the multidimensional knowledge space (such as the comparative relationship with the refrigeration performance of similar vehicles) and contextual information (such as performance evaluation in the context of refrigeration demand in the current transportation task).

[0035] The first parsed feature vector is concatenated with the target dependent state feature of the refrigeration unit to generate a joint feature vector. The attention mechanism is used to dynamically adjust the weight of the first parsed feature vector in the joint feature vector according to the importance of the target dependent state feature. For example, if the feature of the degree of impact of the refrigeration unit failure on the refrigeration effect of the refrigeration truck is very critical in the current task, then in the joint feature vector, the part related to this feature will be given a higher weight.

[0036] The fused joint feature vector is enhanced through the residual network to increase the expressive power of the feature. This fused and enhanced joint feature representation is then input into a pre-built mapping conversion model (such as a decision tree model). The decision tree model converts this feature representation into the initial element knowledge feature corresponding to the refrigeration unit based on the learned mapping relationship. This feature reflects the attributes, state or behavior pattern of the refrigeration unit under a given refrigeration performance dependency state of the refrigeration truck. For example, it is possible to obtain a recommendation for adjusting the current performance of the refrigeration unit (such as whether maintenance is required to adapt to the refrigeration needs of the refrigeration truck).

[0037] By comparing the preset standard features (such as the optimal operating state features of the refrigeration unit) or using domain knowledge (such as the industry standard performance indicators of the refrigeration unit), the initial element knowledge features are verified and adjusted to generate the first intermediate element knowledge features. For example, if the performance adjustment suggestions in the initial element knowledge features do not match the maintenance cycle in the industry standard, then adjustments are made.

[0038] The particle swarm optimization algorithm is used to perform global feature optimization on the first intermediate factor knowledge feature to generate the optimized second intermediate factor knowledge feature. For example, in the particle swarm optimization algorithm, each possible performance adjustment scheme is regarded as a particle, and the optimal performance adjustment scheme is found as the second intermediate factor knowledge feature through information sharing and iterative optimization between particles.

[0039] The second intermediate element knowledge feature is input into the semantic consistency verification model, and the semantic consistency verification model performs consistency verification based on semantic similarity calculation and preset semantic standards (such as the semantic description of the normal operation of the refrigeration unit). If the verification passes, the verified third intermediate element knowledge feature is output. If the verification fails, corrections and adjustments are made according to the feedback information until the semantic consistency requirements are met.

[0040] Analyze the contextual information of the third intermediate factor knowledge feature in the multidimensional knowledge space. For example, consider the impact of contextual information such as the urgency of the current transportation task and the value of the transported goods on the performance adjustment of the refrigeration unit. After context association enhancement, obtain the fourth intermediate factor knowledge feature, verify and evaluate it, and generate the fifth intermediate factor knowledge feature.

[0041] Monitor and analyze the state changes (such as real-time temperature adjustment data, operating noise changes, etc.) and behavior evolution (such as the switching frequency of the cooling mode, etc.) of the refrigeration unit, and dynamically update the fifth intermediate element knowledge features based on these results to generate the sixth intermediate element knowledge features.

[0042] The domain analysis module is used to compare and analyze the features or entities in different domains. For example, the technical performance domain of the refrigeration unit is compared with the transportation cost domain, and the commonalities (such as both are related to transportation efficiency) and differences (such as technical performance focuses on refrigeration effect, and transportation cost focuses on economic cost) between them are identified and analyzed. The domain analysis results are obtained and a cross-domain mapping model is built based on them. Using this cross-domain mapping model, the sixth intermediate element knowledge feature is mapped from the technical performance domain to the transportation cost domain, while maintaining the consistency of key information and semantic meaning, and finally the element knowledge feature corresponding to the refrigeration unit is obtained.

[0043] The target training error portion corresponding to the AI ​​robot network is determined based on the second sample multidimensional knowledge space characteristics corresponding to the second cold chain transport element and the deviation of the element knowledge characteristics. Assume that the second sample multidimensional knowledge space characteristics corresponding to the refrigeration unit include indicators such as actual operating temperature and temperature adjustment frequency. Compare these indicators with the element knowledge characteristics obtained previously (such as the ideal operating state characteristics corresponding to the performance adjustment suggestions) and calculate the deviation. For example, the difference between the actual operating temperature and the ideal operating temperature, the difference between the temperature adjustment frequency and the recommended adjustment frequency, etc. Determine the first error parameter based on these deviations.

[0044] The linkage transport elements corresponding to the refrigerated trucks in the sample pharmaceutical cold chain logistics data are obtained, such as the road slope in the transport route. The deviation between the third sample multidimensional knowledge space characteristics corresponding to the road slope (such as the impact index of the slope on the vehicle speed, the indirect impact index on the refrigeration effect, etc.) and the first sample multidimensional knowledge space characteristics corresponding to the refrigerated truck (such as refrigeration efficiency, etc.) is calculated to determine the second error parameter.

[0045] Obtain the first influence coefficient (for example, the influence weight of the refrigerated truck's own performance on the overall transportation effect determined based on expert experience) and the second influence coefficient (for example, the influence weight of external factors such as road slope on the transportation effect). Update the first error parameter based on the first influence coefficient, update the second error parameter based on the second influence coefficient, and then sum the updated first error parameter and the second error parameter to generate the target training error corresponding to the AI ​​robot network.

[0046] Step S150, updating the network weight information of the AI ​​robot network based on the target training error to generate an updated AI robot network.

[0047] In the above vaccine transportation example, once the target training error corresponding to the AI ​​robot network is determined, this error can be used to update the network's weight information. Assume that the AI ​​robot network is a neural network based on the back-propagation algorithm. According to the target training error, calculate the contribution of each weight to the error (by calculating the partial derivative of the error with respect to the weight). For example, for the weights involved in calculating the relevant features of the first cold chain transport element (refrigerated truck), according to the size and direction of the target training error, adjust the values ​​of these weights according to a certain learning rate (a pre-set small value used to control the step size of the weight update).

[0048] If the target training error indicates that the AI ​​robot network over-predicted the refrigeration performance of the refrigerated truck (i.e., the actual refrigeration effect is not as good as the network predicted), then the weights corresponding to the input data related to the refrigerated truck (such as the age of the vehicle, refrigeration efficiency, etc.) may be appropriately reduced, so that the network's output will be closer to the actual situation the next time similar data is processed. Similarly, the weights involved in calculating the features related to the second cold chain transportation element (refrigeration unit) are also adjusted according to their contribution to the target training error.

[0049] By adjusting all relevant weights in the network, an updated AI robot network is generated. When processing the next batch of sample pharmaceutical cold chain logistics data, the updated network will be able to more accurately perform multi-dimensional knowledge space conversion on cold chain transportation elements, analyze element dependency status, and determine element knowledge characteristics, thereby continuously improving the analysis and processing capabilities of pharmaceutical cold chain logistics data and better ensuring the quality and efficiency of pharmaceutical cold chain transportation.

[0050] In another example, consider a pharmaceutical company transporting a batch of special drugs, which involves special packaging materials and transportation routes. Packaging materials are the first cold chain transportation element, and transportation routes are the second cold chain transportation element. The first sample multidimensional knowledge space features corresponding to packaging materials may include indicators such as thermal insulation performance and moisture resistance, and the target-dependent state features corresponding to transportation routes may include special requirements for packaging under climatic conditions on the route.

[0051] Following the process of determining the element knowledge features in the previous steps, through a series of operations such as deep neural network models, feature encoding layers, and semantic parsing modules, the element knowledge features corresponding to the transportation route are obtained, such as the risk assessment of the transportation route under specific packaging materials (such as the assessment of the risk of packaging getting damp in humid climate areas). Then, the first error parameter is determined based on the deviation between the second sample multidimensional knowledge space features corresponding to the transportation route (such as actual weather condition records, road conditions, etc.) and the element knowledge features. At the same time, the linkage transportation elements related to the packaging materials (such as the impact of the vibration of the transport vehicle on the packaging) are considered to calculate the corresponding second error parameter.

[0052] After determining the target training error according to the first error parameter and the second error parameter, the weight information of the AI ​​robot network is updated according to the back propagation algorithm. For example, if it is found that the network's risk assessment of the transportation route is inaccurate (the target training error is large), then adjust the weights related to the calculation of transportation route characteristics, such as the weights corresponding to data such as climate conditions and road conditions, so that the updated AI robot network can more accurately analyze various situations in the transportation process of special drugs and ensure the quality and safety of drugs.

[0053] Based on the above steps, the embodiment of the present application obtains a sample pharmaceutical cold chain logistics data sequence and determines the corresponding cold chain transport element sequence, and uses the initialized AI robot network to perform multi-dimensional knowledge space conversion on the cold chain transport elements to generate sample multi-dimensional knowledge space features. Furthermore, by identifying the first cold chain transport element and the second cold chain transport element that have a target element dependency state, and obtaining the corresponding target dependency state characteristics, the element knowledge characteristics of the second cold chain transport element can be accurately determined. Based on the deviation between the element knowledge characteristics and the second sample multi-dimensional knowledge space characteristics, the target training error of the AI ​​robot network is calculated, and the network weight information is updated accordingly to generate an updated AI robot network. This method effectively improves the accuracy and efficiency of pharmaceutical cold chain logistics data analysis, optimizes the application performance of the AI ​​robot network in the field of pharmaceutical cold chain logistics, and thus contributes to the intelligent management and decision-making of pharmaceutical cold chain logistics.

[0054] In a possible implementation, step S130 includes: determining a target dependency state category corresponding to the target element dependency state, determining the corresponding target dependency state feature based on the target dependency state category, and the target dependency state category is determined based on the element category of the first cold chain transport element in the target logistics scenario.

[0055] In a possible implementation, the determining the target dependency state category corresponding to the target element dependency state and determining the corresponding target dependency state feature based on the target dependency state category includes: determining the target dependency state category corresponding to the target element dependency state. When the target dependency state category is an inclusion dependency state category, determining the dependency state feature corresponding to the inclusion dependency state category as the target dependency state feature.

[0056] In this embodiment, the previously mentioned pharmaceutical cold chain logistics scenario is taken as an example, such as the case of transporting vaccines. Assume that the means of transport (refrigerated truck) is the first cold chain transportation element, and the temperature control equipment (refrigeration unit) is an element that has a dependency relationship with the means of transport. In this scenario, first determine the target dependency state category corresponding to the target element dependency state. The element category of the means of transport (refrigerated truck) in the entire pharmaceutical cold chain logistics target scenario is the basic transportation facility category. Based on this, its dependency state category with the temperature control equipment (refrigeration unit) can be determined as including a dependency state category. This is because the refrigeration unit is an indispensable part of the refrigeration function of the refrigeration truck and is included in the refrigeration system of the refrigeration truck.

[0057] For this type of dependent state category, the corresponding dependent state feature needs to be determined based on the relationship between the two. From a technical perspective, the dependent state feature can start with the contribution rate of the refrigeration unit in the overall refrigeration function of the refrigerated truck. This contribution rate can be reflected by the ratio of the refrigeration power of the refrigeration unit to the total refrigeration capacity of the refrigerated truck. For example, if the refrigeration power of the refrigeration unit is 10 kilowatts and the total refrigeration capacity of the refrigerated truck is 15 kilowatts, then its contribution rate is 10 / 15 = 2 / 3. In addition, the degree of influence of the refrigeration unit failure on the refrigeration effect of the refrigerated truck is also an important part of the dependent state feature. Through actual tests or analysis of previous failure cases, if the refrigeration unit fails, the speed of the temperature rise inside the refrigerated truck, the time to reach the dangerous temperature that may affect the quality of the vaccine, and other data can be monitored to quantify the degree of this influence. For example, after testing, after the refrigeration unit fails, the temperature inside the refrigerated truck rises by 5 degrees Celsius per hour, and reaches the dangerous temperature that may affect the quality of the vaccine after 3 hours. These data clearly reflect the degree of influence of the refrigeration unit failure on the refrigeration effect of the refrigerated truck, thereby determining the target dependent state feature corresponding to the dependent state category.

[0058] Let's take the transportation route as the first cold chain transportation element and the packaging material as the element with which it has a dependency relationship. The element category of the transportation route in the target scenario of pharmaceutical cold chain logistics is the transportation path planning category. Its dependency state category with the packaging material can also be determined as an inclusive dependency state category, because the selection of packaging materials needs to consider the characteristics of the transportation route. For this inclusive dependency state category, its dependency state characteristics can be determined from the additional protection capabilities that packaging materials need to have in different geographical environments. If the transportation route passes through mountainous areas, the packaging materials need better shock resistance; if it passes through areas with changeable climates, better moisture resistance is required. By analyzing historical transportation data and comparing data such as the damage rate and moisture rate of the same vaccine packaging on different routes, the demand for this protection capability can be quantified, and then the target dependency state characteristics can be determined. For example, on a transportation route that passes through mountainous areas, statistics show that when vaccines are transported with packaging materials without special shockproof measures, the packaging damage rate reaches 10%, which indicates that packaging materials on this route need to have corresponding shockproof capabilities as part of the target-dependent state characteristics; on a route that passes through a climate-changing and humid area, it was found that when vaccines are transported with packaging materials without special moisture-proof measures, the moisture rate reaches 8%, which reflects the moisture-proof capabilities that packaging materials need to have on this route in the target-dependent state characteristics. In this way, the corresponding target-dependent state characteristics are determined based on the target element dependency state category.

[0059] In a possible implementation, step S130 may further include:

[0060] Step S131, obtaining a logistics element relationship graph, wherein the logistics element relationship graph includes feature members corresponding to each relationship graph element, and feature links representing element dependency states between the feature members.

[0061] Step S132, obtaining the first cold chain transport element in the cold chain transport element sequence, and determining the second cold chain transport element in the cold chain transport element sequence that has a characteristic link association with the first cold chain transport element from the logistics element relationship map.

[0062] Step S133, obtaining the element dependency state corresponding to the characteristic link of the first cold chain transportation element and the second cold chain transportation element, and outputting it as the target element dependency state.

[0063] In this embodiment, in the pharmaceutical cold chain logistics scenario, for example, during the transportation of vaccines, there are many cold chain transportation elements, such as transportation vehicles, temperature control systems, packaging materials, transportation personnel, transportation routes, and the goods (vaccines) themselves.

[0064] First, building a logistics element relationship map is the foundation. This map covers in detail the characteristic members corresponding to each relationship map element and the characteristic links that represent the element dependency status between these characteristic members. Taking the transport vehicle (refrigerated truck) as the first cold chain transport element as an example, in the logistics element relationship map, the characteristic members corresponding to the transport vehicle relationship map element include the vehicle model, refrigeration capacity, load capacity, mileage, etc. The characteristic links in the map describe the dependency relationship between the transport vehicle and other elements.

[0065] After determining the first cold chain transport element, the transport vehicle, from the cold chain transport element sequence, the second cold chain transport element associated with the transport vehicle through a feature link can be determined by querying the logistics element relationship map. For example, the temperature control system (refrigeration unit) is associated with the transport vehicle because the refrigeration unit is a key device to ensure the low temperature environment in the transport vehicle. This association between the two is the element dependency relationship represented by the feature link. For another example, packaging materials are also associated with transport vehicles because the characteristics of packaging materials need to match the transportation conditions of the transport vehicle. For example, the degree of bumpiness of the vehicle may affect the requirements for the shockproof performance of the packaging materials. These associations are all reflected in the logistics element relationship map through feature links.

[0066] Obtain the element dependency state corresponding to the characteristic link between the transport vehicle and the temperature control system (or other related elements such as packaging materials). This element dependency state is the target element dependency state. For transport vehicles and refrigeration units, the element dependency state may be that the refrigeration capacity of the refrigeration unit must meet the temperature control requirements of the internal space of the transport vehicle and the goods (vaccines); for transport vehicles and packaging materials, the element dependency state may be that the protective performance of the packaging materials must adapt to the transportation environment of the transport vehicle.

[0067] In a possible implementation, the second cold chain transportation element is a relationship graph element in the logistics element relationship graph, which has a target element dependency state with the first cold chain transportation element.

[0068] Step S140 may include:

[0069] Step S141, determining a first error parameter based on the second sample multidimensional knowledge space feature corresponding to the second cold chain transportation element and the deviation of the element knowledge feature.

[0070] Step S142, obtaining the linkage transport element corresponding to the first cold chain transport element in the sample pharmaceutical cold chain logistics data.

[0071] Step S143: determining a second error parameter based on the degree of deviation between the third sample multidimensional knowledge space feature corresponding to the coordinated transportation element and the first sample multidimensional knowledge space feature.

[0072] Step S144: determine the target training error corresponding to the AI ​​robot network based on the first error parameter and the second error parameter.

[0073] Taking the temperature control system (refrigeration unit) as the second cold chain transport element as an example, its corresponding second sample multidimensional knowledge space features include data such as refrigeration efficiency, temperature fluctuation range, and energy consumption of the refrigeration unit in actual operation. The element knowledge features corresponding to the refrigeration unit determined by the previous steps may be ideal refrigeration performance indicators, optimal energy consumption range, etc. based on relevant factors such as the transport vehicle (the first cold chain transport element). The deviation between the two is calculated to determine the first error parameter. For example, if the actual refrigeration efficiency of the refrigeration unit is lower than the ideal refrigeration efficiency, the deviation is positive, and the specific value of the deviation can be determined based on the ratio of the difference between the two to the ideal refrigeration efficiency, and this value becomes part of the first error parameter.

[0074] Then, the linkage transport element corresponding to the transport vehicle (first cold chain transport element) in the sample pharmaceutical cold chain logistics data is obtained. For example, the transport route is a linkage transport element. The third sample multidimensional knowledge space features corresponding to the transport route include the altitude change of the route, road conditions (flatness, slope, etc.), meteorological conditions along the way, etc. The first sample multidimensional knowledge space features corresponding to the transport vehicle include the driving speed, fuel consumption, and refrigeration efficiency of the refrigeration unit (because the conditions of the transport route will affect the working state of the refrigeration unit). The deviation between the third sample multidimensional knowledge space features corresponding to the transport route and the first sample multidimensional knowledge space features corresponding to the transport vehicle is calculated to determine the second error parameter. For example, if the altitude change of the transport route is large, resulting in a decrease in the driving speed of the transport vehicle and an increase in fuel consumption, there is a deviation from the ideal state on a flat route, and this deviation is quantified as part of the second error parameter.

[0075] Finally, the target training error corresponding to the AI ​​robot network is determined based on the first error parameter and the second error parameter. For example, the weight of the first error parameter is set to 0.6, and the weight of the second error parameter is set to 0.4 (these weights can be determined based on actual conditions or experience). Multiply the first error parameter by its weight, multiply the second error parameter by its weight, and then add the two together to get the target training error. If the first error parameter is 0.2 and the second error parameter is 0.3, then the target training error is 0.2×0.6 + 0.3×0.4 = 0.24. The target training error will be used to subsequently update the network weight information of the AI ​​robot network to improve the accuracy of the network's processing and analysis of pharmaceutical cold chain logistics data, thereby better ensuring the reasonable configuration and efficient operation of various elements in the pharmaceutical cold chain transportation process, and ensuring the quality and safety of vaccines and other drugs during transportation.

[0076] Taking another scenario as an example, when transporting special medicines, the transport personnel serve as the first cold chain transport element. In the logistics element relationship graph, the characteristic members corresponding to the transport personnel relationship graph element include personnel qualifications, driving experience, fatigue driving risk assessment, etc. From the graph, it can be determined that the second cold chain transport element, such as the transport vehicle, has a characteristic link association with the transport personnel. Because the driving operation of the transport personnel directly affects the driving status of the transport vehicle. The element dependency state between them may be that the driving skills of the transport personnel must adapt to the operating requirements of the transport vehicle, and the performance of the transport vehicle will also affect the operating difficulty of the transport personnel.

[0077] For transport vehicles (the second cold chain transport element), the corresponding second sample multidimensional knowledge space features include the actual driving speed, steering flexibility, braking performance, etc. of the vehicle. The element knowledge features corresponding to the transport vehicle determined based on factors such as transport personnel may be the ideal driving speed range, optimal steering and braking performance indicators, etc. The deviation between the two is calculated to determine the first error parameter. For example, if the actual driving speed often exceeds the ideal driving speed range, the deviation is positive, and the first error parameter is obtained through specific numerical calculations.

[0078] The linkage transport element corresponding to the transport personnel may be packaging materials. The third sample multidimensional knowledge space characteristics corresponding to the packaging materials include the protective effect of the packaging materials, the adaptability to changes in the transportation environment, etc. The operational stability in the first sample multidimensional knowledge space characteristics corresponding to the transport personnel (stable driving helps protect the drugs in the packaging materials), etc. The deviation between the two is calculated to determine the second error parameter. Finally, the target training error is determined according to the weight calculation method mentioned above to optimize the AI ​​robot network, thereby improving the accuracy of the analysis and processing of cold chain transportation elements in special drug transportation scenarios.

[0079] In a possible implementation, step S144 may include:

[0080] Step S1441, obtaining a first influence coefficient and a second influence coefficient.

[0081] Step S1442: Update the first error parameter based on the first influence coefficient to generate an updated first error parameter, and update the second error parameter based on the second influence coefficient to generate an updated second error parameter.

[0082] Step S1443, summing the updated first error parameter and the updated second error parameter to generate the target training error corresponding to the AI ​​robot network.

[0083] In this embodiment, the determination of the first influence coefficient and the second influence coefficient needs to be based on a deep understanding of the importance of each element in the pharmaceutical cold chain logistics system.

[0084] Taking the relationship between transport vehicles (refrigerated trucks) and temperature control systems (refrigeration units) as an example, it is assumed that the first error parameter is calculated based on the deviation of the second sample multidimensional knowledge space characteristics and element knowledge characteristics corresponding to the refrigeration unit. This deviation reflects the difference between the actual operation and the ideal operation state of the refrigeration unit. If the performance of the transport vehicle (refrigerated truck) has a more critical impact on the overall cold chain transportation effect, such as the vehicle's refrigeration capacity and driving stability, which are directly related to the temperature stability and safety of the vaccine during transportation, then the first impact coefficient may be set to a relatively large value, assuming it is 0.7. The setting of this coefficient is based on the analysis of a large amount of transportation data and industry experience, indicating that the performance of the transport vehicle occupies a relatively important position in the entire cold chain transportation link.

[0085] The second error parameter is determined based on the deviation of the relevant characteristics of the linkage transport elements (such as transport routes) corresponding to the transport vehicles in the sample pharmaceutical cold chain logistics data. If the conditions of the transport route (such as road conditions, altitude changes, meteorological conditions, etc.) have a certain impact on the coordinated work of the transport vehicle and the refrigeration unit, but the impact is slightly smaller than the performance of the transport vehicle itself, then the second impact coefficient may be set to 0.3.

[0086] The first error parameter is updated based on the first influence coefficient, that is, the first error parameter is multiplied by the first influence coefficient to generate an updated first error parameter. For example, if the first error parameter is 0.2, then the updated first error parameter is 0.2 * 0.7 = 0.14. Similarly, the second error parameter is updated based on the second influence coefficient. Assuming that the second error parameter is 0.3, the updated second error parameter is 0.3 * 0.3 = 0.09.

[0087] Then the updated first error parameter and the updated second error parameter are summed up, 0.14 + 0.09 = 0.23. This 0.23 is the target training error corresponding to the AI ​​robot network. The target training error reflects the comprehensive deviation between different elements in the entire system and will be used to adjust the AI ​​robot network to improve its analysis and prediction capabilities for pharmaceutical cold chain logistics data.

[0088] In a possible implementation, the method further includes:

[0089] A feature update vector is determined based on the target training error.

[0090] Based on the feature update vector, the target dependency state feature is updated to generate an updated target dependency state feature.

[0091] In this embodiment, the feature update vector is determined based on the target training error of 0.23 calculated above. The determination of the feature update vector needs to consider the relationship between the target training error and each cold chain transportation element. For example, in the relationship between the transport vehicle and the refrigeration unit, the target training error may indicate that the deviation between the actual operating state of the refrigeration unit and the ideal state has an impact on the entire cold chain transportation. Then the feature update vector may be adjusted for the key features of the refrigeration unit, such as refrigeration power, temperature adjustment speed, etc.

[0092] The target dependent state feature is updated based on this feature update vector. Assume that the target dependent state feature includes the degree of influence of the refrigeration unit failure on the refrigeration effect of the refrigerated truck and the contribution rate of the refrigeration unit in the overall refrigeration function of the refrigerated truck. If the feature update vector indicates that the refrigeration power of the refrigeration unit needs to be adjusted, this may affect the contribution rate of the refrigeration unit in the overall refrigeration function of the refrigerated truck. For example, if the refrigeration power is reduced, the contribution rate may be reduced from the original 0.6 to 0.5. At the same time, the degree of influence of the refrigeration unit failure on the refrigeration effect of the refrigerated truck may also increase due to the reduction of the refrigeration power, such as the temperature rise rate during the failure changes from the original 5 degrees Celsius per hour to 6 degrees Celsius per hour. In this way, the target dependent state feature is updated based on the feature update vector to generate the updated target dependent state feature, so that the AI ​​robot network can more accurately adapt to the actual situation in the pharmaceutical cold chain logistics system, improve the analysis and processing capabilities of cold chain transportation elements, and ensure the quality and safety of vaccines and other drugs during transportation.

[0093] Taking the transportation of special medicines as an example, when the transportation personnel are the first cold chain transportation element and the transportation vehicle is the second cold chain transportation element. The first error parameter is calculated based on the deviation between the second sample multidimensional knowledge space characteristics and the element knowledge characteristics corresponding to the transportation vehicle. Assume that the first error parameter is 0.15. If the operation of the transportation personnel has a greater impact on the overall transportation effect, for example, the driving experience and operation stability of the transportation personnel are directly related to the safe transportation of special medicines, the first influence coefficient is set to 0.6. Then the updated first error parameter is 0.15 * 0.6 = 0.09.

[0094] The linkage transport element corresponding to the transport personnel is the packaging material, and the second error parameter determined based on the deviation of its related characteristics is assumed to be 0.2. If the packaging material has a smaller impact on the overall transport effect than the transport personnel operation, the second influence coefficient is set to 0.4, and the updated second error parameter is 0.2 * 0.4 = 0.08. The target training error is summed up to 0.09 + 0.08 = 0.17.

[0095] Based on this target training error, a feature update vector is determined, such as the driving speed and steering flexibility of the transport vehicle. Then, based on this feature update vector, the target dependent state feature is updated, such as the dependency between the transport personnel's operation and the transport vehicle's performance. If the feature update vector indicates that the steering flexibility of the transport vehicle needs to be adjusted, this may affect the difficulty of the transport personnel's operation, and then change the dependency between the transport personnel's operation and the transport vehicle's performance, thereby generating an updated target dependent state feature to improve the AI ​​robot network's ability to analyze and process the relationship between cold chain transport elements in the transportation process of special drugs.

[0096] In a possible implementation, step S140 may include:

[0097] Step A110, using a preset deep neural network model to perform preliminary feature extraction on the multidimensional knowledge space features of the first sample to generate a low-level feature representation. The deep neural network model is composed of multiple layers of nonlinear processing units and is used to learn and extract abstract information from the multidimensional knowledge space features of the first sample.

[0098] In this embodiment, taking the transportation of vaccines as an example, the transportation vehicle (refrigerated truck) is the first cold chain transportation element, the temperature control equipment (refrigeration unit) is the second cold chain transportation element, and the first sample multidimensional knowledge space feature includes the purchase time, maintenance record, refrigeration capacity, mileage, load capacity and other information of the transportation vehicle. The deep neural network model is composed of multiple layers of nonlinear processing units. For example, the first layer of nonlinear processing units receives the data of the purchase time of the transportation vehicle, processes it according to the pre-set weight and its own activation function (such as ReLU function), and converts the purchase time into the relatively abstract information of the vehicle's service life. For maintenance records, the model may analyze the frequency of maintenance, the type of maintenance (whether it is routine maintenance or fault repair) and other information to comprehensively obtain the reliability index of the vehicle. In terms of refrigeration capacity, it may be combined with factors such as the size of the internal space of the vehicle to obtain abstract information such as the refrigeration efficiency per unit space. Data such as mileage and load capacity will also be processed similarly, and finally generate a low-level feature representation, which integrates the preliminary abstract results of various aspects of the transportation vehicle information.

[0099] Step A120, inputting the initially extracted low-level feature representation into the feature coding layer, and performing compression encoding and decoding reconstruction on the low-level features through the feature coding layer to generate compressed coding features, wherein the feature coding layer adopts an autoencoder structure.

[0100] For example, the low-level feature representation containing multi-dimensional information such as vehicle age, reliability index, and unit space cooling efficiency is compressed into a lower-dimensional encoding feature. This process removes some redundant information. Then, the decoder part decodes and reconstructs this encoding feature to obtain a compressed encoding feature that is similar to the original low-level feature representation in terms of information volume but more compact.

[0101] Step A130, semantically parse the compressed coding features through a semantic parsing module to map the compressed coding features to a predefined semantic space, and generate a first parsed feature vector with corresponding semantic meaning, wherein the first parsed feature vector includes basic attribute information of the first cold chain transport element and its semantic association and context information in the multidimensional knowledge space.

[0102] On this basis, the predefined semantic space is constructed according to the knowledge in the field of pharmaceutical cold chain logistics. For example, in this semantic space, the refrigeration efficiency of the vehicle may be associated with the semantic concept of "cold chain guarantee capability", and the reliability index of the vehicle is associated with "transport stability", etc. Through this mapping, a first analytical feature vector with corresponding semantic meaning is generated. The first analytical feature vector contains the basic attribute information of the transport vehicle, such as the actual refrigeration efficiency value of the vehicle, the quantitative value of reliability, etc., as well as its semantic association in the multidimensional knowledge space, such as the association between refrigeration efficiency and cold chain guarantee capability, and also contains contextual information, such as the level of the refrigeration efficiency of the transport vehicle relative to other similar vehicles in the same task in the current transport task.

[0103] Step A140, concatenate the first analytical feature vector and the target dependent state feature to generate a joint feature vector, and use an attention mechanism or a gating mechanism to perform feature fusion processing on the joint feature vector, dynamically adjust the weight of the first analytical feature vector in the joint feature vector according to the importance of the target dependent state feature, and obtain a fused joint feature vector.

[0104] Assume that the target dependent state feature contains information such as the contribution rate of the refrigeration unit in the overall refrigeration function of the refrigerated truck and the degree of influence of the refrigeration unit failure on the refrigeration effect of the refrigerated truck. When splicing, these two types of features are combined into a joint feature vector. The attention mechanism is used to perform feature fusion processing on the joint feature vector. For example, if the degree of influence of the refrigeration unit failure on the refrigeration effect of the refrigerated truck is a very critical factor in the current task, then the attention mechanism will dynamically adjust the weight of the first analytical feature vector in the joint feature vector according to this importance. If the first analytical feature vector contains information such as the refrigeration efficiency of the vehicle, since the refrigeration unit failure has a great impact on the refrigeration effect, then during the fusion process, some weights related to the refrigeration efficiency may be adjusted accordingly according to the degree of influence of the refrigeration unit failure, and the fused joint feature vector is obtained.

[0105] Step A150, performing feature enhancement on the fused joint feature vector through a residual network or a densely connected network to generate a fused and enhanced joint feature representation.

[0106] The structural feature of the residual network is that it allows information to be transmitted more efficiently in the network. It adds the input information to the output information of the network layer, allowing the network to learn more complex feature representations. For example, for the refrigeration efficiency-related information in the fused joint feature vector, the residual network may further explore the relationship between it and other features (such as the contribution rate of the refrigeration unit) to generate a fused and enhanced joint feature representation that contains richer and deeper feature information.

[0107] Step A160, input the fused and enhanced joint feature representation into a pre-built mapping conversion model, so as to convert the fused and enhanced joint feature representation into the initial element knowledge features corresponding to the second cold chain transport element according to the learned mapping relationship through the mapping conversion model, and the initial element knowledge features reflect the attributes, states or behavior patterns of the second cold chain transport element under a given dependency state, wherein the mapping conversion model is a machine learning model such as a deep neural network, a support vector machine or a decision tree, which is used to map the fused and enhanced joint feature representation to the element knowledge feature space of the second cold chain transport element.

[0108] The fused and enhanced joint feature representation is input into the pre-built mapping conversion model, where it is assumed that the mapping conversion model is a decision tree model. The decision tree model converts the fused and enhanced joint feature representation into the initial element knowledge features corresponding to the refrigeration unit based on the learned mapping relationship. For example, the decision tree model maps the initial element knowledge features of the refrigeration unit based on the vehicle refrigeration efficiency, refrigeration unit contribution rate, and the relationship between them in the joint feature representation. This feature may reflect the attributes, states, or behavior patterns of the refrigeration unit under the refrigeration performance dependency state of a given transport vehicle, such as whether the refrigeration unit needs maintenance and adjustment, whether certain parts need to be replaced to adapt to the refrigeration needs of the transport vehicle, etc.

[0109] Step A170, by comparing with preset standard features or using domain knowledge, verify and adjust the initial element knowledge features, generate the first intermediate element knowledge features corresponding to the second cold chain transportation element, use genetic algorithm or particle swarm optimization algorithm to perform global feature optimization on the first intermediate element knowledge features, generate optimized second intermediate element knowledge features, input the second intermediate element knowledge features into the semantic consistency verification model, so as to perform semantic consistency verification on the second intermediate element knowledge features according to predefined semantic rules and standards through the semantic consistency verification model, determine whether the second intermediate element knowledge features meet the expected semantic meaning and association relationship, and generate a verification result. Wherein, the semantic consistency verification model verifies the consistency between the second intermediate element knowledge features and the preset semantic standards or domain ontology based on semantic similarity calculation or ontology reasoning.

[0110] For example, the preset standard features may be formulated based on the optimal operating state of the refrigeration unit, such as the standard refrigeration efficiency range of the refrigeration unit, the normal component wear cycle, etc. If the maintenance recommendations for the refrigeration unit in the initial element knowledge features do not match the standard refrigeration efficiency range, or the proposed component replacement cycle does not conform to the domain knowledge in the industry, then adjustments need to be made to generate the first intermediate element knowledge features corresponding to the second cold chain transport element (refrigeration unit).

[0111] In the particle swarm optimization algorithm, each possible performance adjustment scheme of the refrigeration unit is regarded as a particle, and each particle has its own position (representing an adjustment scheme) and speed. Particles continuously adjust their positions and speeds by sharing information (such as the current optimal solution of each particle and the global optimal solution). After multiple iterations, the optimal performance adjustment scheme is found and the optimized second intermediate factor knowledge feature is generated.

[0112] The semantic consistency verification model verifies the consistency between the preset semantic standards or domain ontology based on semantic similarity calculation or ontology reasoning. For example, the semantic standard may specify the semantic description of the refrigeration unit under normal operating conditions, including the normal operating conditions of each component of the refrigeration unit, the semantic expression of the overall refrigeration effect, etc. The semantic consistency verification model calculates the semantic similarity between the second intermediate element knowledge feature and these semantic standards. If the semantic similarity reaches a certain threshold, it is considered to have passed the verification and the verified third intermediate element knowledge feature is output. If the verification fails, based on the feedback information of the model, such as indicating which part of the semantics does not meet the standard, the second intermediate element knowledge feature is corrected and adjusted until the semantic consistency requirements are met.

[0113] Step A180, processing and analyzing the verification result, if the second intermediate element knowledge feature passes the semantic consistency check, then outputting the verified third intermediate element knowledge feature. If it fails the check, then correcting and adjusting the second intermediate element knowledge feature according to the feedback information of the semantic consistency check model until the second intermediate element knowledge feature meets the semantic consistency requirement, analyzing the context information of the verified third intermediate element knowledge feature in the multidimensional knowledge space, and after contextually enhancing the third intermediate element knowledge feature according to the context information, verifying and evaluating the contextually enhanced fourth intermediate element knowledge feature, and generating the contextually enhanced fifth intermediate element knowledge feature.

[0114] For example, consider the context information such as the urgency of the current transportation task, the type and quantity of vaccines to be transported, and the requirements for the refrigeration unit. If the transportation task is urgent, it may be necessary to optimize the performance adjustment plan of the refrigeration unit to ensure that the transportation needs can be met in a short time. After the third intermediate element knowledge feature is contextually enhanced according to these contextual information, the fourth intermediate element knowledge feature is obtained, which is then verified and evaluated, such as evaluating whether the adjusted performance plan is feasible in actual operation, whether it can meet the transportation task requirements while ensuring the quality and safety of vaccines, and generating the fifth intermediate element knowledge feature after contextual association enhancement.

[0115] Step A190, monitor and analyze the state changes and behavior evolution of the second cold chain transport element, and dynamically update the fifth intermediate element knowledge features based on the monitoring and analysis results to generate the dynamically updated sixth intermediate element knowledge features, compare and analyze the features or entities in different domains through the domain analysis module, identify and analyze the commonalities and differences between different domains, obtain domain analysis results, and build a cross-domain mapping model based on the domain analysis results, use the cross-domain mapping model to cross-domain map the sixth intermediate element knowledge features, and generate the element knowledge features corresponding to the second cold chain transport element. The cross-domain mapping model is used to map the dynamically updated sixth intermediate element knowledge features from one domain to another, while maintaining the consistency of key information and semantic meaning of the sixth intermediate element knowledge features.

[0116] Monitor and analyze the state changes and behavior evolution of the refrigeration unit, such as real-time monitoring of the temperature adjustment data, energy consumption data, and component operation status data of the refrigeration unit, as well as analyzing its switching frequency of refrigeration modes, performance changes in different operation stages, and other behavioral evolutions. Based on these monitoring and analysis results, dynamically update the fifth intermediate element knowledge feature. If it is found that the refrigeration efficiency of the refrigeration unit decreases within a certain period of time, it may be necessary to adjust the maintenance recommendations or performance adjustment plans in the fifth intermediate element knowledge feature to generate a dynamically updated sixth intermediate element knowledge feature.

[0117] The domain analysis module is used to compare and analyze the features or entities in different domains, such as comparing the technical performance domain of the refrigeration unit with the transportation cost domain. In the technical performance domain, the focus is on technical indicators such as refrigeration efficiency and reliability of the refrigeration unit; in the transportation cost domain, the focus is on economic indicators such as energy consumption cost and maintenance cost of the refrigeration unit. Identify and analyze the commonalities between them, such as the improvement of refrigeration efficiency may reduce the cost of vaccine loss during transportation, which is the commonality with the transportation cost domain; the difference is that the technical performance domain focuses more on the optimization of technical indicators, while the transportation cost domain focuses more on the control of economic costs. Based on the domain analysis results, a cross-domain mapping model is constructed. This model can map the dynamically updated sixth intermediate element knowledge features from the technical performance domain to the transportation cost domain while maintaining the consistency of key information and semantic meaning. Using this cross-domain mapping model, the sixth intermediate element knowledge features are cross-domain mapped, and finally the element knowledge features corresponding to the refrigeration unit (the second cold chain transportation element) are generated. This element knowledge feature comprehensively considers multiple factors such as technical performance, transportation cost, and transportation task context, and can provide a comprehensive and accurate basis for the management and optimization of refrigeration units in pharmaceutical cold chain logistics.

[0118] Taking the transportation of special medicines as an example, when the transportation personnel are the first cold chain transportation element and the transportation vehicles are the second cold chain transportation element. For the first sample multi-dimensional knowledge space features of the transportation personnel, such as driving experience, qualification level, fatigue driving risk assessment and other information, the deep neural network model performs preliminary feature extraction, converts driving experience into abstract information such as operating proficiency, and converts qualification level into professional ability level indicators. The feature encoding layer compresses and decodes the low-level features, and the semantic parsing module maps them to the semantic space to generate the first parsed feature vector, which contains the basic attribute information of the transportation personnel, the semantic association in the multi-dimensional knowledge space (such as the association between operating proficiency and transportation safety) and context information (such as the level of operating performance in the transportation of special medicines relative to other transportation tasks).

[0119] The first analytical feature vector is concatenated with the target dependent state feature (such as the degree of influence of the transport personnel's operation on the performance of the transport vehicle) into a joint feature vector. The weight is adjusted through the attention mechanism to obtain the fused joint feature vector, and then the feature is enhanced through the residual network. The mapping conversion model (such as the deep neural network model) converts it into the initial element knowledge feature corresponding to the transport vehicle, such as the driving state adjustment suggestion and safety performance evaluation of the transport vehicle under the transport personnel operation dependent state.

[0120] The first intermediate element knowledge feature is obtained by comparing with preset standards (such as safe driving standards for transport vehicles) or domain knowledge for verification and adjustment, and the second intermediate element knowledge feature is obtained by global feature optimization using genetic algorithms. The semantic consistency verification model verifies it, and the third intermediate element knowledge feature is corrected according to the verification results. The fourth intermediate element knowledge feature is obtained by analyzing the context information (such as the special requirements of the nature of special drugs for transport vehicles) for association enhancement, and the fifth intermediate element knowledge feature is obtained by verification and evaluation.

[0121] Monitor the status changes (such as real-time changes in driving speed and braking performance) and behavior evolution (such as driving behavior patterns under different road conditions) of transport vehicles, and dynamically update the sixth intermediate element knowledge characteristics. Analyze the relationship between the technical performance domain and the transportation efficiency domain of transport vehicles through the domain analysis module, build a cross-domain mapping model, and perform cross-domain mapping to obtain the element knowledge characteristics corresponding to the transport vehicle. This feature comprehensively considers multiple factors such as the operation of transport personnel, the performance of the transport vehicle itself, and the transportation requirements of special drugs, which helps to optimize the cold chain transportation process of special drugs.

[0122] In a possible implementation, the method further includes:

[0123] Step S160, obtaining input target pharmaceutical cold chain logistics data.

[0124] In this embodiment, the input target pharmaceutical cold chain logistics data comes from a wide range of sources, covering multiple aspects of the transportation process. Including relevant data of transport vehicles (refrigerated trucks), such as vehicle model, purchase time, maintenance records, mileage, real-time location information, speed information, etc.; data of temperature control equipment (refrigeration unit), such as refrigeration unit model, refrigeration power, real-time temperature adjustment data, energy consumption data, fault records, etc.; data of packaging materials, such as packaging material type, thickness, thermal insulation performance index, shockproof and moisture-proof performance index, etc.; data of transportation personnel, such as driver qualification information, driving experience, fatigue driving risk assessment results, working hours record, etc.; data of transportation routes, including starting point, destination, passing points, route road conditions (such as road slope, flatness, traffic congestion, etc.), meteorological conditions (such as temperature, humidity, weather conditions, etc.); and data of goods (vaccines), such as vaccine types, batches, quantities, and sensitivity requirements to temperature and humidity. These data together constitute the target pharmaceutical cold chain logistics data.

[0125] Step S170, loading the target pharmaceutical cold chain logistics data into the updated AI robot network for multi-dimensional knowledge space conversion, and generating a multi-dimensional knowledge space feature vector corresponding to each target element in the target pharmaceutical cold chain logistics data.

[0126] In this embodiment, the updated AI robot network is obtained through a series of training and optimization steps, and can effectively process the input data. Taking the transport vehicle as an example, the network will convert the vehicle's purchase time, mileage and other data into a part of the multidimensional knowledge space feature vector such as the vehicle's aging degree and reliability index. For the refrigeration unit, its refrigeration power, energy consumption data, etc. will be converted into the representation of refrigeration efficiency, energy efficiency, etc. in the multidimensional knowledge space, becoming the elements of the corresponding multidimensional knowledge space feature vector. Data such as the thickness of the packaging material and the thermal insulation performance index will be converted into feature vector elements such as the thermal insulation effect and the ability to adapt to environmental changes. The qualification information and driving experience of the transport personnel will be converted into components of the multidimensional knowledge space feature vector such as the proficiency of operation and the ability to deal with emergencies. The road condition information and meteorological conditions of the transport route will be converted into feature vector elements such as the difficulty of transportation and environmental risks. The type of goods (vaccines), the sensitivity requirements to temperature and humidity, etc. will be converted into elements of the multidimensional knowledge space feature vector such as the fragility of the goods and special protection needs. In this way, the multidimensional knowledge space feature vector corresponding to each target element in the target pharmaceutical cold chain logistics data is generated.

[0127] Step S180, based on the multidimensional knowledge space feature vector corresponding to each target element in the target pharmaceutical cold chain logistics data, determine the multidimensional analysis result corresponding to the target pharmaceutical cold chain logistics data.

[0128] In a possible implementation, step S180 includes:

[0129] Step S181, cluster analysis is performed on all feature vectors of the multidimensional knowledge space according to the dimension information of the feature vectors and the similarity measurement method to generate a clustered feature vector group. For each clustered feature vector group, the relationship between the feature vectors in the feature vector group is mined to generate a feature relationship graph within each feature vector group. The nodes in the feature relationship graph represent the target elements corresponding to each feature vector, and the edges represent the relationship between the elements. The weight of the edge is set according to the strength of the relationship.

[0130] For example, in terms of dimensional information, if the feature vector contains dimensions related to equipment performance (such as the refrigeration efficiency of the refrigeration unit, the driving speed of the transport vehicle, etc.), then these feature vectors may be classified into one category; if it is a dimension related to environmental factors (such as the meteorological conditions of the transportation route, the moisture-proof performance of the packaging material, etc.), it may be classified into another category. The similarity measurement method can use calculation methods such as Euclidean distance. Through this clustering analysis, a clustered feature vector group is generated.

[0131] Then, taking the equipment performance cluster group as an example, the nodes in the feature relationship graph represent the target elements corresponding to each feature vector, such as refrigeration unit, transport vehicle, etc. The edge represents the relationship between the elements. For example, there may be a relationship between the refrigeration efficiency of the refrigeration unit and the driving speed of the transport vehicle, because the refrigeration effect of the refrigeration unit will affect the internal environment of the transport vehicle, which may affect the driving speed of the vehicle (for example, insufficient refrigeration in a high temperature environment may cause the vehicle to slow down to ensure the temperature of the goods). The weight of the edge is set according to the strength of the relationship. If the refrigeration efficiency has a greater impact on the driving speed, then the weight of this edge will be relatively high.

[0132] Step S182, performing cross-group feature association analysis based on the feature relationship graph within each feature vector group, determining whether there is a correlation relationship between feature vectors of different feature vector groups, and constructing a cross-group feature association matrix by comparing the features of nodes and edges in the feature relationship graphs of different feature vector groups. The elements in the cross-group feature association matrix represent the strength of feature associations between different feature vector groups, and the rows and columns correspond to different feature vector groups, respectively, thereby comprehensively reflecting the association between the various groups, wherein the association relationship is caused by the mutual influence between different links or different factors in the pharmaceutical cold chain logistics process.

[0133] For example, cross-group analysis between equipment performance clustering groups and environmental factor clustering groups. Compare the features of nodes (such as refrigeration units) in the equipment performance clustering group feature relationship graph and nodes (such as meteorological conditions of the transportation route) in the environmental factor clustering group feature relationship graph as well as the edges. If it is found that the refrigeration efficiency of the refrigeration unit is associated with the temperature in the meteorological conditions of the transportation route (such as the refrigeration unit requires a higher refrigeration power in hot weather), then it is determined that the feature vectors between different feature vector groups are associated. Through such comparison, a cross-group feature association matrix is ​​constructed. The elements in the cross-group feature association matrix represent the strength of feature associations between different feature vector groups, and the rows and columns correspond to different feature vector groups. For example, the association strength element between the equipment performance clustering group and the environmental factor clustering group may be high because there are more mutual influences between the two, while the association strength element between the equipment performance clustering group and the transportation personnel clustering group may be relatively low. The cross-group feature association matrix comprehensively reflects the association between each group.

[0134] Step S183, based on the cross-group feature association matrix, identify the target elements that have an influencing effect in the target pharmaceutical cold chain logistics data, generate a key target element set, and for the key target element set, use the corresponding multidimensional knowledge space feature vectors in the target pharmaceutical cold chain logistics data to perform trend analysis, determine the change trend of each key target element in the key target element set in the cold chain logistics process, and generate key element trend analysis results.

[0135] For example, if it is found that there is a strong correlation between the refrigeration efficiency of the refrigeration unit and multiple factors such as the meteorological conditions of the transportation route, the driving speed of the transportation vehicle, the temperature stability of the cargo (vaccine), etc., then the refrigeration unit will be identified as a key target element and enter the key target element set. For this key target element set, the corresponding multi-dimensional knowledge space feature vector in the target pharmaceutical cold chain logistics data is used for trend analysis. For the refrigeration unit, its refrigeration efficiency can be analyzed over time (based on refrigeration power, energy consumption and other data). If it is found that the refrigeration efficiency has a downward trend in the near future, this is part of the key element trend analysis results.

[0136] Step S184, performing anomaly detection according to the key element trend analysis result to obtain an abnormal target element set, and performing risk assessment on the abnormal target element set to obtain a risk assessment result.

[0137] If the refrigeration efficiency of the refrigeration unit decreases and exceeds a preset threshold (this threshold can be determined based on historical data and industry standards), the refrigeration unit will be judged as an abnormal target element and enter the abnormal target element set. Perform a risk assessment on the abnormal target element set to obtain a risk assessment result. For the case of a decrease in the refrigeration efficiency of the refrigeration unit, the risk assessment may take into account the impact on the temperature stability of the goods (vaccines). If the risk of causing the vaccine temperature to exceed the safe range is high, the risk assessment result will show a high risk level.

[0138] Step S185, generating optimization suggestions for the pharmaceutical cold chain logistics process based on the abnormal target element set and the risk assessment result.

[0139] If the refrigeration unit is an abnormal target element and the risk level is high, optimization suggestions may include emergency maintenance of the refrigeration unit, adjusting the transportation route to avoid high temperature areas (if meteorological conditions are one of the influencing factors), reducing the speed of the transportation vehicle to reduce the load on the refrigeration unit, etc.

[0140] Step S186, summarizes the optimization suggestions for the pharmaceutical cold chain logistics process, the abnormal target element set, the risk assessment results, the key element trend analysis results, the cross-group feature association matrix, the feature relationship diagram within each feature vector group, the clustered feature vector group and the multidimensional knowledge space feature vector corresponding to each target element in the target pharmaceutical cold chain logistics data to obtain the multidimensional analysis results corresponding to the target pharmaceutical cold chain logistics data.

[0141] In this embodiment, the multi-dimensional analysis results provide a comprehensive basis for the management and optimization of pharmaceutical cold chain logistics, which helps to ensure the quality and safety of pharmaceutical products during cold chain transportation.

[0142] Taking the transportation of special medicines as an example, special medicines may have more stringent requirements for temperature, humidity, light and other conditions. In the acquired target pharmaceutical cold chain logistics data, the data related to the special properties of special medicines becomes an important part. These data are loaded into the updated AI robot network to generate multi-dimensional knowledge space feature vectors corresponding to special medicines. For example, the sensitivity of special medicines to light may be converted into special protection demand indicators.

[0143] In cluster analysis, cluster groups related to special requirements of special drugs may be formed. For example, feature vectors related to the protection requirements of special drugs (such as light protection, precise temperature control, etc.) are grouped together. In the construction of the feature relationship graph, the relationship between the protection requirements of special drugs and target elements such as packaging materials and temperature control equipment will be clarified. When performing cross-group feature association analysis, the association between the special drug protection requirements cluster group and other cluster groups (such as the meteorological conditions cluster group of the transportation route) will be given priority consideration to construct a cross-group feature association matrix.

[0144] When identifying the key target element set, the special properties of special drugs will make the elements closely related to them (such as specific packaging materials, high-precision temperature control equipment, etc.) become key target elements. Trend analysis is performed on these key target elements, such as the trend of temperature control accuracy of specific temperature control equipment over time. In anomaly detection, if the temperature control accuracy decreases, the equipment enters the abnormal target element set. After risk assessment, if it is found that the risk of special drugs deteriorating is extremely high, the generated optimization suggestions may include replacing temperature control equipment, adjusting transportation plans, etc. Finally, all relevant information is summarized to obtain the multi-dimensional analysis results corresponding to the transportation of special drugs to ensure the safety and effectiveness of special drugs during cold chain transportation.

[0145] Figure 2 The hardware structure of the multi-dimensional analysis system 100 for pharmaceutical cold chain logistics data based on AI robots provided in an embodiment of the present invention is shown as follows: Figure 2 As shown, the AI ​​robot-based pharmaceutical cold chain logistics data multidimensional analysis system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0146] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the AI ​​robot-based pharmaceutical cold chain logistics data multidimensional analysis system 100 uses to execute or use to complete the exemplary method described in the present invention.

[0147] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the multidimensional analysis method of pharmaceutical cold chain logistics data based on AI robots in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0148] The specific implementation process of the processor 110 can be found in the various method embodiments executed by the above-mentioned AI robot-based pharmaceutical cold chain logistics data multidimensional analysis system 100. The implementation principles and technical effects are similar and will not be repeated in this embodiment.

[0149] In addition, an embodiment of the present invention also provides a readable storage medium, in which computer executable instructions are preset. When the processor executes the computer executable instructions, the above-mentioned multidimensional analysis method of pharmaceutical cold chain logistics data based on AI robots is implemented.

[0150] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, various features are sometimes combined into one embodiment, drawing or description thereof.

Claims

1. A multidimensional analysis method for pharmaceutical cold chain logistics data based on AI robots, characterized in that: The method comprises: Acquire a sample pharmaceutical cold chain logistics data sequence, and determine a cold chain transportation element sequence corresponding to the sample pharmaceutical cold chain logistics data sequence; Based on the sample pharmaceutical cold chain logistics data in the sample pharmaceutical cold chain logistics data sequence and the initialized AI robot network, each cold chain transportation element in the cold chain transportation element sequence is converted into a multidimensional knowledge space to generate a sample multidimensional knowledge space feature; Determine a first cold chain transport element and a second cold chain transport element in the cold chain transport element sequence that have a target element dependency state, and obtain a target dependency state feature corresponding to the target element dependency state; Based on the first sample multidimensional knowledge space feature corresponding to the first cold chain transportation element and the target dependent state feature, determine the element knowledge feature corresponding to the second cold chain transportation element, and determine the target training error corresponding to the AI ​​robot network based on the second sample multidimensional knowledge space feature corresponding to the second cold chain transportation element and the deviation of the element knowledge feature; The network weight information of the AI ​​robot network is updated based on the target training error to generate an updated AI robot network.

2. The multidimensional analysis method of pharmaceutical cold chain logistics data based on AI robot according to claim 1 is characterized in that: The step of obtaining the target dependency state feature corresponding to the target element dependency state includes: Determine the target dependency state category corresponding to the target element dependency state, and determine the corresponding target dependency state feature based on the target dependency state category, wherein the target dependency state category is determined based on the element category of the first cold chain transport element in the target logistics scenario.

3. The multidimensional analysis method of pharmaceutical cold chain logistics data based on AI robot according to claim 2 is characterized in that: The determining of the target dependency state category corresponding to the target element dependency state and determining the corresponding target dependency state feature based on the target dependency state category includes: Determine the target dependency state category corresponding to the target element dependency state; When the target dependency state category is an inclusion dependency state category, the dependency state feature corresponding to the inclusion dependency state category is determined as the target dependency state feature.

4. The multidimensional analysis method of pharmaceutical cold chain logistics data based on AI robot according to claim 1 is characterized in that: The step of determining the first cold chain transportation element and the second cold chain transportation element in the cold chain transportation element sequence that are in a target element dependent state comprises: Acquire a logistics element relationship graph, wherein the logistics element relationship graph includes feature members corresponding to each relationship graph element and feature links representing element dependency states between the feature members; Acquire the first cold chain transport element in the cold chain transport element sequence, and determine, from the logistics element relationship graph, the second cold chain transport element in the cold chain transport element sequence that has a characteristic link association with the first cold chain transport element; The element dependency state corresponding to the characteristic link of the first cold chain transport element and the second cold chain transport element is obtained, and the output is the target element dependency state.

5. The multidimensional analysis method of pharmaceutical cold chain logistics data based on AI robot according to claim 1 is characterized in that: The second cold chain transportation element is a relational graph element in the logistics element relational graph, which has a target element dependency state with the first cold chain transportation element; The determining of the target training error corresponding to the AI ​​robot network based on the second sample multidimensional knowledge space feature corresponding to the second cold chain transportation element and the deviation of the element knowledge feature includes: Determine a first error parameter based on the second sample multidimensional knowledge space feature corresponding to the second cold chain transportation element and the deviation of the element knowledge feature; Acquire the linkage transport element corresponding to the first cold chain transport element in the sample pharmaceutical cold chain logistics data; Determining a second error parameter based on the degree of deviation between the third sample multidimensional knowledge space feature corresponding to the linkage transport element and the first sample multidimensional knowledge space feature; The target training error corresponding to the AI ​​robot network is determined based on the first error parameter and the second error parameter.

6. The multidimensional analysis method of pharmaceutical cold chain logistics data based on AI robot according to claim 5 is characterized in that: Determining the target training error corresponding to the AI ​​robot network based on the first error parameter and the second error parameter includes: Obtaining a first influence coefficient and a second influence coefficient; Update the first error parameter based on the first influence coefficient to generate an updated first error parameter, and update the second error parameter based on the second influence coefficient to generate an updated second error parameter; The updated first error parameter and the updated second error parameter are summed to generate the target training error corresponding to the AI ​​robot network.

7. The multidimensional analysis method of pharmaceutical cold chain logistics data based on AI robot according to claim 1 is characterized in that: The method further comprises: Determining a feature update vector based on the target training error; Based on the feature update vector, the target dependency state feature is updated to generate an updated target dependency state feature.

8. The multidimensional analysis method of pharmaceutical cold chain logistics data based on AI robot according to claim 1 is characterized in that: The determining, based on the first sample multidimensional knowledge space feature corresponding to the first cold chain transportation element and the target dependent state feature, the element knowledge feature corresponding to the second cold chain transportation element comprises: Performing preliminary feature extraction on the multidimensional knowledge space features of the first sample using a preset deep neural network model to generate a low-level feature representation, wherein the deep neural network model is composed of multiple layers of nonlinear processing units and is used to learn and extract abstract information from the multidimensional knowledge space features of the first sample; Inputting the initially extracted low-level feature representation into the feature coding layer, compressing and encoding the low-level features and decoding and reconstructing them through the feature coding layer to generate compressed coding features, wherein the feature coding layer adopts an autoencoder structure; The compressed coding features are semantically parsed by a semantic parsing module to map the compressed coding features to a predefined semantic space, and a first parsed feature vector having a corresponding semantic meaning is generated, wherein the first parsed feature vector includes basic attribute information of the first cold chain transport element and its semantic association and context information in a multidimensional knowledge space; The first analytical feature vector and the target dependent state feature are concatenated to generate a joint feature vector, and an attention mechanism or a gating mechanism is used to perform feature fusion processing on the joint feature vector, and the weight of the first analytical feature vector in the joint feature vector is dynamically adjusted according to the importance of the target dependent state feature to obtain a fused joint feature vector; Performing feature enhancement on the fused joint feature vector through a residual network or a densely connected network to generate a fused and enhanced joint feature representation; The fused and enhanced joint feature representation is input into a pre-built mapping conversion model, so as to convert the fused and enhanced joint feature representation into an initial element knowledge feature corresponding to the second cold chain transportation element according to the learned mapping relationship through the mapping conversion model, and the initial element knowledge feature reflects the attribute, state or behavior mode of the second cold chain transportation element under a given dependency state, wherein the mapping conversion model is a machine learning model such as a deep neural network, a support vector machine or a decision tree, and is used to map the fused and enhanced joint feature representation to the element knowledge feature space of the second cold chain transportation element; By comparing the preset standard features or using domain knowledge, the initial element knowledge features are verified and adjusted to generate the first intermediate element knowledge features corresponding to the second cold chain transportation element; Using a genetic algorithm or a particle swarm optimization algorithm, globally optimizing the first intermediate element knowledge feature to generate an optimized second intermediate element knowledge feature; The second intermediate element knowledge feature is input into a semantic consistency verification model, so as to perform semantic consistency verification on the second intermediate element knowledge feature according to predefined semantic rules and standards through the semantic consistency verification model, determine whether the second intermediate element knowledge feature conforms to the expected semantic meaning and association relationship, and generate a verification result; wherein the semantic consistency verification model verifies the consistency between the second intermediate element knowledge feature and the preset semantic standard or domain ontology based on semantic similarity calculation or ontology reasoning; Processing and analyzing the verification result, if the second intermediate element knowledge feature passes the semantic consistency check, outputting the verified third intermediate element knowledge feature; if it fails the check, correcting and adjusting the second intermediate element knowledge feature according to feedback information of the semantic consistency check model until the second intermediate element knowledge feature meets the semantic consistency requirement; Analyzing the context information of the verified third intermediate element knowledge feature in the multidimensional knowledge space, and after contextually enhancing the third intermediate element knowledge feature according to the context information, verifying and evaluating the fourth intermediate element knowledge feature after the contextual association enhancement, and generating a fifth intermediate element knowledge feature after the contextual association enhancement; Monitor and analyze the state changes and behavior evolution of the second cold chain transportation element, and dynamically update the fifth intermediate element knowledge feature according to the monitoring and analysis results to generate a dynamically updated sixth intermediate element knowledge feature; Comparing and analyzing features or entities in different domains through a domain analysis module, identifying and analyzing commonalities and differences between different domains, obtaining domain analysis results, and constructing a cross-domain mapping model based on the domain analysis results, the cross-domain mapping model being used to map the dynamically updated sixth intermediate element knowledge features from one domain to another domain while maintaining consistency in key information and semantic meaning of the sixth intermediate element knowledge features; The cross-domain mapping model is used to perform cross-domain mapping on the sixth intermediate element knowledge feature to generate element knowledge features corresponding to the second cold chain transportation element.

9. The multidimensional analysis method of pharmaceutical cold chain logistics data based on AI robot according to claim 1 is characterized in that: The method further comprises: Obtain input target pharmaceutical cold chain logistics data; Loading the target pharmaceutical cold chain logistics data into the updated AI robot network for multi-dimensional knowledge space conversion, generating a multi-dimensional knowledge space feature vector corresponding to each target element in the target pharmaceutical cold chain logistics data; Determine the multidimensional analysis result corresponding to the target pharmaceutical cold chain logistics data based on the multidimensional knowledge space feature vector corresponding to each target element in the target pharmaceutical cold chain logistics data; Wherein, the step of determining the multidimensional analysis result corresponding to the target pharmaceutical cold chain logistics data based on the multidimensional knowledge space feature vector corresponding to each target element in the target pharmaceutical cold chain logistics data comprises: Perform cluster analysis on all feature vectors of the multidimensional knowledge space according to the dimension information of the feature vector and the similarity measurement method to generate clustered feature vector groups. For each clustered feature vector group, explore the relationship between the feature vectors in the feature vector group to generate a feature relationship graph within each feature vector group. The nodes in the feature relationship graph represent the target elements corresponding to each feature vector, and the edges represent the relationship between the elements. The weight of the edge is set according to the strength of the relationship. Based on the feature relationship graph within each feature vector group, a cross-group feature association analysis is performed to determine whether there is an association relationship between the feature vectors of different feature vector groups. By comparing the features of the nodes and edges in the feature relationship graphs of different feature vector groups, a cross-group feature association matrix is ​​constructed. The elements in the cross-group feature association matrix represent the strength of the feature association between different feature vector groups, and the rows and columns correspond to different feature vector groups, respectively, thereby comprehensively reflecting the association between the various groups, wherein the association relationship is caused by the mutual influence between different links or different factors in the pharmaceutical cold chain logistics process; According to the cross-group feature association matrix, target elements that have an influencing effect in the target pharmaceutical cold chain logistics data are identified, and a key target element set is generated. For the key target element set, a trend analysis is performed using the corresponding multidimensional knowledge space feature vector in the target pharmaceutical cold chain logistics data to determine the change trend of each key target element in the key target element set in the cold chain logistics process, and generate a key element trend analysis result; Performing anomaly detection according to the key element trend analysis result to obtain an abnormal target element set, and performing risk assessment on the abnormal target element set to obtain a risk assessment result; Generate optimization suggestions for the pharmaceutical cold chain logistics process based on the abnormal target element set and the risk assessment result; The optimization suggestions for the pharmaceutical cold chain logistics process, the abnormal target element set, the risk assessment results, the key element trend analysis results, the cross-group feature association matrix, the feature relationship diagram within each feature vector group, the clustered feature vector group and the multidimensional knowledge space feature vector corresponding to each target element in the target pharmaceutical cold chain logistics data are summarized to obtain the multidimensional analysis results corresponding to the target pharmaceutical cold chain logistics data.

10. A multidimensional analysis system for pharmaceutical cold chain logistics data based on AI robots, characterized in that: The AI ​​robot-based pharmaceutical cold chain logistics data multidimensional analysis system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the AI ​​robot-based pharmaceutical cold chain logistics data multidimensional analysis method as described in any one of claims 1 to 9.

Images