Whole-course tracking method and system for medicine cold-chain logistics transportation

Abstract

The invention provides a medicine cold-chain logistics transportation whole-course tracking method and system, and relates to the technical field of logistics transportation monitoring, and the method comprises the steps: firstly obtaining an original sensing tracking data flow outputted by a sensor cluster in a transportation carrier in a continuous collection time window, and then generating a transportation carrier moving trajectory according to a geographic positioning parameter; physical parameters at the same moment as the geographic position coordinate points are combined to generate a multi-dimensional environment state mark set, and meanwhile, a logistics transfer node topology connection network comprising logistics transfer node units and directed transfer path edges is constructed; and inputting the moving trajectory, the multi-dimensional environment state mark set and the topological connection network into a pre-constructed space-time state feature fusion tracking model, generating a whole-course state tracking feature tensor, performing abnormal state mode matching on the whole-course state tracking feature tensor to obtain an abnormal state type identifier and a space-time distribution feature, and generating an abnormal early warning instruction set. According to the invention, the whole medicine cold chain transportation process can be comprehensively and accurately tracked, and abnormity can be intelligently identified and early warned in time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics transportation monitoring, in particular to a medicine cold-chain logistics transportation whole-process tracking method and system. BACKGROUND

[0002] Medicine products are extremely sensitive to environmental conditions such as temperature and humidity, and any abnormal fluctuation of environmental parameters may cause medicine products to deteriorate and fail, thereby affecting the health and safety of patients. Therefore, it is a key requirement for the industry to accurately track and effectively monitor the whole process of medicine cold-chain logistics transportation.

[0003] At present, the existing medicine cold-chain logistics transportation tracking methods have many limitations. Some methods only rely on single sensor data for monitoring, such as only focusing on temperature parameters, while ignoring humidity, vibration and other factors that may affect the quality of medicine products, and cannot fully reflect the environmental state during transportation. In addition, although some methods collect multiple environmental parameter data, they only simply record and display the data without deep analysis and fusion processing, making it difficult to dig out the potential risks and abnormal patterns hidden behind the data.

[0004] In terms of transportation trajectory tracking, most existing technologies can only provide approximate location information of the transportation carrier, cannot accurately construct the running trajectory line of the transportation carrier, and cannot clearly show the various logistics transfer nodes passed through during transportation and their connection relationship. This makes it difficult to quickly locate the specific location and link where the problem occurs when a problem occurs, which is not conducive to taking effective measures in a timely manner. In addition, the existing tracking methods lack intelligent identification and early warning mechanisms for abnormal states. When environmental parameter abnormalities or transportation trajectory deviations occur during transportation, timely warnings are often not issued, resulting in problems not being handled in a timely manner and increasing the quality risk of medicine products. SUMMARY

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a medicine cold-chain logistics transportation whole-process tracking method, which comprises:

[0006] Obtaining the original sensor tracking data stream output by the sensor cluster deployed inside the transportation carrier in the continuous collection time window during the medicine cold-chain logistics transportation process, the original sensor tracking data stream comprising a temperature sensor parameter subsequence, a humidity sensor parameter subsequence, a vibration sensor parameter subsequence and a geolocation parameter subsequence marked with a collection timestamp;

[0007] generate a transportation carrier running trajectory line according to the moving sequence of the geographic position coordinate points in the geographic positioning parameter subsequence, and generate a multi-dimensional environmental state marker set associated with each of the geographic position coordinate points in the transportation carrier running trajectory line according to the physical parameters with the same time stamp as each of the geographic position coordinate points in the temperature sensing parameter subsequence, the humidity sensing parameter subsequence, and the vibration sensing parameter subsequence;

[0008] construct a logistics transfer node topological connection network through which the transportation carrier running trajectory line passes, the logistics transfer node topological connection network including a plurality of logistics transfer node units and directed transfer path edges connecting the logistics transfer node units, the logistics transfer node units being obtained after the geographic position coordinate points in the transportation carrier running trajectory line are matched with a preset logistics transfer node database;

[0009] input the transportation carrier running trajectory line, the multi-dimensional environmental state marker set, and the logistics transfer node topological connection network into a pre-constructed spatiotemporal state feature fusion tracking model for joint analysis and processing, to generate a medical cold chain logistics transportation whole-process state tracking feature tensor;

[0010] perform abnormal state mode matching processing on the medical cold chain logistics transportation whole-process state tracking feature tensor, to obtain an abnormal state type identifier and an abnormal state spatiotemporal distribution feature corresponding to the medical cold chain logistics transportation whole-process state tracking feature tensor, and generate a medical cold chain logistics transportation abnormal early warning instruction set according to the abnormal state type identifier and the abnormal state spatiotemporal distribution feature.

[0011] In still another aspect, an embodiment of the present application further provides a medical cold chain logistics transportation whole-process tracking system, comprising:

[0012] a processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-mentioned medical cold chain logistics transportation whole-process tracking method by executing the machine-executable instructions.

[0013] In still another aspect, an embodiment of the present application further provides a computer program product, the computer program product comprising machine-executable instructions stored in a computer-readable storage medium, a processor of a medical cold chain logistics transportation whole-process tracking system reading the machine-executable instructions from the computer-readable storage medium, and the processor executing the machine-executable instructions, so that the medical cold chain logistics transportation whole-process tracking system executes the above-mentioned medical cold chain logistics transportation whole-process tracking method.

[0014] Based on the above aspects, by acquiring the original sensor tracking data stream containing multiple parameters output by the sensor cluster inside the transportation carrier, generating the transportation carrier operation trajectory line based on the geographic positioning parameters, and combining the multi-dimensional environment state label set associated with each geographic position coordinate point, a logistics transfer node topology connection network is constructed, which shows each logistics transfer node and the directed transfer path relationship between them that the transportation carrier passes through. The operation trajectory line, multi-dimensional environment state label set and logistics transfer node topology connection network are input into the space-time state feature fusion tracking model for joint analysis and processing, generating a medical cold chain logistics transportation whole-process state tracking feature tensor, realizing deep fusion and feature extraction of multiple data, and mining potential information and risk patterns hidden behind the data. The state tracking feature tensor is subjected to abnormal state pattern matching processing, which can accurately identify the abnormal state type and space-time distribution characteristics in the transportation process, and generate corresponding abnormal early warning instruction set, realizing intelligent perception and timely warning of abnormal conditions of medical cold chain logistics transportation, which helps relevant personnel to take measures quickly, ensures the quality and safety of medical products, and improves the reliability and management level of medical cold chain logistics transportation. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is the execution flow diagram of the medical cold chain logistics transportation whole-process tracking method provided by the embodiment of the present application.

[0016] Figure 2 is a schematic diagram of exemplary hardware and software components of the medical cold chain logistics transportation whole-process tracking system provided by the embodiment of the present application. DETAILED DESCRIPTION

[0017] Figure 1 is a flow diagram of the medical cold chain logistics transportation whole-process tracking method provided by an embodiment of the present application, which will be described in detail below.

[0018] Step S110, acquiring the original sensor tracking data stream output by the sensor cluster deployed inside the transportation carrier in the continuous acquisition time window in the medical cold chain logistics transportation process, the original sensor tracking data stream containing temperature sensor parameter subsequence, humidity sensor parameter subsequence, vibration sensor parameter subsequence and geographic positioning parameter subsequence with acquisition time stamp label.

[0019] In this embodiment, the transportation carrier is a well-equipped vaccine cold chain transportation refrigerated vehicle. The refrigerated vehicle is internally deployed with an integrated sensor cluster, which contains multiple sensors of different types. Before the start of the transportation task, the sensor cluster is initialized and a unified continuous acquisition time window is set, for example, the entire transportation period from the task start time T_start to the task end time T_end.

[0020] The sensor cluster continuously outputs the original sensor tracking data stream according to a preset collection frequency, for example, collecting data once every 1 second. The temperature sensor collects the temperature value inside the carriage at a frequency of 1 per second, and adds a timestamp mark to each temperature sample value, generating a temperature sensor parameter sub-sequence. The humidity sensor collects the relative humidity value inside the carriage at the same frequency, and adds a corresponding timestamp mark to each humidity sample value, generating a humidity sensor parameter sub-sequence. The vibration sensor is installed inside the refrigerated vehicle chassis or cargo box to sense mechanical vibration during transportation, and also collects vibration amplitude values at a frequency of 1 per second, and adds a timestamp mark, generating a vibration sensor parameter sub-sequence. The global positioning system module obtains the current geographic position coordinates of the refrigerated vehicle once every second, including the longitude value and the latitude value, and adds a timestamp mark to each geographic position coordinate point, generating a geographic positioning parameter sub-sequence. The above four sub-sequences together constitute the original sensor tracking data stream, and the number of parameters in each sub-sequence is the total number of seconds from T_start to T_end plus 1.

[0021] Step S120, generating a transport carrier running trajectory line according to the moving order of the geographic position coordinate points in the geographic positioning parameter sub-sequence, and generating a multi-dimensional environmental state marker set associated with each geographic position coordinate point in the transport carrier running trajectory line according to the physical parameters in the temperature sensor parameter sub-sequence, the humidity sensor parameter sub-sequence and the vibration sensor parameter sub-sequence with the same time stamp as each geographic position coordinate point.

[0022] Step S121, extracting a plurality of geographic position coordinate points arranged in order of collection time from the geographic positioning parameter sub-sequence, and performing a redundant point elimination process on the geographic position coordinate points to obtain an effective geographic position coordinate point sequence for constructing the transport carrier running trajectory line, the redundant point elimination process being realized by calculating the displacement distance parameter between adjacent geographic position coordinate points and filtering the geographic position coordinate points with a displacement distance parameter less than a preset static discrimination threshold.

[0023] A sub-sequence of geolocation parameters is extracted from the raw sensor tracking data stream. This sub-sequence contains 1 geolocation coordinate point per second between time T_start and time T_end, forming an initial list of coordinate points with a length of total seconds plus 1. The list is traversed, and for each pair of adjacent coordinate points in the list, such as the i-th geolocation coordinate point P_i and the i+1-th geolocation coordinate point P_{i+1}, the Euclidean distance between the two points is calculated as a displacement distance parameter D_i. The displacement distance parameter is calculated by converting the longitude and latitude values of P_i and P_{i+1} into actual distance projections on the surface of the Earth, obtaining the distance difference in the longitude direction and the distance difference in the latitude direction, and then calculating the square root of the sum of the squares of the two differences by the Pythagorean theorem to obtain the straight-line distance D_i. A static discrimination threshold D_static is set, which can be set according to actual needs, for example, set to 5 meters. If the calculated displacement distance parameter D_i is less than the static discrimination threshold of 5 meters, it is considered that the vehicle is in a static state during this time period, and the latter point P_{i+1} is a redundant point, which is removed from the initial list. If the displacement distance parameter D_i is greater than or equal to 5 meters, the point P_{i+1} is retained. The next point pair is processed until the entire list is traversed. After traversing and removing the entire initial list, a sequence containing only the effective position points in the actual movement of the vehicle is obtained, that is, the effective geolocation coordinate point sequence, denoted as sequence Q, and the length of the effective geolocation coordinate sequence is M, which is much smaller than the length of the original sequence.

[0024] In step S122, according to the collection time stamp corresponding to each effective geolocation coordinate point in the effective geolocation coordinate point sequence, the time interval parameter and the displacement direction angle parameter between adjacent effective geolocation coordinate points are calculated; the time interval parameter and the displacement direction angle parameter are normalized respectively, and a motion state change description vector between adjacent effective geolocation coordinate points is generated according to the normalized time interval parameter and the displacement direction angle parameter. The motion state change description vector contains a normalized motion duration component and a normalized motion direction change component.

[0025] For each adjacent point pair in the sequence of valid geographic position coordinate points Q, such as the jth valid geographic position coordinate point Q_j and the (j+1)th valid geographic position coordinate point Q_{j+1}, the time instants T_j and T_{j+1} attached to the two points respectively are extracted, and the difference between T_{j+1} and T_j is calculated to obtain the time interval parameter Δt_j. Then, the displacement direction from the point Q_j to the point Q_{j+1} is calculated. The displacement direction is referenced to the north direction as 0 degree, and is obtained by calculating the clockwise angle of the line connecting the two points relative to the north direction, which ranges from 0 degree to 360 degree, as the displacement direction angle parameter θ_j. In order to eliminate the dimensional influence and make the subsequent processing more stable, the time interval parameter and the displacement direction angle parameter are normalized respectively. For the time interval parameter Δt_j, it is divided by the maximum time interval value Δt_max appearing in the entire transportation process to obtain the standardized motion duration component α_j, which ranges from 0 to 1, and α_j is equal to Δt_j divided by Δt_max. For the displacement direction angle parameter θ_j, the sine and cosine values of the angle are used for normalization, and sin(θ_j) and cos(θ_j) are calculated, both of which range from -1 to 1, and together constitute the standardized motion direction change component. The standardized motion duration component α_j, the standardized motion direction sine component sin(θ_j) and the standardized motion direction cosine component cos(θ_j) are combined to form a 3-dimensional motion state change description vector V_j, V_j is equal to (α_j, sin(θ_j), cos(θ_j)), which is used to represent the duration and direction change characteristics of the motion from the point Q_j to the point Q_{j+1}.

[0026] In step S123, the sequence of valid geographic position coordinate points and the motion state change description vectors between adjacent valid geographic position coordinate points are associated and combined to generate an initial trajectory segment set with time sequence coherence, which is composed of the continuous valid geographic position coordinate points in the sequence of valid geographic position coordinate points and the corresponding motion state change description vectors.

[0027] Each point in the sequence of valid geolocation coordinate points Q is associated with a motion state change description vector connecting the point and the next point. Specifically, for point Q_j, the motion state change description vector V_j from itself to Q_{j+1} is associated. For the last point Q_M, there is no subsequent motion vector, so it only contains the point coordinate. A series of data units are generated, each containing a point coordinate and a motion vector (except for the last point), and the above units are arranged in the original time sequence to form an initial trajectory segment set with time sequence continuity. The set describes the motion state change details of the vehicle on different road segments, and each element in the set corresponds to a trajectory segment.

[0028] In step S124, the initial trajectory segment set is subjected to trajectory segment splicing processing. According to the geolocation coordinate point matching degree between the end points of adjacent trajectory segments in the initial trajectory segment set and the continuity parameter calculated based on the motion state change description vectors, the adjacent trajectory segments that satisfy the preset splicing condition are connected to generate the transport carrier running trajectory line.

[0029] Since the redundant points are removed, the segments in the initial trajectory segment set are already continuous, but in some special cases, such as data interruption caused by loss of global positioning system signals, there may be truly discontinuous segments. It is necessary to ensure the final integrity of the trajectory. The initial trajectory segment set is traversed to check the connection between adjacent segment end points. For the last point Q_k of the previous segment and the first point Q_{k+1} of the next segment, the geolocation coordinate point matching degree between the two points, i.e., the displacement distance parameter D_k between them, is calculated. At the same time, the similarity of the last motion vector V_k of the previous segment and the first motion vector V_{k+1} of the next segment is compared, and the cosine similarity between the two vectors is calculated as the continuity parameter C_k, which is equal to the dot product of V_k and V_{k+1} divided by the modulus of V_k multiplied by the modulus of V_{k+1}. If the displacement distance parameter D_k is less than a preset splicing threshold, for example, 10 meters, and the continuity parameter C_k is higher than a preset similarity threshold, for example, 0.95, then the two segments are considered to be continuous in space and motion trend, and the next segment is directly connected after the previous segment. If D_k is greater than or equal to 10 meters or C_k is less than or equal to 0.95, it is determined that there is a data interruption, and the place is marked as a breakpoint. After checking and splicing all adjacent segments, the final transport carrier running trajectory line is generated, which is a complete sequence of valid point coordinates and their associated motion vectors.

[0030] Step S125, for each trajectory point in the transportation carrier running trajectory line, a logistics node proximity marking process is performed, the spatial distance parameter between each trajectory point and all logistics transfer nodes in the preset logistics transfer node database is calculated, and the trajectory point with a spatial distance parameter less than a preset proximity distance threshold is marked as a logistics node proximity point, and the identifier of the logistics transfer node closest to the trajectory point is recorded as the node association identifier.

[0031] For each trajectory point in the transportation carrier running trajectory line, such as trajectory point Q_j, its geographic location coordinates are obtained. At the same time, the information of all logistics transfer nodes is read from a preset logistics transfer node database, each node containing a node identifier and the geographic location coordinates of the node. The logistics transfer node database is pre-constructed and contains all possible warehouses, transfer stations, distribution centers, etc. along the way, such as node A, node B, node C, and node H. For trajectory point Q_j, the spatial distance parameter between it and each logistics transfer node, such as node A, is calculated in turn, and the calculation method is the same as the displacement distance parameter in step S121, that is, the Euclidean distance between the two points is calculated. The distance values d(Q_j, A), d(Q_j, B), and d(Q_j, H) are obtained. A proximity distance threshold is set, for example, 500 meters. All calculated distance values are traversed, and if there is any distance value less than 500 meters, such as d(Q_j, C) equal to 300 meters, then the trajectory point Q_j is marked as a logistics node proximity point. At the same time, the minimum value is found from all distance values less than 500 meters, and the logistics transfer node corresponding to the minimum value is the node closest to the trajectory point Q_j, and the identifier of the node, such as node C, is recorded as the node association identifier of the trajectory point Q_j. If all distance values are greater than or equal to 500 meters, then the trajectory point Q_j is not marked as a proximity point.

[0032] Step S126, according to the distribution density and order of the logistics node proximity points on the transportation carrier running trajectory line, the logistics transfer node sequence passed through by the transportation carrier running trajectory line is determined, and the logistics transfer node sequence is composed of the logistics transfer node identifiers arranged in the direction of travel of the transportation carrier running trajectory line.

[0033] Traverse the entire transportation carrier operation trajectory line, identify all the trajectory points marked as logistics node proximity points. According to the order of these proximity points on the trajectory line, form a preliminary node access list. But because the vehicle may stay near the same node for a long time, thus generating multiple consecutive proximity points corresponding to the same node, de-duplication processing is needed. For example, trajectory points Q_5, Q_6, Q_7, Q_8 are all marked as proximity points, and their node association identifiers are all node C. Then, when determining the passing node sequence, only one node C is retained, representing that the vehicle has passed through node C. According to the direction of the trajectory line, each different node identifier is processed in turn, and finally an ordered logistics transfer node sequence is generated, for example, starting from node A, then passing through node C, then to node E, and finally to node H. The sequence represents the actual key logistics node sequence passed by the vehicle.

[0034] Step S127, generating a multi-dimensional environmental state marker set associated with each of the geographical position coordinate points in the transportation carrier operation trajectory line according to the physical parameters in the temperature sensing parameter subsequence, the humidity sensing parameter subsequence, and the vibration sensing parameter subsequence with the same time stamp as each of the geographical position coordinate points.

[0035] Step S1271, extracting temperature instantaneous sampling values with the same collection time stamp as each of the geographical position coordinate points from the temperature sensing parameter subsequence, and arranging the temperature instantaneous sampling values according to the order of appearance of the geographical position coordinate points on the transportation carrier operation trajectory line, to generate a temperature parameter time sequence association list corresponding to the transportation carrier operation trajectory line.

[0036] For each trajectory point in the transportation carrier operation trajectory line, such as trajectory point Q_j, obtain its collection time stamp T_j. From the temperature sensing parameter subsequence, find the temperature instantaneous sampling value Temp_j with the time stamp equal to T_j. Because the collection frequency is consistent, in theory, each time stamp has a corresponding temperature value. Arrange the found Temp_j according to the order of trajectory point Q_j on the trajectory line, to generate a temperature parameter time sequence association list Temp_List corresponding to the trajectory line one by one, with a list length of M, and each element corresponding to a temperature value of a trajectory point.

[0037] Step S1272, extracting humidity instantaneous sampling values with the same collection time stamp as each of the geographical position coordinate points from the humidity sensing parameter subsequence, and arranging the humidity instantaneous sampling values according to the order of appearance of the geographical position coordinate points on the transportation carrier operation trajectory line, to generate a humidity parameter time sequence association list corresponding to the transportation carrier operation trajectory line.

[0038] Similarly, for the trajectory point Q_j, find the humidity instantaneous sampling value Hum_j from the humidity sensing parameter subsequence with the timestamp equal to T_j. Arrange Hum_j in the order of the trajectory point Q_j to generate the humidity parameter time sequence association list Hum_List, and the list length is also M.

[0039] Step S1273, extract the vibration instantaneous sampling value from the vibration sensing parameter subsequence with the same collection timestamp as each of the geographic position coordinate points, and arrange the vibration instantaneous sampling value in the order of the appearance of the geographic position coordinate points on the transportation carrier running trajectory line to generate the vibration parameter time sequence association list corresponding to the transportation carrier running trajectory line.

[0040] Similarly, for the trajectory point Q_j, find the vibration instantaneous sampling value Vib_j from the vibration sensing parameter subsequence with the timestamp equal to T_j. Arrange Vib_j in the order of the trajectory point Q_j to generate the vibration parameter time sequence association list Vib_List, and the list length is also M.

[0041] Step S1274, perform temperature state interval mapping processing on each temperature instantaneous sampling value in the temperature parameter time sequence association list, and convert each temperature instantaneous sampling value into a corresponding temperature state label according to a preset temperature state division rule, the temperature state label being used to represent the deviation degree category of the internal environment temperature of the transportation carrier relative to the temperature range required for drug storage.

[0042] The preset temperature state division rule is as follows. Assuming that the storage requirement temperature range of the vaccine is 2 degrees Celsius to 8 degrees Celsius. According to the range, several temperature state categories are defined: for example, when the temperature instantaneous sampling value is lower than 2 degrees Celsius, it is marked as a “low temperature deviation” state; when the temperature instantaneous sampling value is between 2 degrees Celsius and 8 degrees Celsius, it is marked as a “normal” state; when the temperature instantaneous sampling value is higher than 8 degrees Celsius but lower than 10 degrees Celsius, it is marked as a “mild over-temperature” state; and when the temperature instantaneous sampling value is higher than 10 degrees Celsius, it is marked as a “serious over-temperature” state. For each temperature value Temp_j in the temperature parameter time sequence association list Temp_List, determine it according to the above rule to convert it into a corresponding temperature state label, such as Temp_Label_j. Traverse the entire list to generate a temperature state label sequence Temp_Label_List.

[0043] Step S1275, perform humidity state interval mapping processing on each humidity instantaneous sampling value in the humidity parameter time sequence association list, and convert each humidity instantaneous sampling value into a corresponding humidity state label according to a preset humidity state division rule, the humidity state label being used to represent the deviation degree category of the internal environment humidity of the transportation carrier relative to the humidity range required for drug storage.

[0044] A preset humidity state classification rule is assumed that the storage requirement of the vaccine is a humidity range of 35% to 75%. According to the range, several humidity state categories are defined: for example, when the humidity instantaneous sampling value is lower than 35%, it is marked as a “dry” state; when the humidity instantaneous sampling value is between 35% and 75%, it is marked as a “normal” state; when the humidity instantaneous sampling value is higher than 75%, it is marked as a “humid” state. For each humidity value Hum_j in the humidity parameter time series correlation list Hum_List, the above rule is used for judgment, and the corresponding humidity state label Hum_Label_j is converted. The entire list is traversed to generate a humidity state label sequence Hum_Label_List.

[0045] Step S1276, vibration intensity level mapping processing is performed on each vibration instantaneous sampling value in the vibration parameter time series correlation list, and each vibration instantaneous sampling value is converted into a corresponding vibration intensity level label according to a preset vibration intensity level classification rule, the vibration intensity level label is used to represent the level of mechanical impact intensity that the transport carrier bears in the transportation process.

[0046] A preset vibration intensity level classification rule is defined according to the amplitude of the vibration instantaneous sampling value, for example, levels 1 to 5. For example, when the vibration value is less than threshold A, it is level 1, indicating slight vibration; when the vibration value is between threshold A and threshold B, it is level 2; when the vibration value is between threshold B and threshold C, it is level 3; when the vibration value is between threshold C and threshold D, it is level 4; when the vibration value is greater than threshold D, it is level 5, indicating severe impact. Thresholds A, B, C, and D can be dynamically set according to historical data and drug tolerance. For each vibration value Vib_j in the vibration parameter time series correlation list Vib_List, the above rule is used for judgment, and the corresponding vibration intensity level label Vib_Label_j is converted. The entire list is traversed to generate a vibration intensity level label sequence Vib_Label_List.

[0047] Step S1277, the temperature state label, the humidity state label, and the vibration intensity level label are combined and packaged according to the association relationship with the geographic location coordinate point to generate a multi-dimensional environmental state label set corresponding to each geographic location coordinate point in the transport carrier operation trajectory line.

[0048] For each trajectory point Q_j, its corresponding temperature state label Temp_Label_j, humidity state label Hum_Label_j and vibration intensity level label Vib_Label_j are combined together to form a 3-tuple, i.e. (Temp_Label_j, Hum_Label_j, Vib_Label_j). The 3-tuple is the multi-dimension environment state label of the trajectory point Q_j. All the 3-tuples of the trajectory points are arranged in the order of the trajectory points to generate a multi-dimension environment state label set which completely corresponds to the transportation carrier running trajectory line. Each element in the set is a label vector which contains the state information of three dimensions of temperature, humidity and vibration.

[0049] Step S130, constructing a logistics transfer node topology connection network which the transportation carrier running trajectory line passes through, the logistics transfer node topology connection network contains a plurality of logistics transfer node units and directed transfer path edges connecting the logistics transfer node units, the logistics transfer node units are obtained by matching the geographical position coordinate points in the transportation carrier running trajectory line with a preset logistics transfer node database.

[0050] Step S131, extracting all trajectory points marked as logistics node proximity points from the transportation carrier running trajectory line, and clustering the logistics node proximity points belonging to the same logistics transfer node identifier according to the node association identifier corresponding to each logistics node proximity point to obtain a node trajectory point set corresponding to each logistics transfer node identifier.

[0051] From the transportation carrier running trajectory line, all trajectory points marked as logistics node proximity points in step S125 are found. According to the node association identifier attached to each proximity point, proximity points with the same node association identifier are classified into the same category. For example, all proximity points with the node association identifier of node A form the trajectory point set S_A of node A; all proximity points with the node association identifier of node B form the trajectory point set S_B of node B, and so on. Each set S_X contains all trajectory points of the vehicle when it stays or passes through the vicinity of the corresponding node X.

[0052] Step S132, calculating the average value of the geographical position coordinates of all logistics node proximity points in each node trajectory point set to generate the actual passing position coordinates of the logistics transfer node unit represented by each logistics transfer node identifier in the transportation carrier running trajectory line.

[0053] For each node trajectory point set, for example, the set S_A of node A, the set contains several trajectory points, each point has its own longitude value and latitude value. The arithmetic mean of these longitude values is calculated to obtain the average longitude value; the arithmetic mean of these latitude values is calculated to obtain the average latitude value. The geographic position coordinates formed by the average longitude value and the average latitude value are the actual passing position coordinates P_A of node A in this transportation trajectory. The same calculation method is applied to node B, node C and all other nodes passed to obtain the actual passing position coordinates corresponding to each node.

[0054] Step S133, according to the time sequence of each logistics transfer node unit being sequentially passed on the transportation carrier running trajectory, arranging the logistics transfer node units in time sequence to generate a logistics transfer node sequence actually passed by the transportation carrier.

[0055] According to the earliest occurrence time of the trajectory point in each node trajectory point set, the time when the vehicle arrives at the node area can be determined. All the nodes passed are sorted according to the chronological order of the earliest time stamp in their corresponding trajectory point set. For example, the earliest time stamp in the node A set is T_A, the earliest time stamp in the node C set is T_C, and T_A is earlier than T_C. Node A is arranged before node C, and finally an ordered logistics transfer node sequence is generated, denoted as Node_Sequence, for example (node A, node C, node E, node H).

[0056] Step S134, for two adjacent logistics transfer node units in the logistics transfer node sequence, according to the actual passing position coordinates corresponding to the previous logistics transfer node unit and the actual passing position coordinates corresponding to the next logistics transfer node unit, the spatial straight line distance between the two logistics transfer node units is calculated as the path distance parameter.

[0057] For each pair of adjacent nodes in the logistics transfer node sequence Node_Sequence, for example, node A and node C, the actual passing position coordinates P_A of node A and the actual passing position coordinates P_C of node C are obtained. The spatial straight line distance between P_A and P_C, i.e. the Euclidean distance, is calculated in the same way as the displacement distance parameter in step S121. The path distance parameter D_AC is obtained, which represents the geographical distance between node A and node C.

[0058] Step S135, according to the chronological order of adjacent logistics transfer node units in the logistics transfer node sequence, a directed transfer path edge connecting the previous logistics transfer node unit to the next logistics transfer node unit is determined, and the path distance parameter is attached to the directed transfer path edge as edge attribute information.

[0059] According to the order in the node sequence, node A is in front and node C is in back, so a directed transfer path edge Edge_AC is defined from node A to node C. The path distance parameter D_AC calculated in step S134 is attached to the edge as attribute information of the edge. For the next pair of nodes C and E, a directed edge Edge_CE is also defined, and the path distance parameter D_CE is attached.

[0060] Step S136, all the logistics transfer node units are taken as network nodes, all the directed transfer path edges are taken as directed edges connecting the network nodes, and the edge attribute information is combined to construct the logistics transfer node topology connection network.

[0061] All the logistics transfer node units identified in step S133, i.e. node A, node C, node E, and node H, are taken as nodes in the network. All the directed transfer path edges defined in step S135, i.e. Edge_AC, Edge_CE, and Edge_EH, are taken as directed edges connecting these nodes. Each edge is accompanied by the corresponding path distance attribute. Thus, a directed graph structure, i.e. the logistics transfer node topology connection network, is constructed. The network clearly reflects the actual logistics node sequence and the spatial distance relationship between the nodes.

[0062] Step S140, the transport carrier running trajectory line, the multi-dimensional environment state label set, and the logistics transfer node topology connection network are input into the pre-constructed spatio-temporal state feature fusion tracking model for joint analysis processing to generate a medical cold chain logistics transportation whole-process state tracking feature tensor.

[0063] Step S141, the transport carrier running trajectory line is input into the trajectory encoding module of the spatio-temporal state feature fusion tracking model, the sequence of geographical position coordinate points in the transport carrier running trajectory line is subjected to spatial position embedding representation processing to generate a sequence of spatial trajectory embedding feature vectors corresponding to the transport carrier running trajectory line.

[0064] The spatio-temporal state feature fusion tracking model is a pre-constructed and trained deep learning model. First, the transportation carrier running trajectory line, i.e., the effective geographic position coordinate point sequence Q, is input into the trajectory encoding module of the model. The module first normalizes the longitude value and the latitude value of each trajectory point Q_j, mapping the original longitude range and latitude range to a unified numerical interval, such as between -1 and 1. Then, through a linear transformation layer, the normalized two-dimensional coordinates (longitude value, latitude value) are mapped to a higher-dimensional vector, such as a spatial trajectory embedding feature vector E_pos_j with a dimension of D1. The linear transformation layer is a fully connected network, and its weight parameters are learned during the model training process. For all M points in the trajectory line, after processing by the trajectory encoding module, a spatial trajectory embedding feature vector sequence E_pos is generated. The spatial trajectory embedding feature vector sequence is a matrix with a shape of M times D1, and each row corresponds to a spatial position embedding representation of a trajectory point.

[0065] In step S142, the multi-dimensional environment state label set is input into the state encoding module of the spatio-temporal state feature fusion tracking model, and the temperature state label, humidity state label, and vibration intensity level label corresponding to each geographic position coordinate point in the multi-dimensional environment state label set are jointly embedded and represented to generate an environment state embedding feature vector corresponding to each geographic position coordinate point. All environment state embedding feature vectors are arranged in the order of the appearance of the geographic position coordinate points to generate an environment state embedding feature vector sequence.

[0066] Next, the multi-dimensional environment state label set, i.e., the (Temp Label j, Hum Label j, Vib Label j) triplet corresponding to each trajectory point Q j, is input into the state encoding module of the model. The temperature state label, humidity state label, and vibration intensity level label are all discrete categorical variables. The state encoding module first establishes an independent embedding lookup table for each categorical variable. For example, for the temperature state label, an embedding table is established to map the four categories of "low temperature deviation", "normal", "mild over-temperature", and "severe over-temperature" to a vector of dimension D2. Similarly, embedding tables are established for the humidity state label and the vibration intensity level label, mapping them to vectors of the same dimension D2. For trajectory point Q j, its temperature embedding vector e temp j, humidity embedding vector e hum j, and vibration embedding vector e vib j are obtained by looking up the tables respectively. The dimensions of these three vectors are all D2. Then, the three vectors are concatenated in the feature channel dimension to form an environment state embedding feature vector E env j of dimension 3 times D2. The same operation is performed on all M trajectory points to obtain an environment state embedding feature vector sequence E env, which is a matrix of shape M times (3*D2).

[0067] In step S143, the logistics transfer node topology connection network is input into a topology encoding module of the spatio-temporal state feature fusion tracking model, graph structure embedding representation processing is performed on the logistics transfer node units and directed transfer path edges in the logistics transfer node topology connection network, and a network topology embedding feature representation corresponding to the logistics transfer node topology connection network is generated. The network topology embedding feature representation includes a node embedding feature vector of each logistics transfer node unit and an edge embedding feature vector of each directed transfer path edge.

[0068] The constructed logistics transfer node topology connection network, i.e., the graph including nodes A, C, E, H and edges Edge_AC, Edge_CE, Edge_EH, is input into the topology encoding module. The module first generates an initial node feature vector for each node in the network, such as node A. The initial vector can be based on the inherent properties of the node, such as the type of the node (e.g., provincial hub, municipal distribution center) or the historical throughput of the node, etc. If there is no such information, random initialization or zero vector can be used. Then, through a graph neural network layer, such as a graph convolution network layer or a graph attention network layer, the information of neighboring nodes is aggregated to update the feature representation of each node. In the graph neural network layer, the updated features of node A will consider the information of its neighbor node C and the edge attribute (path distance parameter D_AC) of the edge Edge_AC between them. After processing by the graph neural network layer, each node is encoded into a node embedding feature vector with a dimension of D3, such as the embedding vector of node A Node_A_emb. At the same time, for each edge, such as Edge_AC, its path distance parameter D_AC can be mapped to an edge embedding feature vector Edge_AC_emb with a dimension of D4 through a linear transformation. Finally, the topology encoding module outputs the network topology embedding feature representation, which includes the set of embedding feature vectors of all nodes and the set of embedding feature vectors of all edges.

[0069] Step S144, feature alignment and splicing processing of the space trajectory embedding feature vector sequence and the environment state embedding feature vector sequence on the time axis are performed to generate a primary joint feature tensor that fuses space-time information. Each time step of the primary joint feature tensor contains space position information and environment state information at the corresponding time point.

[0070] Step S1441, the original collection time stamp corresponding to each space trajectory embedding feature vector in the space trajectory embedding feature vector sequence is obtained, and the original collection time stamp corresponding to each environment state embedding feature vector in the environment state embedding feature vector sequence is obtained.

[0071] Since the space trajectory embedding feature vector sequence E_pos and the environment state embedding feature vector sequence E_env are both generated in the order of the trajectory points Q_j, they naturally have the same order and one-to-one correspondence. Each vector is associated with the original collection time stamp T_j of the same trajectory point Q_j. Therefore, the alignment of this step is direct.

[0072] Step S1442, according to the original collection timestamp, the spatial trajectory embedding feature vector sequence and the environment state embedding feature vector sequence are aligned and checked in the time axis, and it is confirmed that the spatial trajectory embedding feature vector sequence and the environment state embedding feature vector sequence have a one-to-one corresponding relationship in the time dimension.

[0073] By simply comparing whether the lengths of the two sequences are equal, both are M, and whether the vectors corresponding to each position are associated with the same trajectory point, it can be confirmed that they are one-to-one corresponding in the time dimension.

[0074] Step S1443, the spatial trajectory embedding feature vector and the environment state embedding feature vector at each time point after alignment are spliced and combined in the feature channel dimension, to generate a spatio-temporal combined feature vector corresponding to each time point.

[0075] For each trajectory point Q_j, the spatial trajectory embedding feature vector E_pos_j with a dimension of D1 is spliced with the environment state embedding feature vector E_env_j with a dimension of 3 times D2 in the feature channel dimension. The dimension of the spliced new vector F_j is D1 plus 3 times D2. The F_j is the spatio-temporal combined feature vector corresponding to the trajectory point Q_j, which simultaneously fuses the spatial position information and the multi-dimensional environment state information of the point.

[0076] Step S1444, the spatio-temporal combined feature vectors corresponding to all time points are stacked and arranged in time sequence to form a two-dimensional feature matrix as the primary joint feature tensor, and the row dimension of the two-dimensional feature matrix corresponds to the time step, and the column dimension corresponds to the total number of spliced feature channels.

[0077] The spatio-temporal combined feature vectors F_1 to F_M corresponding to all M trajectory points are arranged in a two-dimensional matrix in the order from 1 to M. The first row of the matrix is F_1, the second row is F_2, and so on, and the last row is F_M. The shape of the matrix is M times (D1 plus 3 times D2), which is called the primary joint feature tensor T_primary.

[0078] Step S1445, the primary joint feature tensor is subjected to feature dimension normalization processing, the numerical range of each feature channel in the primary joint feature tensor is adjusted to a preset model input interval, and a normalized primary joint feature tensor with uniform scale representation is generated.

[0079] In order to stabilize the subsequent model processing, the primary joint feature tensor T primary is subjected to layer normalization processing. For each row in the matrix (i.e., the feature vector at each time step), the mean and variance of all eigenvalues in the row are calculated, and then each eigenvalue in the row is subtracted from the mean and divided by the square root of the variance, so that the mean of each row feature is 0 and the variance is 1. After layer normalization processing, the normalized primary joint feature tensor T primary norm is obtained as the input of the subsequent module.

[0080] Step S145, inputting the primary joint feature tensor and the network topology embedding feature representation into a cross-attention fusion module of the spatio-temporal state feature fusion tracking model, taking the primary joint feature tensor as the query basis and taking the network topology embedding feature representation as the key-value basis, calculating the association weight of each time step feature in the primary joint feature tensor and each node and edge in the logistics transfer node topology connection network through a multi-head cross-attention mechanism, and generating an enhanced feature tensor with network topology context information by weighting and aggregating the network topology embedding feature representation according to the association weight.

[0081] Step S1451, linearly mapping the primary joint feature tensor into a query matrix required by the multi-head cross-attention mechanism, the number of rows of the query matrix corresponding to the time step length of the primary joint feature tensor, and the number of columns of the query matrix corresponding to a preset query feature dimension.

[0082] The normalized primary joint feature tensor T primary norm, which has a shape of M by (D1+3D2), is input into a linear mapping layer (a fully connected network) to map its feature dimension from (D1+3D2) to a preset query feature dimension D q. The mapping result is a matrix with a shape of M by D q, which is used as the query matrix Q of the multi-head cross-attention mechanism.

[0083] Step S1452, integrating all node embedding feature vectors and edge embedding feature vectors contained in the network topology embedding feature representation, and linearly mapping them into a key matrix and a value matrix required by the multi-head cross-attention mechanism, the number of rows of the key matrix and the value matrix corresponding to the total number of node embedding feature vectors and edge embedding feature vectors, and the number of columns of the key matrix and the value matrix corresponding to a preset key feature dimension and a value feature dimension, respectively.

[0084] All node embedding feature vectors in the network topology embedding feature representation, such as Node_A_emb, Node_C_emb, and all edge embedding feature vectors, such as Edge_AC_emb, Edge_CE_emb, are integrated into a list, which contains L elements (L equals the number of nodes plus the number of edges). The list is input into another linear mapping layer to map to a preset key feature dimension D_k, obtaining a key matrix K with a shape of L by D_k. At the same time, the same list is input into another linear mapping layer to map to a preset value feature dimension D_v, obtaining a value matrix V with a shape of L by D_v.

[0085] Step S1453, performing dot product operation on the query matrix and the key matrix to obtain an attention score matrix, each element in the attention score matrix representing the original score of the association strength between the feature of one time step in the primary joint feature tensor and one node or one edge in the network topology embedding feature representation.

[0086] The dot product of the query matrix Q (M by D_q) and the transpose of the key matrix K (D_k by L) is calculated. The result of the dot product operation is an attention score matrix S with a shape of M by L. The element S_ij in the i-th row and j-th column of the matrix S represents the original association strength score between the trajectory feature of the i-th time step and the j-th network topology element (a node or an edge).

[0087] Step S1454, performing scaling processing and softmax normalization processing on the attention score matrix to generate an attention weight matrix, the sum of each row of elements in the attention weight matrix being one, used to represent the attention distribution of the feature of each time step to all nodes and edges.

[0088] Each element in the attention score matrix S is scaled by dividing by the square root of D_k to prevent the gradient from vanishing due to the large dot product result. Then, the softmax function is applied to each row of the scaled matrix to normalize it. After the softmax processing, the attention weight matrix A is obtained, which still has a shape of M by L. Each row of the matrix A, such as the i-th row, has a sum of all L elements equal to one, and each element has a value between 0 and 1, representing the attention weight of the trajectory feature of the i-th time step to each of the L network topology elements.

[0089] Step S1455, performing weighted summation operation on the attention weight matrix and the value matrix to obtain a weighted network topology context feature matrix, the number of rows of the weighted network topology context feature matrix being the same as the time step length of the primary joint feature tensor, and the number of columns being the same as the value feature dimension.

[0090] The attention weight matrix A (M by L) is multiplied with the value matrix V (L by D_v). The result is a matrix C with shape M by D_v, which is the weighted network topology context feature matrix. The i-th row in matrix C is the weighted sum of all L row vectors in the value matrix V with the weights in the i-th row of the attention weight matrix A. Therefore, each row in matrix C contains the most relevant context information from the entire logistics transfer node topology connection network to the trajectory feature at the i-th time step.

[0091] Step S1456, the primary joint feature tensor is spliced and fused with the weighted network topology context feature matrix in the feature channel dimension to generate the enhanced feature tensor with network topology context information.

[0092] The original primary joint feature tensor T_primary_norm (shape M by (D1+3D2)) is spliced with the weighted network topology context feature matrix C (shape M by D_v) just generated in the feature channel dimension. The shape of the spliced new tensor T_enhanced is M by ((D1+3D2)+D_v). The T_enhanced is the enhanced feature tensor with network topology context information, which fuses the original spatio-temporal trajectory environment information and the context information obtained from the logistics node network.

[0093] Step S146, the enhanced feature tensor is input into the time sequence convolutional coding layer of the spatio-temporal state feature fusion tracking model. The time sequence convolutional coding layer performs multi-scale time receptive field feature extraction processing on the enhanced feature tensor through time sequence convolution kernels with different expansion rates to generate the pharmaceutical cold chain logistics transportation whole-process state tracking feature tensor.

[0094] The enhanced feature tensor T_enhanced is input into a temporal convolutional encoding layer. The layer is composed of multiple parallel temporal convolutional blocks, each using a convolutional kernel with a different dilation rate. For example, the first convolutional block uses a convolutional kernel with a dilation rate of 1, has a small receptive field, and is used to capture local changes over a short period of time; the second convolutional block uses a convolutional kernel with a dilation rate of 2, has an expanded receptive field, and is used to capture patterns over a slightly longer period of time; the third convolutional block uses a convolutional kernel with a dilation rate of 4, has an even larger receptive field, and is used to capture periodic trends over a longer period of time. Each temporal convolutional block performs convolutional operations on the input tensor T_enhanced along the time dimension (M time steps) to extract features. The output of each convolutional block is a new feature map that retains the same time step length (still M) but may have a different number of feature channels. Finally, the outputs of all convolutional blocks are concatenated along the feature channel dimension to obtain a final output tensor. This final output tensor is the medicine cold-chain logistics transportation whole-process state tracking feature tensor T_final. The shape of T_final is M by D_final, where D_final is the sum of the output channel numbers of all convolutional blocks. This tensor densely encodes the deep feature representation of the spatial, environmental, and network topology context information at each time step throughout the entire transportation process.

[0095] Step S150, performing abnormal state pattern matching processing on the medicine cold-chain logistics transportation whole-process state tracking feature tensor to obtain an abnormal state type identifier and an abnormal state spatio-temporal distribution feature corresponding to the medicine cold-chain logistics transportation whole-process state tracking feature tensor, and generating a medicine cold-chain logistics transportation abnormality early warning instruction set according to the abnormal state type identifier and the abnormal state spatio-temporal distribution feature.

[0096] Step S151, inputting the medicine cold-chain logistics transportation whole-process state tracking feature tensor into a pre-trained abnormal state pattern recognition classifier, the abnormal state pattern recognition classifier performing global feature extraction and pattern classification processing on the medicine cold-chain logistics transportation whole-process state tracking feature tensor, and outputting an abnormal state type identifier to which the medicine cold-chain logistics transportation whole-process state tracking feature tensor belongs.

[0097] The whole-process state tracking feature tensor T_final generated in step S156, which has a shape of M by D_final, is input into a pre-trained abnormal state pattern recognition classifier. The classifier first performs a global average pooling operation on T_final along the time dimension M through a global average pooling layer, i.e., for each feature channel, it calculates the arithmetic mean of the feature values of that channel over all M time steps to obtain a global feature vector G with a dimension of D_final. This global feature vector G condenses the deep state features of the entire transportation process.

[0098] Then, the global feature vector G is input into a classification head of the classifier. The classification head is stacked by multiple fully connected layers, for example, the first layer maps G of D_final dimensions to an intermediate vector of 512 dimensions and passes through a ReLU activation function, the second layer maps 512 dimensions to 256 dimensions and passes through a ReLU activation function, and the last layer maps 256 dimensions to an output vector of C dimensions, where C is equal to the total number of predefined abnormal state types, which are not simply single-parameter over-limit categories, but a complex abnormal category system defined based on spatiotemporal coupling features of multi-source sensing data. The system includes but is not limited to the following complex abnormal types: for example, the type of "refrigeration unit progressive failure accompanied by compartment heat load imbalance", which is characterized by a coordinated mode of continuous enhancement of energy in a specific frequency band of the vibration spectrum and nonlinear acceleration of the temperature rise rate in the feature space; for example, the type of "compartment sealing damage accompanied by external hot and humid air intrusion", which is characterized by the coupling features of saturated humidity parameter and fluctuating temperature parameter, and is often accompanied by geographical location parameter showing that the vehicle is in a high-humidity external environment; for example, the type of "transport path bumping leading to accumulation of drug packaging micro-damage", which is characterized by the impact energy accumulation value of the vibration parameter exceeding the dynamic threshold and having statistical correlation with the road surface grade index of the vehicle driving section. The C-dimensional vector output by the last layer is normalized by a softmax function to obtain a C-dimensional probability distribution vector P, each element value in the vector is between 0 and 1, and the sum of all elements is 1, indicating the probability of T_final belonging to each complex abnormal state type. The identifier of the type with the highest probability value is selected as the final abnormal state type identifier, for example, if the probability of "refrigeration unit progressive failure accompanied by compartment heat load imbalance" is the highest, the identifier "refrigeration unit progressive failure accompanied by compartment heat load imbalance" is output.

[0099] In step S152, the medicine cold-chain logistics transportation whole-process state tracking feature tensor is input into a pre-trained abnormal spatiotemporal positioning network. The abnormal spatiotemporal positioning network performs time-step-by-time-step feature analysis processing on the medicine cold-chain logistics transportation whole-process state tracking feature tensor to generate an abnormal confidence score sequence corresponding to each time step.

[0100] Meanwhile, the same full-path state tracking feature tensor T_final is input into a parallel abnormal spatio-temporal localization network. The network adopts an architecture combining temporal convolution and bidirectional recurrent neural network. Specifically, T_final is first passed through a one-dimensional temporal convolution layer, which uses multiple convolution kernels with a size of 3 to perform convolution operations along the time dimension to capture feature changes within a local time window, and outputs a feature map T_conv, which maintains the shape of M by a certain feature dimension. Then, T_conv is input into a bidirectional long short-term memory network layer. The bidirectional long short-term memory network layer contains two long short-term memory network sub-layers, a forward sub-layer and a backward sub-layer. The forward sub-layer processes T_conv in time order from step 1 to step M, and outputs a forward hidden state sequence; the backward sub-layer processes T_conv in time reverse order from step M to step 1, and outputs a backward hidden state sequence. For each time step j, the forward hidden state and the backward hidden state are spliced in the feature channel dimension to obtain a hidden state vector H_j that integrates the forward and backward context information. All H_j of all time steps form a sequence, and the sequence length is M. Then, the hidden state sequence is input into an output layer, which is a fully connected layer plus a sigmoid activation function, which maps H_j of each time step to a scalar value s_j between 0 and 1 as the abnormal confidence score of the time step. Finally, the abnormal spatio-temporal localization network outputs an abnormal confidence score sequence S_list with a length of M, S_list equals (s_1, s_2,..., s_M), and each element s_j in the sequence corresponds to the abnormal confidence score of the trajectory point Q_j at the time, which represents the probability of the occurrence of the above-mentioned composite abnormal state within the local time window in which the time point is located.

[0101] In step S153, according to the abnormal confidence score of each time step in the abnormal confidence score sequence, a continuous time step interval whose abnormal confidence score exceeds a preset abnormal judgment threshold is screened out, and a time range corresponding to the continuous time step interval on the running trajectory line of the transport carrier is determined as an abnormal state occurrence time period.

[0102] An abnormality determination threshold Th abnorma is set, which is not a fixed constant, but an adaptive threshold dynamically adjusted according to the abnormality state type identification. For example, for the abnormality type of "progressive failure of refrigeration unit accompanied by imbalance of heat load in vehicle cabin", which develops slowly but has serious consequences, the threshold is set to a relatively low 0.7 to capture the early abnormality germination; for the abnormality type of "damage to the sealing of the vehicle cabin accompanied by the intrusion of external hot and humid air", which is strong in suddenness, the threshold is set to a higher 0.9 to avoid frequent false positives. Traverse each score s j in the abnormality confidence score sequence S list, and mark those time steps with s j greater than the corresponding dynamic threshold as candidate abnormal time steps. Then, the continuous segments in the above candidate abnormal time steps are merged into an interval. For example, assuming in a certain transportation scenario, for the abnormality type of "progressive failure of refrigeration unit accompanied by imbalance of heat load in vehicle cabin", all s j from the 800th second to the 1200th second are greater than 0.7, then a continuous interval [800, 1200] is formed; all s j from the 2000th second to the 2150th second are greater than 0.7, then another continuous interval [2000, 2150] is formed. For each of the above continuous intervals, according to its starting time step and ending time step, it corresponds to a specific time range on the transportation carrier running trajectory line. Assuming that the starting time step is the a th step, the corresponding timestamp is T a, and the ending time step is the b th step, the corresponding timestamp is T b, then an abnormal state occurrence time period is determined from T a to T b. In the above example, two abnormal state occurrence time periods are obtained: time period 1 is from T 800 to T 1200, and time period 2 is from T 2000 to T 2150.

[0103] Step S154, according to the abnormal state occurrence time period, extracts the trajectory segment corresponding to the abnormal state occurrence time period from the transportation carrier running trajectory line as an abnormal trajectory segment, and extracts the environmental state label subset corresponding to the abnormal state occurrence time period from the multi-dimensional environmental state label set as an abnormal environmental state label subset.

[0104] For each abnormal state occurrence time period, for example, time period 1 from T_800 to T_1200, all the corresponding trajectory points in this time period are extracted from the trajectory line of the transport carrier. The trajectory points are from the point Q_800 corresponding to the 800th time step to the point Q_1200 corresponding to the 1200th time step, forming an abnormal trajectory segment 1, denoted as Traj_seg_1. At the same time, from the multi-dimensional environment state label set, all the corresponding environment state label triples in this time period are extracted, from (Temp_Label_800, Hum_Label_800, Vib_Label_800) of the 800th time step to (Temp_Label_1200, Hum_Label_1200, Vib_Label_1200) of the 1200th time step, forming an abnormal environment state label subset 1, denoted as Env_subset_1. For time period 2, the above operations are also performed to obtain abnormal trajectory segment 2 and abnormal environment state label subset 2.

[0105] Step S155, according to the sequence of geographic position coordinate points in the abnormal trajectory segment, calculate the geographic area range covered by the abnormal state in the abnormal state occurrence time period, and according to the distribution of each state label in the abnormal environment state label subset, generate the abnormal change trend description of the environment state in the abnormal state occurrence time period.

[0106] For each abnormal trajectory segment, e.g. Traj_seg_1, extract the longitude values of all trajectory points, find the minimum value Lon_min_1 and the maximum value Lon_max_1; extract the latitude values of all trajectory points, find the minimum value Lat_min_1 and the maximum value Lat_max_1. Draw a rectangular geographic area with the four extreme points (Lon_min_1, Lat_min_1) and (Lon_max_1, Lat_max_1) as the geographic area range Area_1 covered by the abnormal state. Perform multi-dimensional time series analysis on the corresponding abnormal environmental state label subset Env_subset_1. Take temperature state label as an example, map it to a numerical sequence (e.g. “normal” to 0, “mild over-temperature” to 1, “severe over-temperature” to 2), calculate the first-order difference of the sequence to get the temperature change rate sequence; calculate the sliding window variance of the sequence to get the temperature volatility sequence. Perform similar processing on humidity state label and vibration intensity level label. Then extract key features from these derived sequences: e.g. temperature change rate is positive and acceleration exceeds threshold, indicating temperature is accelerating upward; periodic impact pattern appears in vibration level sequence, indicating vibration has specific frequency characteristics. Combine the above features to generate structured abnormal change trend description Trend_1, e.g. “in time period 1, temperature state rises from normal to severe over-temperature at an increasing rate, temperature rising acceleration is positive; humidity state maintains in normal range but presents weak negative correlation fluctuation with temperature change; vibration level changes from level 2 steady state to level 3 to level 4 periodic impact pattern, impact interval is about 15 to 20 seconds”.

[0107] Step S156, combine and package the abnormal state occurrence time period, the abnormal trajectory segment, the abnormal environmental state label subset, the geographic area range covered by the abnormal state, and the abnormal change trend description to generate the abnormal state spatio-temporal distribution feature.

[0108] All the time periods of abnormal states, the corresponding abnormal trajectory segments for each time period, the subset of abnormal environmental state markers for each time period, the geographical area for each time period, and the description of the abnormal change trends for each time period are organized chronologically to form a structured data object. For example, for time period 1, a record item Item_1 is generated, which equals (time period 1, Traj_seg_1, Env_subset_1, Area_1, Trend_1); for time period 2, Item_2 is generated, which equals (time period 2, Traj_seg_2, Env_subset_2, Area_2, Trend_2). All these record items are arranged chronologically to constitute the spatiotemporal distribution characteristics of abnormal states. This characteristic fully describes when, where, and what kind of environmental anomalies with complex internal relationships occurred, as well as the multidimensional evolutionary patterns of the anomalies.

[0109] Step S157: Generate a set of early warning instructions for abnormal medical cold chain logistics transportation based on the abnormal state type identifier and the spatiotemporal distribution characteristics of the abnormal state.

[0110] Step S1571: Parse the abnormal state type identifier and search for the warning level code corresponding to the abnormal state type identifier from the preset mapping relationship library. The warning level code is used to indicate the severity level of the abnormal state.

[0111] The abnormal state type identifier output in step S151 is analyzed, assuming it is "progressive failure of refrigeration unit accompanied by unbalanced heat load in the carriage". A search is performed from a pre-defined mapping database. This database is a multi-dimensional lookup table built based on historical accident severity statistics and drug loss risk assessment. It predefines the correspondence between each composite abnormality type and multiple warning levels, including: drug quality impact level, emergency response time window level, and economic loss risk level. For example, the drug quality impact level corresponding to "progressive failure of refrigeration unit accompanied by unbalanced heat load in the carriage" is "high impact" (may lead to a large-scale reduction in drug efficacy), the emergency response time window level is "urgent" (intervention must be carried out within 2 hours), and the economic loss risk level is "significant loss". Combining these three dimensions, a pre-defined fusion rule (e.g., taking the highest level or a weighted average) is used to obtain the final warning level code, "Level 1 Warning" (the highest level).

[0112] Step S1572: Extract the start and end time points of the time period during which the abnormal state occurs from the spatiotemporal distribution features of the abnormal state, and extract the start and end geographical coordinates of the abnormal trajectory segment from the spatiotemporal distribution features of the abnormal state as the warning trigger location coordinates.

[0113] From the abnormal state spatio-temporal distribution characteristics, the starting time point and the ending time point of the first abnormal time period (i.e., the earliest occurring time period) are extracted, for example, the starting time point T_start_alarm equal to T_800 and the ending time point T_end_alarm equal to T_1200 are extracted from the time period 1. At the same time, from the abnormal trajectory segment Traj_seg_1 corresponding to the time period, the geographic position coordinates of the first trajectory point Q_800 are extracted as the starting warning position P_start_alarm, and the geographic position coordinates of the last trajectory point Q_1200 are extracted as the ending warning position P_end_alarm.

[0114] In step S1573, an abnormal environment state marker subset is extracted from the abnormal state spatio-temporal distribution characteristics, and an environment state abnormal evolution description vector is generated according to the change sequence of the temperature state marker, the humidity state marker, and the vibration intensity level marker in the abnormal state occurrence time period in the abnormal environment state marker subset.

[0115] From the abnormal state spatio-temporal distribution characteristics, the abnormal environment state marker subset Env_subset_1 corresponding to the first abnormal time period is extracted. The subset contains the temperature state marker, the humidity state marker, and the vibration intensity level marker of each time step from time step 800 to 1200. First, the discrete state markers are converted into numerical vectors according to the preset complex coding rules. For the temperature state marker, not only its category (such as "normal" for 0, "mild over-temperature" for 1, and "severe over-temperature" for 2) is encoded, but also its change trend feature is encoded, for example, the change direction coding (up for +1, stable for 0, and down for -1) is obtained by calculating the difference of consecutive time steps, and the volatility coding (high volatility for 1 and low volatility for 0) is obtained by calculating the variance in the local window. Similar multi-element coding rules are established for the humidity state marker and the vibration intensity level marker. In this way, the original three-tuple of each time step is expanded into a higher-dimensional numerical vector, for example, an 8-dimensional vector, which contains the original category coding, the change direction coding, and the volatility coding. Then, these high-dimensional vectors of all time steps in the time period are stacked in time order to form a two-dimensional numerical matrix, with the number of rows being 401 and the number of columns being 8. The matrix is flattened to form a one-dimensional sequence as an environment state abnormal evolution description vector Evo_Vector, with a dimension of 401 times 8.

[0116] In step S1574, the warning level code, the starting time point, the ending time point, the warning trigger position coordinates, and the environment state abnormal evolution description vector are combined to generate a basic warning instruction unit for the currently detected abnormal state.

[0117] Combine the information obtained in steps S1571 to S1673: the warning level is encoded as "Level 1 warning", the start time point is T_start_alarm, the end time point is T_end_alarm, the warning trigger position coordinates are (P_start_alarm, P_end_alarm), and the evolution description vector is Evo_Vector. Serialize and package these five parts according to the predefined data packaging format to form a basic warning instruction unit Unit_base.

[0118] Step S1575, determine whether the abnormal state type identifier belongs to a preset chain reaction abnormal type set, if the abnormal state type identifier belongs to the chain reaction abnormal type set, predict the downstream logistics transfer node units that the abnormal state may affect according to the logistics transfer node topology connection network, and generate predictive warning instruction units for the downstream logistics transfer node units.

[0119] Check whether the abnormal state type identifier "refrigeration unit progressive failure accompanied by car heat load imbalance" output in step S151 is in a preset chain reaction abnormal type set. The chain reaction abnormal type set is a composite abnormal category system constructed based on post-accident analysis of historical transportation accident data and the physical mechanism of pharmaceutical cold chain transportation, and contains composite abnormal types that have a clear causal propagation path or a high probability of triggering secondary abnormalities. "Refrigeration unit progressive failure accompanied by car heat load imbalance" is pre-classified into this set because it is highly likely to cause a series of subsequent secondary abnormalities: first, the uneven temperature distribution in the car intensifies, which in turn triggers temperature overrun in local hot spots; temperature overrun can cause performance degradation of some pharmaceutical packaging materials, releasing trace amounts of moisture, which in turn causes humidity to rise abnormally; the abnormal rise in humidity coupled with continuous vibration can cause the sealing performance of pharmaceutical packaging to decline, ultimately leading to the risk of pharmaceutical contamination.

[0120] According to the logistics transfer node topology connection network constructed in step S146, the prediction is made. The current abnormality occurs between node A and node C in the logistics transfer node sequence, and the vehicle is driving from node A to node C. The attribute information of node C is obtained from the network structure, including that the type of node C is "regional distribution center", the temperature control capability level is "level 2", the average stay time of the same type of pharmaceuticals passing through this node is "several hours", and the temperature field distribution uniformity historical statistical index of the internal storage area of node C is "standard deviation 0.3 degrees Celsius". At the same time, the attribute information of the downstream node of node C, i.e. node E, is obtained, and the type of node E is "terminal distribution station", the temperature control capability level is "level 3", the storage space type is "small refrigerated warehouse", and the "temperature overrun event frequency" in the historical temperature and humidity fluctuation record of node E is "several times per year".

[0121] Based on the above information, a predictive early warning instruction unit Unit_pred_C is generated for node C. This unit contains the following specific content:

[0122] First, the predicted abnormal type identification sequence, which is a vector containing the possible secondary abnormal types and their occurrence probabilities. For example, the first element is "temperature field uniformity deterioration", the occurrence probability P_temp_field is calculated based on the refrigeration failure degree parameter in the current abnormal type, the node C temperature control level parameter, and the node C historical temperature field uniformity index, P_temp_field is equal to the sigmoid function acting on the weighted linear combination of the above parameters; the second element is "local hotspot area temperature overrun", the occurrence probability P_hotspot is calculated based on the temperature field uniformity deterioration probability and the node C internal goods stacking density parameter (extracted from historical data); the third element is "humidity abnormality caused by packaging material moisture absorption", the occurrence probability P_humidity is calculated based on the local hotspot area temperature overrun probability and the node C environmental basic humidity parameter (obtained from meteorological data).

[0123] Second, the predicted arrival time window at node C, which is an interval [T_arrive_C_early, T_arrive_C_late], calculated based on the spatial distance between the current vehicle position and the actual passing position coordinates of node C, the current average speed, and the historical travel time fluctuation range of the same route transportation, for example, T_arrive_C_early is the current time plus 3.5 hours, T_arrive_C_late is the current time plus 4.2 hours.

[0124] Third, the recommended measure code set, which contains specific operation suggestions for different functional positions of node C. For example, the measure code for the temperature control technician is "start the standby refrigeration unit of node C in advance and preheat to the set temperature interval, and check the calibration status of the temperature sensor array"; the measure code for the warehouse management personnel is "clear the dedicated isolation area before the arrival of the medicine, which should have independent temperature monitoring and recording functions"; the measure code for the quality inspection personnel is "prepare portable temperature and humidity recorders, packaging integrity detection equipment, and sampling tools, and immediately perform online detection after the arrival of the medicine".

[0125] Similarly, a predictive early warning instruction unit Unit_pred_E is generated for node E. This predictive early warning instruction unit contains:

[0126] First, the predicted abnormal type identification sequence is further propagated on the basis of the node C prediction result. For example, the first element is "drug packaging sealability decline", and the occurrence probability P seal is calculated on the basis of the humidity abnormality probability in the node C prediction result, the transportation distance parameter from the node E to the node C, and the historical road surface grade index of the road section. The second element is "drug contamination risk", and the occurrence probability P contamination is calculated on the basis of the packaging sealability decline probability and the storage environment cleanliness level parameter (obtained from the facility file) of the node E.

[0127] Second, the time window for arriving at the node E is predicted, which is calculated on the basis of the time window for arriving at the node C, the spatial distance from the node C to the node E, and the average stay duration of the node C. For example, T arrive E early is T arrive C early plus the average stay duration of 4 hours of the node C plus the road travel time, and T arrive E late is T arrive C late plus the maximum stay duration of 6 hours of the node C plus the road travel time.

[0128] Third, the recommended measure code set is generated. For example, the measure code for the receiving personnel of the node E is "prepare a special isolated storage area, and increase the sampling rate of the batch of drugs before entering the warehouse to 3 times the standard rate, and focus on detecting the packaging sealability and water activity". The measure code for the quality traceability personnel is "create an abnormal batch tracking label in the quality traceability system, and automatically associate the summary of the whole process sensing data of this transportation, so as to facilitate subsequent quality investigation".

[0129] Step S1576, all generated basic early warning instruction units and predictive early warning instruction units are summarized and sorted according to the time sequence and spatial correlation, to form the medical cold chain logistics transportation abnormal early warning instruction set.

[0130] The base early warning instruction unit Unit_base is aggregated with all predictive early warning instruction units Unit_pred_C and Unit_pred_E. In chronological order, Unit_base corresponds to the current immediately occurring anomaly, the earliest in time, and is placed first; Unit_pred_C corresponds to the next node that is about to be reached, the second in time; and Unit_pred_E corresponds to a node further downstream, the last in time. At the same time, in space, they are associated with the corresponding node identifiers, and the path distance information between nodes is attached, and finally the early warning instruction set list is formed. Each element in the early warning instruction set list is a structured early warning instruction unit, which contains complete early warning levels, time and space information, anomaly evolution characteristics, prediction information, and specific recommended measures. The instruction set can be serialized into a standard format data packet, sent to the monitoring center's large screen display system, the transport driver's mobile terminal application program, the early warning receiving equipment of the on-site management personnel of node C and node E, and the quality management department system of the drug receiving party, so that all parties can take coordinated disposal actions in advance according to the early warning level and the prediction information, to minimize drug loss and quality risk.

[0131] For example, the method further comprises:

[0132] Step S210, a plurality of historical original sensor tracking data samples collected in a historical medical cold chain logistics transportation process are obtained, and a corresponding historical full-process state tracking feature tensor true value and a historical abnormal state type identification true value are labeled for each historical original sensor tracking data sample.

[0133] In the model training phase, a large number of historical original sensor tracking data samples are first obtained from the historical database. Each sample contains a temperature sensor parameter subsequence, a humidity sensor parameter subsequence, a vibration sensor parameter subsequence, and a geographic positioning parameter subsequence in a complete medical cold chain transportation process. The above samples cover diversified scenarios of different seasons, different transportation routes, different drug categories, and different transportation durations.

[0134] At the same time, domain experts jointly analyze and judge with post-mortem in-depth analysis and multi-dimensional data to label the corresponding full-process state tracking feature tensor true value of each historical sample under the ideal state, and the abnormal state type identification true value actually occurring in the entire transportation process.

[0135] The abnormal state type identifies a complex abnormal type including but not limited to the following: for example, a "temperature gradient anomaly coupled with high-frequency low-amplitude vibration" type, which is manifested as a continuous and nonlinear upward trend in temperature within a certain time window, while the vibration sensor captures persistent micro-vibration in a certain frequency range, which often indicates that the refrigeration unit of the refrigerated vehicle is gradually failing; for another example, a "path deviation coupled with periodic temperature fluctuations" type, which is manifested as the transport carrier deviating from the planned path, and the temperature presents periodic fluctuations during the deviation, and the fluctuation amplitude exceeds the normal threshold, which may mean that the transport carrier stays in the non-authorized area for a long time and repeatedly opens and closes the compartment door; for another example, a "humidity oversaturation coupled with impact vibration" type, which is manifested as the humidity parameter continuously rises and reaches saturation state, while multiple transient high-amplitude impact signals appear in the vibration parameter, which often indicates that the compartment sealing is damaged and encounters serious road bumps, which may cause the medicine package to be wet and damaged. The abnormal state type identifier of each historical sample is selected from a complex composite abnormal category system.

[0136] In step S220, a historical transport carrier running trajectory sample is generated according to the historical geolocation parameter subsequence in each historical raw sensor tracking data sample, and a historical multi-dimensional environmental state marker set sample associated with each historical geolocation coordinate point in the historical transport carrier running trajectory sample is generated according to the historical temperature sensor parameter subsequence, the historical humidity sensor parameter subsequence, and the historical vibration sensor parameter subsequence in each historical raw sensor tracking data sample.

[0137] For each historical sample, the complete process of steps S121 to S137 is repeatedly executed. Specifically, the historical geolocation coordinate point sequence is extracted from the historical geolocation parameter subsequence, and after the redundant points are removed, the historical effective geolocation coordinate point sequence is obtained, and then the motion state change description vector is calculated and spliced to generate the historical transport carrier running trajectory sample. At the same time, the temperature instantaneous sampling value, the humidity instantaneous sampling value and the vibration instantaneous sampling value with the same timestamp as each historical effective geolocation coordinate point are extracted from the historical temperature sensor parameter subsequence, the historical humidity sensor parameter subsequence and the historical vibration sensor parameter subsequence.

[0138] Then, the instantaneous sampling values are converted into corresponding historical temperature state labels, historical humidity state labels and historical vibration intensity level labels according to a preset complex state division rule in steps S134 to S137. The state division rule is not a simple threshold judgment, but a complex mapping based on statistical distribution and dynamic benchmark. For example, the temperature state label includes not only basic categories such as “low temperature deviation”, “normal”, “mild over-temperature” and “severe over-temperature”, but also categories based on change trend such as “temperature rapid rise”, “temperature slow rise” and “temperature periodic fluctuation”; the humidity state label includes categories such as “dry”, “normal”, “humid” and “saturated condensation”; and the vibration intensity level label is refined into multiple levels such as “smooth”, “slight jolt”, “moderate impact”, “severe impact” and “high frequency resonance”. Finally, the above labels are combined and packaged in time sequence to generate a historical multi-dimensional environmental state label set sample associated with each historical geographic location coordinate point in the historical transportation carrier running trajectory line sample.

[0139] Step S230, constructing a historical logistics transfer node topology connection network sample passed by each of the historical transportation carrier running trajectory line samples, the historical logistics transfer node topology connection network sample including a plurality of historical logistics transfer node units and historical directed transfer path edges connecting the historical logistics transfer node units.

[0140] For each historical sample, the complete process of steps S141 to S146 is repeatedly executed. First, all trajectory points marked as historical logistics node proximity points are extracted from the historical transportation carrier running trajectory line sample, and clustered according to the historical node identifier associated with each proximity point to obtain a historical node trajectory point set corresponding to each historical logistics transfer node. Then, the mean value of the geographic location coordinates of all trajectory points in each set is calculated to generate the actual passing location coordinates of each historical logistics transfer node in this historical transportation. Next, according to the appearance order of these actual passing location coordinates on the time axis, a historical logistics transfer node sequence actually passed by the historical transportation is generated. For each pair of adjacent nodes in the sequence, the spatial straight-line distance between their actual passing location coordinates is calculated as a historical path distance parameter, and a directed historical transfer path edge connecting the front and rear nodes is defined based on this, with the historical path distance parameter attached as an attribute of the edge. Finally, all historical logistics transfer nodes are taken as network nodes, and all historical directed transfer path edges are taken as directed edges connecting the nodes, to construct a historical logistics transfer node topology connection network sample corresponding to each historical sample. The network sample not only reflects the node sequence passed in the historical transportation process, but also contains spatial distance information between nodes, and possibly implicit node type, node size and other attribute information.

[0141] Step S240, the historical transportation carrier running trajectory line sample, the historical multi-dimensional environment state label set sample and the historical logistics transfer node topology connection network sample are taken as inputs, and the corresponding historical full-process state tracking feature tensor true value is taken as output. The initialized spatio-temporal state feature fusion tracking model is subjected to first-stage supervised learning training, the network parameters of the spatio-temporal state feature fusion tracking model are updated, and a first intermediate state spatio-temporal state feature fusion tracking model is obtained.

[0142] The historical sample data generated in steps S220 and S230, i.e., the historical transportation carrier running trajectory line sample, the historical multi-dimensional environment state label set sample and the historical logistics transfer node topology connection network sample, are taken as input data and fed into the initialized spatio-temporal state feature fusion tracking model. The model performs forward propagation calculation according to the architecture described in steps S151 to S156: the trajectory encoding module encodes the historical trajectory line into a historical spatial trajectory embedding feature vector sequence, the state encoding module encodes the historical multi-dimensional environment state label set into a historical environment state embedding feature vector sequence, and the topology encoding module encodes the historical logistics transfer node topology connection network into a historical network topology embedding feature representation. Then, a historical primary joint feature tensor is generated through feature alignment splicing, and a historical enhanced feature tensor is generated through the cross-attention fusion module by fusing network topology information. Finally, a predicted historical full-process state tracking feature tensor is output through the time sequence convolution encoding layer. The predicted historical feature tensor is compared with the historical full-process state tracking feature tensor true value obtained in step S210, and the reconstruction error between the two is calculated, for example, using a mean square error loss function. Using the back propagation algorithm and an optimizer, for example, the Adam optimizer, all trainable parameters inside the model are updated according to the loss function value. After tens of thousands of iterations of historical samples, the predicted feature tensor output by the model becomes more and more close to the labeled true value. After this stage of training is completed, the first intermediate state spatio-temporal state feature fusion tracking model is obtained.

[0143] Step S250, the historical transportation carrier running trajectory line sample, the historical multi-dimensional environment state label set sample and the historical logistics transfer node topology connection network sample are input into the first intermediate state spatio-temporal state feature fusion tracking model to generate a predicted full-process state tracking feature tensor, and the predicted full-process state tracking feature tensor is input into a pre-set abnormal state pattern recognition classifier to obtain a predicted abnormal state type identification.

[0144] The same historical training samples are again subjected to forward propagation calculation using the first intermediate state model to generate a predicted full-range state tracking feature tensor. Then, the predicted feature tensor is input into the abnormal state pattern recognition classifier described in step S151. The architecture of the classifier is consistent with that described in step S151, including a global average pooling layer, multiple fully connected layers, and a softmax output layer. The classifier processes the input predicted feature tensor and outputs a probability distribution vector, with each element of the vector corresponding to the predicted probability of a composite abnormal class (such as "temperature gradient anomaly coupled with high-frequency low-amplitude vibration", "path deviation coupled with periodic temperature fluctuation", "humidity oversaturation coupled with impact vibration", etc.). The class with the highest probability value is selected from the probability distribution to obtain the predicted abnormal state type identifier, such as "temperature gradient anomaly coupled with high-frequency low-amplitude vibration".

[0145] Step S260, a classification loss function is constructed according to the difference between the predicted abnormal state type identifier and the historical abnormal state type identifier true value, and a reconstruction loss function is constructed according to the difference between the predicted full-range state tracking feature tensor and the historical full-range state tracking feature tensor true value.

[0146] The classification loss function is calculated, and a cross-entropy loss function is used to measure the difference between the predicted composite abnormal class probability distribution and the real abnormal class identifier annotated by the historical sample. Specifically, the historical abnormal state type identifier true value is converted into a one-hot encoding vector, and cross-entropy calculation is performed with the probability distribution vector output by the classifier to obtain the classification loss value L cls. At the same time, the reconstruction loss between the predicted full-range state tracking feature tensor and the historical full-range state tracking feature tensor true value is calculated again, for example, using the mean square error loss to obtain the reconstruction loss value L rec. The two loss functions measure the quality of the model output from different angles.

[0147] Step S270, the classification loss function and the reconstruction loss function are combined by weighting, a joint optimization objective function is generated, and a back propagation algorithm and an optimizer are used to perform second-stage joint optimization training on the network parameters of the first intermediate state spatio-temporal state feature fusion tracking model until the joint optimization objective function converges, obtaining the trained spatio-temporal state feature fusion tracking model.

[0148] The classification loss L cls and the reconstruction loss L rec are weighted and summed according to a preset weight coefficient to generate a joint optimization objective function L total, which is equal to a multiplied by L rec plus β multiplied by L cls. The values of α and β are dynamically adjusted according to the actual training effect, for example, in the initial stage, α can be set to 0.9 and β can be set to 0.1, so that the model prioritizes learning feature reconstruction; as the training progresses, the proportion of α and β is gradually adjusted, for example, finally adjusted to α is 0.6 and β is 0.4, so that the model takes into account both the accuracy of feature representation and the adaptability to the downstream anomaly classification task. Using the joint objective function, the entire spatio-temporal state feature fusion tracking model (including the bottom layer network layer that may be shared with the classifier) is subjected to back propagation and parameter updating. After multiple rounds of iterative optimization, when the joint objective function value tends to be stable and no longer decreases significantly, or when the preset maximum number of iterations is reached, the training process is terminated, and the trained spatio-temporal state feature fusion tracking model is finally obtained. The model has the ability to extract deep fusion features from complex multi-source sensor data.

[0149] In an exemplary embodiment, a medical cold-chain logistics transportation whole-process tracking system can be provided, which can be a terminal, a server, etc., and its internal structure diagram can be as shown in Figure 2 The medical cold-chain logistics transportation whole-process tracking system includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. The processor is used to provide computing and control capabilities. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface is used to exchange information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals, and wireless communication can be achieved through WIFI, mobile cellular network, near field communication, or other technologies. The computer program is executed by the processor to implement a medical cold-chain logistics transportation whole-process tracking method. The display unit is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device can be a touch layer overlaid on the display screen, or buttons, trackballs, or touchpads provided on the shell of the medical cold-chain logistics transportation whole-process tracking system, or an external keyboard, touchpad, or mouse, etc.

[0150] It should be noted that the foregoing description of embodiments of the application has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the application to the precise form described, and many modifications, variations, and alternatives are possible.

Claims

1. A method for end-to-end tracking of pharmaceutical cold chain logistics transportation, characterized in that, The method includes: The raw sensor tracking data stream is obtained from the sensor cluster deployed inside the transport vehicle during the cold chain logistics transportation of pharmaceuticals within a continuous acquisition time window. The raw sensor tracking data stream includes a temperature sensor parameter subsequence, a humidity sensor parameter subsequence, a vibration sensor parameter subsequence, and a geolocation parameter subsequence marked with acquisition timestamps. The transport vehicle's trajectory is generated based on the movement order of the geographic location coordinates in the geographic location parameter subsequence, and a multi-dimensional environmental state marker set associated with each geographic location coordinate in the transport vehicle's trajectory is generated based on the physical parameters in the temperature sensor parameter subsequence, the humidity sensor parameter subsequence, and the vibration sensor parameter subsequence that have the same timestamp as each geographic location coordinate. Construct a logistics transfer node topology connection network along the trajectory of the transport vehicle. The logistics transfer node topology connection network includes multiple logistics transfer node units and directed transfer path edges connecting the logistics transfer node units. The logistics transfer node units are obtained by matching the geographical coordinates of the transport vehicle's trajectory with a preset logistics transfer node database. The transport vehicle's trajectory line, the multi-dimensional environmental state marker set, and the logistics transfer node topology connection network are input into a pre-constructed spatiotemporal state feature fusion tracking model for joint analysis and processing to generate a pharmaceutical cold chain logistics transportation full-process state tracking feature tensor. Anomaly pattern matching processing is performed on the feature tensor of the entire state tracking of pharmaceutical cold chain logistics transportation to obtain the anomaly type identifier and spatiotemporal distribution characteristics of the anomaly state corresponding to the feature tensor of the entire state tracking of pharmaceutical cold chain logistics transportation. Anomaly warning instruction set for pharmaceutical cold chain logistics transportation is generated based on the anomaly type identifier and the spatiotemporal distribution characteristics of the anomaly state.

2. The method for full-process tracking of pharmaceutical cold chain logistics transportation according to claim 1, characterized in that, The step of generating the transport vehicle trajectory line based on the movement order of geographical coordinate points in the geographic positioning parameter subsequence includes: Multiple geographic location coordinate points arranged in chronological order of collection time are extracted from the geographic location parameter subsequence, and redundant points are removed from the geographic location coordinate points to obtain an effective geographic location coordinate point sequence for constructing the operating trajectory of the transport vehicle. The redundant point removal process is achieved by calculating the displacement distance parameter between adjacent geographic location coordinate points and filtering out geographic location coordinate points whose displacement distance parameter is less than a preset static discrimination threshold. Based on the acquisition timestamp corresponding to each valid geographic location coordinate point in the valid geographic location coordinate point sequence, calculate the time interval parameter and displacement direction angle parameter between adjacent valid geographic location coordinate points; normalize the time interval parameter and displacement direction angle parameter respectively, and generate a motion state change description vector between adjacent valid geographic location coordinate points based on the normalized time interval parameter and displacement direction angle parameter. The motion state change description vector includes a standardized motion duration component and a standardized motion direction change component. The effective geographic location coordinate point sequence is associated and combined with the motion state change description vector between the adjacent effective geographic location coordinate points to generate an initial trajectory segment set with temporal continuity. The initial trajectory segment set consists of consecutive effective geographic location coordinate points in the effective geographic location coordinate point sequence and their corresponding motion state change description vectors. The initial trajectory segment set is subjected to trajectory segment splicing processing. Based on the matching degree of the geographical coordinates of the endpoints of adjacent trajectory segments in the initial trajectory segment set and the continuity parameter calculated based on the motion state change description vector, the adjacent trajectory segments that meet the preset splicing conditions are connected to generate the transport vehicle running trajectory line. For each trajectory point in the operation trajectory line of the transport vehicle, logistics node proximity marking processing is performed. The spatial distance parameter between each trajectory point and all logistics transfer nodes in the preset logistics transfer node database is calculated. The trajectory points whose spatial distance parameter is less than the preset proximity distance threshold are marked as logistics node proximity points. At the same time, the identifier of the logistics transfer node closest to the trajectory point is recorded as the node association identifier. Based on the distribution density and order of the logistics node proximity points on the transport vehicle's trajectory line, the sequence of logistics transfer nodes along the transport vehicle's trajectory line is determined. The sequence of logistics transfer nodes is composed of logistics transfer node identifiers arranged sequentially according to the travel direction of the transport vehicle's trajectory line.

3. The method for full-process tracking of pharmaceutical cold chain logistics transportation according to claim 1, characterized in that, The step of generating a multi-dimensional environmental state marker set associated with each geographical coordinate point in the transport vehicle's trajectory line based on physical parameters in the temperature sensing parameter subsequence, humidity sensing parameter subsequence, and vibration sensing parameter subsequence that have the same timestamp as each geographical coordinate point includes: Extract instantaneous temperature sampling values ​​that are the same as the collection timestamp of each geographical coordinate point from the temperature sensing parameter subsequence, and arrange the instantaneous temperature sampling values ​​according to the order in which the geographical coordinate points appear on the transport vehicle's running trajectory to generate a time-series association list of temperature parameters corresponding to the transport vehicle's running trajectory. Extract the instantaneous humidity sampling value that is the same as the collection timestamp of each of the geographical coordinate points from the humidity sensing parameter subsequence, and arrange the instantaneous humidity sampling values ​​according to the order in which the geographical coordinate points appear on the transport vehicle's running trajectory to generate a humidity parameter time sequence association list corresponding to the transport vehicle's running trajectory. Extract the instantaneous vibration sampling value that is the same as the acquisition timestamp of each of the geographical coordinate points from the vibration sensing parameter subsequence, and arrange the instantaneous vibration sampling values ​​according to the order in which the geographical coordinate points appear on the transport vehicle's running trajectory to generate a vibration parameter time sequence association list corresponding to the transport vehicle's running trajectory. Each instantaneous temperature sample value in the time-series association list of temperature parameters is subjected to temperature state interval mapping processing. According to the preset temperature state division rules, each instantaneous temperature sample value is converted into a corresponding temperature state label. The temperature state label is used to characterize the degree of deviation of the internal ambient temperature of the transport carrier from the required temperature range for drug storage. Each instantaneous humidity sample value in the humidity parameter time-series association list is subjected to humidity state interval mapping processing. Each instantaneous humidity sample value is converted into a corresponding humidity state label according to the preset humidity state division rules. The humidity state label is used to characterize the degree of deviation of the humidity inside the transport carrier from the humidity range required for drug storage. Each instantaneous vibration sample value in the vibration parameter time-series association list is subjected to vibration intensity level mapping processing. According to the preset vibration intensity level classification rules, each instantaneous vibration sample value is converted into a corresponding vibration intensity level label. The vibration intensity level label is used to characterize the level of mechanical impact intensity that the transport vehicle is subjected to during transportation. The temperature status marker, humidity status marker, and vibration intensity level marker are combined and encapsulated according to their association with the geographical coordinate points to generate a multi-dimensional environmental status marker set corresponding to each geographical coordinate point in the trajectory line of the transport vehicle.

4. The method for full-process tracking of pharmaceutical cold chain logistics transportation according to claim 1, characterized in that, The construction of the logistics transfer node topology connection network along the trajectory of the transport vehicle includes: Extract all trajectory points marked as logistics node proximity points from the operation trajectory line of the transport vehicle, and cluster the logistics node proximity points belonging to the same logistics transfer node identifier according to the node association identifier corresponding to each logistics node proximity point to obtain the node trajectory point set corresponding to each logistics transfer node identifier. Calculate the average of the geographical coordinates of all neighboring points of the logistics node in each set of node trajectory points, and generate the actual passing position coordinates of the logistics transfer node unit represented by each logistics transfer node identifier in the running trajectory line of the transport vehicle; Based on the time sequence in which each logistics transfer node unit is passed on the trajectory line of the transport vehicle, the logistics transfer node units are arranged in chronological order to generate a sequence of logistics transfer nodes actually passed by the transport vehicle. For two adjacent logistics transfer node units in the logistics transfer node sequence, the spatial straight-line distance between the two logistics transfer node units is calculated as the path distance parameter based on the actual passing position coordinates corresponding to the previous logistics transfer node unit and the actual passing position coordinates corresponding to the next logistics transfer node unit. Based on the sequential order of adjacent logistics transfer node units in the logistics transfer node sequence, the directed transfer path edge connecting the previous logistics transfer node unit to the next logistics transfer node unit is determined, and the path distance parameter is appended to the directed transfer path edge as edge attribute information. All the logistics transfer node units are treated as network nodes, all the directed transfer path edges are treated as directed edges connecting the network nodes, and the topology connection network of the logistics transfer nodes is constructed by combining the edge attribute information.

5. The method for full-process tracking of pharmaceutical cold chain logistics transportation according to claim 1, characterized in that, The process of inputting the transport vehicle's trajectory line, the multi-dimensional environmental state marker set, and the logistics transfer node topology network into a pre-constructed spatiotemporal state feature fusion tracking model for joint analysis and processing generates a pharmaceutical cold chain logistics transportation end-to-end state tracking feature tensor, including: The transport vehicle's trajectory line is input into the trajectory encoding module of the spatiotemporal state feature fusion tracking model. The sequence of geographical location coordinates in the transport vehicle's trajectory line is processed by spatial location embedding representation to generate a spatial trajectory embedding feature vector sequence corresponding to the transport vehicle's trajectory line. The multi-dimensional environmental state label set is input into the state encoding module of the spatiotemporal state feature fusion tracking model. The temperature state label, humidity state label and vibration intensity level label corresponding to each geographic location coordinate point in the multi-dimensional environmental state label set are jointly embedded and represented to generate an environmental state embedding feature vector corresponding to each geographic location coordinate point. All environmental state embedding feature vectors are arranged in the order of appearance of the geographic location coordinate points to generate an environmental state embedding feature vector sequence. The logistics transfer node topology connection network is input into the topology coding module of the spatiotemporal state feature fusion tracking model. The logistics transfer node units and directed transfer path edges in the logistics transfer node topology connection network are processed by graph structure embedding representation to generate the network topology embedding feature representation corresponding to the logistics transfer node topology connection network. The network topology embedding feature representation includes the node embedding feature vector of each logistics transfer node unit and the edge embedding feature vector of each directed transfer path edge. The spatial trajectory embedded feature vector sequence and the environmental state embedded feature vector sequence are aligned and spliced ​​on the time axis to generate a primary joint feature tensor that integrates spatiotemporal information. Each time step of the primary joint feature tensor contains both spatial location information and environmental state information at the corresponding time point. The primary joint feature tensor and the network topology embedding feature representation are input into the cross-attention fusion module of the spatiotemporal state feature fusion tracking model. The cross-attention fusion module uses the primary joint feature tensor as the query basis and the network topology embedding feature representation as the key value basis. It calculates the association weight between the features of each time step in the primary joint feature tensor and each node and edge in the logistics transfer node topology connection network through a multi-head cross-attention mechanism. Based on the association weight, it performs weighted aggregation on the network topology embedding feature representation to generate an enhanced feature tensor with network topology context information. The enhanced feature tensor is input into the temporal convolutional coding layer of the spatiotemporal state feature fusion tracking model. The temporal convolutional coding layer performs multi-scale temporal receptive field feature extraction processing on the enhanced feature tensor through multiple temporal convolutional kernels with different dilation rates to generate the full-process state tracking feature tensor of the pharmaceutical cold chain logistics transportation.

6. The method for full-process tracking of pharmaceutical cold chain logistics transportation according to claim 5, characterized in that, The step of aligning and concatenating the spatial trajectory embedded feature vector sequence and the environmental state embedded feature vector sequence on the time axis to generate a primary joint feature tensor that fuses spatiotemporal information includes: Obtain the original acquisition timestamp corresponding to each spatial trajectory embedding feature vector in the spatial trajectory embedding feature vector sequence, and obtain the original acquisition timestamp corresponding to each environmental state embedding feature vector in the environmental state embedding feature vector sequence; Based on the original acquisition timestamp, the spatial trajectory embedded feature vector sequence and the environmental state embedded feature vector sequence are aligned and verified on the time axis to confirm that the spatial trajectory embedded feature vector sequence and the environmental state embedded feature vector sequence have a one-to-one correspondence in the time dimension. The spatial trajectory embedding feature vector and the environmental state embedding feature vector at each time point after alignment are concatenated and combined in the feature channel dimension to generate a spatiotemporal combination feature vector corresponding to each time point. The spatiotemporal combined feature vectors corresponding to all time points are stacked in chronological order to form a two-dimensional feature matrix as the primary joint feature tensor. The row dimension of the two-dimensional feature matrix corresponds to the time step, and the column dimension corresponds to the total number of feature channels after concatenation. The primary joint feature tensor is subjected to feature dimension normalization processing, and the numerical range of each feature channel in the primary joint feature tensor is adjusted to the preset model input range to generate a normalized primary joint feature tensor with a unified scale representation.

7. The method for full-process tracking of pharmaceutical cold chain logistics transportation according to claim 5, characterized in that, The step of inputting the primary joint feature tensor and the network topology embedding feature representation into the cross-attention fusion module of the spatiotemporal state feature fusion tracking model to generate an enhanced feature tensor with network topology context information includes: The primary joint feature tensor is linearly mapped to the query matrix required by the multi-head cross-attention mechanism. The number of rows in the query matrix corresponds to the number of time steps of the primary joint feature tensor, and the number of columns in the query matrix corresponds to the preset query feature dimension. All node embedding feature vectors and edge embedding feature vectors contained in the network topology embedding feature representation are integrated and linearly mapped to the key matrix and value matrix required by the multi-head cross-attention mechanism. The number of rows in the key matrix and the value matrix corresponds to the total number of node embedding feature vectors and edge embedding feature vectors, and the number of columns in the key matrix and the value matrix corresponds to the preset key feature dimension and value feature dimension, respectively. Perform a dot product operation between the query matrix and the key matrix to obtain an attention score matrix. Each element in the attention score matrix represents the original score of the association strength between a feature at a time step in the primary joint feature tensor and a node or an edge in the network topology embedding feature representation. The attention score matrix is ​​scaled and normalized using softmax to generate an attention weight matrix. The sum of the elements in each row of the attention weight matrix is ​​one, which is used to represent the distribution of the degree of attention of the features at each time step to all nodes and edges. The attention weight matrix and the value matrix are weighted and summed to obtain a weighted network topology context feature matrix. The number of rows in the weighted network topology context feature matrix is ​​the same as the number of time steps in the primary joint feature tensor, and the number of columns is the same as the dimension of the value feature. The primary joint feature tensor and the weighted network topology context feature matrix are concatenated and fused along the feature channel dimension to generate the enhanced feature tensor with network topology context information.

8. The method for full-process tracking of pharmaceutical cold chain logistics transportation according to claim 1, characterized in that, The abnormal state pattern matching processing of the feature tensor for the entire state tracking of pharmaceutical cold chain logistics transportation is performed to obtain the abnormal state type identifier and the spatiotemporal distribution characteristics of the abnormal state corresponding to the feature tensor for the entire state tracking of pharmaceutical cold chain logistics transportation, including: The pharmaceutical cold chain logistics transportation status tracking feature tensor is input into a pre-trained abnormal state pattern recognition classifier. The abnormal state pattern recognition classifier performs global feature extraction and pattern classification processing on the pharmaceutical cold chain logistics transportation status tracking feature tensor and outputs the abnormal state type identifier to which the pharmaceutical cold chain logistics transportation status tracking feature tensor belongs. Simultaneously, the feature tensor of the entire state tracking of the pharmaceutical cold chain logistics transportation is input into the pre-trained anomaly spatiotemporal localization network. The anomaly spatiotemporal localization network performs time-step feature analysis on the feature tensor of the entire state tracking of the pharmaceutical cold chain logistics transportation to generate an anomaly confidence score sequence corresponding to each time step. Based on the abnormal confidence score of each time step in the abnormal confidence score sequence, the continuous time step intervals in which the abnormal confidence score exceeds the preset abnormal judgment threshold are selected, and the time range corresponding to the continuous time step intervals on the operating trajectory line of the transport vehicle is determined as the time period of abnormal state occurrence. Based on the time period of the abnormal state occurrence, the trajectory segment corresponding to the time period of the abnormal state occurrence is extracted from the operating trajectory line of the transport vehicle as the abnormal trajectory segment, and the environmental state marker subset corresponding to the time period of the abnormal state occurrence is extracted from the multi-dimensional environmental state marker set as the abnormal environmental state marker subset. The geographical area covered by the abnormal state during the time period of the abnormal state is calculated based on the sequence of geographical coordinate points in the abnormal trajectory segment, and a description of the abnormal change trend of the environmental state during the time period of the abnormal state is generated based on the distribution of each state marker in the subset of abnormal environmental state markers. The time period of the abnormal state, the abnormal trajectory segment, the subset of abnormal environmental state markers, the geographical area covered by the abnormal state, and the description of the abnormal change trend are combined and encapsulated to generate the spatiotemporal distribution characteristics of the abnormal state.

9. The method for full-process tracking of pharmaceutical cold chain logistics transportation according to claim 1, characterized in that, The step of generating a set of early warning instructions for pharmaceutical cold chain logistics transportation anomalies based on the anomaly type identifier and the spatiotemporal distribution characteristics of the anomaly includes: The abnormal state type identifier is parsed, and the warning level code corresponding to the abnormal state type identifier is searched from the preset mapping relationship library. The warning level code is used to indicate the severity level of the abnormal state. The start and end times of the time period during which the abnormal state occurs are extracted from the spatiotemporal distribution features of the abnormal state, and the starting and ending geographical coordinates of the abnormal trajectory segments are extracted from the spatiotemporal distribution features of the abnormal state as early warning trigger location coordinates. Extract a subset of abnormal environmental state markers from the spatiotemporal distribution features of the abnormal state, and generate an environmental state abnormal evolution description vector based on the change sequence of temperature state markers, humidity state markers and vibration intensity level markers in the subset of abnormal environmental state markers during the time period of the abnormal state occurrence. The warning level code, the start time point, the end time point, the warning trigger location coordinates, and the environmental state abnormal evolution description vector are combined to generate a basic warning instruction unit for the currently detected abnormal state. Determine whether the abnormal state type identifier belongs to a preset chain reaction abnormal type set. If the abnormal state type identifier belongs to the chain reaction abnormal type set, predict the downstream logistics transfer node units that the abnormal state may affect based on the logistics transfer node topology connection network, and generate a predictive early warning instruction unit for the downstream logistics transfer node units. All the generated basic early warning instruction units and predictive early warning instruction units are summarized and sorted according to time sequence and spatial correlation to form the pharmaceutical cold chain logistics transportation anomaly early warning instruction set.

10. A pharmaceutical cold chain logistics transportation end-to-end tracking system, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the pharmaceutical cold chain logistics transportation end-to-end tracking method according to any one of claims 1 to 9 by executing the machine-executable instructions.

Images