Virtual personnel placeholder physical examination demand scheduling method based on deep learning

Abstract

This invention discloses a deep learning-based method for scheduling virtual personnel for physical examinations, comprising the following steps: collecting and standardizing operational data from physical examination institutions; inputting the input data set into an improved DA-RNN model for joint decoding processing, outputting a physical examination demand guidance matrix; generating virtual personnel based on the physical examination demand guidance matrix; mapping the virtual personnel to a digital twin scheduling environment constructed based on the input data set; generating virtual occupancy information in the digital twin scheduling environment and writing it into the physical examination department queue; generating physical examination path planning results for physical examination personnel and outputting physical examination guidance; triggering rolling adjustments based on real-time scheduling feedback data, dynamically recalculating the scheduling status, and outputting updated physical examination guidance. This invention, based on deep learning prediction and virtual personnel occupancy scheduling, achieves dynamic planning and rolling updates of physical examination paths, possessing advantages such as accurate scheduling, high resource utilization, and strong execution stability.

Description

Technical Field

[0001] This invention relates to the field of intelligent physical examination scheduling, and in particular to a method for scheduling virtual personnel occupancy physical examination needs based on deep learning. Background Technology

[0002] With the development of medical informatization and smart hospital construction, physical examination institutions are gradually introducing information systems to manage the physical examination process. Existing technologies are usually based on appointment systems, queue management systems, or simple rule-based scheduling methods to arrange the check-in order, departmental queuing, and examination routes of examinees. These technologies mostly rely on historical average service times or static queuing rules to generate physical examination guidance, which can alleviate the pressure of manual scheduling to a certain extent. However, their adaptability to fluctuations in physical examination demand, dynamic changes in departmental resources, and individual differences among examinees is limited, making it difficult to achieve refined scheduling in complex operating environments.

[0003] In actual physical examinations, frequent occurrences of arrival time discrepancies, changes in departmental service rhythms, and unexpected queue adjustments mean that existing technologies often lack effective modeling and utilization of real-time feedback information. This prevents the pre-determining of queue positions or continuous correction of generated examination guidelines, leading to extended waiting times, frequent changes in examination routes, and even resource idleness or localized congestion. Furthermore, traditional methods typically fail to link predicted examination demand with actual queue occupancy mechanisms, resulting in a lack of closed-loop feedback between scheduling decisions and execution status, making it difficult to achieve stable and sustainable dynamic physical examination scheduling.

[0004] Therefore, how to provide a deep learning-based method for scheduling virtual personnel for physical examinations is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a deep learning-based virtual personnel occupancy scheduling method for physical examinations. This invention, based on deep learning prediction and virtual personnel occupancy scheduling, achieves dynamic planning and rolling updates of physical examination paths, and has the advantages of precise scheduling, high resource utilization, and strong execution stability.

[0006] A method for scheduling virtual personnel space for physical examinations based on deep learning according to an embodiment of the present invention includes the following steps: Collect operational data from medical examination institutions and perform standardized processing to obtain an input data set; The input dataset is fed into the improved DA-RNN model for joint decoding, and the output is a physical examination demand guidance matrix covering the prediction time window; For each person who has completed the registration for a physical examination, a virtual person is generated based on the physical examination needs guidance matrix; A digital twin scheduling environment containing the queue status of each physical examination department is constructed based on the input data set, and virtual personnel are mapped into the digital twin scheduling environment; Based on the physical examination demand guidance matrix and virtual personnel attributes, virtual queuing information is generated in the digital twin scheduling environment and written into the corresponding physical examination department queue to lock the queuing position; Based on virtual occupancy information, generate physical examination route planning results for examinees and output physical examination guidance; During the physical examination, rolling adjustments are triggered based on real-time scheduling feedback data, the scheduling status is dynamically recalculated, and updated physical examination guidelines are output.

[0007] Optionally, the operational data of the physical examination institution includes physical examination resource data, physical examination layout data, historical queuing data, and physical examination personnel data. The standardization process includes: unifying the units and normalizing the values ​​of the service capacity parameters and examination duration parameters in the physical examination resource data; unifying the spatial coordinates of the department location coordinates in the physical examination layout data and establishing a path constraint index; aligning the timestamps in the historical queuing data and physical examination personnel data with time references and resampling; performing anomaly identification processing on missing or abnormal data records and removing invalid records; and encapsulating the standardized data into an input data set according to a unified data format.

[0008] Optionally, the output of the physical examination demand guidance matrix covering the prediction time window specifically includes: Extract the data of the current day's physical examination personnel from the historical queuing data and the physical examination personnel data from the input dataset, and perform time slice alignment processing on the queue arrival timestamp, queue number, completion timestamp, physical examination personnel arrival timestamp and physical examination personnel completion timestamp according to the time slice dimension to obtain the historical queuing time slice sequence and the current day's physical examination personnel time slice sequence. Extract the personnel identifier, physical examination package identifier, physical examination item combination and arrival timestamp from the time slice sequence of the physical examination personnel on the same day. Determine the constraint attributes between items based on the physical examination item combination, and determine the individual preference attributes based on the personnel identifier and the doctor identifier in the physical examination resource data. Combine the above information to construct the physical examination personnel demand sequence. Construct a physical examination resource status sequence based on physical examination resource data and historical queuing time slice sequences; Based on the improved DA-RNN model, the demand encoder in the dual encoder cross-attention bridging structure is used to encode the demand sequence of the physical examination personnel and output the demand hidden state sequence. At the same time, the resource encoder in the dual encoder cross-attention bridging structure is used to encode the physical examination resource state sequence and output the resource hidden state sequence. For each prediction time slice, a decoding state is generated. Based on the decoding state and the demand hidden state sequence, a temporal attention weight is calculated. The demand hidden state sequence is then weighted and converged using the temporal attention weight to obtain the demand temporal context representation. In the decoding phase, a cross-attention connection is established between the decoding state and the resource hidden state sequence. The cross-attention weights are calculated and the resource hidden state sequence is weighted and converged to obtain the resource-aligned context representation. The demand time context representation, resource alignment context representation, and decoding status are fused to obtain the decoding fusion representation. Based on the decoding fusion representation, the queuing priority score is calculated for each index position in the personnel dimension, department dimension, project portfolio dimension, and time slice dimension. For the same person, the queuing priority score value of each index position under the same prediction time slice is probabilistically processed to obtain the physical examination demand guidance matrix corresponding to the current prediction time slice. Based on the queuing priority probability of each index position in the physical examination demand guidance matrix, the index position corresponding to the maximum queuing priority probability is selected to determine the occupancy status vector. The placeholder state vector is injected back into the decoding stage corresponding to the prediction time slice, participating in the decoding state update and cross-attention weight calculation corresponding to the prediction time slice. After iteratively covering the prediction time window, the physical examination demand guidance matrix corresponding to each prediction time slice is collected, and the physical examination demand guidance matrix covering the prediction time window is output.

[0009] Optionally, the generation of the virtual personnel specifically includes: For each person who has completed the physical examination, the corresponding person index position is located based on the physical examination demand guidance matrix. The physical examination package identifier, physical examination item combination and arrival timestamp associated with the person index position are read. The queuing priority probability of the person index position in the department dimension, item combination dimension and time slice dimension is extracted to obtain the physical examination demand guidance data corresponding to the person. Candidate department sequences are generated based on the queuing priority probability order of physical examination demand guidance data at the department level, and candidate time slice sequences are generated based on the queuing priority probability order of the candidate department sequences at the time slice level. Based on the queuing priority probability order of the physical examination demand guidance data, a candidate item combination sequence is generated, and the physical examination item combination attributes are determined from the physical examination item combinations of the examinees according to the candidate item combination sequence. Based on the personnel identifier associated with the personnel index position in the physical examination demand guidance matrix, the doctor preference identifier is determined by combining the doctor identifier in the physical examination resource data, and the inter-item constraint attribute corresponding to the combination attribute of the physical examination items is read. The inter-item constraint attribute and the doctor preference identifier are respectively used as the inter-item constraint attribute and individual preference attribute of the virtual personnel. Arrival time attributes are determined based on arrival timestamps. The combination attributes of physical examination items, inter-item constraint attributes, arrival time attributes, individual preference attributes, candidate department sequences, and candidate time slice sequences are combined and encapsulated to generate virtual personnel. After establishing a one-to-one binding relationship between virtual personnel and corresponding physical examination personnel, they are written into the scheduling data structure.

[0010] Optionally, the mapping of the virtual personnel specifically includes: Extract physical examination resource data and physical examination layout data from the input dataset, establish a physical examination department resource mapping table based on the physical examination resource data, and establish a physical examination department space mapping table based on the physical examination layout data; A digital twin scheduling environment data structure is constructed based on the physical examination department resource mapping table and the physical examination department spatial mapping table. The digital twin scheduling environment data structure includes a set of physical examination departments, a set of department queue data structures, and reachability relationship data used to represent the access constraints between departments. Initialize the department queue data structure for each physical examination department in the digital twin scheduling environment; Virtual personnel are mapped into the digital twin scheduling environment. Based on the combined attributes of physical examination items, the set of target departments associated with the virtual personnel is determined. A mapping relationship between the virtual personnel and the set of target departments is established in the digital twin scheduling environment. The arrival time attribute of the virtual personnel is registered to complete the virtual personnel mapping process.

[0011] Optionally, locking the queuing position specifically includes: In the digital twin scheduling environment, for each virtual person, the attributes of the physical examination item combination, the constraint attributes between items, the arrival time attributes and the individual preference attributes are obtained. In the physical examination demand guidance matrix, the personnel index position corresponding to the physical examination personnel bound to the virtual personnel is located according to the personnel dimension. The queuing priority probability set of the personnel index position in the department dimension, the item combination dimension and the time slice dimension is extracted. The target department identifier is determined based on the combination attributes of the queuing priority probability set and the physical examination items. The effective timestamp of the placeholder is determined based on the arrival time attribute of the virtual personnel, and the effective timestamp of the placeholder is used as the time base to locate the insertable placeholder time slice interval in the department queue data structure corresponding to the target department identifier. In the department queue data structure corresponding to the target department identifier, the existing queue order is matched and compared by combining the placeholder effective timestamp and the placeholder time slice interval to determine the placeholder queue number used for virtual placeholder, and the expected service end time corresponding to the placeholder queue number is determined as the placeholder invalidation timestamp. The target department identifier, the queue number, the timestamp of the queue taking effect, and the timestamp of the queue taking effect are encapsulated in a structured manner to generate virtual queue information. The virtual queue information is then written into the department queue data structure corresponding to the target department identifier to lock the queue position.

[0012] Optionally, the output of the physical examination guidelines specifically includes: In the digital twin scheduling environment, a virtual person is obtained for each physical examination person and bound to the physical examination person. The combination attributes of physical examination items, the constraint attributes between items and the arrival time attributes of the virtual person are read. At the same time, the virtual occupancy information of the virtual person in each target department is read. The set of physical examination items to be executed is obtained by parsing the combination attributes of the physical examination items, and the execution order of the physical examination items is obtained by sorting the execution order of the set of physical examination items according to the order constraint relationship of the items in the constraint attributes between the items. Based on the execution order of the physical examination items and the target department identifiers corresponding to each physical examination item, a sequence of physical examination departments is generated. Then, based on the department accessibility data in the digital twin scheduling environment, access constraint verification processing is performed on the sequence of physical examination departments to obtain a sequence of physical examination departments that meets the access path constraints. The arrival timestamp corresponding to the arrival time attribute is used as the expected start time of the first item. Combined with the effective timestamp and queue number of the placeholder corresponding to each target department in the physical examination department sequence, the expected start time and expected end time of each physical examination item in the execution order of the physical examination items are calculated in turn. The physical examination process includes the order of examination items, the sequence of examination departments, and the estimated start and end times of each examination item. These are then packaged into a physical examination pathway planning result, which is output as a physical examination guide.

[0013] Optionally, the output of the updated physical examination guidelines specifically includes: Real-time scheduling feedback data is collected during the physical examination process, and the real-time scheduling feedback data is standardized and packaged according to timestamps to form a feedback data stream. The feedback data stream is parsed to obtain the target physical examination department identifier, event type identifier, and event occurrence timestamp. Based on the event type identifier, the corresponding mapping process is triggered to generate service time deviation trigger identifier, queue change trigger identifier, and arrival change trigger identifier. When a service time deviation trigger flag is detected, the department queue data structure corresponding to the target physical examination department flag is located in the digital twin scheduling environment. The actual service start time and actual service end time are written to update the queue status. Based on the actual service start time and actual service end time, deviation correction processing is performed on the expected service start time and expected service end time to obtain the updated expected service start time and expected service end time. When a queue change trigger flag is detected, the queue number set and the department queue length are updated in the department queue data structure. Based on the updated queue number set, a consistency check is performed on the placeholder queue number to obtain the consistency check result. When an arrival change trigger flag is detected, the arrival time attribute is updated in the virtual personnel data structure, the occupancy effective timestamp is updated in the virtual occupancy information, and the estimated start time of the first item is updated in the physical examination route planning results. When the consistency check result indicates that the queue number is inconsistent, the virtual queue information is updated based on the consistency check result, and the updated virtual queue information is written back to the queue data structure corresponding to the target physical examination department identifier to re-lock the queue position. Based on the updated digital twin scheduling environment and the updated virtual occupancy information, the physical examination route planning results are rolled over to obtain the updated physical examination route planning results, and the updated physical examination guidance is output.

[0014] The beneficial effects of this invention are: This invention constructs a deep learning-based mechanism for predicting physical examination needs and scheduling virtual personnel, achieving effective linkage between the predicted results and the execution process of physical examinations. By introducing an improved DA-RNN model, the system separately models and jointly decodes the sequence of physical examination personnel's needs and the sequence of physical examination resource states. This enables the output of a physical examination needs guidance matrix within the predicted time window, allowing the scheduling system to have a holistic understanding of the multi-dimensional relationships between personnel, departments, project combinations, and time slices before the physical examinations are executed. The virtual personnel generated on this basis not only characterize the project combinations, constraint attributes, and individual preferences of the physical examination personnel, but also map the predicted results into executable occupancy decisions, providing a stable and quantifiable decision-making basis for subsequent scheduling.

[0015] Based on the mapping of virtual personnel in a digital twin scheduling environment and the writing of virtual queuing information, this invention achieves pre-locking of queuing positions before the physical examination is performed. This ensures that the physical examination route planning results no longer rely on static rules or post-event adjustments, but are based on the synergy of predicted demand and real-time queue status. By introducing queuing effective timestamps and queuing ineffective timestamps into the departmental queue data structure, the time range of queuing can be effectively constrained, avoiding invalid queuing or long-term resource occupation, thereby improving the overall utilization rate of physical examination resources and reducing waiting time and repeated route adjustments for examinees during the actual execution process.

[0016] During the physical examination process, this invention further enhances the accuracy of scheduling status changes by continuously collecting and analyzing real-time scheduling feedback data, introducing service time deviation trigger flags, queue change trigger flags, and arrival change trigger flags. When a trigger condition is detected, the system can perform rolling updates on queue status, occupancy information, and physical examination path planning results within the digital twin scheduling environment, forming a closed-loop scheduling mechanism that integrates prediction, occupancy, execution, and feedback. Through this mechanism, physical examination guidance can dynamically evolve with the operational status, significantly improving the stability and continuity of scheduling results, reducing the risk of scheduling failures due to sudden changes, and thus achieving a more efficient and reliable scheduling effect for physical examination needs. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a deep learning-based virtual personnel occupancy physical examination demand scheduling method proposed in this invention; Figure 2 This is a flowchart illustrating the process of generating a physical examination demand guidance matrix for a deep learning-based virtual personnel occupancy physical examination demand scheduling method proposed in this invention. Figure 3 This is a flowchart illustrating the queuing position locking process of a deep learning-based virtual personnel occupancy physical examination demand scheduling method proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-3 A deep learning-based method for scheduling virtual personnel for physical examinations includes the following steps: Collect operational data from medical examination institutions and perform standardized processing to obtain an input data set; The input dataset is fed into the improved DA-RNN model for joint decoding, and the output is a physical examination demand guidance matrix covering the prediction time window; For each person who has completed the registration for a physical examination, a virtual person is generated based on the physical examination needs guidance matrix; A digital twin scheduling environment containing the queue status of each physical examination department is constructed based on the input data set, and virtual personnel are mapped into the digital twin scheduling environment; Based on the physical examination demand guidance matrix and virtual personnel attributes, virtual queuing information is generated in the digital twin scheduling environment and written into the corresponding physical examination department queue to lock the queuing position; Based on virtual occupancy information, generate physical examination route planning results for examinees and output physical examination guidance; During the physical examination, rolling adjustments are triggered based on real-time scheduling feedback data, the scheduling status is dynamically recalculated, and updated physical examination guidelines are output.

[0020] In this embodiment, the operational data of the medical examination institution includes medical examination resource data, medical examination layout data, historical queuing data, and medical examination personnel data. The medical examination resource data includes the clinic identifiers, equipment identifiers, doctor identifiers, service capacity parameters, and examination duration parameters for each medical examination department. The medical examination layout data includes floor identifiers, department location coordinates, and path constraints. The historical queuing data includes queue arrival timestamps, queue numbers, and completion timestamps. The medical examination personnel data includes personnel identifiers, medical examination package identifiers, medical examination item combinations, arrival timestamps, and completion timestamps. The standardization process includes: unifying the units and normalizing the values ​​of the service capacity parameters and examination duration parameters in the medical examination resource data; unifying the spatial coordinates of the department location coordinates in the medical examination layout data and establishing a path constraint index; aligning the timestamps in the historical queuing data and medical examination personnel data with a time reference and resampling; performing anomaly identification processing on missing or abnormal data records and removing invalid records; and encapsulating the standardized data into an input data set according to a unified data format.

[0021] In this embodiment, the output of the physical examination demand guidance matrix covering the prediction time window specifically includes: Extract the data of the current day's physical examination personnel from the historical queuing data and the physical examination personnel data from the input dataset, and perform time slice alignment processing on the queue arrival timestamp, queue number, completion timestamp, physical examination personnel arrival timestamp and physical examination personnel completion timestamp according to the time slice dimension to obtain the historical queuing time slice sequence and the current day's physical examination personnel time slice sequence. Extract the personnel identifier, physical examination package identifier, physical examination item combination and arrival timestamp from the time slice sequence of physical examination personnel on the same day. Determine the constraint attributes between items based on the physical examination item combination, and determine the individual preference attributes based on the personnel identifier and the doctor identifier in the physical examination resource data. Combine the above information to construct the physical examination personnel demand sequence. The physical examination personnel demand sequence includes personnel identifier, physical examination package identifier, physical examination item combination, arrival timestamp, constraint attributes between items and individual preference attributes in each time slice. A physical examination resource status sequence is constructed based on physical examination resource data and historical queuing time slice sequences. The physical examination resource status sequence includes clinic identifier, equipment identifier, doctor identifier, service capacity parameters, examination duration parameters, and queue arrival rhythm parameters and queue completion rhythm parameters obtained from historical queuing data in each time slice. Based on the improved DA-RNN model, the demand encoder in the dual encoder cross-attention bridging structure is used to encode the demand sequence of the physical examination personnel and output the demand hidden state sequence. At the same time, the resource encoder in the dual encoder cross-attention bridging structure is used to encode the physical examination resource state sequence and output the resource hidden state sequence. The improvement of the DA-RNN model is that the original DA-RNN model uses a single encoder and a one-way decoding structure to perform time series modeling of the input time series, while the improved model introduces a dual encoder cross-attention bridging structure to separate the modeling of demand information and resource state information and perform joint modeling in the decoding stage. At the same time, a placeholder state back-injection structure is introduced to inject the placeholder state formed by the prediction results back into the decoding stage to participate in the decoding state update and attention calculation. The output of the demand hidden state sequence and the resource hidden state sequence includes: inputting the corresponding medical examinee demand sequence and medical examinee resource state sequence into the corresponding encoder one by one according to the time slice order; performing feature mapping and vectorization processing on the data corresponding to the current time slice within each time slice to form the input vector of the current time slice; inputting the input vector and the hidden state corresponding to the previous time slice into the recurrent neural network computing unit of the corresponding encoder; performing linear transformation and nonlinear transformation processing on the input vector and the hidden state of the previous time slice respectively; and calculating the candidate state representation of the current time slice according to the state update rule of the recurrent neural network; performing state update and information retention processing based on the candidate state representation and the hidden state of the previous time slice to obtain the hidden state corresponding to the current time slice; repeating the above processing until the encoding of all time slices is completed, and storing and arranging the hidden states corresponding to each time slice in the order of the time slices to obtain the demand hidden state sequence and the resource hidden state sequence respectively. For each prediction time slice, a decoding state is generated. Based on the decoding state and the sequence of hidden demand states, a temporal attention weight is calculated. The temporal attention weight is then used to perform weighted aggregation on the sequence of hidden demand states to obtain a demand temporal context representation. The generation of the decoding state includes: in each prediction time slice, using a recurrent neural network structure in the decoding stage, performing state recursion update processing based on the decoding state of the previous prediction time slice to generate the decoding state corresponding to the current prediction time slice. The calculation of the temporal attention weight includes performing correlation mapping calculation between the decoding state and the hidden demand states corresponding to each time slice in the sequence of hidden demand states to obtain the attention score of each time slice and performing normalization processing to obtain the corresponding temporal attention weight. In the decoding phase, a cross-attention connection is established between the decoding state and the resource hidden state sequence. The cross-attention weights are calculated and the resource hidden state sequence is weighted and converged to obtain the resource-aligned context representation. The demand time context representation, resource alignment context representation, and decoding state are fused to obtain a decoded fusion representation. Based on the decoded fusion representation, queuing priority scores are calculated for each index position in the personnel dimension, department dimension, project portfolio dimension, and time slice dimension. The queuing priority scores are obtained by performing feature mapping and weighted calculation on the decoded fusion representation by the scoring calculation structure. For the same person in the same predicted time slice, the queuing priority score value of each index position is probabilistically processed to obtain the physical examination demand guidance matrix corresponding to the current predicted time slice. Based on the queuing priority probability of each index position in the physical examination demand guidance matrix, the index position corresponding to the maximum queuing priority probability is selected to determine the occupancy state vector. The probabilistic processing is to perform exponential mapping on each queuing priority score value and normalize it by the sum of all exponential mapping values ​​of the person in the predicted time slice to obtain the queuing priority probability of each index position and form the physical examination demand guidance matrix corresponding to the current predicted time slice. The occupancy state vector represents the corresponding department index, project combination index and time slice index. The placeholder state vector is injected back into the decoding stage corresponding to the prediction time slice, participates in the decoding state update and cross-attention weight calculation corresponding to the prediction time slice, and after iteratively covering the prediction time window, the physical examination demand guidance matrix corresponding to each prediction time slice is collected, and the physical examination demand guidance matrix covering the prediction time window is output. The participation includes: at each prediction time slice, inputting a placeholder state vector into the decoding stage, which, together with the decoding state corresponding to the current prediction time slice, serves as the input data for the decoding stage; based on the placeholder state vector and the decoding state, performing a fusion process on the internal state of the decoding stage to update the decoding input state corresponding to the current prediction time slice; using the updated decoding input state to participate in the state recursive calculation of the decoding stage to generate the decoding state update result for the current prediction time slice; simultaneously, using the placeholder state vector as the input for cross-attention calculation, together with the corresponding hidden state in the demand hidden state sequence or resource hidden state sequence, to participate in the calculation of cross-attention weights, so that the calculation process of cross-attention weights is affected by the placeholder state vector; based on the updated decoding state and the context representation weighted by the cross-attention weights, completing the decoding calculation for the current prediction time slice, and entering the processing flow of the next prediction time slice.

[0022] In this embodiment, the generation of the virtual personnel specifically includes: For each person who has completed the physical examination, the corresponding person index position is located based on the physical examination demand guidance matrix. The physical examination package identifier, physical examination item combination and arrival timestamp associated with the person index position are read. The queuing priority probability of the person index position in the department dimension, item combination dimension and time slice dimension is extracted to obtain the physical examination demand guidance data corresponding to the person. Candidate department sequences are generated based on the queuing priority probability order of physical examination demand guidance data at the department level, and candidate time slice sequences are generated based on the queuing priority probability order of the candidate department sequences at the time slice level. Based on the queuing priority probability order of the physical examination demand guidance data, a candidate item combination sequence is generated, and the physical examination item combination attributes are determined from the physical examination item combinations of the examinees according to the candidate item combination sequence. Based on the personnel identifier associated with the personnel index position in the physical examination demand guidance matrix, the doctor preference identifier is determined by combining the doctor identifier in the physical examination resource data, and the inter-item constraint attribute corresponding to the combination attribute of the physical examination items is read. The inter-item constraint attribute and the doctor preference identifier are respectively used as the inter-item constraint attribute and individual preference attribute of the virtual personnel. The inter-item constraint attribute includes the fasting constraint identifier and the item sequence constraint relationship. Arrival time attributes are determined based on arrival timestamps. The combination attributes of physical examination items, inter-item constraint attributes, arrival time attributes, individual preference attributes, candidate department sequences, and candidate time slice sequences are combined and encapsulated to generate virtual personnel. After establishing a one-to-one binding relationship between virtual personnel and corresponding physical examination personnel, they are written into the scheduling data structure.

[0023] In this embodiment, the mapping of virtual personnel specifically includes: Extract physical examination resource data and physical examination layout data from the input dataset. Establish a physical examination department resource mapping table based on the physical examination resource data and a physical examination department space mapping table based on the physical examination layout data. The physical examination department space mapping table associates floor identifiers, department location coordinates and access path constraint indexes at the floor level, and establishes department access reachability relationship data based on the access path constraint indexes. A digital twin scheduling environment data structure is constructed based on the physical examination department resource mapping table and the physical examination department spatial mapping table. The digital twin scheduling environment data structure includes a set of physical examination departments, a set of department queue data structures, and reachability relationship data used to represent the access constraints between departments. For each physical examination department in the digital twin scheduling environment, the department queue data structure is initialized. The initialization includes reading the current queue number set from the department queue data structure and calculating the department queue length. Based on the examination duration parameter and service capacity parameter, the expected service start time and expected service end time corresponding to each queue number in the queue number set are calculated sequentially to obtain the department queue length, queue number set, expected service start time and expected service end time. Virtual personnel are mapped into the digital twin scheduling environment. Based on the combined attributes of physical examination items, the set of target departments associated with the virtual personnel is determined. A mapping relationship between the virtual personnel and the set of target departments is established in the digital twin scheduling environment. The arrival time attribute of the virtual personnel is registered to complete the virtual personnel mapping process.

[0024] In this embodiment, locking the queuing position specifically includes: In the digital twin scheduling environment, for each virtual person, the attributes of the physical examination item combination, the constraint attributes between items, the arrival time attributes and the individual preference attributes are obtained. In the physical examination demand guidance matrix, the personnel index position corresponding to the physical examination personnel bound to the virtual personnel is located according to the personnel dimension. The queuing priority probability set of the personnel index position in the department dimension, the item combination dimension and the time slice dimension is extracted. The target department identifier is determined based on the queuing priority probability set and the combination attribute of the physical examination items. The determination of the target department identifier includes selecting the department index corresponding to the maximum queuing priority probability in the candidate department set associated with the combination attribute of the physical examination items and mapping it to obtain the target department identifier. The effective time stamp of the placeholder is determined based on the arrival time attribute of the virtual personnel, and the effective time stamp of the placeholder is used as the time base to locate the insertable placeholder time slice interval in the department queue data structure corresponding to the target department identifier. The determination of the effective time stamp of the placeholder includes taking the arrival time stamp corresponding to the arrival time attribute as the effective time stamp of the placeholder. In the department queue data structure corresponding to the target department identifier, the existing queue order is matched and compared by combining the placeholder effective timestamp and the placeholder time slice interval to determine the placeholder queue number used for virtual placeholder, and the expected service end time corresponding to the placeholder queue number is determined as the placeholder invalidation timestamp. The determination of the placeholder queue number includes comparing the execution order of the expected service start time and expected service end time in the queue number set to determine the insertion position that meets the placeholder effective timestamp, and setting the number corresponding to the insertion position as the placeholder queue number. The target department identifier, the queue number, the timestamp of the queue taking effect, and the timestamp of the queue taking effect are encapsulated in a structured manner to generate virtual queue information. The virtual queue information is then written into the department queue data structure corresponding to the target department identifier to lock the queue position.

[0025] In this embodiment, the output of the physical examination guidance specifically includes: In the digital twin scheduling environment, a virtual person is obtained for each physical examination person and bound to the physical examination person. The combination attributes of physical examination items, the constraint attributes between items and the arrival time attributes of the virtual person are read. At the same time, the virtual occupancy information of the virtual person in each target department is read. The set of physical examination items to be executed is obtained by parsing the combination attributes of the physical examination items, and the execution order of the physical examination items is obtained by sorting the execution order of the set of physical examination items according to the order constraint relationship of the items in the constraint attributes between the items. Based on the execution order of the physical examination items and the target department identifiers corresponding to each physical examination item, a sequence of physical examination departments is generated. Then, based on the department accessibility data in the digital twin scheduling environment, access constraint verification processing is performed on the sequence of physical examination departments to obtain a sequence of physical examination departments that meets the access path constraints. The arrival timestamp corresponding to the arrival time attribute is used as the estimated start time of the first item. Combined with the effective timestamp and queue number of the target department identifier in the physical examination department sequence, the estimated start time and estimated end time of each physical examination item in the execution order are calculated in sequence. The calculation of the estimated start time includes selecting the earliest executable time that satisfies the validity of the reservation within the effective time interval defined by the effective timestamp of the target department corresponding to the current physical examination item, the estimated end time of the previous physical examination item, and the invalid timestamp of the reservation. The effective time interval is defined by the effective timestamp of the reservation to the invalid timestamp of the reservation. The calculation of the estimated end time includes adding the estimated start time of the current physical examination item with the examination duration parameter corresponding to the current physical examination item to obtain the estimated end time of the current physical examination item. The physical examination process includes the order of examination items, the sequence of examination departments, and the estimated start and end times of each examination item. These are then packaged into a physical examination pathway planning result, which is output as a physical examination guide.

[0026] In this embodiment, the output of the updated physical examination guidelines specifically includes: Real-time scheduling feedback data is collected during the physical examination process, and the real-time scheduling feedback data is standardized and packaged according to timestamps to form a feedback data stream. The real-time scheduling feedback data includes the actual service start time, actual service end time, queue change events, and physical examination personnel arrival change events of each physical examination department. The feedback data stream is parsed to obtain the target physical examination department identifier, event type identifier, and event occurrence timestamp. Based on the event type identifier, the corresponding mapping process is triggered to generate service time deviation trigger identifier, queue change trigger identifier, and arrival change trigger identifier. The mapping of the service time deviation triggering identifier includes: reading the expected service start time and expected service end time corresponding to the target physical examination department identifier from the digital twin scheduling environment; calculating the time difference between the actual service start time and the corresponding expected service start time in the real-time scheduling feedback data; calculating the time difference between the actual service end time and the corresponding expected service end time; and generating a service time deviation triggering identifier when either time difference exceeds a preset time deviation threshold. The mapping of the queue change trigger identifier includes: parsing queue change events in real-time scheduling feedback data, identifying personnel insertion, personnel removal, or queue order change events that occur in the queue of the physical examination department, and mapping the parsed queue structure change information to queue change trigger identifiers; The mapping of the arrival change trigger identifier includes: reading the actual arrival time of the examinee from the arrival change event of the examinee, and performing time comparison processing between the actual arrival time and the arrival time attribute recorded in the virtual personnel data structure. When the actual arrival time of the examinee is inconsistent with the recorded arrival time attribute, an arrival change trigger identifier is generated. When a service time deviation trigger flag is detected, the department queue data structure corresponding to the target physical examination department flag is located in the digital twin scheduling environment. The actual service start time and actual service end time are written to update the queue status. Based on the actual service start time and actual service end time, deviation correction processing is performed on the expected service start time and expected service end time to obtain the updated expected service start time and expected service end time. When a queue change trigger flag is detected, the queue number set and the department queue length are updated in the department queue data structure. Based on the updated queue number set, a consistency check is performed on the placeholder queue number to obtain the consistency check result. When an arrival change trigger flag is detected, the arrival time attribute is updated in the virtual personnel data structure, the occupancy effective timestamp is updated in the virtual occupancy information, and the estimated start time of the first item is updated in the physical examination route planning results. When the consistency check result indicates that the queue number is inconsistent, the virtual queue information is updated based on the consistency check result, and the updated virtual queue information is written back to the department queue data structure corresponding to the target physical examination department to re-lock the queuing position. The queue update process includes recalculating and updating the queue number, the queue effective timestamp, and the queue ineffective timestamp. Based on the updated digital twin scheduling environment and updated virtual occupancy information, a rolling update process is performed on the physical examination route planning results to obtain updated physical examination route planning results and output updated physical examination guidance. The rolling update process includes recursively updating the estimated start time and estimated end time of each physical examination item according to the execution order of the physical examination items.

[0027] Example 1: To verify the feasibility of this invention in practice, it was applied to the daily health checkup scheduling scenario of a large comprehensive health checkup institution. This institution's departments span multiple floors, with a high density of examinees and significant differences in the combinations of examination items. There are also constraints on travel paths and examination sequences between different departments. In actual operation, problems frequently arise such as uneven queuing times, congestion in some departments while other departments are idle, and examinees frequently waiting or having to return. Traditional scheduling methods based on manual experience or static rules are unable to respond promptly to dynamic changes, thus affecting overall health checkup efficiency and the examinee experience.

[0028] In this scenario, after personnel check-in, the medical examination institution uniformly collects and standardizes data on medical examination resources, layout, historical queuing, and personnel, and inputs this data into an improved deep learning model. The model generates a medical examination demand guidance matrix covering the predicted time window through joint decoding, characterizing the queuing priority relationships of different personnel in different departments and time slices. Based on this guidance matrix, the system generates a unique virtual person for each checked-in personnel. This virtual person includes attributes such as the combination of medical examination items, inter-item constraints, arrival time, and individual preferences. Subsequently, a digital twin scheduling environment is constructed based on the spatial layout and resource status of the medical examination institution, and the virtual persons are mapped to the corresponding departmental queue structures.

[0029] In a digital twin scheduling environment, the system combines a physical examination demand guidance matrix with virtual personnel attributes to pre-generate virtual queuing information for virtual personnel in the target department queue. Queuing intervals are defined by the queuing activation and deactivation times, thus locking in reasonable queuing positions without affecting the real queue order. Based on this, the system further automatically generates physical examination path planning results for examinees according to the virtual queuing information, clarifying the execution order of examination items, the order of visits to corresponding departments, and the estimated time for each step. This result is provided to examinees and on-site staff in the form of examination guidance, making the examination process clearer and more controllable.

[0030] During the physical examination process, the system continuously collects information on the actual service time of each department, queue changes, and arrival changes of examinees. When service rhythm deviations, queue structure changes, or abnormal arrivals are detected, the system can promptly trigger a rolling adjustment mechanism to dynamically recalculate the digital twin scheduling environment, virtual occupancy information, and physical examination path planning results, and output updated physical examination guidance. Through this method, the invention effectively alleviates the problem of uneven queuing between departments in practical applications, reduces unnecessary waiting and back-and-forth trips for examinees, makes the physical examination process more seamless and smooth, and improves the overall utilization level of physical examination resources, demonstrating significant practical application value.

[0031] Table 1. Performance Comparison of the Invention and Traditional Site-Occupied Physical Examination Demand Scheduling Method

[0032] As can be clearly seen from Table 1, the method of the present invention is superior to the traditional method in many indicators.

[0033] In terms of average waiting time, the average waiting time for physical examinations under traditional scheduling methods is 42.6 minutes, while the time is reduced to 35.8 minutes using the method of this invention, a reduction of 6.8 minutes. This improvement mainly stems from the comprehensive modeling of queuing priority relationships in the physical examination demand guidance matrix across both the personnel and time slice dimensions. This allows physical examination personnel to participate in the queue in advance through a virtual personnel mechanism after registration, reducing the time consumption caused by disorderly arrivals and temporary queuing.

[0034] Regarding peak department queue length, traditional scheduling methods reach a peak queue length of 28 people during peak hours, while the method of this invention reduces this value to 23 people. This change indicates that by distributing virtual occupancy information across time slices, the entry time of examinees into the queue is more evenly distributed, thereby effectively alleviating the problem of concentrated queuing within a single time period and reducing the risk of instantaneous congestion in the department.

[0035] Regarding the total completion time of the physical examination, the average total completion time for examinees under the traditional scheduling method is 196 minutes, while the method of this invention reduces this time to 174 minutes. This 22-minute difference is mainly due to the fact that the physical examination route planning result takes into account the order constraints of the items, the accessibility of departments, and the effective time interval of virtual occupancy during the generation process. This reduces waiting and backtracking for examinees during the flow of multiple departments, thereby compressing the overall physical examination process time.

[0036] Regarding the resource utilization rate of departments, the traditional scheduling method achieves a utilization rate of 71.4%, while the method of this invention improves it to 78.9%. This improvement demonstrates that the digital twin scheduling environment, based on real-time reflection of queue status, dynamically corrects virtual occupancy information through a rolling adjustment mechanism, enabling departments with previously low utilization rates to achieve more reasonable personnel allocation, thereby improving overall resource utilization efficiency.

[0037] The number of times the physical examination route was reversed was reduced from 1.32 times per person using the traditional method to 0.91 times per person. This result indicates that the present invention introduces a placeholder failure timestamp constraint when generating the physical examination route planning results, ensuring that the placeholder remains valid when the examinee arrives at the target department, significantly reducing route reversals and repeated waiting caused by queue failures.

[0038] Regarding scheduling recalculation response latency, the average response latency of traditional scheduling methods is 4.8 seconds, while the method of this invention reduces it to 3.4 seconds. This is because the rolling adjustment mechanism only updates the local scheduling state corresponding to the trigger flag, without having to recalculate the entire scheduling scheme, thus significantly improving the real-time response capability of the system.

[0039] The congestion rate during peak hours decreased from 18.6% to 13.2%, while the stability fluctuation range of the scheduling scheme decreased from 16.4 minutes to 11.7 minutes. This indicates that under high load conditions, the present invention can suppress frequent and large fluctuations in scheduling results by linking the virtual occupancy information with the path planning results, making the overall scheduling process more stable and controllable.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A deep learning-based method for scheduling virtual personnel space for physical examinations, characterized in that, Includes the following steps: Collect operational data from medical examination institutions and perform standardized processing to obtain an input data set; The input dataset is fed into the improved DA-RNN model for joint decoding, and the output is a physical examination demand guidance matrix covering the prediction time window; For each person who has completed the registration for a physical examination, a virtual person is generated based on the physical examination needs guidance matrix; A digital twin scheduling environment containing the queue status of each physical examination department is constructed based on the input data set, and virtual personnel are mapped into the digital twin scheduling environment; Based on the physical examination demand guidance matrix and virtual personnel attributes, virtual queuing information is generated in the digital twin scheduling environment and written into the corresponding physical examination department queue to lock the queuing position. Based on virtual occupancy information, generate physical examination route planning results for examinees and output physical examination guidance; During the physical examination, rolling adjustments are triggered based on real-time scheduling feedback data, the scheduling status is dynamically recalculated, and updated physical examination guidelines are output.

2. The method for scheduling virtual personnel space for physical examinations based on deep learning according to claim 1, characterized in that, The operational data of the physical examination institution includes physical examination resource data, physical examination layout data, historical queuing data, and physical examination personnel data. The standardization process includes: unifying the units and normalizing the values ​​of service capacity parameters and examination duration parameters in the physical examination resource data; unifying the spatial coordinates of the department location coordinates in the physical examination layout data and establishing a path constraint index; aligning the timestamps in the historical queuing data and physical examination personnel data with time references and resampling; performing anomaly identification processing on missing or abnormal data records and removing invalid records; and encapsulating the standardized data into an input data set according to a unified data format.

3. The method for scheduling virtual personnel space for physical examinations based on deep learning according to claim 1, characterized in that, The output of the physical examination demand guidance matrix covering the predicted time window specifically includes: Extract the data of the current day's physical examination personnel from the historical queuing data and the physical examination personnel data from the input dataset, and perform time slice alignment processing on the queue arrival timestamp, queue number, completion timestamp, physical examination personnel arrival timestamp and physical examination personnel completion timestamp according to the time slice dimension to obtain the historical queuing time slice sequence and the current day's physical examination personnel time slice sequence. Extract the personnel identifier, physical examination package identifier, physical examination item combination and arrival timestamp from the time slice sequence of the physical examination personnel on the same day. Determine the constraint attributes between items based on the physical examination item combination, and determine the individual preference attributes based on the personnel identifier and the doctor identifier in the physical examination resource data. Combine the above information to construct the physical examination personnel demand sequence. Construct a physical examination resource status sequence based on physical examination resource data and historical queuing time slice sequences; Based on the improved DA-RNN model, the demand encoder in the dual encoder cross-attention bridging structure is used to encode the demand sequence of the physical examination personnel and output the demand hidden state sequence. At the same time, the resource encoder in the dual encoder cross-attention bridging structure is used to encode the physical examination resource state sequence and output the resource hidden state sequence. For each prediction time slice, a decoding state is generated. Based on the decoding state and the demand hidden state sequence, a temporal attention weight is calculated. The demand hidden state sequence is then weighted and converged using the temporal attention weight to obtain the demand temporal context representation. In the decoding phase, a cross-attention connection is established between the decoding state and the resource hidden state sequence. The cross-attention weights are calculated and the resource hidden state sequence is weighted and converged to obtain the resource-aligned context representation. The demand time context representation, resource alignment context representation, and decoding status are fused to obtain the decoding fusion representation. Based on the decoding fusion representation, the queuing priority score is calculated for each index position in the personnel dimension, department dimension, project portfolio dimension, and time slice dimension. For the same person, the queuing priority score value of each index position under the same prediction time slice is probabilistically processed to obtain the physical examination demand guidance matrix corresponding to the current prediction time slice. Based on the queuing priority probability of each index position in the physical examination demand guidance matrix, the index position corresponding to the maximum queuing priority probability is selected to determine the occupancy status vector. The placeholder state vector is injected back into the decoding stage corresponding to the prediction time slice, participating in the decoding state update and cross-attention weight calculation corresponding to the prediction time slice. After iteratively covering the prediction time window, the physical examination demand guidance matrix corresponding to each prediction time slice is collected, and the physical examination demand guidance matrix covering the prediction time window is output.

4. The method for scheduling virtual personnel space for physical examinations based on deep learning according to claim 1, characterized in that, The generation of the virtual personnel specifically includes: For each person who has completed the physical examination, the corresponding person index position is located based on the physical examination demand guidance matrix. The physical examination package identifier, physical examination item combination and arrival timestamp associated with the person index position are read. The queuing priority probability of the person index position in the department dimension, item combination dimension and time slice dimension is extracted to obtain the physical examination demand guidance data corresponding to the person. Candidate department sequences are generated based on the queuing priority probability order of physical examination demand guidance data at the department level, and candidate time slice sequences are generated based on the queuing priority probability order of the candidate department sequences at the time slice level. Based on the queuing priority probability order of the physical examination demand guidance data, a candidate item combination sequence is generated, and the physical examination item combination attributes are determined from the physical examination item combinations of the examinees according to the candidate item combination sequence. Based on the personnel identifier associated with the personnel index position in the physical examination demand guidance matrix, the doctor preference identifier is determined by combining the doctor identifier in the physical examination resource data, and the inter-item constraint attribute corresponding to the combination attribute of the physical examination items is read. The inter-item constraint attribute and the doctor preference identifier are respectively used as the inter-item constraint attribute and individual preference attribute of the virtual personnel. Arrival time attributes are determined based on arrival timestamps. The combination attributes of physical examination items, inter-item constraint attributes, arrival time attributes, individual preference attributes, candidate department sequences, and candidate time slice sequences are combined and encapsulated to generate virtual personnel. After establishing a one-to-one binding relationship between virtual personnel and corresponding physical examination personnel, they are written into the scheduling data structure.

5. The method for scheduling virtual personnel space for physical examinations based on deep learning according to claim 1, characterized in that, The mapping of the virtual personnel specifically includes: Extract physical examination resource data and physical examination layout data from the input dataset, establish a physical examination department resource mapping table based on the physical examination resource data, and establish a physical examination department space mapping table based on the physical examination layout data; A digital twin scheduling environment data structure is constructed based on the physical examination department resource mapping table and the physical examination department spatial mapping table. The digital twin scheduling environment data structure includes a set of physical examination departments, a set of department queue data structures, and reachability relationship data used to represent the access constraints between departments. Initialize the department queue data structure for each physical examination department in the digital twin scheduling environment; Virtual personnel are mapped into the digital twin scheduling environment. Based on the combined attributes of physical examination items, the set of target departments associated with the virtual personnel is determined. A mapping relationship between the virtual personnel and the set of target departments is established in the digital twin scheduling environment. The arrival time attribute of the virtual personnel is registered to complete the virtual personnel mapping process.

6. The method for scheduling virtual personnel space for physical examinations based on deep learning according to claim 1, characterized in that, The locking of the queue position specifically includes: In the digital twin scheduling environment, for each virtual person, the attributes of the physical examination item combination, the constraint attributes between items, the arrival time attributes and the individual preference attributes are obtained. In the physical examination demand guidance matrix, the personnel index position corresponding to the physical examination personnel bound to the virtual personnel is located according to the personnel dimension. The queuing priority probability set of the personnel index position in the department dimension, the item combination dimension and the time slice dimension is extracted. The target department identifier is determined based on the combination attributes of the queuing priority probability set and the physical examination items. The effective timestamp of the placeholder is determined based on the arrival time attribute of the virtual personnel, and the effective timestamp of the placeholder is used as the time base to locate the insertable placeholder time slice interval in the department queue data structure corresponding to the target department identifier. In the department queue data structure corresponding to the target department identifier, the existing queue order is matched and compared by combining the placeholder effective timestamp and the placeholder time slice interval to determine the placeholder queue number used for virtual placeholder, and the expected service end time corresponding to the placeholder queue number is determined as the placeholder invalidation timestamp. The target department identifier, the queue number, the timestamp of the queue taking effect, and the timestamp of the queue taking effect are encapsulated in a structured manner to generate virtual queue information. The virtual queue information is then written into the department queue data structure corresponding to the target department identifier to lock the queue position.

7. The method for scheduling virtual personnel space-occupying physical examination needs based on deep learning according to claim 1, characterized in that, The output of the physical examination guidelines specifically includes: In the digital twin scheduling environment, a virtual person is obtained for each physical examination person and bound to the physical examination person. The combination attributes of physical examination items, the constraint attributes between items and the arrival time attributes of the virtual person are read. At the same time, the virtual occupancy information of the virtual person in each target department is read. The set of physical examination items to be executed is obtained by parsing the combination attributes of the physical examination items, and the execution order of the physical examination items is obtained by sorting the execution order of the set of physical examination items according to the order constraint relationship of the items in the constraint attributes between the items. Based on the execution order of the physical examination items and the target department identifiers corresponding to each physical examination item, a sequence of physical examination departments is generated. Then, based on the department accessibility data in the digital twin scheduling environment, access constraint verification processing is performed on the sequence of physical examination departments to obtain a sequence of physical examination departments that meets the access path constraints. The arrival timestamp corresponding to the arrival time attribute is used as the expected start time of the first item. Combined with the effective timestamp and queue number of the placeholder corresponding to each target department in the physical examination department sequence, the expected start time and expected end time of each physical examination item in the execution order of the physical examination items are calculated in turn. The physical examination process includes the order of examination items, the sequence of examination departments, and the estimated start and end times of each examination item. These are then packaged into a physical examination pathway planning result, which is output as a physical examination guide.

8. The method for scheduling virtual personnel space-occupying physical examination needs based on deep learning according to claim 1, characterized in that, The output of the updated physical examination guidelines specifically includes: Real-time scheduling feedback data is collected during the physical examination process, and the real-time scheduling feedback data is standardized and packaged according to timestamps to form a feedback data stream. The feedback data stream is parsed to obtain the target physical examination department identifier, event type identifier, and event occurrence timestamp. Based on the event type identifier, the corresponding mapping process is triggered to generate service time deviation trigger identifier, queue change trigger identifier, and arrival change trigger identifier. When a service time deviation trigger flag is detected, the department queue data structure corresponding to the target physical examination department flag is located in the digital twin scheduling environment. The actual service start time and actual service end time are written to update the queue status. Based on the actual service start time and actual service end time, deviation correction processing is performed on the expected service start time and expected service end time to obtain the updated expected service start time and expected service end time. When a queue change trigger flag is detected, the queue number set and the department queue length are updated in the department queue data structure. Based on the updated queue number set, a consistency check is performed on the placeholder queue number to obtain the consistency check result. When an arrival change trigger flag is detected, the arrival time attribute is updated in the virtual personnel data structure, the occupancy effective timestamp is updated in the virtual occupancy information, and the estimated start time of the first item is updated in the physical examination route planning results. When the consistency check result indicates that the queue number is inconsistent, the virtual queue information is updated based on the consistency check result, and the updated virtual queue information is written back to the queue data structure corresponding to the target physical examination department identifier to re-lock the queue position. Based on the updated digital twin scheduling environment and the updated virtual occupancy information, the physical examination route planning results are rolled over to obtain the updated physical examination route planning results, and the updated physical examination guidance is output.

Images