A method and system for medical task allocation and management based on intelligent scheduling
By employing an intelligent scheduling-based medical task allocation method, which utilizes semantic encoding and multi-head attention mechanisms for adaptive matching of tasks and personnel, the problem of low task allocation efficiency in existing technologies is solved, achieving efficient and accurate allocation of medical tasks and rational utilization of resources.
Patent Information
- Application Number
- CN202611295455.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-25
- Publication Date
- 2026-11-10
AI Technical Summary
Current methods for allocating tasks in pharmaceutical R&D and clinical trials rely on manual scheduling or rule engines, resulting in highly subjective decision-making, low efficiency, and an inability to meet the pharmaceutical industry's requirements for accuracy and timeliness in task allocation. This leads to uneven resource utilization, untimely responses to high-priority tasks, and negatively impacts the efficiency and compliance of R&D projects.
A method based on intelligent scheduling is adopted. Task feature vectors are generated through a semantic coding model in the medical field. Combined with the global task topology map and the state data of candidate medical personnel, a multi-head attention mechanism and a temporal feature extraction network are used for matching. The allocation threshold is dynamically adjusted to achieve adaptive matching between medical tasks and personnel.
It has improved the timeliness and accuracy of pharmaceutical task allocation, ensured the rational use of resources and timely response to high-priority tasks, and improved the overall efficiency and compliance of pharmaceutical R&D projects.
Smart Images

Figure CN123070599A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of task allocation and management technology, and more specifically, to a method and system for medical task allocation and management based on intelligent scheduling. Background Technology
[0002] Currently, task allocation in pharmaceutical R&D and clinical trials primarily relies on manual scheduling or rule-engine-based automated allocation systems. Manual scheduling typically involves project managers assigning tasks based on their understanding of team members and a rough assessment of workload, resulting in a highly subjective and inefficient decision-making process. Rule-engine-based automated allocation systems usually set several fixed allocation rules. These rules are mostly based on simple matching of task type and personnel qualifications, with a single matching dimension. Furthermore, once set, these rules operate in a fixed manner and cannot be dynamically adjusted according to the actual execution status of tasks and the real-time work status of personnel.
[0003] As pharmaceutical R&D projects become increasingly complex and regulatory requirements become more stringent, the existing allocation methods are no longer sufficient to meet the pharmaceutical industry's demands for accuracy and timeliness in task allocation. Current technologies lack a comprehensive approach to understanding inter-task dependencies, accurately assessing dynamic personnel workload, and tracking personnel capability evolution trends. This leads to a disconnect between pharmaceutical task allocation decisions and actual circumstances, resulting in frequent problems such as uneven resource utilization, delayed responses to high-priority tasks, and high task handover costs, impacting the overall efficiency and compliance assurance of pharmaceutical R&D projects. Therefore, achieving adaptive intelligent matching and allocation between pharmaceutical tasks and personnel to improve the timeliness and accuracy of pharmaceutical task allocation is a key challenge facing the industry. Summary of the Invention
[0004] This application provides a method and system for managing medical task allocation based on intelligent scheduling, which can realize adaptive intelligent matching and allocation between medical tasks and medical personnel, thereby improving the timeliness and accuracy of medical task allocation.
[0005] Firstly, this application provides a medical task allocation and management method based on intelligent scheduling, the task allocation and management method comprising the following steps: In response to a submitted medical task, extract the task attribute information contained in the medical task; Based on the task attribute information, the medical task is analyzed to generate a task feature vector for the medical task; Obtain the status data of each candidate medical personnel in the candidate medical personnel set, and match the task feature vector with the status data of each candidate medical personnel to generate a task matching value for each candidate medical personnel for the medical task. Based on the task matching value, target medical personnel who meet the preset allocation threshold are selected, and the medical task is assigned to the target medical personnel.
[0006] In this embodiment, the task attribute information includes: task type, task priority, expected completion time, and required resource tags.
[0007] In this embodiment, the process of performing feature parsing on the medical task based on the task attribute information to generate the task feature vector of the medical task specifically includes: A semantic coding model for the pharmaceutical field is used to semantically encode each field in the task attribute information, generating a field semantic vector corresponding to each field. The field semantic vectors are then concatenated according to a preset dimension to generate an initial feature vector. Construct a global task topology graph, which uses all in-transit medical tasks in the system as nodes and the temporal dependencies and resource competition relationships between tasks as edges. The temporal dependencies include hard dependencies where a successor task cannot start if the predecessor task is not completed, and the resource competition relationships include soft dependencies formed when two or more tasks need to compete for the same scarce medical resource. Obtain the node position information and associated edge weights of the medical task in the global task topology graph, and fuse the node position information and associated edge weights into the initial feature vector in a graph embedding manner to generate a task feature vector.
[0008] In this embodiment, matching the task feature vector with the state data of each candidate medical professional to generate a task matching value for each candidate medical professional for the medical task specifically includes: Capability status data is extracted from the status data of the candidate medical personnel. The task feature vector and the capability status data are input into a multi-head attention-based matching network to determine the alignment score between the task requirements and each capability sub-dimension of the capability status data. The scores are then aggregated and used as the first matching sub-value. Load status data is extracted from the status data of the candidate medical personnel. The expected completion time limit in the task feature vector is used as the prediction window. The remaining execution time of each task in the current task queue in the load status data is superimposed on the time axis to predict the idle time window distribution of the corresponding candidate medical personnel in the prediction window. A second matching sub-value is generated according to the temporal overlap between the idle time window distribution and the expected execution time of the medical task. Behavioral status data and corresponding task type attribution are extracted from the status data of the candidate medical personnel. The multidimensional historical execution trajectory with the same medical task type attribution is input into the temporal feature extraction network to generate a third matching sub-value. The first matching sub-value, the second matching sub-value, and the third matching sub-value are weighted and fused to generate a task matching value.
[0009] In this embodiment, the alignment score is further corrected by a capability timeliness decay factor, which is determined according to the half-life function of the interval between the candidate medical personnel and the last time they performed the same type of task.
[0010] In this embodiment, the task matching value is a numerical value used to measure the degree of fit between medical tasks and candidate medical personnel.
[0011] In this embodiment, the method for determining the idle time window in the generation method of the second matching sub-value specifically includes: Obtain the remaining execution time and current time progress anchor point of each task in the current task queue of the candidate medical personnel, and construct a prediction time axis; The distribution of idle time windows is determined based on the current time progress anchor point and the predicted time axis.
[0012] In this embodiment, selecting target medical personnel who meet the preset allocation threshold based on the task matching value specifically includes: The task matching value of each candidate medical personnel is compared with the preset allocation threshold to construct a preliminary screening set; When the initial screening personnel set contains only one candidate medical personnel, that person is identified as the target medical personnel. When the initial screening personnel set contains two or more candidate medical personnel, the initial screening personnel set is sorted according to the task matching value, and then the target medical personnel are determined. When the initial screening personnel set is empty, candidate personnel are determined according to the task matching value, triggering the manual review and approval process.
[0013] In this embodiment, the preset method for the allocation threshold specifically includes: Based on the average historical task matching value of all candidate medical personnel under each task type, initial allocation thresholds are set for each task type. The preset allocation threshold is dynamically adjusted up or down during system operation based on the real-time load status of the system. When the overall system load rate is lower than the preset light load threshold, the allocation threshold is lowered to reduce the task allocation threshold. When the overall system load rate is higher than the preset heavy load threshold, the allocation threshold is raised to increase the task allocation threshold.
[0014] Secondly, this application provides a medical task allocation management system based on intelligent scheduling, used to execute a medical task allocation management method based on intelligent scheduling, the task allocation management system comprising: The task response module is used to respond to submitted medical tasks and extract the task attribute information contained in the medical tasks. The feature parsing module is used to perform feature parsing on the medical task based on the task attribute information and generate the task feature vector of the medical task. The task matching module is used to obtain the status data of each candidate medical personnel in the candidate medical personnel set, match the task feature vector with the status data of each candidate medical personnel, and generate a task matching value for each candidate medical personnel for the medical task. The task allocation module is used to select target medical personnel who meet the preset allocation threshold based on the task matching value, and to allocate the medical task to the target medical personnel.
[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: In response to a submitted medical task, the system extracts task attribute information contained in the medical task; based on the task attribute information, it performs feature parsing on the medical task to generate a task feature vector; it obtains the status data of each candidate medical personnel in the candidate medical personnel set, and matches the task feature vector with the status data of each candidate medical personnel to generate a task matching value for each candidate medical personnel for the medical task; based on the task matching value, it selects a target medical personnel who meet a preset allocation threshold and assigns the medical task to the target medical personnel.
[0016] Therefore, this application firstly provides a standardized data input foundation for subsequent feature analysis and matching calculations by responding to submitted medical tasks and extracting task attribute information. Secondly, it generates task feature vectors from task attribute information using a semantic coding model in the medical field, enabling a global perception of temporal dependencies and resource competition relationships between tasks during the subsequent matching process. Thirdly, it acquires the status data of candidate medical personnel and, based on this, achieves fine-grained non-equilibrium matching of task requirements and personnel capabilities through a multi-head attention mechanism. It transforms static load counting into dynamic spatiotemporal availability analysis by predicting the distribution of idle time windows through temporal overlay. It captures the direction of capability changes by extracting the evolution trend of the network encoding historical trajectory through temporal feature extraction, so that the matching results comprehensively reflect the overall suitability between personnel and tasks. Finally, it filters all candidate medical personnel through a preset allocation threshold and selects target medical personnel, ensuring that the allocation decision has a reasonable processing path under various circumstances, improving the timeliness and accuracy of medical task allocation.
[0017] In summary, the technical solution adopted in this application can realize adaptive intelligent matching and allocation between medical tasks and medical personnel, thereby improving the timeliness and accuracy of medical task allocation. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is an exemplary flowchart of a medical task allocation and management method based on intelligent scheduling provided in this application; Figure 2 This is a schematic diagram of a device scenario for medical task allocation and management based on intelligent scheduling, according to the present application. Figure 3 This is a module structure diagram of the intelligent scheduling-based medical task allocation management system provided in this application. Detailed Implementation
[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0021] This application provides a method and system for managing pharmaceutical task allocation based on intelligent scheduling. The core of this method involves responding to a submitted pharmaceutical task and extracting task attribute information from the task; performing feature analysis on the task based on the task attribute information to generate a task feature vector; obtaining the status data of each candidate pharmaceutical personnel in a candidate pharmaceutical personnel set; matching the task feature vector with the status data of each candidate pharmaceutical personnel to generate a task matching value for each candidate pharmaceutical personnel for the pharmaceutical task; selecting target pharmaceutical personnel who meet a preset allocation threshold based on the task matching value; and allocating the pharmaceutical task to the target pharmaceutical personnel.
[0022] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of a medical task allocation and management method based on intelligent scheduling according to this embodiment of the present application. The task allocation and management method includes the following steps: In step S1, in response to the submitted medical task, the task attribute information contained in the medical task is extracted.
[0023] In practical implementation, responding to submitted pharmaceutical tasks, the task attribute information contained within the task is extracted. That is, a pharmaceutical task is a structured data packet containing task attribute information fields filled in by the task initiator according to a preset task template. After format and integrity checks on the received pharmaceutical task, task attribute information is extracted from the task according to field mapping rules. Task attribute information includes: task type, task priority, expected completion deadline, and required resource tags. The task type field is selected by the task initiator from a system-maintained task type enumeration set. Task types include: clinical trial data verification tasks, pharmacovigilance case assessment tasks, subject enrollment screening tasks, adverse drug reaction report review tasks, and clinical trial protocol deviation judgment tasks. The task priority field is selected by the task initiator based on a comprehensive judgment of the task's time sensitivity, safety impact scope, and regulatory reporting deadline. When creating a task, the task initiator automatically prompts the regulatory reporting deadline requirements for that type of task based on the task type. The task initiator determines the priority level based on this prompt and the task's own urgency. The expected completion deadline includes: the latest absolute completion time and the recommended relative completion time. The task initiator determines the latest absolute completion time based on the subject follow-up nodes, data locking nodes, or regulatory reporting deadlines involved in the task. It automatically generates a recommended relative completion time based on the historical average processing time for this task type for the task initiator's reference and confirmation. Required resource tags include: required professional field tags, required qualification certificate tags, required operation permission level tags, and required data access scope tags. The task initiator selects tags matching the task requirements from the system-maintained resource tag system.
[0024] In step S2, based on the task attribute information, the medical task is subjected to feature parsing to generate a task feature vector for the medical task.
[0025] In this embodiment, based on the task attribute information, feature parsing is performed on the medical task to generate the task feature vector of the medical task, including: A semantic coding model for the pharmaceutical field is used to semantically encode each field in the task attribute information, generating a field semantic vector corresponding to each field. The field semantic vectors are then concatenated according to a preset dimension to generate an initial feature vector. Construct a global task topology graph, which uses all in-transit medical tasks in the system as nodes and the temporal dependencies and resource competition relationships between tasks as edges. The temporal dependencies include hard dependencies where a successor task cannot start if the predecessor task is not completed, and the resource competition relationships include soft dependencies formed when two or more tasks need to compete for the same scarce medical resource. Obtain the node position information and associated edge weights of the medical task in the global task topology graph, and fuse the node position information and associated edge weights into the initial feature vector in a graph embedding manner to generate a task feature vector.
[0026] In practical implementation, a medical domain semantic coding model can be used to semantically encode each field in the task attribute information, generating a corresponding field semantic vector. These semantic vectors are then concatenated according to a preset dimension to generate an initial feature vector. In other words, each field in the task attribute information is encoded separately. For discrete fields such as task type, task priority, and required resource tags, a pre-built medical domain semantic coding model is invoked for semantic encoding. This model is a semantic coding model pre-trained based on a medical domain corpus, capable of mapping unstructured text tags into dense vector representations. For the latest absolute completion time node and the recommended relative completion time in the expected completion time limit, normalized encoding is performed, converting the time value into a time offset relative to the current moment and mapping it to a preset numerical range suggested by experts. After each field is encoded, a corresponding field semantic vector is generated. All field semantic vectors are then concatenated according to a preset dimension order to generate the initial feature vector. The preset dimension order is a field arrangement order predefined by expert experience to ensure that the feature vectors of all tasks under the same task type have consistent semantic meaning at the same dimension position. Secondly, a global task topology graph is constructed. This graph uses all in-transit medical tasks within the system as nodes, and the temporal dependencies and resource competition relationships between tasks as edges. Temporal dependencies include hard dependencies where a successor task cannot start until the predecessor task is completed, and resource competition relationships include soft dependencies formed when two or more tasks compete for the same scarce medical resource. All in-transit medical tasks include tasks that have not yet started and tasks that are currently being executed but have not yet finished, excluding historical tasks that have been completed. For each pair of tasks within the system, the existence of a temporal dependency or resource competition relationship between them is checked. If either relationship exists, an edge is constructed between the pair of tasks. Temporal dependencies include hard dependencies and soft dependencies. Hard dependencies indicate that a subsequent task cannot start until the preceding task is completed. For example, a clinical trial data verification task can only start after the subject enrollment screening task is completed. Hard dependencies are determined based on the task type in the task attribute information and the dependent task identifier declared by the task initiator during task creation. Soft dependencies indicate that the completion of a preceding task provides convenient but not necessary conditions for the subsequent task. For example, a task references the output of another task but does not require that output for execution. Soft dependencies are determined based on the semantic relevance in the task description text. Resource competition relationships represent constraints formed when two or more tasks compete for the same scarce medical resource. Scarce medical resources include, but are not limited to, specific testing equipment, specific data access channels, and specific qualified reviewers. When two or more tasks have the same scarce resource tag in their required resource tags, a resource competition relationship is determined between the two or more tasks, and an edge is constructed between the corresponding nodes. The edge weight is determined based on the intensity of the resource competition.After completing the node and edge construction of all in-transit tasks, a global task topology graph is generated. The global task topology graph contains all in-transit medical task nodes as well as temporal dependency edges and resource competition edges between tasks.
[0027] Furthermore, in the specific implementation, the node position information and associated edge weights of the medical task in the global task topology graph are obtained. These node position information and associated edge weights are then fused into the initial feature vector using graph embedding to generate a task feature vector. Specifically, the node position information includes the in-degree, out-degree, and topological level of the corresponding node in the global task topology graph. The in-degree represents the number of other tasks that depend on the completion of this task to be started; the out-degree represents the number of other tasks that this task depends on; and the topological level represents the dependency distance of this node from the starting node of the topology graph. The associated edge weights are the weight values of each edge. The weights of temporal dependency edges are determined based on the inter-task dependency strength coefficient, which is jointly determined by the hardness or softness of the dependency type and the dependency tightness declared by the initiator. The weights of resource competition edges are determined based on the resource competition intensity coefficient, which is determined by the number of tasks competing for the same resource. The above node position information and associated edge weights are processed according to a preset graph embedding algorithm to generate a structural embedding vector of the task node in the graph. The graph embedding algorithm employs a random walk-based graph embedding method. It simulates a random walk starting from the current node and proceeding along the edges to generate a node sequence. This node sequence is then input into the graph embedding model to generate a structural embedding vector. The structural embedding vector is concatenated to the end of the initial feature vector to generate a task feature vector containing both the task's own semantic information and the task's global context semantics. This task feature vector simultaneously represents the task attribute characteristics of the medical task and its structural position characteristics within the entire in-transit task network.
[0028] In step S3, the status data of each candidate medical personnel in the candidate medical personnel set is obtained, and the task feature vector is matched with the status data of each candidate medical personnel to generate a task matching value for each candidate medical personnel for the medical task.
[0029] In practical implementation, the status data of each candidate medical professional in the candidate medical professional personnel set is obtained. That is, the candidate medical professional personnel set is the set of all currently registered personnel capable of undertaking medical tasks. Status data includes capability status data, workload status data, and behavioral status data. Capability status data characterizes the task-taking capability boundaries of candidate medical professionals and includes at least professional field tags, qualification certificate tags, and task type tags. The professional capability tag vectors of each candidate medical professional are read from the personnel capability vector library. These vectors are generated by fusion after semantic encoding and multi-hot encoding of the original capability profile data of each candidate medical professional. Workload status data characterizes the real-time workload saturation of candidate medical professionals and includes at least the task queue currently being executed by the candidate medical professional and the remaining execution time of each task. The task list currently being executed by each candidate medical professional and the remaining execution time of each task are read from the task execution management system. Behavioral status data characterizes the historical performance level of candidate medical professionals on various tasks and includes at least the completion quality score sequence, processing time sequence, and mid-process return rate sequence organized by task type. Completed task records for each candidate medical personnel within a preset historical time window are extracted from the historical task execution database, grouped by task type, and then extracted into sequences. The capability status data, workload status data, and behavioral status data of each candidate medical personnel are correlated and aggregated to form the status data for each candidate medical personnel.
[0030] In this embodiment, the task feature vector is matched with the state data of each candidate medical personnel to generate a task matching value for each candidate medical personnel for the medical task. This can be achieved through the following steps: Capability status data is extracted from the status data of the candidate medical personnel. The task feature vector and the capability status data are input into a multi-head attention-based matching network to determine the alignment score between the task requirements and each capability sub-dimension of the capability status data. The scores are then aggregated and used as the first matching sub-value. Load status data is extracted from the status data of the candidate medical personnel. The expected completion time limit in the task feature vector is used as the prediction window. The remaining execution time of each task in the current task queue in the load status data is superimposed on the time axis to predict the idle time window distribution of the corresponding candidate medical personnel in the prediction window. A second matching sub-value is generated according to the temporal overlap between the idle time window distribution and the expected execution time of the medical task. Behavioral status data and corresponding task type attribution are extracted from the status data of the candidate medical personnel. The multidimensional historical execution trajectory with the same medical task type attribution is input into the temporal feature extraction network to generate a third matching sub-value. The first matching sub-value, the second matching sub-value, and the third matching sub-value are weighted and fused to generate a task matching value.
[0031] In practical implementation, firstly, capability status data can be extracted from the status data of candidate medical personnel. The task feature vector and capability status data are then input into a multi-head attention-based matching network to determine the alignment scores between the task requirements and each capability sub-dimension of the capability status data. These scores are then aggregated and used as the first matching sub-value. In other words, capability status data can be extracted from the candidate medical personnel's status data, and the task feature vector and the professional capability label vector from the capability status data are input into the multi-head attention-based matching network. It should be noted that before inputting the task feature vector into the multi-head attention-based matching network, the internal structure of the task feature vector is analyzed. Specifically, the task feature vector is composed of an initial feature vector and a structural embedding vector. The initial feature vector carries the semantic information of the task attributes and is located in the lower-dimensional segment of the vector; the structural embedding vector carries the structural position information of the task in the global task topology graph and is located in the higher-dimensional segment of the vector. Based on this structure, the multi-head attention-based matching network does not directly perform global attention calculation between the overall task feature vector and the capability status data. Instead, it divides the task feature vector into semantic feature segments and graph structure feature segments according to the dimensional segmentation points of the vector. Therefore, the multi-head attention-based matching network comprises multiple parallel attention heads. Each attention head independently calculates the sub-dimensional alignment score between the task feature vector and the professional competence label vector. Specifically, the task feature vector contains the task's requirement representation in each competence requirement sub-dimensional, and the professional competence label vector contains the personnel's competence representation in the corresponding competence sub-dimensional. Each attention head maps the task feature vector to a query matrix and the professional competence label vector to a key matrix and a value matrix, respectively, and calculates the alignment score between each sub-dimensional using scaled dot product attention. The alignment scores output by each attention head are concatenated and input into a feedforward network for aggregation, outputting the first matching sub-value. The first matching sub-value represents the overall degree of fit between the candidate medical personnel's professional competence and the competence requirements of the medical task.
[0032] Furthermore, in the specific implementation, load status data is extracted from the status data of candidate medical personnel. Using the expected completion time limit in the task feature vector as the prediction window, the remaining execution time of each task in the current task queue in the load status data is superimposed on the time axis to predict the idle time window distribution of the corresponding candidate medical personnel within the prediction window. A second matching sub-value is generated based on the temporal overlap between the idle time window distribution and the expected execution time of the medical task. That is, the idle time window distribution can be elaborated in detail in subsequent steps. The expected execution time of the medical task is obtained, and the temporal overlap between each idle window and the expected execution time in the idle time window distribution is calculated. Specifically, each idle window is traversed, and the intersection duration between each idle window and the time window corresponding to the expected execution time of the medical task is calculated. The ratio of the sum of all intersection durations to the expected execution time is taken as the temporal overlap degree. When the time overlap is greater than or equal to 1, it indicates that the candidate medical personnel has at least one consecutive idle window with a length not less than the expected execution time within the prediction window, and the second matching sub-value is set to 1. When the time overlap is greater than 0 and less than 1, it indicates that the candidate medical personnel has several idle windows within the prediction window, but these windows only partially overlap with the expected execution time, and the value of this time overlap is directly used as the second matching sub-value. When the time overlap is equal to 0, it indicates that the candidate medical personnel does not have any idle time windows overlapping with the expected execution time of the medical task within the expected completion time limit, and the second matching sub-value is set to 0. The value of the second matching sub-value ranges from 0 to 1, representing the confidence that the candidate medical personnel has a sufficient execution window within the expected completion time limit of the medical task. Furthermore, the remaining execution time of each task in the current task queue is replaced by a dynamic remaining time prediction value instead of a static value. That is, the execution time of the task, the historical average processing time of the task type at the system level, and the personal historical average processing time of the candidate medical personnel for the task type are obtained. The dynamic remaining execution time is calculated by weighted interpolation. That is, the execution time of the task is used as the base value, the difference between the historical average processing time and the execution time is used as the system correction item, and the difference between the personal historical average processing time and the execution time is used as the personal correction item. The system correction item and the personal correction item are assigned different weights and then superimposed to generate the remaining execution time. The weights can be preset by expert experience.
[0033] Furthermore, in the specific implementation, behavioral state data and corresponding task type attribution are extracted from the state data of candidate medical personnel. The multidimensional historical execution trajectory, which corresponds to the medical task type attribution, is input into the temporal feature extraction network to generate a third matching sub-value. That is, the multidimensional historical execution trajectory includes a completion quality score sequence, a processing time sequence, and a mid-course return rate sequence, with the time nodes in each sequence aligned. The third matching sub-value is directly output as the predicted completion quality score, whose value range is consistent with the scoring scale of the completion quality score sequence in the behavioral state data. It represents the task completion quality score predicted based on the candidate medical personnel's historical performance if the medical task is assigned to them. The temporal feature extraction network uses the score value of the last time node in the historical completion quality score sequence as the baseline value, and uses the evolutionary trend encoding of the processing time sequence and the evolutionary trend encoding of the mid-course return rate sequence as correction terms, outputting the predicted quality score value through a fully connected layer. Specifically, when the processing time sequence shows a downward trend, it indicates that the person's efficiency in performing this type of task is improving, and the correction term is positive; when the processing time sequence shows an upward trend, it indicates that efficiency is decreasing, and the correction term is negative. When the return rate sequence shows a downward trend, it indicates that the person's work quality stability is improving, and the correction term is positive; when the return rate sequence shows an upward trend, it indicates that stability is decreasing, and the correction term is negative. If the predicted quality score exceeds the candidate medical personnel's historical average quality score, it indicates that the person's ability in this type of task is on an upward trend, and the third matching sub-value is correspondingly high; conversely, it indicates that the ability is on a downward trend, and the third matching sub-value is correspondingly low. The multi-dimensional historical execution trajectory is input into the temporal feature extraction network, which is a feature encoder built based on a temporal convolutional neural network. The temporal feature extraction network performs temporal encoding on each sequence, extracting the evolution trend of each sequence over time, including the direction and rate of change of ability. The evolution trend encodings of each sequence are aggregated after attention pooling and output as the third matching sub-value. The third matching subvalue represents the candidate medical personnel's prediction of the quality of medical task completion based on their historical performance.
[0034] Furthermore, before being input into the temporal feature extraction network, the multidimensional historical execution trajectory is processed by a reference calibration layer. Specifically, this involves obtaining a set of all personnel performing the same task type as the medical task within a preset time window, extracting the completion quality score sequence, processing time sequence, and mid-process return rate sequence for each person in the set within the preset time window, and calculating the mean and standard deviation of the overall performance for each time slice. The original score value of the candidate medical personnel in each time slice is subtracted from the mean of all personnel in the corresponding time slice, and then divided by the standard deviation to generate a standardized score sequence. This standardized score sequence replaces the original score sequence and is input into the temporal feature extraction network. Even further, when there are no execution records of the same task type as the medical task in the candidate medical personnel's historical execution records, a cross-type transfer matching mechanism is activated. This involves extracting the execution trajectory encoding values corresponding to each existing task type in the candidate medical personnel's historical execution records, and calculating the semantic similarity between the task feature vector of the medical task and the task type prototype vector corresponding to each existing task type. The task type prototype vector is generated by averaging the task feature vectors of multiple historical tasks under each task type. The existing task type with the highest semantic similarity is selected as the source task type. The execution trajectory encoding value of the candidate medical personnel on the source task type is multiplied by a preset cross-type transfer coefficient and used as the third matching sub-value. Specifically, the semantic similarity between the task feature vector of the medical task and the corresponding task type prototype vector of each existing task type can be calculated. The existing task type with the highest semantic similarity is selected as the source task type, and this semantic similarity is used as the cross-type transfer coefficient.
[0035] In practical implementation, the first, second, and third matching sub-values are weighted and fused to generate the task matching value. Specifically, the first matching sub-value is multiplied by a first weight coefficient to obtain a first weighted value; the second matching sub-value is multiplied by a second weight coefficient to obtain a second weighted value; and the third matching sub-value is multiplied by a third weight coefficient to obtain a third weighted value. The first, second, and third weighted values are then summed to obtain the task matching value. All three weight coefficients are preset positive numbers, and their sum equals one. Each weight coefficient is determined through Bayesian optimization based on actual system operating data, with the optimization objective being to maximize the correlation between the ranking results of the matching values and the ranking results of human expert review. The task matching value is a numerical value used to measure the degree of fit between a medical task and a candidate medical professional. The task matching value is a dimensionless score ranging from zero to one; a higher task matching value indicates a higher degree of fit between the candidate medical professional and the medical task.
[0036] In this embodiment, the alignment score is further corrected by a capability timeliness decay factor. The capability timeliness decay factor is calculated based on the interval between the candidate medical personnel and their last performance of the same type of task, using a half-life function. Specifically, this can be achieved through the following steps: In practical implementation, the completion time of the last time a candidate medical professional performed a task of the same type as a medical task can be obtained. The interval is calculated by subtracting the completion time from the current time. The quotient of the interval divided by a preset half-life parameter is used as the exponent to calculate an exponential function value with a base of 1 / 2. The result is used as the decay coefficient. The preset half-life parameter is a time threshold set for different task types, representing the length of time required for a person's ability to decay to half of its initial value for that type of task. For task types with rapid technological updates, the preset half-life parameter is shorter; for task types with relatively stable technologies, the preset half-life parameter is longer. The preset half-life parameter is pre-configured by the system administrator based on medical industry standards and the knowledge update frequency of each task type. When the interval is zero, it indicates that the candidate medical professional is currently still performing or has just completed that type of task, the decay coefficient is one, and the alignment score is not affected by decay. When the interval approaches infinity, the decay coefficient approaches zero, and the alignment score approaches zero, indicating that the person's effective ability has essentially lost its reference value after a long period of inactivity in that type of task.
[0037] In this embodiment, the determination of the idle time window in the generation method of the second matching sub-value can be implemented using the following steps: Obtain the remaining execution time and current time progress anchor point of each task in the current task queue of the candidate medical personnel, and construct a prediction time axis; The distribution of idle time windows is determined based on the current time progress anchor point and the predicted time axis.
[0038] In the specific implementation, firstly, the remaining execution time and current time progress anchor point of each task in the current task queue of the candidate medical personnel are obtained to construct a predicted time axis. That is, the remaining execution time and current time progress anchor point of each task in the current task queue of the candidate medical personnel are obtained, and the predicted time axis is constructed with the current time progress anchor point as the starting point and the expected completion time limit in the task feature vector as the ending point. The current time progress anchor point is the timestamp of the current system moment, and the expected completion time limit is the latest absolute completion time node set by the task initiator for the medical task. Then, the distribution of idle time windows is determined based on the current time progress anchor point and the predicted time axis. That is, with the current time progress anchor point as the starting position, the remaining execution time of the first task in the queue is marked on the time axis as the first occupied time interval. The ending position of the first task is used as the starting position of the second task, and the remaining execution time of the second task is marked as the second occupied time interval, and so on, until the remaining execution time of all tasks in the queue has been marked. After marking, the continuous intervals on the predicted time axis that are not occupied by any task are marked as the idle time window distribution. When the current task queue is empty, there are no occupied intervals on the prediction timeline, and the entire time interval from the current time progress anchor point to the expected completion deadline is marked as a single idle time window.
[0039] In step S4, target medical personnel who meet the preset allocation threshold are selected based on the task matching value, and the medical task is assigned to the target medical personnel.
[0040] In this embodiment, target medical personnel who meet the preset allocation threshold are selected based on the task matching value. This can be achieved through the following steps: The task matching value of each candidate medical personnel is compared with the preset allocation threshold to construct a preliminary screening set; When the initial screening personnel set contains only one candidate medical personnel, that person is identified as the target medical personnel. When the initial screening personnel set contains two or more candidate medical personnel, the initial screening personnel set is sorted according to the task matching value, and then the target medical personnel are determined. When the initial screening personnel set is empty, candidate personnel are determined according to the task matching value, triggering the manual review and approval process.
[0041] In practical implementation, firstly, the task matching value of each candidate medical personnel is compared with a preset allocation threshold to construct a preliminary screening personnel set. That is, the task matching value of each candidate medical personnel is compared with the allocation threshold, and candidates whose task matching value reaches or exceeds the allocation threshold are selected. The preliminary screening personnel set consists of all candidates medical personnel who meet this condition. The preset method of the allocation threshold will be explained in detail in the following process. Secondly, if the preliminary screening personnel set contains only one candidate medical personnel, that person is determined as the target medical personnel. Then, if the preliminary screening personnel set contains two or more candidate medical personnel, the preliminary screening personnel set is sorted according to the task matching value, and the target medical personnel is determined. That is, if the preliminary screening personnel set contains at least two candidate medical personnel, the preliminary screening personnel set is sorted from high to low according to the task matching value, and the candidate medical personnel at the top of the sort is selected as the target medical personnel. Finally, if the preliminary screening personnel set is empty, the candidate receiving personnel is determined according to the task matching value, triggering the manual review and approval process. That is, if the preliminary screening personnel set is empty, it means that there are currently no candidate medical personnel whose task matching value reaches the allocation threshold. In this scenario, the task priority for the medical task is determined as follows: If the task priority is higher than the preset emergency task priority threshold, the person with the highest task matching value in the candidate medical personnel set is directly selected as the target medical personnel, and the assignment is completed. If the task priority is lower than or equal to the preset emergency task priority threshold, the person with the highest task matching value in the candidate medical personnel set is selected as the candidate assignee, triggering a manual review and approval process. The manual review and approval process is as follows: The task attribute information of the medical task, the status data of the corresponding candidate medical personnel, and the task matching value are sent to the management terminal with approval authority. The management terminal reviewer confirms whether to assign the task to the candidate assignee, or designates another medical personnel to undertake the task. The emergency task priority threshold can be preset based on expert experience.
[0042] It should be noted that when the initial screening set contains at least two candidate medical personnel, and the difference in task matching value between the top-ranked candidate and the second-ranked candidate is less than a preset difference threshold, an anti-allocation oscillation processing is implemented. Specifically, the average number of tasks undertaken and the average task completion quality of the top-ranked candidate within a preset historical time window are obtained, as are the average number of tasks undertaken and the average task completion quality of the second-ranked candidate within the same preset historical time window. The number of tasks undertaken is negatively calculated and multiplied by a first balancing weight, and the average task completion quality is multiplied by a second balancing weight. The two are then added together to obtain a comprehensive balancing score, and the candidate with the higher comprehensive balancing score is selected as the target medical personnel. The historical time window and balancing weights can be preset through historical data analysis.
[0043] In this embodiment, the preset method for allocating the threshold can be implemented using the following steps: Based on the average historical task matching value of all candidate medical personnel under each task type, initial allocation thresholds are set for each task type. The preset allocation threshold is dynamically adjusted up or down during system operation based on the real-time load status of the system. When the overall system load rate is lower than the preset light load threshold, the allocation threshold is lowered to reduce the task allocation threshold. When the overall system load rate is higher than the preset heavy load threshold, the allocation threshold is raised to increase the task allocation threshold.
[0044] In practical implementation, firstly, the average historical task matching values of all candidate medical personnel under each task type are used as a benchmark. Initial allocation thresholds are then set for each task type. Specifically, the historical task matching values of all candidate medical personnel under each task type are obtained, statistically analyzed by task type, and the arithmetic mean of all historical task matching values for each task type is calculated. This arithmetic mean is used as the initial allocation threshold for that task type. For newly added task types without accumulated historical matching values, a globally fixed value is used as the initial allocation threshold. This globally fixed value is preset based on industry experience and ranges from zero to one. Then, the preset allocation threshold is dynamically adjusted upwards or downwards during system operation based on the real-time system load. When the overall system load rate is lower than a preset light load threshold, the allocation threshold is lowered to reduce the task allocation threshold. When the overall system load rate is higher than a preset heavy load threshold, the allocation threshold is raised to increase the task allocation threshold. In other words, the ratio of the number of busy personnel among all candidate medical personnel to the total number of candidate medical personnel is monitored in real time, and this ratio is used as the overall system load rate. The system acquires preset light load and heavy load thresholds. The light load threshold is the upper limit of the load rate for determining a low-load system, and the heavy load threshold is the lower limit of the load rate for determining a high-load system. The light load threshold is lower than the heavy load threshold and can be preset using historical data analysis. When the overall system load rate is below the light load threshold, the allocation threshold is lowered to reduce the task allocation threshold, allowing more candidate medical personnel to meet the task matching value and thus expanding the range of available personnel, making full use of the system's idle human resources. When the overall system load rate is above the heavy load threshold, the allocation threshold is raised to increase the task allocation threshold, ensuring that only candidate medical personnel with high task matching values pass the threshold screening, avoiding further task allocation to overloaded personnel when the system is busy. When the overall system load rate is between the light load and heavy load thresholds, the current allocation threshold remains unchanged. Adjustments to the allocation threshold are made gradually according to preset step sizes, with each adjustment not exceeding the preset maximum adjustment step size. The preset step size and maximum adjustment step size are preset by the system administrator.
[0045] In practical implementation, medical tasks are assigned to target medical personnel. Specifically, the task's attribute information, feature vector, and matching value are encapsulated into an assignment instruction, which is sent to the task receiving end via an internal message channel. The task receiving end is the terminal device used by the target medical personnel to receive tasks, view task details, and submit task execution results. A task assignment record is generated, the target medical personnel's load status data is updated, the medical task and its expected execution duration are added to the tail of the personnel's current task queue, and the order of tasks in the personnel's current task queue is updated.
[0046] like Figure 2 The diagram illustrates a scenario of a device for intelligent scheduling-based medical task allocation and management, comprising three-way interaction: a task initiator, an allocation management unit, and a task receiver. The task initiator submits a medical task request to the allocation management unit. The allocation management unit receives the medical task request, executes an intelligent allocation decision, generates an allocation instruction, and sends the allocation instruction to the task receiver. The task receiver receives and executes the medical task specified in the allocation instruction.
[0047] Therefore, this application firstly provides a standardized data input foundation for subsequent feature analysis and matching calculations by responding to submitted medical tasks and extracting task attribute information. Secondly, it generates task feature vectors from task attribute information using a semantic coding model in the medical field, enabling a global perception of temporal dependencies and resource competition among tasks during the subsequent matching process. Then, by acquiring the status data of candidate medical personnel, it achieves fine-grained unbalanced matching of task requirements and personnel capabilities through a multi-head attention mechanism. It transforms static load counting into dynamic spatiotemporal availability analysis by predicting the distribution of idle time windows through temporal overlay. It captures the direction of capability changes by extracting the evolution trend of the network encoding historical trajectory through temporal feature extraction, ensuring that the matching results comprehensively reflect the overall suitability between personnel and tasks. Finally, it filters all candidate medical personnel through a preset allocation threshold, selecting target medical personnel to ensure that allocation decisions have reasonable processing paths under various circumstances, improving the timeliness and accuracy of medical task allocation.
[0048] In summary, the technical solution adopted in this application can realize adaptive intelligent matching and allocation between medical tasks and medical personnel, thereby improving the timeliness and accuracy of medical task allocation.
[0049] Example 2: This application provides a medical task allocation and management system based on intelligent scheduling, referencing... Figure 3 As shown in the figure, this is a module structure diagram of a medical task allocation and management system based on intelligent scheduling according to this embodiment of the present application. The task allocation and management system includes: The task response module 100 is used to respond to the submitted medical task and extract the task attribute information contained in the medical task. The feature parsing module 200 is used to perform feature parsing on the medical task based on the task attribute information and generate a task feature vector for the medical task. The task matching module 300 is used to obtain the status data of each candidate medical personnel in the candidate medical personnel set, match the task feature vector with the status data of each candidate medical personnel, and generate a task matching value for each candidate medical personnel for the medical task. The task allocation module 400 is used to select target medical personnel who meet the preset allocation threshold according to the task matching value, and to allocate the medical task to the target medical personnel.
[0050] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0051] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.
[0052] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
Claims
1. A method for managing medical task allocation based on intelligent scheduling, characterized in that, Includes the following steps: In response to a submitted medical task, extract the task attribute information contained in the medical task; Based on the task attribute information, the medical task is analyzed to generate a task feature vector for the medical task; Obtain the status data of each candidate medical personnel in the candidate medical personnel set, and match the task feature vector with the status data of each candidate medical personnel to generate a task matching value for each candidate medical personnel for the medical task. Based on the task matching value, target medical personnel who meet the preset allocation threshold are selected, and the medical task is assigned to the target medical personnel.
2. The medical task allocation and management method based on intelligent scheduling as described in claim 1, characterized in that, The task attribute information includes: task type, task priority, expected completion time, and required resource tags.
3. The medical task allocation and management method based on intelligent scheduling as described in claim 1, characterized in that, Based on the task attribute information, feature parsing is performed on the medical task to generate the task feature vector of the medical task, specifically including: A semantic coding model for the pharmaceutical field is used to semantically encode each field in the task attribute information, generating a field semantic vector corresponding to each field. The field semantic vectors are then concatenated according to a preset dimension to generate an initial feature vector. Construct a global task topology graph, which uses all in-transit medical tasks in the system as nodes and the temporal dependencies and resource competition relationships between tasks as edges. The temporal dependencies include hard dependencies where a successor task cannot start if the predecessor task is not completed, and the resource competition relationships include soft dependencies formed when two or more tasks need to compete for the same scarce medical resource. Obtain the node position information and associated edge weights of the medical task in the global task topology graph, and fuse the node position information and associated edge weights into the initial feature vector in a graph embedding manner to generate a task feature vector.
4. The medical task allocation and management method based on intelligent scheduling as described in claim 1, characterized in that, Matching the task feature vector with the state data of each candidate medical professional to generate a task matching value for each candidate medical professional for the medical task specifically includes: Capability status data is extracted from the status data of the candidate medical personnel. The task feature vector and the capability status data are input into a multi-head attention-based matching network to determine the alignment score between the task requirements and each capability sub-dimension of the capability status data. The scores are then aggregated and used as the first matching sub-value. Load status data is extracted from the status data of the candidate medical personnel. The expected completion time limit in the task feature vector is used as the prediction window. The remaining execution time of each task in the current task queue in the load status data is superimposed on the time axis to predict the idle time window distribution of the corresponding candidate medical personnel in the prediction window. A second matching sub-value is generated according to the temporal overlap between the idle time window distribution and the expected execution time of the medical task. Behavioral status data and corresponding task type attribution are extracted from the status data of the candidate medical personnel. The multidimensional historical execution trajectory with the same medical task type attribution is input into the temporal feature extraction network to generate a third matching sub-value. The first matching sub-value, the second matching sub-value, and the third matching sub-value are weighted and fused to generate a task matching value.
5. The medical task allocation and management method based on intelligent scheduling as described in claim 4, characterized in that, The alignment score is further corrected by a capability timeliness decay factor, which is determined based on the interval between the candidate medical personnel and their last performance of a similar task, according to a half-life function.
6. The medical task allocation and management method based on intelligent scheduling as described in claim 1, characterized in that, The task matching value is a numerical value used to measure the degree of fit between medical tasks and candidate medical personnel.
7. The medical task allocation and management method based on intelligent scheduling as described in claim 4, characterized in that, In the method for generating the second matching sub-value, the method for determining the idle time window specifically includes: Obtain the remaining execution time and current time progress anchor point of each task in the current task queue of the candidate medical personnel, and construct a prediction time axis; The distribution of idle time windows is determined based on the current time progress anchor point and the predicted time axis.
8. The medical task allocation and management method based on intelligent scheduling as described in claim 7, characterized in that, The target medical personnel selected based on the task matching value and meeting the preset allocation threshold specifically include: The task matching value of each candidate medical personnel is compared with the preset allocation threshold to construct a preliminary screening set; When the initial screening personnel set contains only one candidate medical personnel, that person is identified as the target medical personnel. When the initial screening personnel set contains two or more candidate medical personnel, the initial screening personnel set is sorted according to the task matching value, and then the target medical personnel are determined. When the initial screening personnel set is empty, candidate personnel are determined according to the task matching value, triggering the manual review and approval process.
9. A medical task allocation and management method based on intelligent scheduling as described in claim 1, characterized in that, The preset method for the allocation threshold specifically includes: Based on the average historical task matching value of all candidate medical personnel under each task type, initial allocation thresholds are set for each task type. The preset allocation threshold is dynamically adjusted up or down during system operation based on the real-time load status of the system. When the overall system load rate is lower than the preset light load threshold, the allocation threshold is lowered to reduce the task allocation threshold. When the overall system load rate is higher than the preset heavy load threshold, the allocation threshold is raised to increase the task allocation threshold.
10. A pharmaceutical task allocation and management system based on intelligent scheduling, used to execute a pharmaceutical task allocation and management method based on intelligent scheduling as described in any one of claims 1 to 9, characterized in that, include: The task response module is used to respond to submitted medical tasks and extract the task attribute information contained in the medical tasks. The feature parsing module is used to perform feature parsing on the medical task based on the task attribute information and generate the task feature vector of the medical task. The task matching module is used to obtain the status data of each candidate medical personnel in the candidate medical personnel set, match the task feature vector with the status data of each candidate medical personnel, and generate a task matching value for each candidate medical personnel for the medical task. The task allocation module is used to select target medical personnel who meet the preset allocation threshold based on the task matching value, and to allocate the medical task to the target medical personnel.
