Medical care task scheduling method based on AI

Abstract

The invention discloses an AI-based medical care task scheduling method, and particularly relates to the field of medical care task scheduling, and the method comprises the steps: S1, nursing data collection; s2, generating a nursing task feature data set; s3, outputting a nursing task priority sequence; s4, generating a preliminary nursing task scheduling scheme; s5, generating a final nursing task scheduling scheme; s6, dynamically updating the nursing task scheduling scheme; s7, displaying and feeding back the nursing task scheduling scheme; aI matching calculation and dynamic planning are carried out after the nursing task priority sequence is generated, scientific scheduling of the nursing tasks is achieved, medical resource waste and efficiency reduction caused by unreasonable scheduling are avoided, meanwhile, when the scheme deviation exceeds the threshold value, a rescheduling mechanism is triggered to dynamically update the scheme, and the scheduling efficiency is improved. The time and energy of manual re-planning are saved, and the accuracy and timeliness of nursing task scheduling are improved.

Description

Technical Field

[0001] This invention relates to the field of nursing task scheduling, and more specifically, to an AI-based method for scheduling medical and nursing tasks. Background Technology

[0002] Against the backdrop of the accelerated intelligent transformation of the healthcare industry, the complexity of hospital nursing work is increasing day by day. A single tertiary hospital can handle thousands of nursing tasks every day, covering multiple types such as basic nursing, emergency treatment, and examination assistance. Medical staff spend more than 8 hours a day on average handling these tasks. Nursing work has become a data-intensive and complex core link in the healthcare service system. The massive amount of nursing task information, medical resource data, and patient demand data provides data support for the intelligent scheduling of nursing tasks.

[0003] Currently, the main existing technologies used in the field of nursing task scheduling include manual scheduling methods and some basic information systems. Manual scheduling methods are usually developed by head nurses based on their personal experience, combined with factors such as the number of medical staff, their skills, and the patients' conditions, to create a nursing task allocation table. Basic information systems simply record nursing tasks electronically. In addition, some hospitals also use Excel spreadsheets for task statistics and allocation, and manually update the task execution status.

[0004] However, in practical use, it still has some shortcomings. The manual scheduling method relies too much on personal experience and cannot comprehensively and scientifically consider all influencing factors. It only roughly allocates tasks based on the number of medical staff and the patient's condition, ignoring key information such as the urgency of the task and the real-time status of resources. The rudimentary information system only records simple information and is difficult to achieve efficient matching of tasks and resources, resulting in unreasonable allocation of nursing tasks and some high-priority tasks not being processed in a timely manner, which affects the quality of nursing care. In actual application, nursing tasks and resource status are dynamic and affected by a variety of factors. In this case, the existing technology is difficult to respond to these changes in real time and cannot accurately adjust the scheduling plan in dynamic scenarios. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide an AI-based medical care task scheduling method, and the present invention provides the following technical solution:

[0006] S1: The nursing data acquisition module collects multi-source medical and nursing business data and uploads the raw data to the database;

[0007] S2: The data processing module extracts nursing task feature parameters from the database and generates a nursing task feature dataset;

[0008] S3: The AI ​​assessment module calls the task feature dataset, combines it with the pre-trained nursing task assessment model to assess each nursing task, and outputs a nursing task priority sequence.

[0009] S4: The resource scheduling module collects the real-time status of medical staff, the real-time inventory of medical resources, and the priority sequence of nursing tasks in real time, and performs AI matching and calculation to generate a preliminary nursing task scheduling plan.

[0010] S5: The task scheduling module uses dynamic programming algorithm to select the objective function and generate the final nursing task scheduling plan based on the preliminary task scheduling plan.

[0011] S6: The execution monitoring module collects task execution feedback data and transmits it to the AI ​​optimization module. The AI ​​optimization module calculates the plan deviation index. If the deviation exceeds the threshold, the rescheduling mechanism is triggered to dynamically update the nursing task scheduling plan.

[0012] S7: The results output module displays the final nursing task scheduling plan through the medical staff interactive terminal, supporting medical staff to view and provide feedback in real time.

[0013] The technical effects and advantages of this invention are as follows:

[0014] 1. This invention collects multi-source medical and nursing business data through a nursing data acquisition module, generates a nursing task feature dataset through a data processing module, and outputs a priority sequence through an AI evaluation module combined with a pre-trained model. It is not limited to traditional manual scheduling that relies solely on experience judgment. On the one hand, it can achieve accurate matching between nursing tasks and medical resources. On the other hand, it can comprehensively consider multiple factors such as task urgency and resource status, avoiding the one-sidedness of scheduling based on a single factor and improving the scientific nature of nursing task allocation.

[0015] 2. Based on the priority sequence of nursing tasks, the resource scheduling module and task scheduling module generate a final scheduling plan through AI matching calculation and dynamic programming algorithms. The plan clearly defines the execution order, responsible medical staff, and precise time for each task, improving the rationality and feasibility of the scheduling plan. The execution monitoring module collects feedback data in real time and transmits it to the AI ​​optimization module to achieve dynamic monitoring of the plan. The final plan is displayed through the medical staff interaction terminal to ensure that medical staff can obtain task information in a timely manner, thereby improving the efficiency of nursing task execution. At the same time, when the plan deviation exceeds the threshold, the rescheduling mechanism is triggered to dynamically update the plan, saving the time and effort of manual rescheduling and improving the response speed of scheduling to actual changes. On the other hand, since traditional static plans are difficult to deal with emergencies, the rescheduling mechanism gives the system dynamic adjustment capabilities, which can make the scheduling plan more in line with the actual execution scenario, thereby improving the quality of nursing services and patient satisfaction.

[0016] 3. After generating the final scheduling plan, this invention supports medical staff to view and provide feedback in real time through the result output module, while also allowing patients and their families to view relevant information. This enables multi-party sharing of scheduling information, avoids misunderstandings and conflicts caused by information asymmetry, reduces unnecessary communication costs, and improves the utilization efficiency of medical resources. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0018] Figure 2 This is a flowchart illustrating the overall structure of the present invention. Detailed Implementation

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] As attached Figure 1 The method for scheduling medical and nursing tasks based on AI includes a nursing data acquisition module, a database, a data processing module, an AI evaluation module, an AI optimization module, a resource scheduling module, a task scheduling module, an execution monitoring module, a result output module, and a medical and nursing interaction terminal.

[0021] The nursing data acquisition module is used to collect multi-source medical and nursing business data and upload the raw data to the database.

[0022] The database is used to store the collected raw data, preprocessed data, and nursing task priority sequences.

[0023] The data processing module is used to process raw data in the database;

[0024] The AI ​​assessment module is used to output a priority sequence for nursing tasks;

[0025] The resource scheduling module is used to generate preliminary nursing task scheduling plans;

[0026] The task scheduling module is used to generate the final nursing task scheduling plan;

[0027] The execution monitoring module is used to collect task execution feedback data;

[0028] The AI ​​optimization module is used to calculate the plan deviation index and update the nursing task scheduling plan.

[0029] The results output module is used to output the final nursing task scheduling plan;

[0030] The medical staff interaction terminal is used to present the final nursing task scheduling plan.

[0031] As attached Figure 2 The AI-based medical care task scheduling method shown includes the following steps in its specific implementation:

[0032] S1: The nursing data acquisition module collects multi-source medical and nursing business data and uploads the raw data to the database.

[0033] In this embodiment, it is important to specifically explain that the multi-source medical and nursing business data includes patient vital signs data, nursing needs, medical staff status, and medical resource information. This data, as raw data, is collected from multiple sources. Patient vital signs data is automatically collected through IoT devices, and differentiated devices are used for patients with different conditions. For critically ill patients, bedside multi-parameter monitors are used to monitor heart rate, blood pressure, blood oxygen saturation, and respiratory rate in real time. For ordinary inpatients, data such as baseline heart rate and activity level are collected through smart bracelets or patch sensors and synchronized to the database every 5 minutes. For patients in the recovery period, nurses manually enter daily blood pressure, blood sugar, and other data using portable vital signs monitors to supplement the patient's dynamic record.

[0034] Nursing needs data is generated through electronic medical orders issued by doctors and synchronized to the database via the hospital information system. The order clearly identifies the task type, execution time window, and operational procedures, such as "no food or water for 6 hours post-surgery." Immediate patient needs are initiated via ward calls, recording the call time, patient bed number, and a description of the need, such as "change IV bag" or "assist turning over," and are then pushed to the nurses' station terminal. Healthcare staff status data is collected through a personnel management system linked to location-tracking wristbands. Basic information, such as nurses' titles and skill qualifications, is stored in the hospital's human resources system and synchronized to the database. Real-time status is updated via RFID badges, including current location, task status, and current workload, with data refreshed every 30 seconds.

[0035] Medical resource information is collected in real time through the equipment management system and the inventory management system. Regarding the status of medical equipment, mobile devices such as infusion pumps and defibrillators are equipped with IoT positioning chips to provide real-time feedback on the status of "in use", "idle", and "under maintenance", with data coming from the medical equipment management platform. Fixed resources such as hospital beds and monitors are recorded as "occupied / idle" through the bed management system, for example, "Bed 503 is an empty bed after surgery, and the matching monitor is available". Drug and consumable information is connected to the hospital's SPD system, and inventory data is synchronized through barcode scanning.

[0036] It should be further noted that all relevant data collected in this embodiment is transmitted in encrypted form, and patient vital signs data does not carry patient identification information; sensitive data such as diagnostic information and allergy history in nursing needs are processed through de-identification algorithms to ensure that the data is usable but the identity is not identifiable; the database adopts a separate storage mode, with patient identity information stored separately in encrypted form on a special server and associated with business data through a hash index; access control is implemented on the stored raw data, and only authorized personnel can access the complete dataset through multi-factor authentication, while ordinary nurses can only view the de-identified data of the patients under their care.

[0037] S2: The data processing module extracts nursing task feature parameters from the database and generates a nursing task feature dataset.

[0038] In this embodiment, it should be specifically noted that the nursing task feature dataset includes task type, task operation complexity, task urgency, patient basic information, patient current vital signs, required medical and nursing skill tags, required medical resource type and quantity, and task time constraints.

[0039] It should be explained that the data processing module preprocesses the raw data in the database before extracting the feature parameters of the nursing task: First, data cleaning is performed, removing outliers in vital signs through threshold verification, and filling in missing values ​​using a combination of forward and backward imputation and the KNN algorithm; next, data standardization is performed, using the Z-Score formula to normalize numerical data and converting categorical data into vector representations through one-hot encoding; then, feature engineering is performed, calculating the urgency of the task using the formula Ej = α * medical risk level + β * time window urgency. The nursing dependence is quantified based on the patient's bedridden status and self-care ability score. Finally, data security desensitization is implemented, and the patient's identity information is encrypted using the SHA-256 hash algorithm, mapping the original ID to an irreversible feature code to ensure that the data does not leak privacy during subsequent processing. After preprocessing, the feature parameters of the above dimensions are integrated into a structured dataset by using the task ID as the index, forming a nursing task feature dataset that can be directly called by the AI ​​evaluation module.

[0040] S3: The AI ​​assessment module calls the task feature dataset and combines it with the pre-trained nursing task assessment model to assess each nursing task and output a nursing task priority sequence.

[0041] In this embodiment, it should be specifically noted that the nursing task priority sequence is generated based on a trained nursing task assessment model, and the specific steps are as follows:

[0042] a1: The AI ​​evaluation module calls the nursing task feature dataset and constructs these features into a feature matrix X = [x1, x2, x3, ..., xn] according to the input requirements of the pre-trained model, and its dimensions are consistent with those during model training;

[0043] As a preferred feasible embodiment, X is the feature matrix of the entire dataset, xi represents the feature vector of the i-th task, and the function of the feature matrix X is to feed data into the nursing task evaluation model.

[0044] a2: Load the pre-trained XGBoost nursing task evaluation model, call the model's predict_proba() method, output the multi-class probability distribution P(yi|xi) for each task i, if the confidence of the highest probability class is less than 0.7, then mark the task as "uncertain task" and trigger the manual review process;

[0045] As a preferred and feasible embodiment, the specific training steps of the pre-trained XGBoost nursing task assessment model are as follows:

[0046] b1: Construct the training dataset;

[0047] b2: Process different types of data in the training dataset differently, and divide the dataset into training set, validation set and test set according to a certain ratio;

[0048] b3: The XGBoost algorithm was selected to construct a nursing task assessment model, and the objective function was set. l(yi, xi) is the cross-entropy loss function. Regularization term to prevent overfitting;

[0049] b4: Use the training set to iteratively train the model. After every 50 rounds of training, use the validation set to calculate the model's accuracy, recall, and F1-score.

[0050] b5: When the model's F1-score on the validation set improves by less than 0.01 for 10 consecutive rounds, stop training. Finally, use the test set to evaluate the model's generalization ability. If the test set accuracy is below 85%, return to step b1 to supplement data and retrain.

[0051] a3: Prioritize the model output results and adjust them using hard rules applied by the rule engine;

[0052] It needs to be explained that "hard rules" refer to some inviolable constraints based on medical industry standards, patient safety bottom lines, and basic principles of resource allocation. For example, emergency rescue tasks are directly upgraded to Level I; priority calibration is to upgrade tasks that are about to expire by one priority level in combination with task time constraints; and resource constraints are also taken into account. If the resources required for a task are currently unavailable, its priority is reduced by one level.

[0053] a4: Arrange the calibrated tasks in descending order of priority level, and arrange tasks of the same level in ascending order of expected time, forming the final nursing task priority sequence S = [t1, t2, t3, ..., tn].

[0054] It should be explained that ti represents the i-th task after sorting, and includes a priority label and a suggested execution time window. The sequence output format is a JSON array.

[0055] S4: The resource scheduling module collects real-time data on the status of medical staff, real-time inventory of medical resources, and priority sequence of nursing tasks, and performs AI matching calculations to generate a preliminary nursing task scheduling plan.

[0056] In this embodiment, the specific process of the AI ​​matching calculation needs to be explained as follows:

[0057] c1: Converts real-time medical staff status into skill tag vectors, load vectors, and location vectors; resource availability matrix; task skill requirements, resource requirements, and urgency weights;

[0058] As a preferred feasible embodiment, it should be explained that the skill tag vector includes intravenous puncture and first aid certification; the load vector is... Position vector refers to The resource availability matrix is ​​transformed from the medical resource inventory, including equipment ID, usage duration, and storage location; the task skill requirements are extracted from the priority sequence output in S3; the resource requirement types include ventilators and blood glucose meters; the urgency weight corresponds to different coefficients according to priority levels I-IV.

[0059] c2: Use cosine similarity to roughly screen personnel for skill matching, then use an integer programming model to calculate scheduling costs and generate a preliminary plan;

[0060] It should be explained that the formula for calculating the matching degree between skill tag vectors and task skill requirements using cosine similarity is: A set of medical staff with a matching degree > 0.6 is selected, where Ski is the skill tag vector of the i-th medical staff member, and Ti is the skill requirement vector of the j-th nursing task. The dot product of the vectors is used to calculate the number of skills that overlap between the two. ||Ski|| and ||Ti|| are the magnitudes of the two vectors, respectively. This formula can be used to quantify the skill matching degree into a value between 0 and 1. A matching degree > 0.6 means that the medical staff member has the core skills required for the task, ensuring that the staff's skills meet the basic requirements of the task.

[0061] It should be noted that the formula for the integer programming model is: In the model, xij = 1 indicates that the i-th medical staff member is assigned to the j-th task, and xij = 0 indicates no assignment. cij represents the scheduling cost, which consists of three parts: first, the distance cost, which is the straight-line distance from the medical staff member's current location to the patient's ward multiplied by 0.3, in meters; second, the skill premium cost, which increases by 5% of the base cost for every 0.1 decrease in matching degree when the matching degree is ≤0.8, to compensate for the operation time that may be caused by skill gaps; and third, the resource coordination cost, which is the distance between the medical staff member and the equipment storage point multiplied by 0.2 if the task requires the use of specific equipment. Constraints include... This means that only one medical staff member will be assigned to each task. This represents the maximum number of tasks that the i-th medical staff member can undertake, where ki is determined by the current workload. By solving this model, the staff member with the lowest cost is matched from the coarsely screened set of medical staff members for each task, while the medical resources required for the task are associated to form a preliminary plan.

[0062] c3: Verify time conflicts, resource exclusivity conflicts, and local neighborhood search to repair conflicts, and output a preliminary nursing task scheduling plan.

[0063] It should be explained that time conflict verification is achieved through task time window overlap calculation: each task is labeled with an expected execution time window [Tstart, Tend], and for multiple tasks assigned to the same medical staff, the time overlap rate is calculated. j and k are two tasks performed by the same person.

[0064] It needs to be further explained that the resource exclusive conflict verification is for non-shareable medical resources. The resource allocation matrix tracks the usage time of the equipment. If the same resource is allocated to two tasks with overlapping time windows, it is judged as a resource exclusive conflict. The resource allocation of the higher-weight task is retained according to the task priority weight.

[0065] When resolving conflicts in a local neighborhood search, for time-conflicting tasks, alternative personnel with a skill matching difference of ≤0.1 and a current load rate of <60% are selected from the coarsely screened set of medical staff. The scheduling cost is recalculated and a suboptimal solution is selected. For resource-exclusive conflict tasks, idle resources of the same type are searched first. If no alternative resources are available, the task time window is adjusted. The conflict rate is re-verified after each resolution until the conflict rate is <5%. The final preliminary solution output includes the task execution order, the binding relationship between medical staff and resources, and the revised time window. A conflict resolution log is generated simultaneously.

[0066] S5: The task scheduling module uses dynamic programming algorithm to select an objective function and perform scheduling path planning based on the preliminary task scheduling plan to generate the final nursing task scheduling plan.

[0067] In this embodiment, it should be specifically explained that the final nursing task scheduling scheme is a structured scheme that includes the execution order of each nursing task, the responsible medical staff, the required medical resources, the precise execution time, the movement path planning, and the emergency backup scheme. It can be directly synchronized to the mobile terminals of medical staff as the basis for the execution of nursing work.

[0068] It needs to be explained that the core of the dynamic programming algorithm is to achieve global path optimization by constructing a state transition equation, the formula being dp[i][k]=MINm∈M{dp[i-1][m]+Cost(m,k,i-1,i)}, where M is the set of all medical staff, and the state is defined as dp[i][k], representing the minimum cumulative cost of the kth medical staff when processing the i-th task. The state transition depends on the execution state dp[i-1][m] of the previous task, where m is the medical staff executing the (i-1)-th task, and the transition cost Cost(m,k,i-1,i) includes the movement time of medical staff m from the (i-1)-th task location to the i-th task location, the actual walking distance calculated based on the hospital floor map, the task switching time, and the resource transfer cost. If the (i-1)-th and i-th tasks share equipment, the time taken to move the equipment is also considered. After processing all n tasks, the global minimum scheduling cost is TotalCost=MINk∈M{dp[n][k]}.

[0069] Further explanation is needed regarding the objective function, which is set as minZ = w1*T + w2*D + w3*R, where T is the total delay time for all tasks, which is the sum of the differences between the actual execution time and the planned time. D represents the total movement distance of medical staff, and R represents the idle time of resources, which is the waiting time from equipment allocation to actual use. The weights are usually set as follows: w1 = 0.5, w2 = 0.3, w3 = 0.2. The algorithm determines the optimal execution order of each task by tracing the optimal state path in reverse. For example, tasks in the same ward are executed in a concentrated manner to reduce movement distance. The optimal movement route for medical staff is determined, such as prioritizing vertical elevators over stairs to shorten the time to cross floors. A 5-minute buffer time is reserved for high-priority tasks to avoid delays in previous tasks affecting subsequent execution.

[0070] S6: The execution monitoring module collects task execution feedback data and transmits it to the AI ​​optimization module. The AI ​​optimization module calculates the plan deviation index. If the deviation exceeds the threshold, the rescheduling mechanism is triggered to dynamically update the nursing task scheduling plan.

[0071] In this embodiment, the calculation process of the scheme deviation index needs to be explained in detail as follows:

[0072] d1: Calculate the task delay deviation Dt:

[0073]

[0074] Dt is the average delay time per task, t is the total number of nursing tasks performed, and T is the total number of nursing tasks performed. 实际完成时间，i T is the actual completion time of the i-th task. 计划完成时间，i It is the planned completion time of the i-th task;

[0075] d2: Calculate resource idle deviation Dr:

[0076]

[0077] Dr is the average idle time per device, T 实际使用，j T is the actual start time of the j-th device, r is the total number of medical devices in use, and T is the total number of devices in use. 计划使用，j This is the planned start date for the j-th device;

[0078] d3: Calculate the manpower load deviation Dh:

[0079]

[0080] Dh is the deviation rate between the actual workload and the plan for each medical staff member, h is the total number of medical staff members on duty, actual working time k is the total actual working time of the kth medical staff member, and planned working time k is the total planned working time of the kth medical staff member.

[0081] d4: Calculate path deviation Dp:

[0082]

[0083] Dp is the deviation rate between the actual distance traveled and the planned distance traveled for each task, p is the total number of nursing tasks performed, actual distance traveled i is the actual distance traveled for the i-th task, and planned distance traveled i is the planned distance traveled for the i-th task.

[0084] d5: Calculate priority deviation Dprio:

[0085]

[0086] Dprio is the proportion of resources that low-priority tasks take from high-priority tasks.

[0087] d6: The five types of deviation indicators are combined into a single plan deviation index I by weighted summation, with the weights adjusted by the hospital according to management needs.

[0088] I=wt*Dt+wr*Dr+wh*Dh+wp*Dp+wprio*dprio,

[0089] I reflects the degree of deviation from the overall scheduling plan. wt, wr, wh, wp, and wprio are the weights of various deviations, and the specific values ​​are determined by the hospital itself.

[0090] It should be explained that if the deviation exceeds the threshold, the AI ​​optimization module will extract the actual execution data, the data of currently incomplete tasks, and the latest resource status, reuse the algorithm framework of preliminary matching and dynamic programming, but only recalculate for the affected subset of tasks to generate an update plan with minimal changes. Subsequently, it will push the change task reminder through the medical staff terminal to realize the dynamic update of the nursing task scheduling plan.

[0091] S7: The results output module displays the final nursing task scheduling plan through the medical staff interactive terminal, supporting medical staff to view and provide feedback in real time.

[0092] The final nursing task scheduling plan adopts a hierarchical visibility mode, which is simultaneously displayed through medical staff interaction terminals, patient and family terminals. Not only can medical staff view it, but patients and their families can also view the corresponding content with patient authorization. It should be explained that when the final nursing task scheduling plan is generated or dynamically updated, the system will automatically trigger hierarchical message push: the medical staff terminal pushes a prompt with a "New task pending confirmation" label through the hospital's internal communication system. Clicking on it will directly jump to the terminal to view the complete plan, which includes full information such as task execution order, time windows accurate to the minute, resource allocation details, and emergency backup plans. It also supports medical staff to provide feedback such as "task completed ahead of schedule" and "temporary resource shortage". The patient and family terminal pushes a simplified prompt through the ward smart screen pop-up, the bound SMS or mini-program, which reads "Your nursing plan has been updated: XX nursing care will be performed at XX time. You can view the details on the ward screen". The displayed content is only the timeline of key nursing actions related to the patient, such as specific nursing items, execution time periods and desensitized medical staff information, hiding professional details such as resource allocation to avoid information overload.

[0093] It needs further explanation that the solution demonstration also supports a two-way feedback loop: medical staff can mark task status and update the execution progress in real time; patients and their families can submit "time conflict applications" through a mini-program, and the AI ​​optimization module will automatically identify the conflict type. If it is a low-priority task, the time will be adjusted directly; if it is a high-priority task, it will trigger manual review. After the nurse confirms, the adjustment result will be fed back, which ensures both the professionalism of scheduling and the patient experience. At the same time, through information anonymization and authorized viewing, medical privacy and information compliance are ensured.

[0094] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0095] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-based method for scheduling medical care tasks, characterized in that, The method includes a nursing data acquisition module, a database, a data processing module, an AI evaluation module, an AI optimization module, a resource scheduling module, a task scheduling module, an execution monitoring module, a result output module, and a medical-nursing interaction terminal. The specific steps of the method are as follows: S1: The nursing data acquisition module collects multi-source medical and nursing business data and uploads the raw data to the database; S2: The data processing module extracts nursing task feature parameters from the database and generates a nursing task feature dataset; S3: The AI ​​assessment module calls the task feature dataset, combines it with the pre-trained nursing task assessment model to assess each nursing task, and outputs a nursing task priority sequence. S4: The resource scheduling module collects the real-time status of medical staff, the real-time inventory of medical resources, and the priority sequence of nursing tasks in real time, and performs AI matching and calculation to generate a preliminary nursing task scheduling plan. S5: The task scheduling module uses dynamic programming algorithm to select the objective function and generate the final nursing task scheduling plan based on the preliminary task scheduling plan. S6: The execution monitoring module collects task execution feedback data and transmits it to the AI ​​optimization module. The AI ​​optimization module calculates the plan deviation index. If the deviation exceeds the threshold, the rescheduling mechanism is triggered to dynamically update the nursing task scheduling plan. S7: The results output module displays the final nursing task scheduling plan through the medical staff interactive terminal, supporting medical staff to view and provide feedback in real time.

2. The AI-based medical care task scheduling method according to claim 1, characterized in that: The nursing task feature dataset includes task type, task operation complexity, task urgency, patient basic information, patient current vital signs, required medical and nursing skill tags, required medical resource type and quantity, and task time constraints.

3. The AI-based medical care task scheduling method according to claim 1, characterized in that: The nursing task priority sequence is generated based on a trained nursing task assessment model, and the specific steps are as follows: a1: The AI ​​assessment module calls the nursing task feature dataset and constructs these features into a feature matrix X = [x1, x2, x3, ..., xn] according to the input requirements of the pre-trained model; a2: Load the pre-trained XGBoost nursing task evaluation model and output a multi-class probability distribution P(yi|xi) for each task i; a3: Prioritize the model output results and adjust them using hard rules applied by the rule engine; a4: Arrange the calibrated tasks in descending order of priority level, and arrange tasks of the same level in ascending order of expected time, forming the final nursing task priority sequence S = [t1, t2, t3, ..., tn].

4. The AI-based medical care task scheduling method according to claim 3, characterized in that: The specific training steps for the pre-trained XGBoost nursing task assessment model are as follows: b1: Construct the training dataset; b2: Process different types of data in the training dataset differently, and divide the dataset into training set, validation set and test set according to a certain ratio; b3: The XGBoost algorithm was selected to construct a nursing task assessment model, and the function was defined. b4: Use the training set to iteratively train the model. After every 50 rounds of training, use the validation set to calculate the model's accuracy, recall, and F1-score. b5: When the model's F1-score on the validation set improves by less than 0.01 for 10 consecutive rounds, stop training and finally use the test set to evaluate the model's generalization ability.

5. The AI-based medical care task scheduling method according to claim 1, characterized in that: The specific process of AI matching calculation is as follows: c1: Convert the real-time status of medical staff into skill tag vectors, load vectors, location vectors, resource availability matrix, task skill requirements, resource requirements, and urgency weights; c2: Use cosine similarity to roughly screen personnel for skill matching, then use an integer programming model to calculate scheduling costs and generate a preliminary plan; c3: Verify time conflicts, resource exclusivity conflicts, and local neighborhood search to repair conflicts, and output a preliminary nursing task scheduling plan.

6. The AI-based medical care task scheduling method according to claim 5, characterized in that: The formula for the integer programming model is: xij = 1 indicates that the i-th medical staff member is assigned to the j-th task, xij = 0 indicates that no task is assigned, and cij is the scheduling cost.

7. The AI-based medical care task scheduling method according to claim 1, characterized in that: The core of the dynamic programming algorithm is to achieve global path optimization by constructing a state transition equation, which is dp[i][k]=MINm∈M{dp[i-1][m]+Cost(m,k,i-1,i)}, where M is the set of medical staff, dp[i][k] is the state definition, dp[i-1][m] is the execution state of the previous task, m is the medical staff executing the (i-1)th task, and the transition cost is Cost(m,k,i-1,i).

8. The AI-based medical care task scheduling method according to claim 1, characterized in that: The calculation process for the scheme deviation index is as follows: d1: Calculate the task delay deviation Dt: t is the total number of nursing tasks performed, T 实际完成时间，i T is the actual completion time of the i-th task. 计划完成时间，i It is the planned completion time of the i-th task; d2: Calculate resource idle deviation Dr: T 实际使用，j T is the actual start time of the j-th device, r is the total number of medical devices in use, and T is the total number of devices in use. 计划使用，j This is the planned start date for the j-th device; d3: Calculate the manpower load deviation Dh: h is the total number of medical staff on duty, actual working hours k is the total actual working time of the kth medical staff, and planned working hours k is the total planned working time of the kth medical staff. d4: Calculate path deviation Dp: p is the total number of nursing tasks performed, actual distance i is the actual distance traveled for the i-th task, and planned distance i is the planned distance traveled for the i-th task. d5: Calculate priority deviation Dprio: d6: The five types of deviation indicators are combined into a single plan deviation index I by weighted summation, with the weights adjusted by the hospital according to management needs. I=wt*Dt+wr*Dr+wh*Dh+wp*Dp+wprio*dprio, wt, wr, wh, wp, and wprio are the weights of various deviations.

Images