A method and system for dynamic priority scheduling of nursing resources in elderly care institutions

Abstract

This invention discloses a dynamic priority scheduling method and system for nursing resources in elderly care institutions, belonging to the field of elderly care resource scheduling technology. It includes: acquiring historical scheduling data and classifying it into scenarios to establish a scenario data set; performing serialized modeling of the scheduling process, generating scheduling behavior trajectories and converting them into a behavior pattern knowledge base; dynamically calculating task priorities through an information flow network; optimizing task execution time windows based on priority and time constraints; establishing a multi-objective scheduling optimization model to solve for the optimal scheduling scheme, and adaptively updating model parameters based on actual execution results; performing cross-scenario adaptation and generating interpretable scheduling decision information. This invention dynamically responds to changes in the state of the nursing recipients, achieving adaptive evolution of task priorities, balanced resource allocation, and visualization of the scheduling process, improving the timeliness of nursing response, resource utilization efficiency, and interpretability of scheduling decisions in elderly care institutions.

Description

Technical Field

[0001] This invention relates to the field of elderly care resource scheduling technology, specifically to a dynamic priority scheduling method and system for elderly care institution nursing resources. Background Technology

[0002] Traditional methods of allocating nursing resources in elderly care institutions often employ fixed priorities or simple first-come-first-served strategies, which make it difficult to respond in real time to dynamic changes in the physiological indicators and behavioral states of those being cared for. This results in delayed responses to high-urgent tasks, uneven utilization of nursing resources, and inconsistent service quality.

[0003] Existing scheduling methods lack the ability to extract contextual information and reuse knowledge from historical scheduling behaviors. Task priority calculation lacks an adaptive evolution mechanism. Fixed time windows cannot adapt to differences in task urgency. Furthermore, the scheduling process lacks interpretability, making it difficult to meet the needs of modern elderly care institutions for refined and intelligent nursing management. Summary of the Invention

[0004] The purpose of this invention is to address the problems in the existing scheduling of nursing resources in elderly care institutions, such as fixed task priorities, inability to dynamically adapt to changes in the status of care recipients, uneven resource allocation, lack of interpretability in scheduling decisions, and poor cross-scenario reusability. Therefore, this invention proposes a dynamic priority scheduling method and system for nursing resources in elderly care institutions.

[0005] The objective of this invention can be achieved through the following technical solution: This invention provides a method for dynamic priority scheduling of nursing resources in elderly care institutions, comprising: S001: Obtain historical scheduling records, task execution logs, and status monitoring data; perform scenario segmentation to obtain a scenario data set; S002: Based on the contextual data set, perform serialization modeling to generate a set of scheduling behavior trajectories, and then obtain a behavior pattern knowledge base through atomic behavior decomposition, four-element data unit encoding and weighted aggregation; S003: Establish a directed information flow network based on the behavior pattern knowledge base, and calculate the basic priority results of each task through the dynamic evolution equation of node information; S004: Based on priority results and time constraints, the optimal execution time is obtained through execution benefit calculation. The time window width is dynamically adjusted and an overlap detection and compression process is established to obtain the optimal execution time window for each task. S005: Establish a multi-objective optimization model that includes weighted completion time, resource load balancing, and priority violation penalties to obtain the optimal scheduling scheme; S006: Obtain the actual execution result data of the scheduling plan, analyze the tasks, and obtain scheduling effect indicators; S007: Online adaptive updates based on scheduling performance metrics; S008: Based on the updated behavior pattern knowledge base, cross-scenario adaptation is performed, and when outputting the scheduling scheme, structured explanation information of task priority composition, resource matching process and rule triggering path is generated and displayed.

[0006] As a preferred embodiment of the present invention, the specific process of scenario division includes: Each data unit includes the task trigger time, the nursing object's state vector, the resource state vector, and the execution result. A state-resource coupling perturbation function is established, which combines the nursing object's state vector and the resource state vector by performing feature cross-interaction and then superimposing the product of the state change gradient and the time sensitivity coefficient. An asymmetric situation potential function is established, which is obtained by weighting the norm square of the coupled feature vector, the resource distribution entropy, and the execution result deviation. Through a situation segmentation boundary function, tasks with potential values ​​less than the first threshold are classified as low-load situations, those between the first and second thresholds are classified as medium-load situations, and those greater than or equal to the second threshold are classified as high-urgency situations. The situation data units are unified and a hierarchical index structure is established to obtain the situation data set.

[0007] As a preferred embodiment of the present invention, the specific process of obtaining the behavior pattern knowledge base includes: Based on the contextual data set, a discrete-time state sequence is generated, including task selection sequence, resource allocation path, and response time sequence; a scheduling path energy function is established; a set of scheduling behavior trajectories is generated through energy convergence; the trajectory is decomposed into atomic behavior units, and four-element data units are established; after normalization and position encoding, a unified encoding function is used to map the data to behavior encoding vectors, and weighted aggregation is used to obtain behavior pattern units, forming a behavior pattern knowledge base.

[0008] As a preferred embodiment of the present invention, the specific process for obtaining the basic priority results of each task includes: The behavior patterns are mapped to information source nodes, and the information source strength is calculated. A directed information flow network is established, and a dynamic evolution equation for node information is established. Based on the information inflow, outflow, and information dissipation coefficient, the process is iterated until convergence to obtain a stable amount of information. The node information is aggregated to the task layer through pattern task space mapping, and after processing by priority function and stabilization function, the basic priority results of each task are obtained.

[0009] As a preferred embodiment of the present invention, the specific process of obtaining the optimal execution time window for each task includes: Obtain the task priority and time constraint set, discretize the time domain into candidate times, calculate the execution benefits to obtain the initial optimal execution time; obtain the precise optimal execution time through parabolic interpolation; dynamically adjust the window width according to priority; based on the optimal execution time window and constrained by the earliest executable time and deadline; establish a window overlap detection mechanism, for tasks with overlapping resource requirements and intersecting windows, compress the window according to priority until the conflict is eliminated.

[0010] As a preferred embodiment of the present invention, the specific process of obtaining the optimal scheduling scheme includes: A binary decision variable is set to represent the allocation relationship between tasks and resources, and a continuous decision variable is set to represent the task start time. A multi-objective optimization model is established, which includes three sub-objectives: weighted completion time, resource load variance, and priority violation penalty. The model is then transformed into a single objective by weighted summation. Tasks are sorted in descending order of priority according to constraints. For each task, feasible resources are traversed, the local objective function increment is calculated, and the resource with the smallest increment is selected for allocation to obtain the optimal scheduling scheme.

[0011] As a preferred embodiment of the present invention, the specific process of online adaptive updating based on scheduling performance indicators includes: To obtain actual execution results, three scheduling performance indicators are established: task completion time, resource utilization, and service quality. These indicators are then integrated to obtain a comprehensive scheduling performance indicator. The weight coefficients in the scheduling path energy function are updated using gradient descent. The information dissipation coefficient is adjusted based on the resource conflict frequency. The compression coefficient is adjusted based on the priority distribution variance, and the parameters are restricted to a stable range.

[0012] As a preferred embodiment of the present invention, the specific process of performing cross-scene adaptation and generating structured interpretation information includes: For new scenarios, extract contextual feature vectors and perform similarity matching with contextual items in the behavior pattern knowledge base. Select the top three behavior patterns with the highest similarity as references, use their scheduling parameters as initial parameters, and quickly fine-tune them. When outputting the scheduling scheme, generate three aspects of explanatory information: task priority composition explanation, listing the behavior pattern nodes that contribute the most and their mapping weights; resource matching process explanation, listing the indicator comparison of candidate resources and the reasons for selection; rule triggering path explanation, tracing back the key rule chain; and display them.

[0013] Another aspect of the present invention provides a dynamic priority scheduling system for nursing resources in elderly care institutions, comprising: a data acquisition and processing module, a scheduling behavior module, a task classification module, a task execution optimization module, a scheduling optimization module, and a cross-scenario adaptation module; The data acquisition module obtains historical scheduling records and status monitoring data, performs scenario segmentation, and obtains a scenario data set.

[0014] The scheduling behavior module performs serialization modeling of tasks based on the contextual data set, generates a set of scheduling behavior trajectories, and then obtains a behavior pattern knowledge base through atomic behavior decomposition.

[0015] The task classification module establishes a directed information flow network based on the behavioral knowledge base, and calculates and processes the basic priority results of each task through the dynamic evolution equation of node information.

[0016] The task execution optimization module calculates the precise optimal execution time based on priority results and time constraints, dynamically adjusts the time window width, and establishes an overlap detection and compression mechanism to obtain the optimal execution time window for each task.

[0017] The scheduling optimization module establishes a multi-objective optimization model, calculates the optimal scheduling scheme, and performs online adaptive updates based on the actual execution results.

[0018] The cross-scenario adaptation module achieves cross-scenario adaptation through context similarity matching and rapid parameter fine-tuning. When outputting the scheduling scheme, it generates and displays structured explanation information on task priority composition, resource matching process, and rule triggering path.

[0019] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention determines the precise optimal execution time by performing revenue calculation and parabolic interpolation, and dynamically adjusts the time window width according to priority, establishing a window overlap detection and compression mechanism to enable high-priority tasks to obtain a narrower execution window, eliminate multi-task resource conflicts, and improve the on-time completion rate of tasks.

[0020] 2. Based on this, the present invention establishes a multi-objective optimization model that includes weighted completion time, resource load balancing and priority violation penalties. Resources are allocated in descending order of priority to match the skill matching of nursing staff with task requirements, avoid overloading some staff while leaving others idle, and improve the overall resource utilization efficiency.

[0021] 3. This invention also introduces an adaptive update mechanism based on actual execution results. It updates the energy function weights through gradient descent, adjusts the information dissipation coefficient according to the conflict rate, and adjusts the compression coefficient according to the priority distribution variance, so that the scheduling strategy can dynamically evolve with the nursing scenario and continuously optimize the scheduling performance.

[0022] 4. This invention generates structured explanations of task priority composition, resource matching process, and rule triggering paths when outputting scheduling schemes, and visualizes them in natural language, solving the black-box problem of scheduling decisions and enhancing caregivers' trust and acceptability of the system. After adopting the above method, the emergency response time for nursing tasks in elderly care institutions is shortened, resource load balancing is improved, service quality indicators are enhanced, and it can quickly adapt to new scenarios of different scales and distributions of nursing needs, demonstrating good portability and practical value. Attached Figure Description

[0023] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0024] Figure 1 This is a diagram illustrating the method steps of the present invention; Figure 2 This is a schematic diagram of the principle of the present invention. Detailed Implementation

[0025] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0026] It should be understood that the terms “comprising” and “including” used in this disclosure and claims indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0027] It should also be understood that the terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the disclosure. As used in this disclosure and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this disclosure and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0028] Please see Figure 1 As shown, this invention provides a method for dynamic priority scheduling of nursing resources in elderly care institutions, comprising: S001: Obtain historical scheduling records, task execution logs, and status monitoring data during the nursing process in elderly care institutions. Perform contextualization and standardization on the data to obtain a contextual data set containing task triggering conditions, resource status, and execution results.

[0029] S002: Based on the contextual data set, the task allocation order, resource selection path and response time in the nursing resource scheduling process are serialized and calculated to obtain the corresponding scheduling behavior trajectory set.

[0030] S003: Perform a structured transformation on the set of scheduling behavior trajectories to obtain the corresponding data unit representation, and encode each data unit uniformly to obtain a behavior pattern knowledge base.

[0031] S004: Based on the behavioral pattern knowledge base, perform association rule mining to extract the mapping relationship between contextual features and task priorities, and establish a task priority analysis function containing rule enhancement items to obtain the basic priority results of each task.

[0032] S005: Based on the task priority results and task time constraint information, perform task execution timing optimization calculations to obtain the optimal execution time window for each task.

[0033] S006: Based on the task priority results and the optimal execution time window, the optimization objective is obtained by performing multi-objective nursing resource scheduling optimization analysis, and the nursing resource allocation relationship is solved to obtain the optimal scheduling scheme.

[0034] S007: Obtain the actual execution result data of the scheduling scheme, analyze the task completion time, resource utilization and service quality, obtain scheduling effect indicators, and adaptively update the rule weights and parameters based on the scheduling effect indicators.

[0035] S008: Based on the updated behavior pattern knowledge base and scheduling analysis process, cross-scenario adaptation applications are performed, and corresponding decision explanation information is generated when outputting scheduling schemes; the task priority composition, resource matching process and rule triggering path are visualized.

[0036] The specific process of performing contextualization and standardization on the data is as follows: Structured mapping is performed on historical scheduling records and status monitoring data to establish the original data sequence D={d i i | i = 1, 2, ..., N}, d i =(t i X i R i Y i ), where d i Let t be the i-th task data; N is the total number of task data;i For the task trigger time, X i Let X be the state vector of the nursing object. i =[x i 1 x i 2 , ..., x i k , ..., x i m ], x i k Let m be the k-th state feature, such as a physiological indicator or behavioral state, and m be the number of dimensions of the state feature; R i Let R be a resource state vector. i =[r i 1 r i 2 , ..., r i j , ..., r i n ], r i j Y represents the j-th resource feature, such as the number of personnel, skill type, or workload; n represents the number of resource feature dimensions; i The result indicates the task completion status.

[0037] Based on this, the state-resource coupling perturbation function is established: Φ i =X i R i +κ· t X i , where Φ i This represents the fused coupled feature vector. Represents the feature cross operator, t X i This represents the gradient of state change over time, where κ is the time sensitivity coefficient, characterizing the strength of the impact of state change on task triggering. A specific numerical example: κ = 0.6, indicating that the state change gradient is superimposed on the feature cross result with 60% intensity. This value is obtained through regression analysis of task triggering delay and state change rate in historical data. It should be noted that X... i R i The result is a vector of dimension m×n, whose (k, j)th component is: x i k ·r i j .

[0038] Subsequently, an asymmetric situational potential function, E, is established based on the coupling characteristics. i =α·||Φ i || 2 2+β·H(R i ) γ·ΔY i , of which E i The potential energy function describes the scheduling stress of the current task in the system, H(R). i The resource distribution entropy, which characterizes the degree of resource equilibrium, is expressed as: , Let ΔY be the proportion of resource type j. i The deviation from the expected result is the difference between the actual result and the expected result. α, β, and γ are adjustment parameters. Specific numerical examples: β=0.3, α=0.4, γ=0.2 (the remaining 0.1 is the system margin). ||Φ i || 2 2 represents the 2-norm square of the fused coupled feature vector, which is the sum of the squares of all elements in the fused coupled feature vector.

[0039] Furthermore, based on the potential energy function, scenarios are divided, and a scenario segmentation boundary function is established: ,in, Let C1, C2, and C3 represent different situation categories, corresponding to low load, medium load, and high urgency, respectively; θ1 and θ2 are the situation classification thresholds; specific numerical examples: θ1 = 0.35, θ2 = 0.70 (potential energy dimension and E...). i (Same as above) These two thresholds are taken from the 30th and 70th percentiles of the potential energy distribution in historical scheduling records, thus obtaining the scenario set C={C1, C2, C3, ...}; After scenario segmentation, the original data undergoes condition reconstruction to establish a task triggering condition vector: Cond i =g(X i , t X i , t i ), Cond i This represents the task trigger condition vector, where the function g(·) integrates state values, trends, and time factors; it establishes a resource state representation: Res i =R i ⊕Ψ(R i ), where Res i Indicates the resource status, Ψ(R) i ) represents the resource load disturbance function, reflecting the degree of uneven resource utilization; ⊕ represents the concatenation operator; the execution result is uniformly represented as: Out i =(T iQ i Among them, T i For response time, Q i For quality indicators.

[0040] A specific numerical example of the function g(·): g(x, ,t)=0.5x+0.3 +0.2×(t / t e ), where t e =300 seconds; when the current state value x=0.8, the trend of change =0.1 units / second, and the elapsed time t=150 seconds, g=0.5×0.8+0.3×0.1+0.2×(150 / 300)=0.4+0.03+0.1=0.53.

[0041] Finally, the above data is reconstructed into a unified contextual data unit: U i =(C k Cond i Res i Out i E i ), and establish a hierarchical index structure based on context categories to create a context data set U={U i This is used for subsequent priority calculation and scheduling model calls.

[0042] The process of sequentially modeling the task allocation sequence, resource selection path, and response time in the nursing resource scheduling process is as follows: Based on the contextual data set, the scheduling process is reconstructed in chronological order to generate a discrete-time state sequence: S={S t |t=1,2,…,T}, where: S t Let S be the system state at time t; T be the total length of the scheduling sequence; the state is defined as: S t =(X t R t Q t ), where: X t R is the current state vector of the nursing object; t Q represents the available resource state vector at the current moment. t This is the queue of tasks to be executed.

[0043] After obtaining the state sequence S, the scheduling decision at each time step is explicitly variableized, and the task selection sequence is defined as: π = {π t |t=1,2,…,T}; Resource allocation path: ρ={ρ t |t=1, 2, …, T}; Response time series: τ={τ t |t=1,2,…,T}.

[0044] After obtaining the scheduling decision sequence variables (π, ρ, τ), the scheduling path energy function is established: ,in, ω1 represents the energy value of the entire scheduling trajectory; the smaller the value, the better. ω2, ω3, and ω4 are weighting coefficients. Choose a cost function for the task, denoted as: , where E(π) t ) represents the contextual potential energy corresponding to the task, calculated based on actual sensor data; Rank(π) t Q t ) represents the original sorting position of the task in the queue; λ1 represents the sorting weight. For example, λ1=0.7 means that the weight of the urgency sub-objective is 70% and the weight of the resource consumption sub-objective is 30%, which is obtained by combining the analytic hierarchy process (AHP) with the scoring of nursing experts. Let the resource matching cost function be expressed as: , where f(ρ t ) is the resource capability vector mapping function; g(π) t ) is the task requirement vector mapping function; The response time cost function is expressed as: ,in For reference response time; Let be the state transition perturbation function, expressed as Where η is the resource change weighting coefficient, for example: η=0.5, indicating that the current resource status and the resource change rate each account for 50% of the weight. This weighting design is suitable for nursing environments with relatively stable resource fluctuations, and the optimal value is obtained through sliding window verification. It represents the squared 2-norm (squared Euclidean distance).

[0045] Based on the established global energy function, the local energy change ΔE is calculated at each moment. t ΔE t =E t E t 1; At each time step, a scheduling sequence is generated by minimizing the path energy function: ;Through a stepwise energy convergence update strategy: at time t, choose: ,in: This indicates the incremental impact of the current choice on the overall energy.

[0046] The obtained scheduling sequences are accumulated step by step to form a complete sequence: (π1, ρ1, τ1) to (π2, ρ2, τ2)……(π T , ρ T , τT ).

[0047] Establish scheduling behavior trajectories based on complete sequences: Γ k1 ={(π t , ρ t , τ t |t=1,2,…,T k1}, then the set of trajectories is: G={Γ k1 |k1=1,2,…,K1}.

[0048] The specific process of performing a structured transformation on the set of scheduling behavior trajectories is as follows: Obtain the set G of scheduling behavior trajectories obtained in the previous step. Further, each scheduling behavior trajectory is represented as: After obtaining the set of scheduling behavior trajectories, each trajectory is analyzed in chronological order, breaking down the continuous trajectory into discrete atomic behavior units: Γ k1，t =(π t , ρ t , τ t S t ), where Γ k1，t S represents the atomic behavior unit of the k1th trajectory at time t. t This indicates the system status information at that moment.

[0049] After completing the trajectory decomposition, based on each atomic behavior unit Γ k，t Establish a four-element data unit: D k1，t =(C k1，t F k1，t A k1，t O k1，t ), where D k1，t The structured data unit representing the k1th trajectory at time t specifically includes: where C k1，t The context term describes the environmental state when scheduling occurs, and is defined as: C k1，t =S t F k1，t These are feature terms that describe the changing relationship between the context and scheduling behavior, and they are generated through feature generation operators. (·) Establish: F k1，t = (S t , π t , ρ t ).

[0050] Furthermore, the feature generation operator is defined as: ,in, A mean vector representing the state of an object. A mean vector representing the resource status. This indicates the task selection indicator function. This indicates the resource selection indicator function.

[0051] A k1，t For decision items, representing scheduling execution actions, it is defined as: A k1，t =(π t , ρ t );O k1，t The execution result item represents the result produced by the scheduling behavior, and is defined as: O k1，t =(τ t ΔS t ), where ΔS t =S t+1 S t , which represents the change in system state.

[0052] After obtaining the four-dimensional data units, the scale differences between the different dimensions of the data are eliminated, and they are uniformly normalized to obtain normalized data units. .

[0053] After normalization, the data units in the same trajectory are rearranged in chronological order to create a sequence set: H k1 ={ |t=1,2,…,T k Furthermore, to enhance the temporal representation capability of the sequence, a positional encoding function is set: P(t) = sin(ωt) + cos(ωt), where ω represents the frequency parameter and t represents the time index; the positional code is concatenated with the data unit to obtain the enhanced data unit. : = ⊕P(t), where ⊕ represents the vector concatenation operation.

[0054] Based on this, the enhanced data unit is mapped using a unified coding function to obtain the behavior coding vector: Z k1，t =Ψ( ), of which Ψ1( ) represents the encoding function, which has the specific form: Ψ1(x)=Wx+b, where W represents the encoding weight matrix, b represents the bias vector, and x represents the input vector.

[0055] Subsequently, all encoded vectors along the same trajectory are weighted and aggregated to obtain the overall behavioral representation of the trajectory: , where α t The time weighting coefficient is defined as follows: That is, the weight is proportional to the time step t; the weight is greater at later times, where i1 is the time step index of the encoded vector in the trajectory.

[0056] After obtaining the overall behavioral representation, establish the behavioral pattern unit: B k1 =(Z k1 , ), where Z k1 Represents the feature vector of behavioral patterns. This indicates the corresponding behavior tag information.

[0057] Finally, all behavioral pattern units are summarized to obtain the behavioral pattern knowledge base: KL={B k1 |k1=1,2,…,K1} Thus, the entire process of transformation from a set of scheduling behavior trajectories to structured data units and then to a behavior pattern knowledge base is completed, realizing the structured expression and knowledge storage of nursing resource scheduling behavior, and providing a data foundation for subsequent priority calculation and scheduling strategy optimization.

[0058] The specific process for obtaining the basic priority results of each task includes: Obtain the behavioral pattern knowledge base (KL) and map each behavioral pattern to an information source node: V={v k1 |k1=1,2,…,K1};Define the information source strength I k1 :I k1 =∥Z k1 ∥ 2 , where: I k1 Let k1 be the initial information quantity of the information source; ∥ 2 It is a norm.

[0059] Based on the set of information source nodes, a directed information flow network is established: G I =(V, E), where: G I For an information flow network, E represents the set of information flow edges. Next, the weight for information transfer between nodes is defined as: W k1→j =exp( ∥Z k1 Z j ∥2 2 ), where: W k1→j Z represents the information transfer weight from node k1 to node j; exp(·) is the natural exponential function; k1 Z j is the difference between the two feature vectors; j is the node index.

[0060] In the information flow network, a dynamic evolution equation for node information is established: ,in, The information content of node k1 at time t; The information content of node k at the next time step; This indicates summing over all nodes; This represents the weight of the flow from node j to node k1; The weight represents the flow from node k1 to node j; μ1 is the information dissipation coefficient, which describes the information loss in the system and is obtained based on the typical update frequency of vital sign data in the nursing scenario.

[0061] Iterate through the calculations until the convergence condition is met: | |<ε1, where ε1 is the convergence threshold; at this point, the steady-state information is obtained. , where T0 represents the convergence time.

[0062] Obtain nursing tasks and establish a set of nursing tasks: R = {r | =1, 2, ..., L}, where: R is the task set; r For the first One task; L is the total number of tasks; To perform pattern task space mapping, a mapping function is established: ,in: For the behavior pattern k1 to the task Mapping weights; Ω For the task The feature center vector.

[0063] Based on the mapping relationship, the node information is aggregated to the task layer: ,in: For the task The amount of information aggregated.

[0064] Based on the amount of aggregated task information, a task priority function is established: P = (1+κ1 ), where: P For the task The priority value; κ1 adjustment coefficient, specific example: κ1=0.9, which indicates high gain feedback, which can make resource adjustment respond quickly to deviations, but it needs to be combined with the system stability boundary, that is, if the gain margin is >6dB, this value reduces the response time by 40% and has no overshoot in the simulation. This represents the gradient term for task information.

[0065] According to priority value P To avoid over-amplification of priority results, a stabilization function is established: ,in, The σ2 value represents the priority of the task after stabilization. σ2 is the compression coefficient. For example, σ2 = 2.0. When the input z changes from -1 to 1, the Sigmoid output changes from 0.12 to 0.88, with the sensitive region concentrated near the zero point. This value makes moderate input changes produce obvious output differences, distinguishing between normal and abnormal states.

[0066] Establish all task priorities as a set: P = { | =1, 2, ..., L}, and sort them to obtain the task priority set.

[0067] Through the above steps, the dynamic evolution calculation of task priorities is realized, which has stronger adaptability and interpretability.

[0068] The specific process of optimizing task execution timing based on task priority results and task time constraint information is as follows: Get the task priority set P={ | =1, 2, ..., L}, and define the task time constraint set: YS={(t s , t d , t e )∣ =1, 2, ..., L}, where: t s The earliest executable time for the task; t d The deadline for the task; t e The standard execution time for the task.

[0069] For each task , time domain H With step size δ t Discretize into M candidate time points, with a step size δ t Choose the smallest time granularity of the system. The candidate time set is represented as: T cand ={t (1) , t (2) , ..., t (M)}, t (m) =t s +(m 1)δ t m = 1, 2, ..., M.

[0070] For each candidate time t (m) Calculate its execution revenue U (t (m) The optimal execution time is selected as the moment with the greatest benefit. .

[0071] At the initial optimal execution time neighborhood[ δ t , +δ t Further optimization using parabolic interpolation is employed. Let the three interpolation points be... δ t , , +δ t The corresponding profit value is , Construct a quadratic interpolation function f(u) = au 2 +bu+c, where u is relative to The time offset. The coefficient is determined by three points: , , The offset of the extreme point is obtained by taking the derivative. When a < 0, the exact optimal execution time is: If a ≥ 0 or If it exceeds the neighborhood range, then directly take the value. .

[0072] Secondly, based on the precise optimal execution time and task priority Establish the task execution time window width Δ : η3 is the window width scaling factor, with a value range of (0, 0.5). Priority The higher, 1 The smaller the window size, the narrower the window width, reflecting the requirement for precise timing of execution for high-priority tasks.

[0073] Then, define the task. The optimal execution time window is: .

[0074] Perform boundary validity checks on the window: if the lower bound is greater than or equal to the upper bound, it indicates that the window has degenerated into a point or is invalid. In this case, perform window expansion correction, letting: , where ε3=2δ t To minimize time margin, ensure the window has a positive length.

[0075] Furthermore, a window overlap detection mechanism is established. For any two tasks... and ',like Furthermore, if the two tasks have overlapping resource requirements, such as sharing the same nursing staff or equipment, then the window size should be adjusted based on priority comparison. Let... < The task If the priority is low, then the task... Compress the window: The compressibility factor γ c ∈(0,1), take γ c =0.7. If overlap still exists after compression, the window of the low-priority task is further shifted to the left or right by a minimum time granularity δ. t Until the conflict is resolved.

[0076] Finally, the optimal execution time windows for all tasks are summed into a set: W = {W | =1, 2, ..., L}. Through the above steps, the dynamic calculation of each task execution window, priority differences, time urgency, and resource avoidance has stronger adaptability and robustness compared to the fixed time window method.

[0077] The multi-objective nursing resource scheduling optimization model is established based on task priority results and the optimal execution time window. The specific process is as follows: For each task For each resource z, by setting a binary decision variable x z ∈{0,1}, representing the task Whether to allocate the execution to resource z, set a continuous decision variable s. ∈R+, representing the task The actual start time of the task; then the task The completion time is c =s +t e , where t e The standard execution time for tasks is determined based on historical statistical data or industry standards and is dynamically updated according to nursing scenarios.

[0078] Under the premise of satisfying resource constraints and time window constraints, three sub-objectives are defined to achieve the comprehensive optimization of multiple performance indicators, including: minimizing the weighted task completion time, maximizing resource load balancing, and minimizing priority violation penalties; Among these, minimizing the weighted task completion time is crucial: high-priority tasks should be completed as early as possible. The objective function f1 is defined as follows: ; Maximizing resource load balancing: Avoiding some resources from being overloaded while others are idle, the total load of resource z is defined as... The average resource load is The objective function f2 for resource load balancing is expressed as minimizing load variance: ; Minimize priority violation penalty: If a high-priority task executes after a low-priority task, a penalty is incurred. The penalty function is defined as follows: , of which 1 (·) θ2 is the indicator function, and θ2 is the unit violation penalty coefficient with a value of 1.

[0079] The above three sub-objectives can be transformed into a single-objective optimization problem by weighted summation: Min F = ω11×f1 + ω22f2 + ω33f3, where ω11, ω22 and ω33 are weight coefficients that satisfy ω11 + ω22 + ω33 = 1 and are dynamically adjusted according to the management preferences of the organization. In this example, the default values ​​are ω11 = 0.5, ω22 = 0.3 and ω33 = 0.2.

[0080] Next, the constraints are obtained, including the uniqueness constraint of task resource allocation, the resource capacity matching constraint, the time window constraint, the resource exclusivity constraint, and the inter-task dependency constraint. Specifically, the uniqueness constraint of task resource allocation states that each task must be allocated to one resource; the resource capacity matching constraint states that the allocated task requirements cannot exceed the resource capacity; the time window constraint states that the task start time must be within its optimal execution time window; the resource exclusivity constraint states that only one task can be executed on the same resource at any given time; and the inter-task dependency constraint states that if tasks... Dependency Task The completion result.

[0081] Next, based on task priority Arrange the tasks in descending order from high to low to obtain the initial task sequence π = (π1, π2, ..., π). X ), where π1 is the highest priority task.

[0082] Establish an available timeline A for each resource z z (t), the initial available time of all resources A z (t)=0 indicates that the resource is available from time 0.

[0083] For each task in the task sequence π =π o In sequence, from o=1 to X, perform the following operation: traverse all resources z=1, ..., Z, and check whether the resource's capabilities meet the task requirements. If not, skip the resource.

[0084] For resources that satisfy the capability constraints, the computational task Earliest feasible start time on this resource : =max(w min A z ), where A z w is the earliest available time for resource z. min For the first The earliest allowed start time for each task is determined; then the completion time is calculated. .

[0085] Check the upper bound constraint of the time window: If >w max If so, then resource z is not feasible. max For the first The latest allowed start time for each task; For all feasible resources, select the one that minimizes the increment of the local objective function. Define the resource selection analysis function. : ,choose .

[0086] The task Allocate resources Update the availability time of this resource: A =s +t e Update resource load L =L +t e Record scheduling decisions: x =1, s =s c =c .

[0087] After the above solution process, the decision variables are obtained. and The optimal value and optimal scheduling scheme are represented as: Sch={( , , , )∣ =1, 2, ..., L}, which represents the resources, start time, and completion time allocated to each task. Through the above multi-objective model calculation process, the optimal allocation of nursing resources under multiple constraints such as task priority, time window, and resource capacity is achieved, ensuring timely response to high-priority tasks and taking into account resource load balancing and scheduling fairness. This has certain advantages compared to single-objective scheduling methods.

[0088] The specific process of obtaining the actual execution result data of the scheduling scheme, establishing scheduling effect evaluation indicators, and adaptively updating the rule weights and model parameters based on the evaluation results is as follows: After obtaining and executing the optimal scheduling plan, the start time, completion time, and response time of each task are collected, along with the actual load and service quality indicator vectors for each resource, such as patient satisfaction and operational compliance scores. Based on this data, three scheduling effectiveness indicators are established: task completion timeliness, resource utilization, and service quality. These three indicators are then linearly weighted and fused into a comprehensive scheduling effectiveness indicator.

[0089] After obtaining the comprehensive scheduling performance evaluation index, the model parameters such as the scheduling path energy function weight coefficients ω1, ω2, ω3, ω4, information dissipation coefficient μ1, and compression coefficient σ2 are adaptively updated. The update strategy adopts an online adjustment process in the form of gradient descent. Define the current integrated scheduling performance index as Ψ (s) The previous cycle was Ψ (s-1) The update rule for each weight coefficient is as follows: ,in The learning rate is 0.05. The partial derivatives are obtained by approximate calculation using the finite difference method; i0 = 1, 2, 3, 4.

[0090] The information dissipation coefficient μ1 is adjusted according to the resource conflict frequency: define the conflict rate. , where N co N represents the number of task pairs that experience resource conflicts during the scheduling process. to This represents the total number of task pairs. The update rule is: , where η μ =0.02 is the update step size. =0.1 is the target conflict rate. If the conflict rate is too high, increase μ1 to enhance information dissipation and reduce the coupling strength of the information flow network.

[0091] For the compression coefficient σ², adjustments are made based on the priority distribution variance: Define the priority distribution variance. If the variance is too small, meaning the priority differentiation is insufficient, then σ² should be appropriately reduced to amplify the priority differences; if the variance is too large, meaning the priority distribution is too extreme, then σ² should be increased to compress the differences. The update rule is: , where η σ =0.01, Var0=0.1 is the objective variance. σ2 is restricted to the interval [0.05, 0.5] to ensure numerical stability.

[0092] After updating the parameters as described above, the updated weight coefficients and model parameters are stored in the system configuration for optimization calculation in the next scheduling cycle. This invention, through its adaptive update mechanism, can dynamically adjust the scheduling strategy based on actual execution results, gradually improving scheduling performance.

[0093] The specific process of adapting applications across scenarios and generating decision explanation information and visualizations based on the updated behavior pattern knowledge base and scheduling model is as follows: Acquire the behavioral pattern knowledge base and updated scheduling model parameters. For new application scenarios, such as elderly care institutions of different sizes, different types of nursing needs, and different human resource configurations, implement cross-scenario adaptation applications, specifically including: For a new scenario, extract its contextual feature vector C. new This includes the distribution of care recipient status, resource availability, and task arrival rate. (C) new Contextual items C of each pattern in the behavioral pattern knowledge base k1 The similarity S is obtained by performing similarity calculation. im (C new C k1 The top three behavioral patterns with the highest similarity are selected as reference patterns, and their corresponding scheduling strategies, such as the ranking weight λ1 in the task selection cost function and the mapping relationship in the resource matching cost function, are used as the initial parameters of the new scenario scheduling model.

[0094] Based on the matched reference pattern parameters, combined with a small amount of actual operational data from the new scenario, such as 1 to 3 days of historical data, the model parameters are quickly fine-tuned according to the adaptive update mechanism in S007, with N rounds of fine-tuning. ft =5, learning rate is ζ ft =0.1, allowing the model to quickly adapt to the statistical characteristics of the new scenario. The fine-tuned model is the scheduling model for the current scenario.

[0095] While outputting the scheduling plan, corresponding decision explanation information is generated for each scheduling decision. The explanation information is expressed in a structured form and includes three aspects: explanation of task priority composition, explanation of resource matching process, and explanation of rule triggering path.

[0096] Explanation of task priority structure: For tasks Its priority Information aggregation With gradient term The decision was made jointly; the explanatory information lists the top three behavioral pattern nodes that contributed the most and their similarity weights. ,in For the task The priority constitutes the set of explanatory information, Ω(k1, () represents the behavioral pattern k1 for the task The mapping weights represent the impact of the behavioral pattern on the task. The contribution level of priority calculation is obtained by normalizing the similarity between the behavioral pattern feature vector and the task feature center vector; Explanation of the resource matching process: For tasks Allocate to resource j The decision-making process, as explained in the information, compares the indicator function values ​​of the candidate resources: ,in, Task The resource matching process is a collection of explanatory information. The task The increment of the local objective function when allocating to resource z; the smaller the value, the better the allocation scheme is in the current local decision; and the selected resource z is labeled. The reasons for selection include, for example, the resource has the lowest load or the highest matching degree.

[0097] Explanation of rule triggering path: This involves tracing back the key rule chain that triggered the current scheduling decision from the behavioral pattern knowledge base. The rule triggering path is defined as follows: ,in, For the task The rule triggering path represents a chain of rules that runs from the initial situation to the final decision action. k1 (1) As the initial situation, A k1 (h) For the h-th step decision action, rule hThis is the identifier for the association rule. The path length H generally does not exceed 5. The explanation information is output in natural language, for example: "Task T001, due to abnormal blood pressure of the patient, scenario C3, triggers high-priority rule R07, matching resource nurse N2, skill matching degree 0.92", and is displayed; where T001 is the task number, C3 is the scenario category number, R07 is the rule number, N2 is the resource number, skill matching degree 0.92, and C3 are scenario category numbers.

[0098] Through the visualization process described above, schedulers can intuitively understand the generation logic of scheduling plans, the basis for resource allocation, and the rule triggering process, thereby enhancing their trust in and interpretability of the decisions.

[0099] Please see Figure 2 As shown, another aspect of the present invention provides a dynamic priority scheduling system for nursing resources in elderly care institutions, including: a data acquisition and processing module, a scheduling behavior module, a task classification module, a task execution optimization module, a scheduling optimization module, and a cross-scenario adaptation module; The data acquisition module obtains historical scheduling records and status monitoring data. It divides the scenarios by using the state-resource coupling perturbation function and the asymmetric scenario potential function to obtain a scenario data set that includes task triggering conditions, resource status and execution results.

[0100] The scheduling behavior module performs serialization modeling of task allocation order, resource selection path and response time based on the context data set, generates a set of scheduling behavior trajectories, and then obtains a behavior pattern knowledge base through atomic behavior decomposition, four-element data unit encoding and weighted aggregation.

[0101] The task classification module establishes a directed information flow network based on the behavioral pattern knowledge base, calculates the stable information volume through the dynamic evolution equation of node information, and obtains the basic priority results of each task through pattern task space mapping and priority function processing.

[0102] Based on priority results and time constraints, the task execution optimization module obtains the precise optimal execution time through execution benefit calculation and parabolic interpolation. It dynamically adjusts the time window width and establishes an overlap detection and compression mechanism to obtain the optimal execution time window for each task.

[0103] The scheduling optimization module establishes a multi-objective optimization model that includes weighted completion time, resource load balancing, and priority violation penalties. It calculates the optimal scheduling scheme and performs online adaptive updates to the energy function weights, information dissipation coefficients, and compression coefficients based on the actual execution results.

[0104] The cross-scenario adaptation module achieves cross-scenario adaptation through context similarity matching and rapid parameter fine-tuning. When outputting the scheduling scheme, it generates and displays structured explanation information on task priority composition, resource matching process, and rule triggering path.

[0105] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A dynamic priority scheduling method for nursing resources in elderly care institutions, characterized in that, include: Obtain historical scheduling records, task execution logs, and status monitoring data; perform scenario segmentation to obtain scenario data sets; Generate discrete-time state sequences based on contextual data sets, including task selection sequences, resource allocation paths, and response time sequences; Establish a scheduling path energy function; generate a set of scheduling behavior trajectories through energy convergence; decompose the trajectory into atomic behavior units, and establish a four-element data unit containing context item, feature item, decision item and execution result item based on each atomic behavior unit; then, after normalization and position encoding, map it to behavior encoding vector through a unified encoding function, and obtain behavior pattern units by weighted aggregation to form a behavior pattern knowledge base; A directed information flow network is established based on a behavioral pattern knowledge base. By calculating the dynamic evolution equation of node information, behavioral patterns are mapped to information source nodes, and the information source strength is calculated. A directed information flow network is established, and a dynamic evolution equation for node information is established. Based on the information inflow, outflow, and information dissipation coefficient, the process is iterated until convergence to obtain a stable amount of information. The node information is aggregated to the task layer through pattern task space mapping, and then processed by priority function and stabilization function to obtain the basic priority results of each task. Based on priority results and time constraints, the optimal execution time is obtained by calculating execution benefits. The time window width is dynamically adjusted and an overlap detection and compression process is established to obtain the optimal execution time window for each task. A multi-objective optimization model incorporating weighted completion time, resource load balancing, and priority violation penalties is established to obtain the optimal scheduling scheme. Obtain the actual execution result data of the scheduling plan, analyze the tasks, and obtain scheduling performance indicators; Online adaptive updates are performed based on scheduling performance metrics; Cross-scenario adaptation is performed based on the updated behavior pattern knowledge base, and structured explanations of task priority composition, resource matching process and rule triggering path are generated and displayed when outputting the scheduling scheme.

2. The dynamic priority scheduling method for nursing resources in elderly care institutions according to claim 1, characterized in that, The specific process of situation segmentation includes: Each data unit includes the task trigger time, the nursing object's state vector, the resource state vector, and the execution result. A state-resource coupling perturbation function is established, which combines the nursing object's state vector and the resource state vector by performing feature cross-interaction and then superimposing the product of the state change gradient and the time sensitivity coefficient. An asymmetric situation potential function is established, which is obtained by weighting the norm square of the coupled feature vector, the resource distribution entropy, and the execution result deviation. Through a situation segmentation boundary function, tasks with potential values ​​less than the first threshold are classified as low-load situations, those between the first and second thresholds are classified as medium-load situations, and those greater than or equal to the second threshold are classified as high-urgency situations. The situation data units are unified and a hierarchical index structure is established to obtain the situation data set.

3. The dynamic priority scheduling method for nursing resources in elderly care institutions according to claim 1, characterized in that, The specific process of obtaining the optimal execution time window for each task includes: Obtain the task priority and time constraint set, discretize the time domain into candidate times, calculate the execution benefits to obtain the initial optimal execution time; obtain the precise optimal execution time through parabolic interpolation; dynamically adjust the window width according to priority; based on the optimal execution time window and constrained by the earliest execution time and deadline; establish a window overlap detection mechanism, for tasks with overlapping resource requirements and intersecting windows, compress the window according to priority until the conflict is eliminated.

4. The dynamic priority scheduling method for nursing resources in elderly care institutions according to claim 3, characterized in that, The specific process of obtaining the optimal scheduling scheme includes: A binary decision variable is set to represent the allocation relationship between tasks and resources, and a continuous decision variable is set to represent the task start time. A multi-objective optimization model is established, which includes three sub-objectives: weighted completion time, resource load variance, and priority violation penalty. The model is then transformed into a single objective by weighted summation. Tasks are sorted in descending order of priority according to constraints. For each task, feasible resources are traversed, the local objective function increment is calculated, and the resource with the smallest increment is selected for allocation to obtain the optimal scheduling scheme.

5. The dynamic priority scheduling method for nursing resources in elderly care institutions according to claim 1, characterized in that, The specific process of online adaptive updating based on scheduling performance metrics includes: To obtain actual execution results, three scheduling performance indicators are established: task completion time, resource utilization, and service quality. These indicators are then integrated to obtain a comprehensive scheduling performance indicator. The weight coefficients in the scheduling path energy function are updated using gradient descent. The information dissipation coefficient is adjusted based on the resource conflict frequency. The compression coefficient is adjusted based on the priority distribution variance, and the parameters are restricted to a stable range.

6. A dynamic priority scheduling method for nursing resources in elderly care institutions according to claim 5, characterized in that, The specific process of performing cross-scene adaptation and generating structured explanatory information includes: For new scenarios, extract contextual feature vectors and perform similarity matching with contextual items in the behavior pattern knowledge base. Select the top three behavior patterns with the highest similarity as references, use their scheduling parameters as initial parameters, and quickly fine-tune them. When outputting the scheduling scheme, generate three aspects of explanatory information: task priority composition explanation, listing the behavior pattern nodes that contribute the most and their mapping weights; resource matching process explanation, listing the indicator comparison of candidate resources and the reasons for selection; rule triggering path explanation, tracing back the key rule chain; and display them.

7. A dynamic priority scheduling system for nursing resources in elderly care institutions, characterized in that, The method for dynamic priority scheduling of nursing resources in elderly care institutions as described in any one of claims 1-6 includes: a data acquisition module, a scheduling behavior module, a task classification module, a task execution optimization module, a scheduling optimization module, and a cross-scenario adaptation module; The data acquisition module acquires historical scheduling records and status monitoring data, performs scenario segmentation, and obtains scenario data sets; The scheduling behavior module performs serialization modeling of tasks based on the contextual data set, generates a set of scheduling behavior trajectories, and then obtains a behavior pattern knowledge base through atomic behavior decomposition. The task classification module establishes a directed information flow network based on the behavioral knowledge base, and calculates and processes the basic priority results of each task through the dynamic evolution equation of node information. The task execution optimization module calculates the precise optimal execution time based on priority results and time constraints, dynamically adjusts the time window width, and establishes an overlap detection and compression mechanism to obtain the optimal execution time window for each task. The scheduling optimization module establishes a multi-objective optimization model, calculates the optimal scheduling scheme, and performs online adaptive updates based on the actual execution results. The cross-scenario adaptation module achieves cross-scenario adaptation through context similarity matching and rapid parameter fine-tuning. When outputting the scheduling scheme, it generates and displays structured explanation information on task priority composition, resource matching process, and rule triggering path.

Images