Nursing service collaborative scheduling method and system based on multi-source data fusion

Abstract

This invention belongs to the field of medical service technology and discloses a method and system for collaborative scheduling of nursing services based on multi-source data fusion. It involves real-time collection of multi-source heterogeneous data from nursing scenarios, cleaning, standardizing, and extracting features from the data to construct multi-dimensional feature vectors. A deep learning prediction model is then constructed, and the multi-dimensional feature vectors are input into the model to output prediction results, which at least include the patient's nursing needs type, intensity, and spatiotemporal distribution. Based on a deep reinforcement learning algorithm combined with the prediction results, a real-time nursing service scheduling strategy is generated, and a co-evolutionary mechanism is used to achieve task allocation and path planning among multiple nursing units. This improves the utilization rate of nursing resources and overall service efficiency throughout the hospital, effectively reducing nursing workload.

Description

Technical Field

[0001] This invention relates to the field of medical service technology, and in particular to a method and system for collaborative scheduling of nursing services based on multi-source data fusion. Background Technology

[0002] Existing nursing scheduling decision-making data is limited in scope, relying solely on basic nursing records and shift data. It fails to integrate multi-source data such as patient vital signs, staff status, medical equipment, and ward environment, resulting in severe data fragmentation and an inability to comprehensively reflect the true supply and demand situation. Traditional prediction models can only simply fit changes in time-series data, failing to consider the temporal dynamics and spatial correlations of nursing scenarios. This makes it difficult to accurately predict the type, intensity, and spatiotemporal distribution of patient nursing needs, leading to a lack of forward-looking scheduling. Most scheduling optimization mechanisms employ single-objective optimization logic, failing to balance response time and resource utilization, resulting in low efficiency in collaborative scheduling of nursing services. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a nursing service collaborative scheduling method and system based on multi-source data fusion.

[0004] To achieve the above objectives, the technical solution of the present invention further includes the following steps in the above-mentioned nursing service collaborative scheduling method based on multi-source data fusion: Real-time acquisition of multi-source heterogeneous data in nursing scenarios; cleaning, standardization and feature extraction of the multi-source heterogeneous data; construction of multi-dimensional feature vectors. A deep learning prediction model is constructed, based on a spatiotemporal feature extraction layer of a Transformer encoder and a hybrid prediction layer integrating an LSTM long short-term memory network and a GNN graph neural network, where the GNN graph neural network is used to model the spatial relationship topology between caregivers and patients. The multi-dimensional feature vector is input into the deep learning prediction model, and the prediction results are output, including at least the patient's nursing needs type, intensity, and spatiotemporal distribution. With the goals of minimizing response latency and maximizing resource utilization, a multi-objective optimization model is constructed. Based on deep reinforcement learning algorithms and prediction results, a real-time nursing service scheduling strategy is generated, and a co-evolutionary mechanism is adopted to realize task allocation and path planning among multiple nursing units.

[0005] Furthermore, in the above-mentioned nursing service collaborative scheduling method based on multi-source data fusion, the real-time collection of multi-source heterogeneous data in the nursing scenario, the cleaning, standardization, and feature extraction of the multi-source heterogeneous data, and the construction of a multi-dimensional feature vector, include: Real-time collection of multi-source heterogeneous data in nursing scenarios, including at least patient health monitoring data, real-time location and work status data of nursing staff, medical equipment usage data, environmental parameter data, and historical service record data; Abnormal noise data caused by sensor failures in multi-source heterogeneous data is removed by threshold verification and time sequence continuity verification to obtain processed multi-source data; Temporal interpolation is used to complete short-term missing data in multi-source data, and duplicate and redundant invalid data are deleted. In the data standardization process, heterogeneous data of different magnitudes and units are normalized to obtain normalized multi-source data. The location coordinates, vital sign values, equipment status and workload in the normalized multi-source data are uniformly mapped to the standard data format, and data desensitization and compliance screening are completed at the same time to obtain cleaned multi-source data. Feature vectors are extracted from the cleaned multi-source data to obtain multi-dimensional feature vectors, which include at least temporal features, spatial features, attribute features, and scene features.

[0006] Furthermore, in the aforementioned collaborative scheduling method for nursing services based on multi-source data fusion, the construction of a deep learning prediction model includes a spatiotemporal feature extraction layer based on a Transformer encoder and a hybrid prediction layer integrating an LSTM (Long Short-Term Memory) network and a GNN (Graph Neural Network). The GNN is used to model the spatial relationship topology between nursing staff and patients, including: The spatiotemporal feature extraction layer based on the Transformer encoder is composed of multiple stacked encoders. Each encoder layer contains a multi-head self-attention module and a feedforward neural network module, and is configured with layer normalization and residual connection structures. The multi-head self-attention mechanism maps the input feature vector into multiple independent query, key, and value vectors, calculates the correlation and dependency between different feature dimensions in parallel, and distinguishes the weight ratio of core features and redundant features.

[0007] Furthermore, in the aforementioned collaborative scheduling method for nursing services based on multi-source data fusion, the construction of a deep learning prediction model includes a spatiotemporal feature extraction layer based on a Transformer encoder and a hybrid prediction layer integrating an LSTM (Long Short-Term Memory) network and a GNN (Graph Neural Network). The GNN is used to model the spatial relationship topology between nursing staff and patients, including: The hybrid prediction layer adopts a dual-branch parallel fusion architecture of LSTM long short-term memory network and GNN graph neural network; The forget gate of the LSTM Long Short-Term Memory network is responsible for filtering and discarding invalid and redundant information in historical nursing data, the input gate is responsible for filtering, retaining effective temporal features and updating unit memory, and the output gate combines the current memory and feature information to output the temporal inference result. The GNN graph neural network uses the GAT dynamic graph attention algorithm to adapt to dynamic nursing scenarios. It adaptively updates the weight coefficients of each node and edge through the attention mechanism, and models the matching relationship between personnel and patients, the matching relationship between personnel and equipment, and the spatial association topology across nursing units in real time.

[0008] Furthermore, in the above-mentioned nursing service collaborative scheduling method based on multi-source data fusion, the step of inputting the multi-dimensional feature vector into the deep learning prediction model and outputting the prediction result includes at least the patient's nursing need type, intensity, and spatiotemporal distribution, including: Nursing needs include basic living care, condition monitoring care, specialized treatment care, emergency resuscitation care, and postoperative rehabilitation care; Nursing demand intensity is quantified using nursing hours, number of staff, equipment usage time, and operational complexity as indicators, and is categorized into routine needs, key needs, and emergency high-risk needs. The spatiotemporal distribution of nursing needs includes the time period when the nursing needs occur, the duration of the needs, the ward to which the needs are located, and the spatial distribution of the locations.

[0009] Furthermore, in the aforementioned collaborative scheduling method for nursing services based on multi-source data fusion, a multi-objective optimization model is constructed with the objectives of minimizing response latency and maximizing resource utilization; a real-time nursing service scheduling strategy is generated based on deep reinforcement learning algorithms combined with prediction results; and a co-evolutionary mechanism is used to realize task allocation and path planning among multiple nursing units, including: Based on the PPO proximal strategy optimization algorithm, the ward nursing scenario is constructed as an intelligent interactive environment, and nursing task allocation, personnel path selection, and equipment allocation are defined as intelligent agent execution actions. With the comprehensive benefit of the multi-objective optimization model as the core reward mechanism, the PPO proximal policy optimization algorithm predicts the future distribution of nursing needs in advance by limiting the policy update range and combining the spatiotemporal prediction results of nursing needs output above during the iteration process.

[0010] Furthermore, in the aforementioned collaborative scheduling method for nursing services based on multi-source data fusion, a multi-objective optimization model is constructed with the objectives of minimizing response latency and maximizing resource utilization; a real-time nursing service scheduling strategy is generated based on deep reinforcement learning algorithms combined with prediction results; and a co-evolutionary mechanism is used to realize task allocation and path planning among multiple nursing units, including: The scheduling schemes of each independent nursing unit are defined as independent subpopulations, and each subpopulation contains multiple sets of differentiated task allocation and path planning individual schemes. Unified planning of personnel movement and equipment transfer routes across units enables the overall optimization of nursing service scheduling across the hospital.

[0011] Furthermore, in the nursing service collaborative scheduling system based on multi-source data fusion, the nursing service collaborative scheduling system includes the following modules: A multi-dimensional data acquisition module is used to collect multi-source heterogeneous data in nursing scenarios in real time, clean, standardize and extract features from the multi-source heterogeneous data, and construct a multi-dimensional feature vector. The prediction model building module is used to construct a deep learning prediction model, which includes a spatiotemporal feature extraction layer based on a Transformer encoder and a hybrid prediction layer integrating an LSTM long short-term memory network and a GNN graph neural network. The GNN graph neural network is used to model the spatial relationship topology between caregivers and patients. The multi-source data fusion module is used to input the multi-dimensional feature vector into the deep learning prediction model and output the prediction results, which include at least the patient's nursing needs type, intensity, and spatiotemporal distribution. The nursing service scheduling module is used to construct a multi-objective optimization model with the goal of minimizing response delay and maximizing resource utilization. Based on deep reinforcement learning algorithms and prediction results, it generates real-time nursing service scheduling strategies and uses a co-evolutionary mechanism to realize task allocation and path planning among multiple nursing units.

[0012] Furthermore, in the nursing service collaborative scheduling system based on multi-source data fusion, the nursing service scheduling module includes the following sub-modules: A sub-module is constructed to build an intelligent interactive environment for ward nursing scenarios based on the PPO proximal strategy optimization algorithm, defining nursing task allocation, personnel path selection, and equipment allocation actions as intelligent agent execution actions. The update submodule is used to use the comprehensive benefit of the multi-objective optimization model as the core reward mechanism. During the iteration process, the PPO proximal policy optimization algorithm predicts the future distribution of nursing needs in advance by limiting the policy update range and combining the spatiotemporal prediction results of nursing needs output above.

[0013] Furthermore, in the nursing service collaborative scheduling system based on multi-source data fusion, the nursing service scheduling module includes the following sub-modules: Define submodules to define the scheduling schemes of each independent nursing unit as independent subpopulations. Each subpopulation contains multiple sets of differentiated task allocation and path planning individual schemes. The unified submodule is used to uniformly plan the movement of personnel and the transfer of equipment across units, so as to achieve the global optimization of nursing service scheduling throughout the hospital.

[0014] Its beneficial effects lie in providing comprehensive, accurate, and high-quality data support for intelligent prediction and scheduling decisions through the construction of standardized multi-dimensional feature vectors, thus ensuring the scientific nature of scheduling decisions from the ground up. An innovative hybrid deep learning model integrating Transformer, LSTM, and GAT is constructed, taking into account global spatiotemporal feature extraction, temporal pattern mining, and spatial topological relationship modeling. This accurately predicts multiple nursing needs, effectively solving the problems of low prediction accuracy and insufficient spatiotemporal perception capabilities of traditional models, and significantly improving the foresight of nursing scheduling. Simultaneously, based on weighted Pareto multi-objective optimization combined with the PPO reinforcement learning algorithm, a bidirectional optimal balance between minimum response latency and maximum resource utilization is achieved under the premise of conforming to clinical business constraints, effectively shortening patient waiting time and balancing the workload of medical staff. Finally, relying on an adaptive co-evolutionary algorithm to break down resource barriers between multiple nursing units, it achieves collaborative task allocation and global path planning across the entire domain, thoroughly improving the problem of resource allocation imbalance, significantly improving the utilization rate of nursing resources and overall service efficiency throughout the hospital, effectively reducing nursing workload and improving the patient experience. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 This is a schematic diagram of the first embodiment of the nursing service collaborative scheduling method based on multi-source data fusion in this invention. Figure 2 This is a schematic diagram of the second embodiment of the nursing service collaborative scheduling method based on multi-source data fusion in this invention. Figure 3 This is a schematic diagram of the first embodiment of the nursing service collaborative scheduling system based on multi-source data fusion in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms "one," "an," and "this" used herein may also include the plural forms. It should be further understood that the terminology used in this specification includes the presence of features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0019] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 As shown, a collaborative scheduling method for nursing services based on multi-source data fusion includes the following steps: Step 101: Collect multi-source heterogeneous data in the nursing scenario in real time, clean, standardize and extract features from the multi-source heterogeneous data, and construct a multi-dimensional feature vector; Specifically, this embodiment collects multi-source heterogeneous data in the nursing scenario in real time, including at least patient health monitoring data, real-time location and work status data of nursing staff, medical equipment usage data, environmental parameter data, and historical service record data. Abnormal noise data caused by sensor malfunctions in the multi-source heterogeneous data is removed through threshold verification and temporal continuity verification to obtain processed multi-source data. Temporal interpolation is used to complete short-term missing data in the multi-source data, and duplicate and redundant invalid data is deleted. In the data standardization stage, heterogeneous data of different magnitudes and units are normalized to obtain normalized multi-source data. The location coordinates, vital sign values, equipment status, and workload in the normalized multi-source data are uniformly mapped to a standard data format, and data anonymization and compliance screening are completed to obtain cleaned multi-source data. Feature vectors are extracted from the cleaned multi-source data to obtain multi-dimensional feature vectors, including at least temporal features, spatial features, attribute features, and scene features.

[0020] First, patient health monitoring data, including real-time vital signs data such as heart rate, blood pressure, blood oxygen, and body temperature; disease severity classification data; dynamic changes in symptoms; medication and treatment progress data; and nursing level requirements data. Second, dynamic data of nursing staff, covering real-time location coordinates of medical staff, location in ward / room, on-duty / off-duty / busy / idle work status, current tasks, remaining workload, skill qualification tags, and shift schedule data. Third, medical equipment usage data, including real-time occupancy status, equipment location, available time, fault status, and equipment-compatible nursing scenario types for nursing equipment such as monitors, infusion pumps, and ventilators. Fourth, ward environmental parameter data, including ward temperature and humidity, environmental cleanliness, bed occupancy status, and ward emergency risk warning signals. Fifth, historical service record data, covering a full range of historical time-series data, including past patient nursing task types, execution duration, personnel matching records, task response time, nursing service evaluations, and abnormal handling records.

[0021] To address the issues of missing, anomalies, redundancy, and inconsistent dimensions in the collected multi-source heterogeneous data, refined data preprocessing was implemented. In the data cleaning stage, threshold verification and temporal continuity verification were used to remove abnormal noise data caused by sensor malfunctions. Temporal interpolation was employed to complete short-term missing data, and duplicate, redundant, and invalid data were deleted. In the data standardization stage, heterogeneous data of different magnitudes and units were uniformly normalized, mapping various data types such as location coordinates, vital sign values, equipment status, and workload to a standardized data format. Simultaneously, data anonymization and compliance screening were completed to ensure the security and compliance of medical data.

[0022] Based on the preprocessed clean dataset, core effective features are mined to complete multi-dimensional feature extraction. The focus is on extracting four main categories of core features: temporal features (patent condition changes, nursing workload fluctuations, equipment usage cycles), spatial features (spatial relationships between personnel, patients, equipment, and wards), attribute features (patient condition severity, personnel skill attributes, equipment functional attributes), scene features (ward environment status, and urgency level of nursing tasks). These features are then fused, encoded, dimensionally aligned, and weighted to construct a standardized multi-dimensional fused feature vector containing spatiotemporal, business, and relational attributes, which serves as the standard input data for subsequent deep learning models.

[0023] Step 102: Construct a deep learning prediction model based on the spatiotemporal feature extraction layer of the Transformer encoder and the hybrid prediction layer integrating the LSTM long short-term memory network and the GNN graph neural network, where the GNN graph neural network is used to model the spatial relationship topology between caregivers and patients. Specifically, in this embodiment, the spatiotemporal feature extraction layer based on the Transformer encoder is composed of multiple stacked encoders. Each encoder layer contains a multi-head self-attention module and a feedforward neural network module, and is configured with layer normalization and residual connection structures. The multi-head self-attention mechanism maps the input feature vector into multiple independent query, key, and value vectors, calculates the correlation and dependency between different feature dimensions in parallel, and distinguishes the weight ratio of core features and redundant features.

[0024] The hybrid prediction layer adopts a dual-branch parallel fusion architecture of LSTM (Long Short-Term Memory) network and GNN (Graph Neural Network). The forget gate of the LSTM network is responsible for filtering and discarding invalid and redundant information in historical nursing data, the input gate is responsible for filtering, retaining effective temporal features and updating unit memory, and the output gate combines the current memory and feature information to output the temporal inference result. The GNN adopts the GAT (Geometry Attention) algorithm to adapt to the dynamic nursing scenario. It adaptively updates the weight coefficients of each node and edge through the attention mechanism, and models the matching relationship between personnel and patients, the matching relationship between personnel and equipment, and the spatial association topology across nursing units in real time.

[0025] This algorithm employs a stacked standard Transformer encoder as its basic feature extraction framework, abandoning the limitations of traditional convolutional and recurrent networks in terms of local features. It leverages the global modeling capabilities of the self-attention mechanism to adapt to feature association mining across all nursing scenarios. The system consists of multiple stacked encoders, each containing a multi-head self-attention module and a feedforward neural network module, configured with layer normalization and residual connection structures to effectively avoid the gradient vanishing problem during deep network training. The multi-head self-attention mechanism is the core algorithm logic, mapping the input feature vector to multiple independent query, key, and value vectors, and simultaneously calculating the dependencies between different feature dimensions, autonomously distinguishing the weight ratios of core and redundant features. For nursing scenarios, it can accurately capture patterns of disease progression over time and resource distribution relationships across space, such as linking short-term patient vital sign fluctuations with long-term care needs and coordinating resource status differences among personnel and equipment in different wards. This allows for deep extraction and reconstruction of global spatiotemporal features, ultimately outputting high-dimensional, refined, and strongly correlated fused latent features, providing high-quality feature input for the backend hybrid prediction network.

[0026] A parallel fusion architecture combining a temporal LSTM network and a graph attention GNN network is adopted to specifically mine the temporal dynamics and spatial topological relationships of nursing scenarios. Through a feature fusion mechanism, complementary information across two dimensions is achieved, enabling accurate prediction of nursing needs. First, the LSTM (Long Short-Term Memory) network branch for temporal prediction relies on the collaborative working mechanism of three gates: input, forget, and output, to address the problem of long-sequence memory failure in traditional recurrent neural networks. The forget gate filters and discards invalid and redundant information from historical nursing data, the input gate filters and retains valid temporal features and updates the unit memory, and the output gate combines the current memory and feature information to output the temporal deduction result. This allows for stable learning of long-term patterns in patient condition evolution, temporal consumption of nursing tasks, and dynamic changes in resource supply and demand, achieving accurate prediction of the temporal dimension of nursing needs. Second, the GNN (Graph Neural Network) branch for spatial modeling specifically employs the dynamic graph attention (GAT) algorithm to adapt to dynamic nursing scenarios. Unlike static graph networks, it can update the scene topology in real time. All static and dynamic entities within the nursing scenario are defined as graph nodes, and the business relationships, spatial distances, and matching relationships between entities are defined as dynamic edges. An attention mechanism adaptively updates the weight coefficients of each node and edge, modeling in real time the personnel-patient matching relationships, personnel-equipment adaptation relationships, and cross-nursing unit spatial relationship topology, dynamically perceiving the global spatial resource distribution characteristics. Finally, through a feature concatenation and fusion mechanism, the LSTM temporal features and GNN spatial topology features are deeply fused, outputting refined nursing demand prediction features that take into account both spatiotemporal dimensions.

[0027] Step 103: Input the multi-dimensional feature vector into the deep learning prediction model and output the prediction results, which should include at least the patient's nursing needs type, intensity, and spatiotemporal distribution. Specifically, in this embodiment, the types of nursing needs include basic living care, condition monitoring care, specialized treatment care, emergency rescue care, and postoperative rehabilitation care; the intensity of nursing needs is quantified by nursing hours, number of staff, equipment occupancy time, and operational complexity, distinguishing between routine needs, key needs, and emergency high-risk needs; the spatiotemporal distribution of nursing needs includes the time period of occurrence, duration, ward, and location of the nursing needs.

[0028] Based on data such as patient condition characteristics, historical nursing records, and dynamic changes in symptoms, we can accurately classify and predict the types of nursing services required by patients, including sub-types such as basic living care, condition monitoring care, specialized treatment care, emergency rescue care, and postoperative rehabilitation care, so as to clarify the differentiated nursing service needs of different patients.

[0029] By combining patient condition level, vital sign fluctuation range, nursing risk coefficient, and nursing resource consumption data of similar cases in the past, the intensity of nursing demand is quantitatively predicted. With nursing hours, number of staff, equipment occupancy time, and operation complexity as core quantitative indicators, routine needs, key needs, and emergency high-risk needs are distinguished to achieve refined classification of nursing needs.

[0030] Based on the model's spatial topology modeling and temporal extrapolation capabilities, the system accurately predicts the occurrence time, duration, ward and location spatial distribution of various nursing needs, clarifies the density of nursing needs in different time periods and regions, and forms a spatiotemporal thermal distribution result of nursing needs across the entire domain, providing a preliminary prediction basis for subsequent dynamic scheduling.

[0031] Step 104: With the goal of minimizing response latency and maximizing resource utilization, construct a multi-objective optimization model; based on deep reinforcement learning algorithm combined with prediction results, generate a real-time nursing service scheduling strategy, and adopt a co-evolutionary mechanism to realize task allocation and path planning among multiple nursing units.

[0032] Specifically, in this embodiment, the ward nursing scenario is constructed as an intelligent interactive environment based on the PPO proximal strategy optimization algorithm. The nursing task allocation, personnel path selection, and equipment allocation actions are defined as intelligent agent execution actions. With the comprehensive benefit of the multi-objective optimization model as the core reward mechanism, during the iteration process, the PPO proximal strategy optimization algorithm predicts the future distribution of nursing needs in advance by limiting the policy update range and combining the spatiotemporal prediction results of nursing needs output above.

[0033] The scheduling schemes of each independent nursing unit are defined as independent sub-species. Each sub-species contains multiple sets of differentiated task allocation and path planning individual schemes. The cross-unit personnel movement and equipment transfer paths are uniformly planned to achieve the global optimization of nursing service scheduling throughout the hospital.

[0034] A constrained weighted Pareto multi-objective optimization algorithm is used to construct a scheduling optimization model, overcoming the limitations of single-objective optimization and adapting to the dual core requirements of timeliness and efficiency in nursing scenarios. The model's core sets two mutually constraining optimization objectives, while also adding multiple hard constraints to align with clinical nursing operations. In terms of objective optimization logic, for the objective of minimizing nursing response delay, priority is given to emergency nursing tasks and near-field service scenarios, compressing the entire process of task assignment, personnel arrival, and equipment deployment. For the objective of maximizing resource utilization, the workload of nursing staff is balanced and controlled to avoid overloading of individual personnel or wards, while also making full use of idle nursing equipment and spare human resources to prevent resource waste. In terms of constraint mechanisms, rigid rules for clinical operations are solidified, including personnel skill matching constraints, specialist nursing tasks are only matched with qualified medical personnel, spatial distance constraints, priority allocation of nearby available resources, task priority constraints, emergency tasks take precedence over routine nursing tasks, equipment compatibility constraints, special nursing equipment is only matched with corresponding disease nursing scenarios, cross-unit scheduling authority constraints, routine tasks are prioritized for allocation within the unit, and cross-unit scheduling is initiated in shortage scenarios. All optimization decisions are generated iteratively within the constraint framework to ensure that the scheduling strategy is scientific, compliant, and in line with clinical reality.

[0035] The Proximal Policy Optimization (PPO) algorithm, a mainstream online reinforcement learning algorithm, is adopted as the core scheduling decision algorithm. It boasts advantages such as stable training, fast convergence speed, and adaptability to dynamic scenarios, enabling real-time adaptation to the dynamically changing needs and resource status of nursing scenarios. The core operating principle of the algorithm is as follows: the entire ward nursing scenario is constructed as an intelligent interactive environment, with nursing staff status, equipment resource status, patient nursing needs, and environmental scenario information uniformly defined as environmental state observations; nursing task allocation, personnel path selection, and equipment deployment actions are defined as agent-executed actions; and the comprehensive benefit of the multi-objective optimization model serves as the core reward mechanism, with higher reward values ​​for lower response latency, higher resource utilization, and better task matching. During iteration, the PPO algorithm avoids sudden policy changes and training instability in a single iteration by limiting the policy update amplitude, continuously iterating and optimizing the action strategy. Combined with the spatiotemporal prediction results of nursing needs output above, the agent can predict the future distribution of nursing needs in advance, dynamically match the optimal executor, optimal service equipment, and shortest service path in real time, and quickly generate a refined, real-time nursing task scheduling scheme within a single nursing unit.

[0036] To address the issues of resource barriers and uneven resource allocation across multiple nursing units, an adaptive co-evolutionary algorithm is introduced to achieve global collaborative scheduling. The core of this algorithm utilizes a multi-subpopulation collaborative iteration and survival-of-the-fittest evolutionary mechanism to achieve optimal task allocation and global path planning across units. The core principle of the algorithm is as follows: Each independent nursing unit's scheduling scheme is defined as an independent subpopulation, with each subpopulation containing multiple sets of differentiated task allocation and path planning individual schemes. First, the population is initialized, generating an initial set of scheduling schemes for multiple units. Then, an adaptive crossover and mutation mechanism optimizes the population. Crossover integrates the advantages of scheduling schemes from different nursing units, exchanging superior strategies for task allocation and path planning. Mutation randomly optimizes local scheduling details, avoiding local optima. Simultaneously, an elite retention strategy is added to retain high-quality scheduling schemes from each iteration while eliminating inferior schemes with high response latency and significant resource waste. After multiple rounds of iterative evolution, the algorithm adaptively converges to the global optimal solution, realizing task splitting, load balancing, and resource complementarity across nursing units. At the same time, combined with the global spatial topology, it uniformly plans the movement paths of personnel and equipment transfer across units, completely solving the problems of resource congestion in single units and resource idleness and imbalance across the entire region, and achieving the global optimality of nursing service scheduling throughout the hospital.

[0037] Its beneficial effects lie in providing comprehensive, accurate, and high-quality data support for intelligent prediction and scheduling decisions through the construction of standardized multi-dimensional feature vectors, thus ensuring the scientific nature of scheduling decisions from the ground up. An innovative hybrid deep learning model integrating Transformer, LSTM, and GAT is constructed, taking into account global spatiotemporal feature extraction, temporal pattern mining, and spatial topological relationship modeling. This accurately predicts multiple nursing needs, effectively solving the problems of low prediction accuracy and insufficient spatiotemporal perception capabilities of traditional models, and significantly improving the foresight of nursing scheduling. Simultaneously, based on weighted Pareto multi-objective optimization combined with the PPO reinforcement learning algorithm, a bidirectional optimal balance between minimum response latency and maximum resource utilization is achieved under the premise of conforming to clinical business constraints, effectively shortening patient waiting time and balancing the workload of medical staff. Finally, relying on an adaptive co-evolutionary algorithm to break down resource barriers between multiple nursing units, it achieves collaborative task allocation and global path planning across the entire domain, thoroughly improving the problem of resource allocation imbalance, significantly improving the utilization rate of nursing resources and overall service efficiency throughout the hospital, effectively reducing nursing workload and improving the patient experience.

[0038] Please see Figure 2 In the collaborative scheduling method for nursing services based on multi-source data fusion, a deep learning prediction model is constructed, consisting of a spatiotemporal feature extraction layer based on a Transformer encoder and a hybrid prediction layer integrating an LSTM (Long Short-Term Memory) network and a GNN (Graph Neural Network). The GNN is used to model the spatial relationship topology between nursing staff and patients. The steps include: Step 201: The hybrid prediction layer adopts a dual-branch parallel fusion architecture of LSTM long short-term memory network and GNN graph neural network; Step 202: The forget gate of the LSTM Long Short-Term Memory Network is responsible for filtering and discarding invalid and redundant information in historical nursing data, the input gate is responsible for filtering, retaining effective temporal features and updating unit memory, and the output gate combines the current memory and feature information to output the temporal inference result. Step 203: The GNN graph neural network adopts the GAT dynamic graph attention algorithm to adapt to the dynamic nursing scenario. Through the attention mechanism, it adaptively updates the weight coefficients of each node and edge, and models the matching relationship between personnel and patients, the matching relationship between personnel and equipment, and the spatial association topology across nursing units in real time.

[0039] The above describes embodiments of the nursing service collaborative scheduling method based on multi-source data fusion of the present invention. Please refer to [link / reference]. Figure 3 In a nursing service collaborative scheduling system based on multi-source data fusion, the system includes the following modules: The multi-dimensional data acquisition module is used to collect multi-source heterogeneous data in nursing scenarios in real time, clean, standardize and extract features from the multi-source heterogeneous data, and construct multi-dimensional feature vectors. The prediction model building module is used to construct a deep learning prediction model, which includes a spatiotemporal feature extraction layer based on a Transformer encoder and a hybrid prediction layer integrating an LSTM long short-term memory network and a GNN graph neural network. The GNN graph neural network is used to model the spatial relationship topology between caregivers and patients. The multi-source data fusion module is used to input multi-dimensional feature vectors into the deep learning prediction model and output prediction results, which include at least the patient's nursing needs type, intensity, and spatiotemporal distribution. The nursing service scheduling module is used to construct a multi-objective optimization model with the goal of minimizing response delay and maximizing resource utilization. Based on deep reinforcement learning algorithms and prediction results, it generates real-time nursing service scheduling strategies and uses a co-evolutionary mechanism to realize task allocation and path planning among multiple nursing units.

[0040] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended technical solutions and equivalents.

Claims

1. A nursing service collaborative scheduling method based on multi-source data fusion, characterized in that, The method for collaborative scheduling of nursing services includes the following steps: Real-time acquisition of multi-source heterogeneous data in nursing scenarios; cleaning, standardization and feature extraction of the multi-source heterogeneous data; construction of multi-dimensional feature vectors. A deep learning prediction model is constructed, based on a spatiotemporal feature extraction layer of a Transformer encoder and a hybrid prediction layer integrating an LSTM long short-term memory network and a GNN graph neural network, where the GNN graph neural network is used to model the spatial relationship topology between caregivers and patients. The multi-dimensional feature vector is input into the deep learning prediction model, and the prediction results are output, including at least the patient's nursing needs type, intensity, and spatiotemporal distribution. With the goals of minimizing response latency and maximizing resource utilization, a multi-objective optimization model is constructed. Based on deep reinforcement learning algorithms and prediction results, a real-time nursing service scheduling strategy is generated, and a co-evolutionary mechanism is adopted to realize task allocation and path planning among multiple nursing units.

2. The nursing service collaborative scheduling method based on multi-source data fusion as described in claim 1, characterized in that, The real-time acquisition of multi-source heterogeneous data in the nursing scenario involves cleaning, standardizing, and extracting features from the multi-source heterogeneous data to construct a multi-dimensional feature vector, including: Real-time collection of multi-source heterogeneous data in nursing scenarios, including at least patient health monitoring data, real-time location and work status data of nursing staff, medical equipment usage data, environmental parameter data, and historical service record data; Abnormal noise data caused by sensor failures in multi-source heterogeneous data is removed by threshold verification and time sequence continuity verification to obtain processed multi-source data; Temporal interpolation is used to complete short-term missing data in multi-source data, and duplicate and redundant invalid data are deleted. In the data standardization process, heterogeneous data of different magnitudes and units are normalized to obtain normalized multi-source data. The location coordinates, vital sign values, equipment status and workload in the normalized multi-source data are uniformly mapped to the standard data format, and data desensitization and compliance screening are completed at the same time to obtain cleaned multi-source data. Feature vectors are extracted from the cleaned multi-source data to obtain multi-dimensional feature vectors, which include at least temporal features, spatial features, attribute features, and scene features.

3. The nursing service collaborative scheduling method based on multi-source data fusion as described in claim 1, characterized in that, The deep learning prediction model is constructed based on a spatiotemporal feature extraction layer of a Transformer encoder and a hybrid prediction layer integrating an LSTM (Long Short-Term Memory) network and a GNN (Graph Neural Network). The GNN is used to model the spatial relationship topology between caregivers and patients, including: The spatiotemporal feature extraction layer based on the Transformer encoder is composed of multiple stacked encoders. Each encoder layer contains a multi-head self-attention module and a feedforward neural network module, and is configured with layer normalization and residual connection structures. The multi-head self-attention mechanism maps the input feature vector into multiple independent query, key, and value vectors, calculates the correlation and dependency between different feature dimensions in parallel, and distinguishes the weight ratio of core features and redundant features.

4. The nursing service collaborative scheduling method based on multi-source data fusion as described in claim 1, characterized in that, The deep learning prediction model is constructed based on a spatiotemporal feature extraction layer of a Transformer encoder and a hybrid prediction layer integrating an LSTM (Long Short-Term Memory) network and a GNN (Graph Neural Network). The GNN is used to model the spatial relationship topology between caregivers and patients, including: The hybrid prediction layer adopts a dual-branch parallel fusion architecture of LSTM long short-term memory network and GNN graph neural network; The forget gate of the LSTM Long Short-Term Memory network is responsible for filtering and discarding invalid and redundant information in historical nursing data, the input gate is responsible for filtering, retaining effective temporal features and updating unit memory, and the output gate combines the current memory and feature information to output the temporal inference result. The GNN graph neural network uses the GAT dynamic graph attention algorithm to adapt to dynamic nursing scenarios. It adaptively updates the weight coefficients of each node and edge through the attention mechanism, and models the matching relationship between personnel and patients, the matching relationship between personnel and equipment, and the spatial association topology across nursing units in real time.

5. The nursing service collaborative scheduling method based on multi-source data fusion as described in claim 1, characterized in that, The process of inputting the multi-dimensional feature vector into a deep learning prediction model and outputting prediction results includes at least the patient's nursing needs type, intensity, and spatiotemporal distribution, including: Nursing needs include basic living care, condition monitoring care, specialized treatment care, emergency resuscitation care, and postoperative rehabilitation care; Nursing demand intensity is quantified using nursing hours, number of staff, equipment usage time, and operational complexity as indicators, and is categorized into routine needs, key needs, and emergency high-risk needs. The spatiotemporal distribution of nursing needs includes the time period when the nursing needs occur, the duration of the needs, the ward to which the needs are located, and the spatial distribution of the locations.

6. The nursing service collaborative scheduling method based on multi-source data fusion as described in claim 1, characterized in that, The goal is to minimize response latency and maximize resource utilization, thereby constructing a multi-objective optimization model. Based on deep reinforcement learning algorithms and prediction results, a real-time nursing service scheduling strategy is generated, and a co-evolutionary mechanism is used to realize task allocation and path planning among multiple nursing units, including: Based on the PPO proximal strategy optimization algorithm, the ward nursing scenario is constructed as an intelligent interactive environment, and nursing task allocation, personnel path selection, and equipment allocation are defined as intelligent agent execution actions. With the comprehensive benefit of the multi-objective optimization model as the core reward mechanism, the PPO proximal policy optimization algorithm predicts the future distribution of nursing needs in advance by limiting the policy update range and combining the spatiotemporal prediction results of nursing needs output above during the iteration process.

7. The nursing service collaborative scheduling method based on multi-source data fusion as described in claim 1, characterized in that, The goal is to minimize response latency and maximize resource utilization, thereby constructing a multi-objective optimization model. Based on deep reinforcement learning algorithms and prediction results, a real-time nursing service scheduling strategy is generated, and a co-evolutionary mechanism is used to realize task allocation and path planning among multiple nursing units, including: The scheduling schemes of each independent nursing unit are defined as independent subpopulations, and each subpopulation contains multiple sets of differentiated task allocation and path planning individual schemes. Unified planning of personnel movement and equipment transfer routes across units enables the overall optimization of nursing service scheduling across the hospital.

8. A nursing service collaborative scheduling system based on multi-source data fusion, characterized in that, The nursing service collaborative scheduling system includes the following modules: A multi-dimensional data acquisition module is used to collect multi-source heterogeneous data in nursing scenarios in real time, clean, standardize and extract features from the multi-source heterogeneous data, and construct a multi-dimensional feature vector. The prediction model building module is used to construct a deep learning prediction model, which includes a spatiotemporal feature extraction layer based on a Transformer encoder and a hybrid prediction layer integrating an LSTM long short-term memory network and a GNN graph neural network. The GNN graph neural network is used to model the spatial relationship topology between caregivers and patients. The multi-source data fusion module is used to input the multi-dimensional feature vector into the deep learning prediction model and output the prediction results, which include at least the patient's nursing needs type, intensity, and spatiotemporal distribution. The nursing service scheduling module is used to build a multi-objective optimization model with the goal of minimizing response latency and maximizing resource utilization. Based on deep reinforcement learning algorithms and prediction results, a real-time nursing service scheduling strategy is generated, and a co-evolutionary mechanism is used to realize task allocation and path planning among multiple nursing units.

9. The nursing service collaborative scheduling system based on multi-source data fusion as described in claim 8, characterized in that, The nursing service scheduling module includes the following sub-modules: A sub-module is constructed to build an intelligent interactive environment for ward nursing scenarios based on the PPO proximal strategy optimization algorithm, defining nursing task allocation, personnel path selection, and equipment allocation actions as intelligent agent execution actions. The update submodule is used to use the comprehensive benefit of the multi-objective optimization model as the core reward mechanism. During the iteration process, the PPO proximal policy optimization algorithm predicts the future distribution of nursing needs in advance by limiting the policy update range and combining the spatiotemporal prediction results of nursing needs output above.

10. The nursing service collaborative scheduling system based on multi-source data fusion as described in claim 8, characterized in that, The nursing service scheduling module includes the following sub-modules: Define submodules to define the scheduling schemes of each independent nursing unit as independent subpopulations. Each subpopulation contains multiple sets of differentiated task allocation and path planning individual schemes. The unified submodule is used to uniformly plan the movement of personnel and the transfer of equipment across units, so as to achieve the global optimization of nursing service scheduling throughout the hospital.

Images