Medical space resource management system and method based on digital twinning

Abstract

The invention discloses a medical space resource management system and method based on digital twinning, and the method comprises the steps: collecting hospital multi-source data, and carrying out the de-identification to generate an initial data set; constructing an improved medical space digital twin model, and performing multi-scale state synchronization based on event driving; the model is operated to output an exogenous feature vector, and a hox process is established in combination with an event type; carrying out the maximum likelihood estimation and online updating of the hox process, and outputting condition intensity and a prediction result; the prediction result drives simulation and queuing deduction to generate an executable resource scheduling scheme; the scheme is implemented, feedback is collected, and the hox process and model parameters are updated. According to the method, the improved medical space digital twin model and the Horkes process are fused, dynamic modeling, event prediction and resource scheduling optimization of hospital multi-source data are achieved, and the medical space resource utilization efficiency and the patient doctor-seeing experience are improved.

Description

Technical Field

[0001] This invention relates to the field of medical information technology and intelligent scheduling, and in particular to a medical spatial resource management system and method based on digital twins. Background Technology

[0002] Currently, medical institutions need to rationally allocate treatment space, equipment, medical staff, and patient flow during daily operations to ensure efficient use of medical resources and improved patient access to care. However, existing medical resource allocation methods largely rely on fixed rules, manual experience, or scheduling methods based on static models, making it difficult to reflect dynamic changes within the hospital in a timely manner.

[0003] Patient traffic fluctuates significantly across different time periods and regions. Sudden events (such as emergency room peaks, equipment malfunctions, and public health emergencies) can cause rapid changes in resource demand structures, making accurate prediction and response difficult using traditional methods in a short timeframe. Current applications of digital twin technology in the healthcare field primarily focus on spatial layout visualization and static resource management, lacking the ability to model event-driven dynamics and effectively capturing the interrelationships between events within the healthcare space. Modeling event patterns often relies on statistical analysis or simple time series methods, failing to introduce models suitable for characterizing the triggering and propagation characteristics of events, resulting in insufficient prediction accuracy and the feasibility of scheduling schemes.

[0004] Existing solutions often rely on batch offline data training for model updates, lacking the ability to perform online incremental updates based on sliding time windows and real-time feedback. This makes it impossible to guarantee the model's sensitivity and adaptability to the latest operating state. Consequently, existing medical resource scheduling methods suffer from delayed response, large prediction bias, and uneven resource allocation when dealing with complex and ever-changing hospital operation scenarios, thus affecting the overall quality of medical services and operational efficiency.

[0005] Therefore, how to provide a medical spatial resource management system and method based on digital twins is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a medical spatial resource management system and method based on digital twins. This invention fully utilizes digital twin technology, Hawkes process event modeling, discrete event simulation, and queuing network deduction to accurately characterize the arrival patterns and interrelationships of various events in the medical space. It also combines historical event data and exogenous feature vectors for online incremental updates, dynamically generating executable medical resource scheduling schemes. This invention possesses advantages such as high prediction accuracy, fast response speed, rational resource allocation, and strong adaptability, effectively improving hospital operational efficiency and the quality of medical services.

[0007] A method for managing medical spatial resources based on digital twins according to an embodiment of the present invention includes: Collect multi-source information data related to hospital operations, including historical event data and real-time status data, and perform de-identification processing on the multi-source information data to generate an initial dataset; An improved digital twin model of medical space is constructed based on the initial dataset. An event-driven multi-scale state synchronization mechanism is introduced to trigger local node refresh, regional structure reconstruction and global parameter recalculation according to different event types and prediction timeliness. The model runs in real time in the digital twin model of the medical space and outputs exogenous feature vectors. Based on the initial dataset and exogenous feature vectors, combined with predefined event types, a Hawkes process is established to characterize the arrival patterns and mutual influence relationships of events in the medical space. Using historical event data and exogenous feature vectors, maximum likelihood estimation is performed on the Hawkes process, and online incremental updates are performed under the conditions of sliding time window and forgetting factor, outputting the conditional intensity time function and event intensity prediction results; The conditional intensity time function and event intensity prediction results are input into the digital twin model of the medical space. Discrete event simulation and queuing network deduction are performed to obtain the distribution results. Under the conditions of capacity constraints, staffing constraints and channel accessibility constraints, an executable medical space resource scheduling scheme is generated. The medical space resource allocation plan is handed over to the relevant management department for implementation. Feedback data during the implementation process is collected and compared with the distribution results. The parameters of the Hawkes process and the medical space digital twin model are updated based on the comparison results.

[0008] Optionally, the historical event data specifically includes timestamped event records of patient arrival, diagnosis, transfer, discharge, and activation and release of medical equipment within the medical space, and the real-time status data specifically includes real-time records of patient flow, medical staff location, medical equipment operating status, and environmental parameters within the medical space.

[0009] Optionally, the de-identification processing of multi-source information data specifically includes deleting or replacing fields that can identify an individual and using data desensitization and encoding techniques to protect privacy information.

[0010] Optionally, the improved medical spatial digital twin model constructed based on the initial dataset introduces an event-driven multi-scale state synchronization mechanism, which triggers local node refresh, regional structure reconstruction, and global parameter recalculation according to different event types and prediction timeliness, including: Based on the initial dataset, construct the virtual spatial topology of the digital twin model of the medical space; An event-driven, multi-scale state synchronization mechanism is introduced into the digital twin model of medical space. Based on the virtual space topology, it receives real-time state data and historical event data, and dynamically updates the state information in the virtual space topology in multiple granularities. Specifically: Local node refresh: When an event only affects a single functional node, update the state vector of the node and its directly adjacent nodes; Regional structure reconstruction: When an event affects a functional area, dynamically adjust the spatial topology subgraph and resource allocation rules within the functional area; Global parameter recalculation: When the event is at the hospital-wide level or crosses regions, the global spatial topology parameters and resource distribution probability matrix are recalculated. A patient flow pattern prediction module is integrated into the digital twin model of medical space, including a spatiotemporal feature extraction unit, a path selection modeling unit, and a congestion trend analysis unit. Based on the state vector, spatial topology subgraph, global spatial topology parameters, and resource distribution probability matrix, the module generates patient location distribution prediction, path selection probability, and congestion trend results. The resource scheduling optimization module is integrated into the digital twin model of medical space, including a demand aggregation unit, a strategy generation unit, and a feasibility verification unit. Based on the spatial topology subgraph, global spatial topology parameters, resource distribution probability matrix, as well as patient location distribution prediction, path selection probability, and congestion trend results, the optimal scheduling scheme is selected. A multi-source data consistency verification module is integrated into the medical space digital twin model, including a data comparison unit, a deviation location unit, and a correction execution unit. The deviation is detected and located according to the optimal scheduling scheme, and when the deviation exceeds the threshold, the state parameters and feature weights of the corrected medical space digital twin model are output. By deploying a digital twin model of the medical space onto a visualization and interactive platform, the operational status of the medical space, patient flow trends, and resource scheduling plans can be dynamically visualized and displayed, and manual intervention and rapid recalculation of the scheduling plans can be supported.

[0011] Optionally, the step of running and outputting exogenous feature vectors in real time within the digital twin model of the medical space, and establishing a Hawkes process to characterize the arrival patterns and interrelationships of events in the medical space based on the initial dataset and exogenous feature vectors, combined with predefined multiple event types, includes: The exogenous feature vector is run in real time in the medical space digital twin model and outputs it. The exogenous feature vector is aligned with the initial dataset according to the event occurrence time to form joint data indexed by the event occurrence time. Based on predefined event types, an event type set is established, and historical events in the joint data are type-bound and time-labeled to generate a time-series feature set. A Hawkes process structure description is established based on a time series feature set. The Hawkes process structure description consists of a baseline arrival component, an exogenous feature vector influence component, and self-excitation and mutual excitation components of historical events. Time decay rules are set according to three triggering levels: intra-node, intra-region, and cross-region. In the Hawkes process structure description, a causal gating mechanism is introduced. With the spatial topological subgraph as a constraint, only the triggering links with valid causal paths from the triggering event to the target event are retained, resulting in a set of gated triggering relationships. A resource constraint suppression mechanism is introduced into the set of gated trigger relationships. Based on the global spatial topology parameters, resource distribution probability matrix and resource occupancy information in the state vector, when the occupancy level exceeds the preset threshold, the corresponding triggering effect is weakened and non-negative truncation is performed to obtain the set of suppressed and corrected trigger relationships. The Hawkes process structure description, the set of gated trigger relationships, the set of suppressed and corrected trigger relationships, and the joint data indexed by the event occurrence time are integrated and solidified into a multi-category condition strength calculation configuration.

[0012] Optionally, the output conditional intensity time function and event intensity prediction results include: Receive joint data and exogenous feature vectors, classify and organize the joint data according to event categories, and arrange the exogenous feature vectors of each event category in chronological order to generate a three-dimensional data structure of event category-time-exogenous feature. Based on the three-dimensional data structure, the initial intensity parameters, exogenous feature influence coefficients, and triggering influence coefficients of each event category are initialized. The values ​​of the initial intensity parameters, exogenous feature influence coefficients, and triggering influence coefficients are calculated based on the historical average occurrence frequency and the mean value of the exogenous features. By utilizing historical event data and corresponding exogenous feature vectors, the optimal parameter combination is obtained based on the maximum likelihood estimation method. In the real-time prediction phase, the latest collected event occurrence time and exogenous feature vector are added to the joint data in chronological order, and the condition strength is calculated within a sliding time window of a set length. The impact of historical event triggering within the window is weighted, and the weighting coefficient is calculated by decreasing the time difference between the event occurrence time and the current prediction time. At each prediction time, based on the calculation rules of the conditional intensity time function, the conditional intensity values ​​of each event category at the current and multiple future time steps are calculated, generating the corresponding event intensity prediction results, which are then classified and summarized according to the event category.

[0013] Optionally, generating an executable medical space resource scheduling scheme includes: The system receives the conditional intensity time function and the event intensity prediction results, maps the event type set to the prediction time period, and establishes the arrival ratio relationship between event types and functional nodes based on the functional nodes of the medical space digital twin model. Within the prediction period, an event arrival sequence is generated using a conditional intensity time function and distributed to each functional node according to the mapping relationship, forming a node arrival data stream; In the digital twin model of medical space, discrete event simulation and queuing network deduction are performed based on the node arrival data flow. The number of service stations, service rate, capacity limit and service discipline are set. The queue length, utilization rate and transfer time and distance of each functional node are recorded to generate samples of waiting time, occupancy rate and transfer distance. Based on the sample data, the distribution and mean of waiting time, the distribution and mean of occupancy rate, and the distribution and mean of overall transfer distance of each node are statistically analyzed to form a resource distribution result set covering multiple dimensions. Based on the resource distribution result set, a resource scheduling plan is formulated, taking bed allocation, time slot arrangement, personnel scheduling and transfer route as joint decision variables. Under the conditions of meeting capacity, personnel and channel constraints, the waiting time, utilization rate and transfer distance are optimized, and the resource scheduling plan including bed allocation, time slot arrangement, personnel and route assignment and execution time is output.

[0014] A medical spatial resource management system based on digital twins according to an embodiment of the present invention includes the following modules: The data acquisition module is used to collect historical events and real-time status data, and generate an initial dataset after de-identification. The twin module is used to build an improved digital twin model of the medical space based on the initial dataset; The modeling module is used to output exogenous feature vectors, and combine them with the initial dataset and predefined event types to establish a Hawkes process to characterize the arrival patterns and mutual influence relationships of events. The calculation module is used to perform maximum likelihood estimation of the Hawkes process based on historical event data and exogenous feature vectors, and output the conditional intensity time function and event intensity prediction results. The scheduling module is used to input the conditional intensity time function and event intensity prediction results into the medical space digital twin model to generate a resource scheduling scheme; The update module is used to implement resource scheduling schemes and collect feedback data, compare it with the distribution results, and update the parameters of the Hawkes process and the digital twin model of the medical space based on the comparison results.

[0015] The beneficial effects of this invention are: This invention effectively solves the problems of delayed event response, insufficient prediction accuracy, and lack of dynamic adaptability in resource scheduling in existing technologies by introducing an event-driven multi-scale state synchronization mechanism, Hawkes process event modeling, and online incremental update method into a digital twin model of medical space, combined with discrete event simulation and queuing network deduction.

[0016] This invention can, based on the collection of multi-source information data, depict the arrival patterns and mutual influence relationships of different event types in real time, and continuously update the model parameters under the conditions of sliding time window and forgetting factor, thereby ensuring the timeliness and accuracy of event prediction results.

[0017] Resource scheduling solutions generated by combining capacity constraints, staffing constraints, and accessibility constraints are more feasible and implementable. They can significantly improve the hospital's operational efficiency during emergencies and peak periods, reduce resource waste, optimize the patient experience, and enhance the overall stability and sustainable operation of the medical service system. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a medical spatial resource management method based on digital twins proposed in this invention; Figure 2 This is a schematic diagram of the structure of a medical spatial resource management system based on digital twin proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1 A digital twin-based method for managing medical spatial resources includes: Collect multi-source information data related to hospital operations, including historical event data and real-time status data, and perform de-identification processing on the multi-source information data to generate an initial dataset; An improved digital twin model of medical space is constructed based on the initial dataset. An event-driven multi-scale state synchronization mechanism is introduced to trigger local node refresh, regional structure reconstruction and global parameter recalculation according to different event types and prediction timeliness. The model runs in real time in the digital twin model of the medical space and outputs exogenous feature vectors. Based on the initial dataset and exogenous feature vectors, combined with predefined event types, a Hawkes process is established to characterize the arrival patterns and mutual influence relationships of events in the medical space. Using historical event data and exogenous feature vectors, maximum likelihood estimation is performed on the Hawkes process, and online incremental updates are performed under the conditions of sliding time window and forgetting factor, outputting the conditional intensity time function and event intensity prediction results; The conditional intensity time function and event intensity prediction results are input into the digital twin model of the medical space. Discrete event simulation and queuing network deduction are performed to obtain the distribution results. Under the conditions of capacity constraints, staffing constraints and channel accessibility constraints, an executable medical space resource scheduling scheme is generated. The medical space resource allocation plan is handed over to the relevant management department for implementation. Feedback data during the implementation process is collected and compared with the distribution results. The parameters of the Hawkes process and the medical space digital twin model are updated based on the comparison results.

[0021] In this embodiment, the historical event data specifically includes timestamped event records of patient arrival, diagnosis, transfer, discharge, and activation and release of medical equipment within the medical space, and the real-time status data specifically includes real-time records of patient flow, medical staff location, medical equipment operating status, and environmental parameters within the medical space.

[0022] In this embodiment, the de-identification processing of multi-source information data specifically includes deleting or replacing fields that can identify an individual and using data desensitization and encoding techniques to protect privacy information.

[0023] In this embodiment, the improved medical spatial digital twin model constructed based on the initial dataset introduces an event-driven multi-scale state synchronization mechanism, which triggers local node refresh, regional structure reconstruction, and global parameter recalculation according to different event types and prediction timeliness, including: Based on the initial dataset, a virtual spatial topology of a digital twin model of medical space is constructed. The virtual spatial topology consists of spatial node units, connecting channel units, and functional label units. The spatial node units are used to represent the virtual mapping of each functional area, the connecting channel units are used to represent the travel paths between functional areas, and the functional label units are used to label the purpose and resource attributes of each spatial node. An event-driven, multi-scale state synchronization mechanism is introduced into the digital twin model of medical space. Based on the virtual space topology, it receives real-time state data and historical event data, and dynamically updates the state information in the virtual space topology in multiple granularities. Specifically: Local node refresh: When an event only affects a single functional node, update the state vector of the node and its directly adjacent nodes; Regional structure reconstruction: When an event affects a functional area, dynamically adjust the spatial topology subgraph and resource allocation rules within the functional area; Global parameter recalculation: When the event is at the hospital-wide level or crosses regions, the global spatial topology parameters and resource distribution probability matrix are recalculated. A patient flow pattern prediction module is integrated into the digital twin model of medical space, including a spatiotemporal feature extraction unit, a path selection modeling unit, and a congestion trend analysis unit, wherein: The spatiotemporal feature extraction unit receives state vectors, spatial topology subgraphs, global spatial topology parameters, and resource distribution probability matrices. It then combines historical event data to extract patient distribution features for each time period and functional area. Specifically, this means aligning and mapping historical patient arrival, stay, and transfer data with state vectors, spatial topology subgraphs, global spatial topology parameters, and resource distribution probability matrices on the time axis and spatial nodes. It then statistically analyzes the arrival rate, number of patients in the hospital, stay duration, transfer probability, and resource occupancy for each time period and functional area. Finally, it performs standardization, smoothing, and missing data completion processing to form a feature vector that reflects the spatiotemporal distribution pattern of patients and the relationship between resource constraints. The path selection construction unit determines the path selection probability distribution of patients between different nodes based on patient distribution characteristics. Specifically, determining the path selection probability distribution of patients between different nodes based on patient distribution characteristics means mapping the extracted patient distribution characteristics of each time period and functional area to the spatial topology of the medical space, calculating the transfer frequency of patients flowing from one functional node to another in a specific time period, and combining historical path data, node capacity limitations, resource occupancy levels, and patient flow patterns to obtain the transfer probability matrix between each pair of nodes through probability statistics and normalization calculations, forming a quantitative probability distribution result that reflects the possible path selection of patients under different time and space conditions. The congestion trend analysis unit, based on path selection results, calculates potential traffic bottlenecks and generates multi-timestep predictions of patient location distribution, path selection probabilities, and congestion trend results, specifically: The probability distribution of path selection is compared with the medical space topology and node capacity data to calculate the expected traffic flow of each path and node at different time periods. When the expected traffic flow exceeds the maximum capacity of a node or path, it is marked as a potential traffic bottleneck, and the location, duration and congestion risk level are recorded. Using the path selection probability and node transition probability matrix as input, combined with historical patient flow patterns, and through multi-time-step iterative calculation, the distribution of patients at each functional node at each future time step is deduced, and the patient location distribution probability matrix at each time step is output to predict the spatial occupancy status of the patient group at future times. Based on the multi-time step location distribution results, the load rate change trend of each node and path is calculated. Combined with the location and duration of traffic bottlenecks, a congestion trend curve is generated and quantified into a congestion index sequence. A resource scheduling and optimization module is integrated into the digital twin model of the medical space, including a demand aggregation unit, a strategy generation unit, and a feasibility verification unit, wherein: The demand aggregation unit receives the spatial topology subgraph, global spatial topology parameters, and resource distribution probability matrix, as well as patient location distribution prediction, path selection probability, and congestion trend results. It then statistically analyzes and merges the resource demands of different regions and time periods to form a resource demand distribution. Specifically, at each prediction time step, the predicted patient location distribution results are matched with the corresponding region's functional type, available resource types, and quantity parameters. Based on the unit resource consumption of different medical service types, the resource demand values ​​of each functional region are calculated. The resource demand values ​​are then aggregated according to time order and spatial location, merging resource quantities with small demand fluctuations in adjacent time periods. Finally, the resource demand trends of different time periods within the same region are weighted and smoothed to form a resource demand distribution matrix covering all functional regions and the entire prediction time period. The strategy generation unit generates multiple candidate scheduling schemes based on the spatial topology subgraph, global spatial topology parameters, resource distribution probability matrix, and set constraints, specifically: The input spatial topology subgraph and global spatial topology parameters are analyzed. The connectivity, spatial distance, passage capacity and available resources in the resource distribution probability matrix between medical functional areas are combined with the set constraints to calculate the feasible solution space and eliminate combinations that violate capacity constraints, path passage constraints or resource availability constraints. Within the feasible solution space, a genetic algorithm is invoked to perform a combined search of resource allocation order, path selection, and time window. The algorithm is scored and ranked according to multiple objective functions such as maximizing service efficiency, improving resource utilization, and minimizing congestion risk, and outputs multiple candidate scheduling schemes that meet different priority requirements. The feasibility verification unit conducts feasibility analysis and performance evaluation of candidate scheduling schemes, and selects the optimal scheduling scheme. Specifically, the feasibility analysis and performance evaluation of candidate scheduling schemes refers to verifying the constraints of the spatial path, resource allocation and time arrangement of the candidate scheduling schemes, eliminating schemes that do not meet the medical space topology, resource capacity, safety specifications and conflict detection, and performing weighted scoring and simulation prediction based on execution time, resource utilization rate, congestion risk index and patient waiting time, comprehensively evaluating the stability and coping ability of candidate scheduling schemes under different operating conditions, and finally selecting the optimal scheduling scheme. A multi-source data consistency verification module is integrated into the digital twin model of the medical space, including a data comparison unit, a deviation location unit, and a correction execution unit, wherein: The data comparison unit receives real-time status data after the execution of the optimal scheduling scheme and prediction data from the medical space digital twin model, and performs a time-series and numerical consistency comparison. Specifically, the time-series and numerical consistency comparison refers to matching the real-time status data collected after the execution of the optimal scheduling scheme with the prediction data generated by the medical space digital twin model on the same time axis, calculating and analyzing the numerical differences and trend deviations at each corresponding time point, and judging the degree of consistency between the prediction and the actual execution results in terms of time sequence and numerical changes. The deviation localization unit identifies the specific source and scope of influence of the deviation based on the comparison results, specifically: By analyzing the numerical differences and trend deviations at various time points, and combining the input data, event triggering conditions, and external interference factors of the medical space digital twin model, we can trace back to the specific links that caused the deviations, including the data acquisition end, the model calculation process, or the scheduling and execution link. The identified sources of deviation are correlated with the medical space topology, patient flow paths, and resource scheduling links to determine their impact on the performance of single-area, multi-area linkage, or global scheduling, and to quantify the potential interference with future predictions and resource allocation. When the deviation exceeds the threshold, the correction execution unit automatically adjusts the state parameters and feature weights of the medical space digital twin model. Specifically, when the deviation exceeds the set threshold, the unit dynamically updates the numerical state quantities and weight coefficients of each input feature in the medical space digital twin model based on the deviation positioning results, so that the calculated output of the medical space digital twin model is highly matched with the actual operating state. By deploying a digital twin model of the medical space onto a visualization and interactive platform, the operational status of the medical space, patient flow trends, and resource scheduling plans can be dynamically visualized and displayed, and manual intervention and rapid recalculation of the scheduling plans can be supported.

[0024] This invention improves the timeliness, accuracy, and feasibility of medical space resource management through an overall architecture that integrates virtual space topology, event-driven multi-scale synchronization, three-unit prediction of patient flow, resource optimization decision-making, and a closed-loop consistency verification. It uses multi-granularity state synchronization to frequently update local, regional, and hospital-wide events to the digital twin, avoiding the lag of traditional medical space digital twin models. By coupling spatiotemporal feature extraction, path selection construction, and congestion trend analysis, it provides advance predictions of patient location distribution and traffic bottlenecks at multiple time steps, more closely reflecting the actual load than rule-based predictions. A genetic algorithm generates multiple executable candidate solutions under capacity, personnel, and channel constraints, and the optimal solution is selected through feasibility testing, achieving a comprehensive improvement that considers waiting time, utilization rate, and congestion risk. Online calibration, consisting of data comparison, deviation location, and correction execution, ensures that the output of the medical space digital twin model continuously aligns with the on-site status, guaranteeing long-term stable operation. This allows for continuous reduction of average waiting time, improvement of node utilization, and reduction of channel congestion during peak and emergency situations. It also supports rapid intervention and recalculation through visual interaction, forming an integrated closed loop of prediction, simulation, scheduling, and feedback, which has strong robustness, interpretability, and engineering feasibility.

[0025] In this embodiment, the step of running and outputting exogenous feature vectors in real time within a digital twin model of the medical space, and establishing a Hawkes process to characterize the arrival patterns and interrelationships of events in the medical space based on the initial dataset and exogenous feature vectors, combined with predefined multiple event types, includes: In the medical space digital twin model, exogenous feature vectors are run in real time and output. The exogenous feature vectors are aligned with the initial dataset according to the event occurrence time to form joint data indexed by the event occurrence time. Specifically, running and outputting exogenous feature vectors in real time in the medical space digital twin model means continuously acquiring the current status of each functional node in the twin at fixed time steps, calculating and organizing queue length, available beds, available equipment time slots, number of on-duty personnel, channel accessibility, local spatial topology identifier, global spatial topology parameters, resource distribution probability and time period identifier, and concatenating them into exogenous feature vectors with timestamps according to the preset field order and writing them into the output buffer. Based on predefined event types, an event type set is established. Historical events in the joint data are type-bound and time-labeled to generate a time-series feature set containing event categories, time labels, and exogenous feature descriptions. A Hawkes process structure description is established based on a time series feature set. The Hawkes process structure description consists of a baseline arrival component, an exogenous feature vector influence component, and self-excitation and mutual excitation components of historical events. Time decay rules are set according to three triggering levels: intra-node, intra-region, and cross-region. In the Hawkes process structure description, a causal gating mechanism is introduced. With the spatial topological subgraph as a constraint, only the triggering links with valid causal paths from the triggering event to the target event are retained, resulting in a set of gated triggering relationships. A resource constraint suppression mechanism is introduced into the set of gated trigger relationships. Based on the global spatial topology parameters, resource distribution probability matrix and resource occupancy information in the state vector, when the occupancy level exceeds the preset threshold, the corresponding triggering effect is weakened and non-negative truncation is performed to obtain the set of suppressed and corrected trigger relationships. The Hawkes process structure description, the set of gated trigger relationships, the set of suppressed and corrected trigger relationships, and the joint data indexed by the event occurrence time are integrated and solidified into a multi-category condition strength calculation configuration.

[0026] This invention generates and aligns exogenous feature vectors in real time within twins, forming joint data indexed by event occurrence time. This eliminates temporal misalignment between features and events, improving the consistency and usability of input data. Based on the Hawkes structure description of "baseline arrival + exogenous influence + self-excitation and mutual excitation," it separates the modeling of external states and historical triggers, enhancing the interpretability of intensity sources. Causal gating is introduced, retaining only trigger links with valid causal paths according to spatial topology, suppressing noise correlation and false triggers. Combined with resource constraint suppression, it proactively reduces the impact of corresponding triggers when capacity is tight or channels are limited, ensuring that intensity evolution is consistent with physical feasibility and avoiding unexecutable high-intensity predictions and numerical instability. Finally, the structure description, gating links, and suppression results are integrated and solidified into a multi-category conditional intensity calculation configuration, facilitating rapid invocation of maximum likelihood and incremental updates, and shortening training and inference latency. This enables more accurate arrival intensity and queuing situation prediction in peak and sudden scenarios, reducing false alarms and overfitting risks, improving the executability and stability of scheduling schemes, and demonstrating comprehensive benefits such as reduced waiting time, increased node utilization, and congestion relief in practical tests.

[0027] In this embodiment, the output conditional intensity time function and event intensity prediction results include: Receive joint data and exogenous feature vectors, classify and organize the joint data according to event categories, and arrange the exogenous feature vectors of each event category in chronological order to generate a three-dimensional data structure of event category-time-exogenous feature. Based on the three-dimensional data structure, the initial intensity parameters, exogenous feature influence coefficients, and triggering influence coefficients of each event category are initialized. The values ​​of the initial intensity parameters, exogenous feature influence coefficients, and triggering influence coefficients are calculated based on the historical average occurrence frequency and the mean value of the exogenous features. Using historical event data and corresponding exogenous feature vectors, the optimal parameter combination is obtained based on the maximum likelihood estimation method, specifically: Historical events are sorted by event type and occurrence time. The corresponding exogenous feature vectors are aligned to each event and adjacent time periods along the same time axis. Based on the established Hawkes structure, the conditional strength at the time of occurrence of each event and the intensity integral within the observation interval are calculated and accumulated to form a log-likelihood target. To prevent overfitting and unexplainable triggering, sparsity or boundary regularization is added and non-negative and capacity-related physical constraints are fixed. Using the baseline strength parameter, exogenous feature weights, mutual excitation coefficients, and time decay coefficients as parameters to be estimated, the system initializes from historical statistics and then iteratively updates using a quasi-Newton method. The conditional strength and objective function are recalculated and the feasibility of the constraints are checked after each update. The process stops when the log-likelihood gain is less than the threshold or the gradient norm is lower than the threshold, thus obtaining the optimal parameter combination that maximizes the log-likelihood under the conditions of historical events and exogenous features. In real-time prediction, the latest event time and exogenous feature vector are added to the joint data in chronological order, and the condition strength is calculated within a sliding time window of a set length. The triggering influence of historical events within the window is weighted in descending order of time difference. Specifically, calculating the condition strength within a sliding time window of a set length means that during real-time prediction, after merging the current moment and its latest exogenous feature vector with the existing joint data, only event records with a time span not exceeding the preset window length are retained as the calculation range. Within the calculation range, relevant historical events are first filtered by event category, and then the trigger contribution value is calculated one by one according to the interval between the current prediction time and the occurrence time of each historical event, based on the weight function that decreases in descending order of time difference. At the same time, the trigger contribution values ​​are accumulated by combining the baseline arrival level and the influence increment of the exogenous feature vector to obtain the instantaneous condition strength for each event category within the current sliding window. At each prediction time, based on the calculation rules of the conditional intensity time function, the conditional intensity values ​​of each event category at the current and multiple future time steps are calculated, generating the corresponding event intensity prediction results, which are then classified and summarized according to the event category.

[0028] In this embodiment, generating an executable medical space resource scheduling scheme includes: The system receives the conditional intensity time function and the event intensity prediction results, maps the event type set to the prediction time period, and establishes the arrival ratio relationship between event types and functional nodes based on the functional nodes of the medical space digital twin model. Within the prediction period, an event arrival sequence is generated using a conditional intensity time function and distributed to each functional node according to the mapping relationship, forming a node arrival data stream; In the digital twin model of medical space, discrete event simulation and queuing network deduction are performed based on the node arrival data flow. The number of service stations, service rate, capacity limit and service discipline are set. The queue length, utilization rate and transfer time and distance of each functional node are recorded to generate samples of waiting time, occupancy rate and transfer distance. Based on the sample data, the distribution and mean of waiting time, the distribution and mean of occupancy rate, and the distribution and mean of overall transfer distance of each node are statistically analyzed to form a resource distribution result set covering multiple dimensions. Based on the resource distribution result set, a resource scheduling plan is formulated, taking bed allocation, time slot arrangement, personnel scheduling and transfer route as joint decision variables. Under the conditions of meeting capacity, personnel and channel constraints, the waiting time, utilization rate and transfer distance are optimized, and the resource scheduling plan including bed allocation, time slot arrangement, personnel and route assignment and execution time is output.

[0029] In this embodiment, the step of entrusting the medical space resource scheduling scheme to relevant management departments for implementation, collecting feedback data during the implementation process, comparing it with the distribution results, and updating the parameters of the Hawkes process and the medical space digital twin model based on the comparison results includes: During the implementation of the medical space resource scheduling plan, feedback data including resource usage, task completion progress, service quality indicators and changes in the external environment are collected and aligned one-to-one with the scheduling distribution results according to timestamps and resource identifiers. The aligned feedback data and scheduling distribution results are analyzed for differences in time and numerical dimensions to identify deviations in resource utilization efficiency, task execution latency, and service response quality, and to locate the key resource units and task links that cause the deviations. Based on the deviation identification results, the event triggering parameters and baseline arrival rate of the Hawkes process model are adjusted, and the state parameters, feature weights, and exogenous feature influence coefficients of the medical space digital twin model are corrected.

[0030] refer to Figure 2 A medical spatial resource management system based on digital twins includes the following modules: The data acquisition module is used to collect historical events and real-time status data, and generate an initial dataset after de-identification. The twin module is used to build an improved digital twin model of the medical space based on the initial dataset; The modeling module is used to output exogenous feature vectors, and combine them with the initial dataset and predefined event types to establish a Hawkes process to characterize the arrival patterns and mutual influence relationships of events. The calculation module is used to perform maximum likelihood estimation of the Hawkes process based on historical event data and exogenous feature vectors, and output the conditional intensity time function and event intensity prediction results. The scheduling module is used to input the conditional intensity time function and event intensity prediction results into the medical space digital twin model to generate a resource scheduling scheme; The update module is used to implement resource scheduling schemes and collect feedback data, compare it with the distribution results, and update the parameters of the Hawkes process and the digital twin model of the medical space based on the comparison results. Example

[0031] To verify the feasibility of this invention in practice, it was applied to the comprehensive emergency and inpatient center of a tertiary hospital. On [date missing], the emergency and inpatient center experienced a peak in winter respiratory disease visits. Patients arrived in large numbers during the morning emergency triage and observation periods, while the afternoon saw an overlap of outpatient and imaging examination demands, leading to a simultaneous increase in pressure on inpatient beds and ICU transfer channels. Existing manual and fixed-rule scheduling was insufficient to respond to instantaneous fluctuations and cross-departmental coordination, often resulting in the contradiction of "long queues in some areas and low overall utilization."

[0032] This invention constructs a digital twin of a medical space in a hospital, using floors, departments, examination rooms, and corridors as basic units. It continuously collects historical event data and real-time status data, forming joint data indexed by event occurrence time. The twin model runs in real-time and outputs exogenous feature vectors. Hawkes processes are established according to predefined event types to characterize arrival patterns and mutual influence relationships. Maximum likelihood estimation is performed based on historical data and exogenous features, combined with online incremental updates using a sliding time window and forgetting factor, to obtain predictions of conditional strength and event intensity over time. The conditional strength and prediction results are fed into the twin, and discrete event simulation and queuing network deduction are performed to generate distribution results such as waiting time, occupancy rate, and transfer distance. Subsequently, an executable resource scheduling scheme is output according to capacity constraints, staffing constraints, and corridor accessibility constraints. During scheme implementation, real-time feedback is collected and compared with the distribution results. If the deviation exceeds a threshold, the twin state parameters and feature weights are automatically corrected, forming a closed-loop optimization.

[0033] Starting at 8:00 AM that day, the system issued an early warning for increased arrivals at the emergency triage station within two hours. The system's twin mechanism suggested expanding the frontline triage station from two work units to three and temporarily adding extra triage lanes. Around 10:00 AM, in response to the shortage of beds in the emergency observation area, the system transferred some mild cases to the general outpatient assessment pathway and released observation beds. In the afternoon, to address the overlapping peaks in CT and MRI examinations, the system fine-tuned the time slots and queuing discipline, staggering the examination batches for cardiology and respiratory departments. During bed shift changes in Ward 7A of the inpatient department around noon, the system pre-arranged the discharge settlement, disinfection, and bed transfer sequences to ensure a smooth ICU transfer route and reduce waiting times in corridors and secondary transfers. The entire process required no additional learning of the new system from medical staff; they only needed to confirm and implement the suggestions at their departmental terminals.

[0034] Table 1. Comparison of Key Functional Node Scheduling Effectiveness on Peak Day (Month xx, 20xx) at a Tertiary Hospital Functional Nodes Time period (20xx / xx / xx) Arrival volume (people / hour) Average wait time (baseline, minutes) Average wait time (system, minutes) Occupancy rate (baseline, %) System utilization (%) Emergency triage desk 08:00–10:00 42 52 34 83 92 Emergency observation area 10:00–12:00 24 61 39 81 90 Respiratory outpatient clinic 14:00–16:00 18 46 31 78 89 Cardiology Outpatient Clinic 09:00–11:00 17 41 28 77 88 CT Imaging Room 13:00–15:00 11 36 25 75 87 MRI Imaging Room 15:00–17:00 6 49 33 72 86 Ward 7A (beds) 12:00–16:00 5 95 62 88 93 ICU transfer corridor 10:00–14:00 2 58 40 85 92 As shown in Table 1, the optimized scheduling and resource allocation based on this system significantly improved the baseline situation across different functional nodes and time periods. For example, at the emergency triage station between 8:00 and 10:00 AM, the arrival rate was 42 people per hour, with a baseline average waiting time of 52 minutes. After system optimization, this decreased to 34 minutes, a reduction of approximately 34.6%, and the occupancy rate increased from 83% to 92%, demonstrating that higher resource utilization was achieved while maintaining high service efficiency. At the emergency observation area between 10:00 and 12:00 PM, the average waiting time decreased from 61 minutes to 39 minutes, and the occupancy rate increased from 81% to 90%, reflecting the dual effectiveness of congestion relief and improved service efficiency.

[0035] Among outpatient clinics, both the respiratory department (14:00–16:00) and the cardiology department (09:00–11:00) showed improved performance: the average waiting time in the respiratory department decreased from 46 minutes to 31 minutes, with an occupancy rate increase of 11 percentage points; the waiting time in the cardiology department decreased from 41 minutes to 28 minutes, with an occupancy rate increase to 88%. This indicates that system optimization not only improved the speed of patient flow but also increased outpatient capacity. Imaging examinations also benefited significantly, with waiting times in the CT imaging room decreasing by approximately 30.6% from 13:00–15:00 and in the MRI imaging room by approximately 32.7%, both with occupancy rates approaching or exceeding 85%, indicating that the bottleneck in imaging examinations has been effectively alleviated.

[0036] For inpatient and intensive care units, the waiting time for beds in Ward 7A of the inpatient department between 12:00 and 16:00 was reduced from 95 minutes to 62 minutes, and the waiting time for ICU transfers between 10:00 and 14:00 was reduced from 58 minutes to 40 minutes, with significant improvements in occupancy rates. This is of great significance in terms of rapid transfer of critically ill patients and improving bed turnover rates. Overall, the optimized scheduling not only significantly reduced patient waiting times but also significantly improved resource utilization, and the improvements covered multiple aspects including emergency, outpatient, imaging, inpatient, and ICU transfers, fully validating the effectiveness and feasibility of this system in medical space resource coordination and patient flow optimization.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for managing medical spatial resources based on digital twins, characterized in that, include: Collect multi-source information data related to hospital operations, including historical event data and real-time status data, and perform de-identification processing on the multi-source information data to generate an initial dataset; An improved digital twin model of medical space is constructed based on the initial dataset. An event-driven multi-scale state synchronization mechanism is introduced to trigger local node refresh, regional structure reconstruction and global parameter recalculation according to different event types and prediction timeliness. The model runs in real time in the digital twin model of the medical space and outputs exogenous feature vectors. Based on the initial dataset and exogenous feature vectors, combined with predefined event types, a Hawkes process is established to characterize the arrival patterns and mutual influence relationships of events in the medical space. Using historical event data and exogenous feature vectors, maximum likelihood estimation is performed on the Hawkes process, and online incremental updates are performed under the conditions of sliding time window and forgetting factor, outputting the conditional intensity time function and event intensity prediction results; The conditional intensity time function and event intensity prediction results are input into the digital twin model of the medical space. Discrete event simulation and queuing network deduction are performed to obtain the distribution results. Under the conditions of capacity constraints, staffing constraints and channel accessibility constraints, an executable medical space resource scheduling scheme is generated. The medical space resource allocation plan is handed over to the relevant management department for implementation. Feedback data during the implementation process is collected and compared with the distribution results. The parameters of the Hawkes process and the medical space digital twin model are updated based on the comparison results.

2. The medical spatial resource management method based on digital twins according to claim 1, characterized in that, The historical event data specifically includes timestamped event records of patient arrival, diagnosis, transfer, discharge, and activation and release of medical equipment within the medical space. The real-time status data specifically includes real-time records of patient flow, medical staff location, medical equipment operating status, and environmental parameters within the medical space.

3. The medical spatial resource management method based on digital twins according to claim 1, characterized in that, The de-identification process for multi-source information data specifically includes deleting or replacing fields that can identify an individual and using data anonymization and encoding techniques to protect privacy information.

4. The medical spatial resource management method based on digital twins according to claim 1, characterized in that, The improved medical spatial digital twin model, constructed based on the initial dataset, introduces an event-driven multi-scale state synchronization mechanism. This mechanism triggers local node refresh, regional structure reconstruction, and global parameter recalculation according to different event types and prediction timeliness, including: Based on the initial dataset, construct the virtual spatial topology of the digital twin model of the medical space; An event-driven, multi-scale state synchronization mechanism is introduced into the digital twin model of medical space. Based on the virtual space topology, it receives real-time state data and historical event data, and dynamically updates the state information in the virtual space topology in multiple granularities. Specifically: Local node refresh: When an event only affects a single functional node, update the state vector of the node and its directly adjacent nodes; Regional structure reconstruction: When an event affects a functional area, dynamically adjust the spatial topology subgraph and resource allocation rules within the functional area; Global parameter recalculation: When the event is at the hospital-wide level or crosses regions, the global spatial topology parameters and resource distribution probability matrix are recalculated. A patient flow pattern prediction module is integrated into the digital twin model of medical space, including a spatiotemporal feature extraction unit, a path selection modeling unit, and a congestion trend analysis unit. Based on the state vector, spatial topology subgraph, global spatial topology parameters, and resource distribution probability matrix, the module generates patient location distribution prediction, path selection probability, and congestion trend results. The resource scheduling optimization module is integrated into the digital twin model of medical space, including a demand aggregation unit, a strategy generation unit, and a feasibility verification unit. Based on the spatial topology subgraph, global spatial topology parameters, resource distribution probability matrix, as well as patient location distribution prediction, path selection probability, and congestion trend results, the optimal scheduling scheme is selected. A multi-source data consistency verification module is integrated into the medical space digital twin model, including a data comparison unit, a deviation location unit, and a correction execution unit. The deviation is detected and located according to the optimal scheduling scheme, and when the deviation exceeds the threshold, the state parameters and feature weights of the corrected medical space digital twin model are output. By deploying a digital twin model of the medical space onto a visualization and interactive platform, the operational status of the medical space, patient flow trends, and resource scheduling plans can be dynamically visualized and displayed, and manual intervention and rapid recalculation of the scheduling plans can be supported.

5. A method for managing medical spatial resources based on digital twins according to claim 1, characterized in that, The process involves running in real-time within a digital twin model of the medical space and outputting exogenous feature vectors. Based on the initial dataset and the exogenous feature vectors, and combined with predefined multiple event types, a Hawkes process is established to characterize the arrival patterns and interrelationships of events in the medical space, including: The exogenous feature vector is run in real time in the medical space digital twin model and outputs it. The exogenous feature vector is aligned with the initial dataset according to the event occurrence time to form joint data indexed by the event occurrence time. Based on predefined event types, an event type set is established. Historical events in the joint data are type-bound and time-labeled to generate a time-series feature set containing event categories, time labels, and exogenous feature descriptions. A Hawkes process structure description is established based on a time series feature set. The Hawkes process structure description consists of a baseline arrival component, an exogenous feature vector influence component, and self-excitation and mutual excitation components of historical events. Time decay rules are set according to three triggering levels: intra-node, intra-region, and cross-region. In the Hawkes process structure description, a causal gating mechanism is introduced. With the spatial topological subgraph as a constraint, only the triggering links with valid causal paths from the triggering event to the target event are retained, resulting in a set of gated triggering relationships. A resource constraint suppression mechanism is introduced into the set of gated trigger relationships. Based on the global spatial topology parameters, resource distribution probability matrix and resource occupancy information in the state vector, when the occupancy level exceeds the preset threshold, the corresponding triggering effect is weakened and non-negative truncation is performed to obtain the set of suppressed and corrected trigger relationships. The Hawkes process structure description, the set of gated trigger relationships, the set of suppressed and corrected trigger relationships, and the joint data indexed by the event occurrence time are integrated and solidified into a multi-category condition strength calculation configuration.

6. The medical spatial resource management method based on digital twin according to claim 1, characterized in that, The output conditional intensity time function and event intensity prediction results include: Receive joint data and exogenous feature vectors, classify and organize the joint data according to event categories, and arrange the exogenous feature vectors of each event category in chronological order to generate a three-dimensional data structure of event category-time-exogenous feature. Based on the three-dimensional data structure, the initial intensity parameters, exogenous feature influence coefficients, and triggering influence coefficients of each event category are initialized. The values ​​of the initial intensity parameters, exogenous feature influence coefficients, and triggering influence coefficients are calculated based on the historical average occurrence frequency and the mean value of the exogenous features. By utilizing historical event data and corresponding exogenous feature vectors, the optimal parameter combination is obtained based on the maximum likelihood estimation method. In real-time prediction, the latest event time and exogenous feature vector are added to the joint data in chronological order, and the condition strength is calculated within a sliding time window of a set length. The impact of historical events triggered within the window is weighted in descending order of time difference. At each prediction time, based on the calculation rules of the conditional intensity time function, the conditional intensity values ​​of each event category at the current and multiple future time steps are calculated, generating the corresponding event intensity prediction results, which are then classified and summarized according to the event category.

7. A method for managing medical spatial resources based on digital twins according to claim 1, characterized in that, The generated executable medical space resource scheduling scheme includes: The system receives the conditional intensity time function and the event intensity prediction results, maps the event type set to the prediction time period, and establishes the arrival ratio relationship between event types and functional nodes based on the functional nodes of the medical space digital twin model. Within the prediction period, an event arrival sequence is generated using a conditional intensity time function and distributed to each functional node according to the mapping relationship, forming a node arrival data stream; In the digital twin model of medical space, discrete event simulation and queuing network deduction are performed based on the node arrival data flow. The number of service stations, service rate, capacity limit and service discipline are set. The queue length, utilization rate and transfer time and distance of each functional node are recorded, and samples of waiting time, occupancy rate and transfer distance are generated. Based on the sample data, the distribution and mean of waiting time, the distribution and mean of occupancy rate, and the distribution and mean of overall transfer distance of each node are statistically analyzed to form a resource distribution result set covering multiple dimensions. Based on the resource distribution result set, a resource scheduling plan is formulated, taking bed allocation, time slot arrangement, personnel scheduling and transfer route as joint decision variables. Under the conditions of meeting capacity, personnel and channel constraints, the waiting time, utilization rate and transfer distance are optimized, and the resource scheduling plan including bed allocation, time slot arrangement, personnel and route assignment and execution time is output.

8. A medical spatial resource management system based on digital twins, comprising the medical spatial resource management method based on digital twins as described in any one of claims 1 to 7, characterized in that, Includes the following modules: The data acquisition module is used to collect historical events and real-time status data, and generate an initial dataset after de-identification. The twin module is used to build an improved digital twin model of the medical space based on the initial dataset; The modeling module is used to output exogenous feature vectors, and combine them with the initial dataset and predefined event types to establish a Hawkes process to characterize the arrival patterns and mutual influence relationships of events. The calculation module is used to perform maximum likelihood estimation of the Hawkes process based on historical event data and exogenous feature vectors, and output the conditional intensity time function and event intensity prediction results. The scheduling module is used to input the conditional intensity time function and event intensity prediction results into the medical space digital twin model to generate a resource scheduling scheme; The update module is used to implement resource scheduling schemes and collect feedback data, compare it with the distribution results, and update the parameters of the Hawkes process and the digital twin model of the medical space based on the comparison results.

Images