Intelligent outpatient service flow scheduling and resource optimization method and system

Abstract

The invention discloses an outpatient service flow intelligent scheduling and resource optimization method and system, and relates to the field of medical information intelligent scheduling and resource optimization, and the method comprises the steps: collecting images of waiting areas of all diagnosis and treatment points, extracting personnel density features, obtaining a personnel movement track through combining with a multi-target tracking algorithm, and constructing a diagnosis and treatment point people flow network topological graph; based on a topological graph, analyzing people flow convergence and dispersion characteristics, and combining historical treatment data to predict a future congestion trend of each treatment point; and according to congestion evolution and real-time reception capability, establishing an optimization model which aims at minimizing average waiting time, generating an optimal shunting path combination, and planning a personalized navigation path for a patient. The dynamic intelligent scheduling of the outpatient service process is realized, and the medical resource utilization rate and the patient treatment experience are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling and resource optimization technology for medical information, specifically to a method and system for intelligent scheduling and resource optimization of outpatient processes. Background Technology

[0002] Currently, outpatient visits commonly suffer from problems such as patient concentration, waiting congestion, and uneven distribution of medical resources. Traditional scheduling methods rely heavily on manual experience or simple queue management, making it difficult to perceive real-time patient flow and waiting conditions at treatment points. This leads to overload in some treatment processes while other resources remain idle, reducing the overall efficiency of hospital services and increasing patient waiting time and the burden on the patient's experience. With the development of video surveillance, image processing, and artificial intelligence technologies, combining real-time patient flow data with optimized medical resource allocation has become an important direction for improving outpatient operational efficiency. This invention proposes an intelligent scheduling and resource optimization method for outpatient processes. Through image acquisition, personnel trajectory analysis, and predictive model construction, it can achieve dynamic perception and optimized diversion of outpatient patient flow, thereby improving the utilization rate of medical resources and reducing average waiting time. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for intelligent scheduling and resource optimization of outpatient processes. Through intelligent patient flow monitoring and prediction models, the outpatient process can be dynamically optimized and scheduled, shortening patient waiting time and improving the efficiency of medical resource utilization.

[0004] An intelligent scheduling and resource optimization method for outpatient processes provided in this embodiment of the invention includes the following steps:

[0005] Images of waiting areas at each clinic are collected, and the population density features within the waiting areas are extracted. A multi-target tracking algorithm is then used to extract the movement trajectories of the individuals.

[0006] Based on the characteristics of population density and population movement trajectory, a topology map of the population flow network at medical treatment sites is constructed.

[0007] Analyze the convergence and dispersion points of the patient flow in the topology of the patient flow network at each treatment point, and calculate the probability of patient flow transfer between each treatment point.

[0008] Based on the probability of patient flow transfer and historical medical data of clinics, predict the congestion evolution trend of each clinic within the future time window;

[0009] Based on the congestion evolution trend and the real-time reception capacity of clinics, a path optimization model is established with the objective function of minimizing the average patient waiting time, and the optimal diversion path combination is calculated.

[0010] Specific diversion scheduling instructions are generated based on the optimal diversion path combination, and personalized navigation paths are planned for patients based on the characteristics of personnel movement trajectories.

[0011] Furthermore, images of the waiting areas at each treatment point were collected, and the population density features within the waiting areas were extracted. Combined with a multi-target tracking algorithm, the movement trajectories of the individuals were extracted, including:

[0012] Adjust the acquisition parameters based on the lighting and pedestrian flow information in the images of the waiting area;

[0013] Distortion correction and perspective transformation are performed on the images of the waiting area to obtain a top-view image, and the top-view image is then standardized to obtain a standardized image.

[0014] The grid cells are divided based on the distribution of people in the standardized image. The number of people in each grid cell is counted to obtain the grid density value. The smoothing radius is determined based on the grid density value. The grid density value is then smoothed using the smoothing radius to obtain a continuous density distribution.

[0015] The spatiotemporal rate of change of the continuous density distribution is calculated to obtain density change characteristics. The continuous density distribution and the density change characteristics are combined to construct the population density characteristics of the waiting area.

[0016] Feature vectors of human targets are extracted from standardized images. A multi-target tracking algorithm is used to calculate the similarity between consecutive frames based on the feature vectors to obtain an association matrix. Multiple human targets are simultaneously tracked using the association matrix to obtain a trajectory sequence. The human movement trajectory is constructed based on the trajectory sequence.

[0017] Furthermore, based on population density characteristics and population movement trajectory characteristics, a topology map of the patient flow network at the treatment center is constructed, including:

[0018] Based on the population density characteristics, obtain the density distribution value and density change value, and calculate the service capacity value of the network node;

[0019] The flow intensity value between network nodes is calculated based on the characteristics of personnel movement trajectories, and the ratio of the flow intensity value to the service capacity value is set as the edge weight value between network nodes.

[0020] The clinics are set as network nodes, and a network topology of clinic flow is constructed based on the service capacity value and edge weight value.

[0021] Furthermore, the convergence and dispersion points of the patient flow in the patient flow network topology diagram of the treatment points are analyzed, and the probability of patient flow transfer between each treatment point is calculated, including:

[0022] Obtain the net inflow rate of people at each network node in the topology diagram of the patient flow at the treatment site within multiple consecutive time windows;

[0023] The attributes of network nodes are identified based on the changing trend of net inflow rate over multiple consecutive time windows. Nodes that meet the first changing trend are marked as inflow convergence points, and nodes that meet the second changing trend are marked as inflow dispersion points.

[0024] The probability of pedestrian transfer between network nodes is calculated based on the marking results of the pedestrian convergence points and pedestrian dispersion points, as well as the net pedestrian inflow rate of the target node in the next time window.

[0025] Furthermore, based on the probability of patient transfer and historical visit data at medical facilities, the predicted congestion trends at each medical facility within the future time window include:

[0026] Based on the probability of patient transfer, combined with the historical medical data of each clinic, the predicted transfer medical data for the target time is calculated.

[0027] The historical medical visit data within a preset time window is analyzed in a time series to obtain periodic characteristics, and the time series prediction medical visit data at the target time is calculated using a neural network based on the periodic characteristics.

[0028] Calculate the prediction error between the transfer prediction medical visit data and the time series prediction medical visit data and the actual medical visit data, and generate dynamic weights based on the prediction error;

[0029] The predicted medical visit data at the target time is obtained by weighting and fusing the predicted medical visit data of the transfer prediction and the time series prediction.

[0030] Based on predicted medical visit data, the capacity occupancy level and the trend of changes in medical visits are calculated to determine the congestion evolution trend of medical sites.

[0031] Furthermore, based on the congestion evolution trend and the real-time reception capacity of treatment centers, a path optimization model is established with the objective function of minimizing the average patient waiting time. The optimal diversion path combinations are calculated to include:

[0032] Based on the congestion evolution trend, determine the congestion correlation between adjacent road segments and generate the road segment congestion index within the target time window;

[0033] Obtain data on the status of medical resources and patient reception efficiency at treatment centers, assess the real-time patient reception capacity of treatment centers, and obtain the real-time patient reception volume of treatment centers;

[0034] The patient's travel time is calculated based on the road congestion index, and the patient's waiting time for treatment is calculated based on the real-time number of patients received at the treatment point. The sum of the travel time and the waiting time for treatment is taken as the average waiting time for the patient.

[0035] A path optimization model is constructed, using the congestion evolution trend and the real-time patient volume at the treatment point as state parameters, and the patient allocation scheme as decision parameters. An objective function is constructed based on the average patient waiting time.

[0036] The path optimization model is iteratively calculated until the optimal triage path combination that satisfies the traffic conditions and patient volume limits is obtained.

[0037] Furthermore, specific triage scheduling instructions are generated based on the optimal triage path combination, and personalized navigation paths are planned for patients based on personnel movement trajectory characteristics, including:

[0038] The regional patient flow allocation ratio is calculated based on the optimal diversion path combination, and the regional transfer threshold is determined based on road capacity and the number of patients received at treatment points, and diversion scheduling instructions are generated.

[0039] Collect patients' historical movement trajectories, extract movement behavior features, and classify patients' mobility capabilities;

[0040] The diversion scheduling instructions are broken down according to mobility capability classification, candidate path combinations are calculated in combination with the road network structure, and differentiated weights are assigned to the path combinations.

[0041] Based on the path combination with the differentiated weights and real-time traffic conditions, intersection turning instructions are generated, and the travel time is calculated by combining signal timing to form a personalized navigation path.

[0042] Furthermore, the method also includes:

[0043] Obtain the patient's real-time location, and when the location deviates from the personalized navigation path, select a new path from the candidate path combination and replan the route;

[0044] The patient's actual travel data is updated to the historical travel trajectory, mobility capability is reclassified, differential weights are adjusted, and personalized navigation paths are updated.

[0045] This invention provides an intelligent scheduling and resource optimization system for outpatient processes, the system comprising:

[0046] The data acquisition unit is used to collect images of the waiting areas of each clinic, extract the personnel density characteristics of the waiting areas, and combine them with a multi-target tracking algorithm to extract the personnel movement trajectory.

[0047] Topology building units are used to construct a topology map of the patient flow network at the clinic based on population density characteristics and population movement trajectory characteristics.

[0048] The probability calculation unit is used to analyze the convergence and dispersion points of people in the topology of the patient flow network at each treatment point, and to calculate the probability of patient flow transfer between each treatment point.

[0049] The trend prediction unit is used to predict the congestion evolution trend of each clinic within a future time window based on the probability of patient flow transfer and historical medical data of clinics.

[0050] The path optimization unit is used to establish a path optimization model with the objective function of minimizing the average patient waiting time, based on the congestion evolution trend and the real-time reception capacity of the treatment points, and to calculate the optimal combination of diversion paths.

[0051] The scheduling execution unit is used to generate specific diversion scheduling instructions based on the optimal diversion path combination, and at the same time plan personalized navigation paths for patients based on the characteristics of personnel movement trajectories.

[0052] One technical solution provided in this embodiment of the invention is an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0053] In this embodiment, by acquiring images of the waiting area and extracting population density and movement trajectory features, the real-time dynamics of pedestrian flow within the treatment center can be accurately reflected. Combining multi-target tracking and pedestrian network topology construction, refined modeling of crowd aggregation, dispersion, and flow patterns is achieved, thereby effectively calculating the probability of pedestrian transfer between treatment centers. Simultaneously, by combining historical treatment data to predict future congestion trends, potential high-load areas can be identified in advance. By establishing a path optimization model aimed at minimizing the average patient waiting time, optimal diversion path combinations can be dynamically generated, achieving rational allocation of medical resources. Furthermore, by combining individualized movement trajectory characteristics to plan precise navigation paths for patients, not only is patient efficiency and experience improved, but the risk of localized congestion is also reduced, enhancing the overall capacity and operational efficiency of medical institutions.

[0054] The above description of the invention is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1A flowchart illustrating an intelligent scheduling and resource optimization method for outpatient processes provided in an embodiment of the present invention;

[0057] Figure 2 This is a schematic diagram of the structure of an outpatient process intelligent scheduling and resource optimization system provided in an embodiment of the present invention. Detailed Implementation

[0058] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. In the following description relating to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements.

[0059] The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0060] like Figure 1 As shown, Figure 1 A flowchart of an intelligent scheduling and resource optimization method for outpatient processes provided by an embodiment of the present invention, the method comprising the following steps:

[0061] Images of waiting areas at each clinic are collected, and the population density features within the waiting areas are extracted. A multi-target tracking algorithm is then used to extract the movement trajectories of the individuals.

[0062] Based on the characteristics of population density and population movement trajectory, a topology map of the population flow network at medical treatment sites is constructed.

[0063] Analyze the convergence and dispersion points of the patient flow in the topology of the patient flow network at each treatment point, and calculate the probability of patient flow transfer between each treatment point.

[0064] Based on the probability of patient flow transfer and historical medical data of clinics, predict the congestion evolution trend of each clinic within the future time window;

[0065] Based on the congestion evolution trend and the real-time reception capacity of clinics, a path optimization model is established with the objective function of minimizing the average patient waiting time, and the optimal diversion path combination is calculated.

[0066] Specific diversion scheduling instructions are generated based on the optimal diversion path combination, and personalized navigation paths are planned for patients based on the characteristics of personnel movement trajectories.

[0067] Furthermore, images of the waiting areas at each treatment point were collected, and the population density features within the waiting areas were extracted. Combined with a multi-target tracking algorithm, the movement trajectories of the individuals were extracted, including:

[0068] Adjust the acquisition parameters based on the lighting and pedestrian flow information in the images of the waiting area;

[0069] Distortion correction and perspective transformation are performed on the images of the waiting area to obtain a top-view image, and the top-view image is then standardized to obtain a standardized image.

[0070] The grid cells are divided based on the distribution of people in the standardized image. The number of people in each grid cell is counted to obtain the grid density value. The smoothing radius is determined based on the grid density value. The grid density value is then smoothed using the smoothing radius to obtain a continuous density distribution.

[0071] The spatiotemporal rate of change of the continuous density distribution is calculated to obtain density change characteristics. The continuous density distribution and the density change characteristics are combined to construct the population density characteristics of the waiting area.

[0072] Feature vectors of human targets are extracted from standardized images. A multi-target tracking algorithm is used to calculate the similarity between consecutive frames based on the feature vectors to obtain an association matrix. Multiple human targets are simultaneously tracked using the association matrix to obtain a trajectory sequence. The human movement trajectory is constructed based on the trajectory sequence.

[0073] When acquiring images of waiting areas at various treatment points, the acquisition parameters can be adjusted according to the real-time scene. For waiting areas with significant lighting variations, adaptive exposure technology dynamically adjusts the camera's exposure parameters. In bright environments, the exposure time is set to 1 / 200 second, the aperture to f / 5.6, and the ISO to 400; in darker environments, the exposure time is automatically adjusted to 1 / 60 second, the aperture to f / 3.5, and the ISO to 1600. Simultaneously, the frame rate parameters are adjusted based on the density of people. When the detected density of people in the waiting area exceeds 3 people per square meter, the camera's frame rate is increased to 25 frames per second to ensure the capture of detailed information about rapidly moving people.

[0074] After acquiring images of the waiting area, distortion correction was performed to eliminate lens distortion. A polynomial model was used to calculate distortion parameters, and the image was corrected using the distortion center point and distortion coefficients. For four marker points within the waiting area, the original image was converted into a top-view image using a perspective transformation matrix, obtained by solving a system of linear equations. The top-view image was then standardized, including size normalization to a resolution of 1280×720 pixels, brightness equalization to maintain an average pixel brightness value of 128, and contrast stretching to expand the dynamic range to 0-255, thus obtaining a standardized image.

[0075] The standardized image was divided into a grid, with the entire waiting area divided into 10×10 grid cells, each corresponding to an area of ​​approximately 2 square meters. Background subtraction and personnel detection algorithms were used to identify individuals within each grid cell. A person was identified when the pixel concentration within a cell exceeded a preset threshold of 0.65 and the shape matched human characteristics. The number of individuals within each grid cell was counted, resulting in a grid density matrix. A smoothing radius was determined based on the grid density values: a smaller smoothing radius (e.g., 1.5 grid cell widths) was used for areas with higher density values, while a larger smoothing radius (e.g., 3 grid cell widths) was used for areas with lower density values. An adaptive Gaussian kernel function was used to smooth the grid density values, generating a continuous density distribution.

[0076] The rate of change of the continuous density distribution in both time and space is calculated. The temporal rate of change is calculated by comparing the density distributions of adjacent time frames, taking the difference between two frames and dividing by the time interval; the spatial rate of change is obtained by calculating the gradients of the density distribution in the horizontal and vertical directions. The temporal and spatial rates of change are combined to construct a three-dimensional feature vector representing the density variation characteristics. Finally, the continuous density distribution and density variation characteristics are combined to construct the population density characteristics of the waiting area, forming a comprehensive feature description that includes spatial distribution, temporal variation, and trend prediction.

[0077] When extracting feature vectors of people in standardized imagery, a candidate region extraction algorithm is first used to detect the location of people in each frame of the image, extracting the appearance and motion features of each person. Appearance features include color histogram features, texture features, and shape features; motion features include displacement vectors and direction of motion. These features are combined to form a 128-dimensional feature vector representing each person. A multi-target tracking algorithm is used to process consecutive video frames, calculating the feature similarity between each detected person in the current frame and all person targets in the previous frame, generating an association matrix. The elements of the association matrix range from 0 to 1; a similarity exceeding 0.75 indicates that the person may be the same. The Hungarian algorithm can be used for optimal matching of the association matrix to solve the target allocation problem and achieve simultaneous tracking of multiple people.

[0078] During tracking, each person is assigned a unique ID, and their position coordinates in consecutive frames are recorded to form a trajectory sequence. Tracking is terminated when a target disappears from the scene for more than 30 frames. The obtained trajectory sequence is smoothed to eliminate noise using a sliding window averaging method with a window size of 5 frames. Based on the smoothed trajectory sequence, movement features are extracted, including movement speed, direction changes, and dwell points. Locations where dwell time exceeds 15 seconds are marked as waiting hotspots, and the main flow direction of movement trajectories is analyzed to determine the main pedestrian flow channels in the waiting area.

[0079] The aforementioned technologies enable the precise extraction of personnel density characteristics and movement trajectories in outpatient waiting areas, effectively improving the scientific nature of intelligent scheduling and resource optimization in outpatient processes. This method can monitor the congestion status of waiting areas at each treatment point in real time. Through analysis of personnel density distribution and movement trajectories, it can predict potential congestion in waiting areas, allowing medical institutions to dynamically adjust the allocation of medical resources, optimize treatment sequences, improve the patient experience, and increase medical efficiency.

[0080] Furthermore, based on population density characteristics and population movement trajectory characteristics, a topology map of the patient flow network at the treatment center is constructed, including:

[0081] Based on the population density characteristics, obtain the density distribution value and density change value, and calculate the service capacity value of the network node;

[0082] The flow intensity value between network nodes is calculated based on the characteristics of personnel movement trajectories, and the ratio of the flow intensity value to the service capacity value is set as the edge weight value between network nodes.

[0083] The clinics are set as network nodes, and a network topology of clinic flow is constructed based on the service capacity value and edge weight value.

[0084] When constructing the topology map of the patient flow network at clinics, it is necessary to process the population density characteristics to obtain density distribution values ​​and density variation values. Density distribution values ​​are extracted from continuous density distributions and obtained by calculating the average population density in the waiting area of ​​each clinic, expressed in person / square meter. For general outpatient clinics, density distribution values ​​typically fluctuate between 0.2 and 2.5 person / square meter. Density variation values ​​are obtained by calculating the rate of change of density distribution values ​​per unit time, expressed in person / (square meter·minute). During the morning peak outpatient period, the density variation value can reach 0.15 person / (square meter·minute), while during the relatively flat period at midday, it is approximately 0.03 person / (square meter·minute). Both density distribution values ​​and density variation values ​​are smoothed using an exponentially weighted moving average method, with a weighting factor set to 0.8 and a smoothing duration of 30 minutes to eliminate the impact of short-term fluctuations.

[0085] Service capacity value is a key indicator characterizing a clinic's ability to handle patients, calculated by combining density distribution and density change values. In the calculation, a weighting coefficient of 0.7 is assigned to the density distribution value, and a weighting coefficient of 0.3 is assigned to the density change value, followed by a weighted sum. When the density distribution value is high and the density change value is negative, it indicates that the clinic is densely populated but slowly dispersing, resulting in a relatively high service capacity value. Conversely, when the density distribution value is high and the density change value is positive, it indicates that the clinic is continuously gathering patients, resulting in a relatively low service capacity value. For example, under normal operating conditions, an internal medicine clinic's service capacity value is approximately 8.5, representing the number of patients it can serve per hour.

[0086] When calculating the flow intensity value between network nodes based on personnel movement trajectory characteristics, the trajectory sequence is first analyzed to identify the patient movement patterns between treatment points. For any two treatment points A and B, the number of patients moving from A to B within the observation time window is counted, recorded as the direct flow number. Indirect flow, i.e., patients starting from A, passing through other treatment points, and finally arriving at B, is also considered, and an attenuation coefficient of 0.6 is assigned to indirect flow. The direct flow number and the weighted indirect flow number are added together to obtain the total flow number. The flow intensity value is defined as the total flow number per unit time, expressed in persons per hour. Taking the flow between the radiology and internal medicine departments as an example, the peak flow intensity value can reach 12 persons per hour, while the off-peak value is approximately 4 persons per hour.

[0087] The edge weight value is calculated as the ratio of flow intensity value to service capacity value, representing the flow intensity per unit of service capacity. Specifically, the flow intensity value is divided by the service capacity value of the target treatment point. A larger edge weight value indicates higher congestion along the corresponding path, potentially leading to longer waiting times for patients. To avoid the influence of extreme values, the calculated edge weight values ​​are normalized to a range between 0 and 1. For example, the edge weight value from internal medicine to the pharmacy can reach 0.85 during peak outpatient hours, indicating high congestion; during off-peak hours, it is approximately 0.25, indicating a smooth path.

[0088] When setting up treatment points as network nodes, all treatment points within the hospital are numbered and assigned unique identifiers, such as N001 for the registration desk, N002 for the internal medicine clinic, N003 for the radiology department, and N004 for the pharmacy. Each network node contains attribute information such as location coordinates, service capacity value, and current service status. The connection relationships between nodes are determined based on the physical spatial layout and the actual mobility probability of patients; not all node pairs are directly connected. An adjacency matrix is ​​constructed based on edge weight values, where the element values ​​in the matrix correspond to the edge weight values ​​between nodes, and the element values ​​for unconnected node pairs are set to infinity. The topology represented by the adjacency matrix can intuitively reflect the flow of people between treatment points.

[0089] The patient flow network topology at the clinic is presented through a visual interface. Node size is proportional to service capacity, node color reflects current congestion level, and line thickness is proportional to edge weight. The topology also supports dynamic display over time, allowing users to review historical data or predict network status changes over a future period. The clinic's patient flow network topology is updated regularly, every 5 minutes, to reflect the latest patient flow status. The topology also marks key nodes and bottleneck paths. Key nodes are defined as those with a service capacity below a threshold of 5 and a connectivity greater than 3, while bottleneck paths are defined as connections with an edge weight greater than 0.7.

[0090] The patient flow network topology map constructed using this method can intuitively reflect the distribution and flow relationships of patients among various outpatient clinics, providing data support for intelligent scheduling and resource optimization of outpatient processes. This method can monitor changes in the service capacity of each clinic and the flow of patients between clinics in real time, dynamically identifying bottlenecks and key nodes in the outpatient process. This allows hospital administrators to rationally allocate medical resources, enabling hospitals to adjust resource allocation in advance, avoid congestion, and significantly improve patient experience and satisfaction.

[0091] Furthermore, the convergence and dispersion points of the patient flow in the patient flow network topology diagram of the treatment points are analyzed, and the probability of patient flow transfer between each treatment point is calculated, including:

[0092] Obtain the net inflow rate of people at each network node in the topology diagram of the patient flow at the treatment site within multiple consecutive time windows;

[0093] The attributes of network nodes are identified based on the changing trend of net inflow rate over multiple consecutive time windows. Nodes that meet the first changing trend are marked as inflow convergence points, and nodes that meet the second changing trend are marked as inflow dispersion points.

[0094] The probability of pedestrian transfer between network nodes is calculated based on the marking results of the pedestrian convergence points and pedestrian dispersion points, as well as the net pedestrian inflow rate of the target node in the next time window.

[0095] When analyzing the patient flow network topology of a clinic, it is necessary to obtain the net inflow rate of each network node within multiple consecutive time windows. The net inflow rate represents the ratio of the difference between the number of people entering and leaving a node per unit time to the node's capacity. To calculate the net inflow rate, first, count the number of people entering and leaving the node within the time window; the difference is the net inflow. Divide the net inflow by the time window length to obtain the net inflow rate. The time window setting is adjusted according to the actual situation of the outpatient clinic, generally set to 15 minutes. For example, in the time window of 8:00-8:15, if 25 patients enter and 10 patients leave the internal medicine clinic, resulting in a net inflow of 15, and the clinic's capacity is 30, then the net inflow rate for this time window is 0.5.

[0096] To obtain accurate trends in pedestrian flow, data from multiple consecutive time windows is needed. Typically, eight consecutive time windows, or two hours of data, are selected for analysis. For each network node, the net pedestrian inflow rate sequence within these eight time windows is recorded. To eliminate the impact of short-term fluctuations, the sequence data is smoothed using a moving average method, with a smoothing window size of 3. The smoothed net inflow rate sequence better reflects the actual trend of pedestrian flow changes. For example, the original net pedestrian inflow rate of a radiology clinic within eight consecutive time windows is [-0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.4, 0.2], and the smoothed sequence is [0.0, 0.17, 0.3, 0.4, 0.5, 0.5, 0.4, 0.3].

[0097] Network node attributes are identified based on the changing trend of net pedestrian inflow rate over multiple consecutive time windows. The first trend is defined as follows: for at least four consecutive time windows, the net pedestrian inflow rate remains positive and shows an upward trend, or gradually rises from a negative value to a positive value and continues to grow. Nodes satisfying the first trend are marked as pedestrian convergence points, indicating that patients are gradually gathering at the clinic. The second trend is defined as follows: for at least four consecutive time windows, the net pedestrian inflow rate remains negative and shows a downward trend, or gradually decreases from a positive value to a negative value and continues to decline. Nodes satisfying the second trend are marked as pedestrian dispersion points, indicating that patients at the clinic are gradually decreasing and transferring to other clinics. Nodes that do not satisfy either the first or second trend are marked as neutral nodes.

[0098] Node attribute labels are updated in real time using a sliding window approach. Whenever new time window data is generated, the oldest time window data is removed, the new data is added, and trend analysis and node attribute labeling are performed again. This method dynamically reflects changes in patient flow at each clinic. For example, during the observation period from 9:00 AM to 11:00 AM, the registration area is initially marked as a patient convergence point, but after 10:30 AM, as the morning peak ends, its attribute may change to a neutral node or a patient dispersion point.

[0099] Based on the marking results of pedestrian convergence and dispersion points and the net pedestrian inflow rate of the target node in the next time window, the pedestrian transfer probability between network nodes is calculated. The pedestrian transfer probability represents the proportion of patients transferred from one treatment point to another. When calculating the transfer probability, the attributes of the source and target nodes, the physical distance between the two nodes, historical transfer data, and the predicted net pedestrian inflow rate of the target node in the next time window are considered. Specifically, if the source node is marked as a pedestrian dispersion point and the target node is marked as a pedestrian convergence point, and there is a direct connection between them, the transfer probability between them is relatively high.

[0100] The transfer probability is calculated using a weighted combination method. Node attribute factors are assigned a weight of 0.4, meaning that when the source node is a point of population dispersion and the target node is a point of population convergence, this factor contributes 0.4; otherwise, it contributes 0. Physical distance is assigned a weight of 0.3; the physical distance between two nodes is converted into a normalized distance factor, and the closer the distance, the closer the contribution of this factor is to 0.3. Historical transfer data is assigned a weight of 0.2; historical transfer probabilities are calculated based on historical transfer data from the same period over the past week, with a maximum contribution of 0.2. The predicted net inflow rate is assigned a weight of 0.1; the higher the predicted net inflow rate of the target node, the closer the contribution of this factor is to 0.1. The final transfer probability is obtained by summing the contributions of all factors.

[0101] During implementation, the weighting of the transfer probability calculation needs to be adjusted for different types of treatment sites. For example, for treatment sites with relatively fixed service hours, such as the laboratory, the weight of historical transfer data can be increased to 0.3; for treatment sites with more temporary needs, such as the emergency room, the weight of node attribute factors can be increased to 0.5. The transfer probability calculation results are stored in matrix form, where the matrix element value represents the probability of a node in a row transferring to the node in a column. For example, the probability of an internal medicine treatment site transferring to the radiology department is 0.35, the probability of transferring to the laboratory is 0.25, and the probability of transferring to the pharmacy is 0.4.

[0102] The analysis of patient convergence points, dispersion points, and transfer probabilities using the above methods provides crucial information for intelligent scheduling of outpatient processes. This method can identify congestion hotspots and impending patient aggregation areas in real time, predict patient flow at each treatment point, and calculate the optimal patient diversion path based on transfer probabilities. Through a patient guidance system, it guides patients to choose less congested routes to the next treatment point, effectively reducing the average waiting time for patients between treatment points. This method is applicable to medical institutions of different sizes and provides strong technical support for optimizing outpatient processes.

[0103] Furthermore, based on the probability of patient transfer and historical visit data at medical facilities, the predicted congestion trends at each medical facility within the future time window include:

[0104] Based on the probability of patient transfer, combined with the historical medical data of each clinic, the predicted transfer medical data for the target time is calculated.

[0105] The historical medical visit data within a preset time window is analyzed in a time series to obtain periodic characteristics, and the time series prediction medical visit data at the target time is calculated using a neural network based on the periodic characteristics.

[0106] Calculate the prediction error between the transfer prediction medical visit data and the time series prediction medical visit data and the actual medical visit data, and generate dynamic weights based on the prediction error;

[0107] The predicted medical visit data at the target time is obtained by weighting and fusing the predicted medical visit data of the transfer prediction and the time series prediction.

[0108] Based on predicted medical visit data, the capacity occupancy level and the trend of changes in medical visits are calculated to determine the congestion evolution trend of medical sites.

[0109] First, the predicted patient transfer data for the target time is calculated using patient transfer probability. This calculation process considers the actual number of patients at each treatment center at the current time and the patient transfer probability matrix. For each treatment center, the current number of patients at that center is counted, and the number of these patients distributed to various treatment centers at the target time is predicted based on the patient transfer probability matrix. For example, if there are 30 patients at the internal medicine treatment center at the current time, according to the transfer probability matrix, it is predicted that 10 patients will be transferred to the radiology department at the next time, 8 patients to the laboratory, 5 patients to the pharmacy, and the remaining 7 patients will remain in internal medicine. Simultaneously, the number of patients transferred from other treatment centers to the target treatment center is considered. For the radiology department, there may be 10 patients transferred from internal medicine, 5 patients transferred from surgery, 3 patients transferred from pediatrics, etc. Adding up the number of all transferred patients, and adding the number of patients currently in the radiology department who are expected to remain there, the predicted patient transfer data for the radiology department at the target time is 25.

[0110] The second prediction method is based on time-series analysis of historical outpatient data within a preset time window to extract periodic features. Historical outpatient data typically exhibits clear periodicity, including intraday, weekly, and seasonal fluctuations. Intraday fluctuations are characterized by peak outpatient visits in the morning and afternoon; weekly fluctuations are characterized by higher visit volumes on Mondays and Fridays; and seasonal fluctuations are characterized by increased internal medicine visits during flu season. When extracting periodic features, a combination of time-domain and frequency-domain analysis is used. Time-domain analysis identifies the periodicity of the data by calculating the autocorrelation function; frequency-domain analysis converts the time-series data to the frequency domain using Fast Fourier Transform (FFT) to identify the main periodic components. For general hospitals, the intraday period is typically 4 hours, the weekly period is 5 days, and the seasonal period is 12 weeks.

[0111] Based on the extracted periodic features, a neural network model is constructed to calculate the time-series predicted medical visit data at the target time. This neural network employs a long short-term memory (LSTM) network structure, which can effectively capture the long-term dependencies in the time-series data. The network input layer has 48 nodes, corresponding to the medical visit data of the previous 48 time windows; the hidden layer uses a two-layer structure, with each layer containing 64 neurons; the output layer has nodes corresponding to the number of prediction time windows, typically set to 6, corresponding to predictions for the next 1.5 hours. During model training, historical medical visit data from the past 3 months is used, with mean squared error as the loss function. The Adam optimizer is used for parameter optimization, with a learning rate of 0.001 and 1000 training iterations. After training, given the medical visit data of the previous 48 time windows at the current time, the model outputs the predicted medical visit data for the next 6 time windows. For example, the predicted time-series medical visit data for the radiology department at the target time is 28 people.

[0112] The prediction errors of transfer prediction and time-series prediction data relative to actual patient visits are calculated to generate dynamic weights. The prediction error is calculated by comparing the differences between historical predictions and actual values. Specifically, the mean absolute error (MAO) of the transfer prediction method and the time-series prediction method is calculated using the prediction data and actual data from the most recent 10 time windows. The MAO of the transfer prediction method is 3.2 patients, and the MAO of the time-series prediction method is 2.1 patients. The dynamic weights are generated based on the reciprocal of the prediction error; that is, the smaller the error, the greater the weight of the corresponding method. Normalization is used to ensure that the sum of the weights of the two methods is 1. For radiology, the weight of the transfer prediction method is 0.4, and the weight of the time-series prediction method is 0.6.

[0113] Dynamic weights are used to weight and fuse the transfer-predicted patient visit data and the time-series predicted patient visit data to obtain the predicted patient visit data at the target time. The weighted fusion process involves multiplying the transfer-predicted patient visit data by its corresponding weight and the time-series predicted patient visit data by its corresponding weight, and then adding the two together. Taking the radiology department as an example, the transfer-predicted patient visit data consists of 25 patients with a weight of 0.4; the time-series predicted patient visit data consists of 28 patients with a weight of 0.6; the weighted fused predicted patient visit data is 25 × 0.4 + 28 × 0.6 = 26.8 patients, rounded to 27. This weighted fusion method can combine the advantages of both prediction methods to improve the accuracy and stability of the prediction.

[0114] Based on predicted patient visit data, the capacity occupancy level and patient visit change trend are calculated to determine the congestion evolution trend of medical facilities. Capacity occupancy level is defined as the ratio of predicted patient visit data to the maximum capacity of the medical facility. Maximum capacity is determined based on the physical space of the medical facility, the number of medical equipment, and the number of medical staff. Taking the radiology department as an example, its maximum capacity is 35 people, and the predicted patient visit data is 27 people, resulting in a capacity occupancy level of 27 / 35 = 0.77. The patient visit change trend is calculated by comparing predicted patient visit data over multiple consecutive time windows. If the predicted patient visit data for the next three time windows shows an upward trend (e.g., 27, 30, 32 people), the patient visit change trend is considered upward; if it shows a downward trend (e.g., 27, 25, 22 people), the trend is considered downward; if the fluctuation is small (e.g., 27, 28, 27 people), the trend is considered stable.

[0115] Based on capacity occupancy and patient visit trends, the congestion evolution trend of treatment centers is categorized into five types: severe congestion worsening, mild congestion worsening, stable congestion, congestion easing, and unimpeded flow. When the capacity occupancy level is greater than 0.85 and the patient visit trend is upward, it is considered severe congestion worsening; when the capacity occupancy level is between 0.7 and 0.85 and the patient visit trend is upward, it is considered mild congestion worsening; when the capacity occupancy level is greater than 0.7 and the patient visit trend is stable, it is considered stable congestion; when the capacity occupancy level is greater than 0.7 and the patient visit trend is downward, it is considered congestion easing; and when the capacity occupancy level is less than 0.7, it is considered unimpeded flow. For the radiology department, the capacity occupancy level is 0.77, and the patient visit trend is upward, so its congestion evolution trend is classified as mild congestion worsening.

[0116] By integrating transfer prediction and time-series prediction, the system achieves accurate forecasting of future congestion at various outpatient clinics, providing strong support for intelligent scheduling of outpatient processes. Based on the prediction results, hospitals can adjust the allocation of medical resources in advance, such as increasing medical staff or temporarily opening backup consultation rooms at clinics predicted to experience congestion. Simultaneously, the system generates patient diversion suggestions based on the prediction results, guiding patients to avoid congested clinics and choose smoother routes through a triage system. The model continuously optimizes its parameters by learning from historical data, adapting to changes in outpatient visit patterns and maintaining the stability and reliability of its predictions.

[0117] Furthermore, based on the congestion evolution trend and the real-time reception capacity of treatment centers, a path optimization model is established with the objective function of minimizing the average patient waiting time. The optimal diversion path combinations are calculated to include:

[0118] Based on the congestion evolution trend, determine the congestion correlation between adjacent road segments and generate the road segment congestion index within the target time window;

[0119] Obtain data on the status of medical resources and patient reception efficiency at treatment centers, assess the real-time patient reception capacity of treatment centers, and obtain the real-time patient reception volume of treatment centers;

[0120] The patient's travel time is calculated based on the road congestion index, and the patient's waiting time for treatment is calculated based on the real-time number of patients received at the treatment point. The sum of the travel time and the waiting time for treatment is taken as the average waiting time for the patient.

[0121] A path optimization model is constructed, using the congestion evolution trend and the real-time patient volume at the treatment point as state parameters, and the patient allocation scheme as decision parameters. An objective function is constructed based on the average patient waiting time.

[0122] The path optimization model is iteratively calculated until the optimal triage path combination that satisfies the traffic conditions and patient volume limits is obtained.

[0123] When determining the congestion correlation between adjacent road segments based on congestion evolution trends, it is necessary to analyze historical congestion data and predicted congestion status of the connecting road segments between each treatment point. Congestion correlation indicates the degree to which the congestion status of one road segment affects adjacent road segments. In the calculation process, the congestion status of each road segment is first quantified, converting the congestion evolution trend into a numerical index: severe congestion worsening is assigned a value of 5, mild congestion worsening is assigned a value of 4, stable congestion is assigned a value of 3, congestion mitigation is assigned a value of 2, and smooth traffic is assigned a value of 1. For example, the current congestion evolution trend of the road segment from the radiology department to the internal medicine department is mild congestion worsening, and its congestion status value is 4. Secondly, the correlation between the congestion status of adjacent road segments is analyzed, using the Pearson correlation coefficient to calculate the correlation degree between the congestion status of adjacent road segments. The higher the correlation degree, the more easily a change in congestion in one road segment affects adjacent road segments. For example, the correlation degree between the road segment from the internal medicine department to the laboratory and the road segment from the radiology department to the internal medicine department is 0.75, indicating a strong correlation between their congestion statuses.

[0124] When generating the road segment congestion index within the target time window, the current congestion status, congestion evolution trend, and congestion correlation are comprehensively considered. The road segment congestion index is represented by a range of 0 to 10, with higher values ​​indicating more severe congestion. The calculation formula considers the base congestion value, trend factor, and correlation effects. The base congestion value is determined by the current congestion status; for example, mild congestion corresponds to a base congestion value of 5. The trend factor is determined by the congestion evolution trend; a congestion worsening trend corresponds to a positive trend factor, and a congestion easing trend corresponds to a negative trend factor. Correlation effects are jointly determined by the congestion status and correlation degree of adjacent road segments. Taking the road segment from the internal medicine department to the radiology department as an example, the current congestion status is mild, with a base congestion value of 5; the congestion evolution trend is stable congestion, with a trend factor of 0; the contribution value from the influence of adjacent road segments is 1.2; the final calculated road segment congestion index is 6.2.

[0125] The real-time patient reception capacity of treatment centers is assessed by acquiring data on the status of medical resources and patient reception efficiency. Medical resource status data includes the number of medical staff, the number of consultation rooms, and the availability of medical equipment; patient reception efficiency data includes the average duration of a single consultation and the workload of medical staff. By combining this data, the real-time patient reception volume of the treatment center is calculated, i.e., the number of patients that can be processed per unit of time. In the calculation process, the basic patient reception capacity is multiplied by a resource adjustment factor and an efficiency adjustment factor. The basic patient reception capacity is determined by the number of medical staff and the number of consultation rooms; the resource adjustment factor reflects the impact of the availability of medical equipment on patient reception capacity; the efficiency adjustment factor reflects the impact of the workload of medical staff on patient reception efficiency. For example, if the basic patient reception capacity of an internal medicine department is 15 patients per hour, and one of the two diagnostic devices is currently under maintenance, the resource adjustment factor is 0.7; and medical staff have been working for 4 hours, the efficiency adjustment factor is 0.85; the calculated real-time patient reception volume is 8.9 patients per hour.

[0126] Patient travel time is calculated based on the road congestion index, and patient waiting time is calculated based on the real-time patient volume at each clinic. Travel time is the time required for a patient to move from one clinic to another, influenced by road length and congestion index. The calculation first determines the baseline travel time under congestion-free conditions, then multiplies it by a congestion factor to obtain the actual travel time. The congestion factor is a function of the road congestion index; the higher the congestion index, the larger the congestion factor. For example, if the road length from internal medicine to radiology is 50 meters, the baseline travel time is 1 minute, the current congestion index is 6.2, and the corresponding congestion factor is 1.8, the calculated actual travel time is 1.8 minutes. Patient waiting time is the time a patient waits at a clinic, influenced by the number of people in queue and the real-time patient volume at the clinic. The calculation divides the number of people in queue by the real-time patient volume at the clinic. For example, if the radiology department currently has 18 people in the queue and the real-time patient volume is 12 patients per hour, or one patient every 5 minutes, the calculated waiting time is 90 minutes. The average patient waiting time is the sum of the travel time and the treatment waiting time.

[0127] When constructing the route optimization model, the congestion evolution trend and the real-time patient volume at treatment points are used as state parameters, and the patient allocation scheme is used as the decision parameter. The patient allocation scheme represents the proportion of patients assigned to each available path. The objective function is to minimize the average patient waiting time, which is the sum of all patient waiting times divided by the total number of patients. Model constraints include: the sum of the patient proportions assigned to each path is 1; the actual patient volume at each treatment point does not exceed its maximum capacity; and the actual traffic volume on each road segment does not exceed its maximum capacity. To solve this optimization problem, a genetic algorithm is used for iterative computation. The population size is set to 100, the crossover probability is 0.8, the mutation probability is 0.1, and the maximum number of iterations is 1000. In each iteration, a new allocation scheme is generated, the corresponding average patient waiting time is calculated, and the scheme with the shortest waiting time is selected and retained for the next generation.

[0128] The path optimization model is iteratively calculated until the optimal triage path combination that satisfies the traffic conditions and patient capacity limits is obtained. During the iteration process, the patient allocation scheme is continuously adjusted, and its impact on the average patient waiting time is evaluated. When the rate of change of the average patient waiting time is less than 0.001 over 50 consecutive iterations, the algorithm is considered to have converged, and the current optimal triage path combination is output. The optimal triage path combination includes the patient allocation ratio between each treatment point. For example, among patients going from the registration desk to the internal medicine department, 70% are advised to go directly to the internal medicine department, and 30% are advised to go to the triage area first and then to the internal medicine department; among patients going from the internal medicine department to the radiology department, 60% are advised to choose the first-floor corridor route, and 40% are advised to choose the second-floor connecting corridor route. The optimal triage path combination also includes the recommended reception order and priority of each treatment point to balance the load of each treatment point.

[0129] To ensure the real-time effectiveness of triage routes, the optimal triage route combination is recalculated every 10 minutes, and the updated triage suggestions are pushed to the triage system and patient mobile applications. In case of emergencies, such as the sudden closure of a treatment center or emergency relocation of medical staff, a recalculation is immediately triggered to ensure that the triage route is always the optimal solution under the current conditions. Simultaneously, the system records triage effectiveness data, including the difference between actual and predicted waiting times, as a basis for subsequent optimization.

[0130] This route optimization method achieves efficient patient triage through precise calculation and dynamic adjustment, significantly reducing patient waiting times. Compared to traditional fixed triage schemes, this method can flexibly adjust based on real-time congestion and patient capacity, resulting in a more balanced load across treatment centers and reducing situations where some centers are overcrowded while others are idle. Hospital administrators can monitor the operational status of each treatment center in real time through the system's monitoring panel and adjust medical resource allocation accordingly. Patients, on the other hand, receive personalized treatment route recommendations through a mobile application, improving their overall healthcare experience.

[0131] Furthermore, specific triage scheduling instructions are generated based on the optimal triage path combination, and personalized navigation paths are planned for patients based on personnel movement trajectory characteristics, including:

[0132] The regional patient flow allocation ratio is calculated based on the optimal diversion path combination, and the regional transfer threshold is determined based on road capacity and the number of patients received at treatment points, and diversion scheduling instructions are generated.

[0133] Collect patients' historical movement trajectories, extract movement behavior features, and classify patients' mobility capabilities;

[0134] The diversion scheduling instructions are broken down according to mobility capability classification, candidate path combinations are calculated in combination with the road network structure, and differentiated weights are assigned to the path combinations.

[0135] Based on the path combination with the differentiated weights and real-time traffic conditions, intersection turning instructions are generated, and the travel time is calculated by combining signal timing to form a personalized navigation path.

[0136] When calculating the patient flow allocation ratio based on the optimal diversion path combination, the hospital's spatial layout and the traffic capacity of different areas need to be considered. In practice, the hospital's outpatient area is divided into several functional blocks, such as registration, waiting, treatment, examination, and dispensing areas. For each block, the proportion of patients passing through that block in the optimal diversion path combination is calculated. The patient proportions on each path are summed to obtain the total flow allocation ratio for each area. Simultaneously, based on the hospital's physical spatial layout, the road traffic capacity of each area is calculated, i.e., the maximum number of patients that can pass through per unit time. Traffic capacity is affected by factors such as passage width, obstacle distribution, and safety distances, and is determined through on-site measurements and flow simulation.

[0137] A regional transfer threshold is determined based on road capacity and clinic capacity. This threshold specifies when patient flow within a region exceeds a certain threshold, necessitating the activation of a diversion mechanism to guide some patients to alternative routes. Determining the transfer threshold requires balancing traffic efficiency and clinic capacity. Typically, the transfer threshold is set as a certain percentage of traffic capacity to allow sufficient buffer space to handle traffic fluctuations. Simultaneously, the capacity limitations of each clinic are considered to ensure that the patient inflow to each clinic after diversion does not exceed its maximum capacity. When generating diversion scheduling instructions, the predicted traffic flow is compared with the transfer threshold. When the predicted traffic flow exceeds the transfer threshold, the proportion of patients requiring diversion and the recommended alternative routes are calculated. The diversion scheduling instructions include information such as region identifier, time period, diversion ratio, and alternative routes.

[0138] The system collects patients' historical movement trajectories and extracts movement behavior features. Patient movement trajectories are collected through the hospital's internal positioning system, recording the sequence of patient location changes within the hospital. The collected trajectory data is cleaned and smoothed to remove outliers and jitter interference. Movement behavior features are extracted from the processed trajectory data, including average movement speed, rate of change of speed, frequency of stops, turning flexibility, and path selection preferences. Average movement speed represents the general speed at which the patient walks within the hospital; the rate of change of speed reflects the stability of the patient's speed; frequency of stops indicates how often the patient stops to rest or wait during movement; turning flexibility reflects the patient's ability and frequency to change direction; and path selection preferences indicate the patient's preference for different types of paths, such as whether they prefer elevators to stairs.

[0139] Patients are classified into mobility categories based on extracted mobility behavior features using clustering algorithms. Generally, patients are categorized into three groups: high mobility, medium mobility, and low mobility. Patients with high mobility have a faster average movement speed, a smaller rate of speed change, a lower frequency of stopping, higher directional flexibility, and a higher acceptance of stairs and longer routes. Patients with medium mobility have a moderate average movement speed, a moderate rate of speed change, a moderate frequency of stopping, and a moderate directional flexibility, and a high acceptance of elevators and medium-distance routes. Patients with low mobility have a slow average movement speed, a large rate of speed change, a high frequency of stopping, lower directional flexibility, and a high need for wheelchair access and shorter routes. The mobility classification results are stored in the patient's electronic record and dynamically updated based on subsequent medical visits.

[0140] The triage instructions are broken down according to mobility categories, and differentiated triage strategies are formulated for patients with different mobility categories. For patients with high mobility, longer but less congested alternative routes can be assigned; for patients with medium mobility, routes balancing distance and congestion are assigned; for patients with low mobility, shorter routes with good accessibility are prioritized, even if these routes are slightly more congested. Based on the hospital's road network structure, candidate route combinations are calculated for each mobility category. The road network structure includes the connectivity relationships between nodes (intersections, treatment points) and edges (corridors, passageways). Using a multi-source shortest path algorithm, considering factors such as distance, congestion, and facility conditions, multiple candidate routes are generated.

[0141] Differential weights are assigned to route combinations, with the weight value of each candidate route determined based on the patient's mobility and route characteristics. Weight calculation considers factors such as route length, congestion level, slope variation, number of turns, and accessibility. For patients with high mobility, the weight ratio for route length and congestion level differs; for patients with moderate mobility, a more balanced approach is taken; and for patients with low mobility, accessibility is emphasized. An additional weight is added for accessibility for patients with low mobility to ensure that recommended routes have sufficient accessibility facilities. Based on the calculated weight values, the route most suitable for the patient is selected as the recommended route.

[0142] The system generates turn instructions based on path combinations with differentiated weights and real-time traffic conditions. At each intersection or decision point, the optimal turning direction is determined according to the weights of each candidate path from the current location to the destination, combined with real-time traffic information. Real-time traffic information is obtained through the hospital's internal monitoring system and updated regularly. Turn instructions are presented in a concise and clear format for easy understanding and execution by patients. Travel time is calculated by incorporating timing information from the hospital's internal signal system. For example, if a patient needs to pass through a passage with an automatic door, the estimated time for the patient to reach the automatic door is calculated, and the travel delay is estimated accordingly. The final personalized navigation path includes a complete path description, the estimated travel time for each segment, the total distance and total time, and key landmarks along the way.

[0143] To ensure the dynamic adaptability of the navigation route, the route recommendations are continuously adjusted based on real-time traffic conditions and patient feedback. When a sudden congestion is detected in a certain area or a patient deviates significantly from the recommended route, the system will recalculate and provide updated navigation suggestions. Simultaneously, for patients with special needs, such as those with mobility impairments or visual impairments, more detailed and targeted navigation guidance is provided, including the location of accessible pathways, and instructions on using ramps and elevators.

[0144] It also provides a collaborative optimization mechanism for group diversion and individual navigation. When multiple patients simultaneously travel from similar origins to similar destinations, the mechanism comprehensively considers both the overall diversion effect and individual needs, avoiding guiding too many patients to the same alternative route and creating new congestion points. This collaborative optimization is achieved through iterative calculations, maximizing the satisfaction of individual patients' navigation needs while ensuring the overall diversion effect.

[0145] This method significantly optimizes outpatient flow efficiency and the patient experience through refined triage scheduling and personalized navigation. Differentiated route planning for patients with varying mobility levels ensures that all types of patients can find the most suitable route, reducing time spent getting lost and wandering around the hospital. It alleviates congestion within the hospital, significantly shortening the average patient travel time. It also prevents some areas from becoming overcrowded while others remain unused. The accuracy of personalized navigation routes reaches a high level, with a significantly reduced deviation rate when patients follow the recommended routes.

[0146] Furthermore, the method also includes:

[0147] Obtain the patient's real-time location, and when the location deviates from the personalized navigation path, select a new path from the candidate path combination and replan the route;

[0148] The patient's actual travel data is updated to the historical travel trajectory, mobility capability is reclassified, differential weights are adjusted, and personalized navigation paths are updated.

[0149] In this embodiment, the positioning system employs a combination of Bluetooth beacons and WiFi signals, deploying beacon devices in key areas of the hospital to form a positioning network covering the entire outpatient area. Patients communicate with the positioning network in real time by carrying a positioning card issued by the hospital or a smart device with the hospital's mobile application installed. The system periodically collects patient location information and transmits it to the workflow scheduling module through the hospital's information system's secure channel.

[0150] When determining whether a patient's location has deviated from the personalized navigation path, the offset distance between the actual location and the planned path needs to be calculated. If the offset distance exceeds a preset threshold, the patient is considered to have deviated from the planned path. The preset threshold is set according to the spatial characteristics of different areas of the hospital; a larger threshold can be set in open areas, while the threshold can be appropriately lowered in narrow corridors. Simultaneously, the movement directions of multiple consecutive patient locations are analyzed. If the angle between the movement directions of multiple consecutive locations and the planned path direction exceeds a certain angle, even if the offset distance does not exceed the threshold, it is still considered a path deviation.

[0151] When replanning a new path from candidate path combinations, the patient's current location is first determined as the new starting point, while the original destination remains unchanged. Based on the patient's mobility classification and current location, feasible paths are selected from pre-calculated candidate path combinations. Differential weights are applied to the selected candidate paths, and the path with the highest weight is selected as the new navigation path. Considering that the patient may choose to deviate from the original path for specific reasons, the weight of candidate paths consistent with the patient's current direction of movement is appropriately increased to increase the likelihood of the new path being accepted.

[0152] The replanned route will be immediately pushed to the patient, and the update will be displayed on the navigation interface. To avoid frequent updates causing inconvenience to patients, a route update cooldown period is set. During the cooldown period, even if a slight deviation is detected, route replanning will not be triggered. For serious deviations, the cooldown period will be ignored, and route replanning will begin immediately.

[0153] Updating the patient's actual travel data to the historical movement trajectory is a crucial step in ensuring continuous system optimization. Actual travel data includes a complete location sequence, timestamps, stops, and the actual walking path. It records the patient's entire movement from origin to destination, including the selected path, walking speed, stop locations and durations, and whether elevators or stairs were used. To ensure data quality, the collected trajectory data undergoes preprocessing, including noise reduction, smoothing, and outlier detection. The processed trajectory data, along with the patient's basic information, is stored in the historical movement trajectory database.

[0154] When reclassifying mobility, a comprehensive analysis of the patient's recent travel history is conducted. Feature indicators are extracted from the processed trajectory data, including average movement speed, speed standard deviation, dwell frequency, and path selection preferences. Based on these features, a clustering algorithm is used to classify mobility. To improve classification accuracy, a time decay factor is introduced, giving higher weight to recent movement data. The classification results are updated in the patient's electronic record as a basis for subsequent navigation route planning.

[0155] Adjusting differentiated weights and updating personalized navigation paths involves dynamically adjusting the weight parameters for path selection based on mobility capability classification results. According to the latest mobility capability classification, the weighting of factors such as path length, congestion level, and facility conditions is adjusted accordingly. Simultaneously, personalized preference factors are adjusted based on the patient's actual path selection behavior. Based on the adjusted differentiated weights, the comprehensive score of candidate paths is recalculated, and the path with the highest score is selected as the updated personalized navigation path.

[0156] This dynamic adjustment and replanning method significantly improves the adaptability and user experience of the outpatient intelligent navigation system. Through real-time location monitoring and path deviation detection, it can promptly identify abnormal patient walking patterns and provide path adjustments, effectively preventing patients from getting lost or arriving at the wrong areas within the hospital. Dynamically updated personalized navigation significantly reduces the average walking time for patients within the hospital, decreasing anxiety caused by searching for examination rooms. Simultaneously, the continuously learning mobility classification mechanism accurately captures changes in patient mobility, optimizing hospital spatial layout and signage systems, and improving overall operational efficiency. This method provides strong support for intelligent scheduling and resource optimization of outpatient processes.

[0157] like Figure 2 As shown, Figure 2 This is a schematic diagram of the structure of an intelligent scheduling and resource optimization system for outpatient processes provided in an embodiment of the present invention. The system includes:

[0158] The data acquisition unit is used to collect images of the waiting areas of each clinic, extract the personnel density characteristics of the waiting areas, and combine them with a multi-target tracking algorithm to extract the personnel movement trajectory.

[0159] Topology building units are used to construct a topology map of the patient flow network at the clinic based on population density characteristics and population movement trajectory characteristics.

[0160] The probability calculation unit is used to analyze the convergence and dispersion points of people in the topology of the patient flow network at each treatment point, and to calculate the probability of patient flow transfer between each treatment point.

[0161] The trend prediction unit is used to predict the congestion evolution trend of each clinic within a future time window based on the probability of patient flow transfer and historical medical data of clinics.

[0162] The path optimization unit is used to establish a path optimization model with the objective function of minimizing the average patient waiting time, based on the congestion evolution trend and the real-time reception capacity of the treatment points, and to calculate the optimal combination of diversion paths.

[0163] The scheduling execution unit is used to generate specific diversion scheduling instructions based on the optimal diversion path combination, and at the same time plan personalized navigation paths for patients based on the characteristics of personnel movement trajectories.

[0164] One technical solution provided in this embodiment of the invention is an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0165] The specific embodiments described above are preferred embodiments of the present invention and are not intended to limit the specific scope of the present invention. The scope of the present invention includes, but is not limited to, these specific embodiments. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. A method for intelligent scheduling and resource optimization of outpatient processes, characterized in that, Includes the following steps: Images of waiting areas at each clinic are collected, and the population density features within the waiting areas are extracted. A multi-target tracking algorithm is then used to extract the movement trajectories of the individuals. Based on the characteristics of population density and population movement trajectory, a topology map of the population flow network at medical treatment sites is constructed. Analyze the convergence and dispersion points of the patient flow in the topology of the patient flow network at each treatment point, and calculate the probability of patient flow transfer between each treatment point. Based on the probability of patient flow transfer and historical medical data of clinics, predict the congestion evolution trend of each clinic within the future time window; Based on the congestion evolution trend and the real-time reception capacity of clinics, a path optimization model is established with the objective function of minimizing the average patient waiting time, and the optimal diversion path combination is calculated. Specific diversion scheduling instructions are generated based on the optimal diversion path combination, and personalized navigation paths are planned for patients based on the characteristics of personnel movement trajectories.

2. The method according to claim 1, characterized in that, Images of waiting areas at each treatment point were collected, and the population density features within the waiting areas were extracted. Combined with a multi-target tracking algorithm, the movement trajectories of individuals were extracted, including: Adjust the acquisition parameters based on the lighting and pedestrian flow information in the images of the waiting area; Distortion correction and perspective transformation are performed on the images of the waiting area to obtain a top-view image, and the top-view image is then standardized to obtain a standardized image. The grid cells are divided based on the distribution of people in the standardized image. The number of people in each grid cell is counted to obtain the grid density value. The smoothing radius is determined based on the grid density value. The grid density value is then smoothed using the smoothing radius to obtain a continuous density distribution. The spatiotemporal rate of change of the continuous density distribution is calculated to obtain density change characteristics. The continuous density distribution and the density change characteristics are combined to construct the population density characteristics of the waiting area. Feature vectors of human targets are extracted from standardized images. A multi-target tracking algorithm is used to calculate the similarity between consecutive frames based on the feature vectors to obtain an association matrix. Multiple human targets are simultaneously tracked using the association matrix to obtain a trajectory sequence. The human movement trajectory is constructed based on the trajectory sequence.

3. The method according to claim 1, characterized in that, Based on population density and movement trajectory characteristics, a topology map of the patient flow network at medical facilities is constructed, including: Based on the population density characteristics, obtain the density distribution value and density change value, and calculate the service capacity value of the network node; The flow intensity value between network nodes is calculated based on the characteristics of personnel movement trajectories, and the ratio of the flow intensity value to the service capacity value is set as the edge weight value between network nodes. The clinics are set as network nodes, and a network topology of clinic flow is constructed based on the service capacity value and edge weight value.

4. The method according to claim 1, characterized in that, Analyze the convergence and dispersal points of the patient flow in the patient flow network topology diagram of the treatment points, and calculate the probability of patient flow transfer between each treatment point, including: Obtain the net inflow rate of people at each network node in the topology diagram of the patient flow at the treatment site within multiple consecutive time windows; The attributes of network nodes are identified based on the changing trend of net inflow rate over multiple consecutive time windows. Nodes that meet the first changing trend are marked as inflow convergence points, and nodes that meet the second changing trend are marked as inflow dispersion points. The probability of pedestrian transfer between network nodes is calculated based on the marking results of the pedestrian convergence points and pedestrian dispersion points, as well as the net pedestrian inflow rate of the target node in the next time window.

5. The method according to claim 1, characterized in that, Based on the probability of patient flow shifts and historical visit data at medical facilities, the predicted congestion trends at each medical facility within the future time window include: Based on the probability of patient transfer, combined with the historical medical data of each clinic, the predicted transfer medical data for the target time is calculated. The historical medical visit data within a preset time window is analyzed in a time series to obtain periodic characteristics, and the time series prediction medical visit data at the target time is calculated using a neural network based on the periodic characteristics. Calculate the prediction error between the transfer prediction medical visit data and the time series prediction medical visit data and the actual medical visit data, and generate dynamic weights based on the prediction error; The predicted medical visit data at the target time is obtained by weighting and fusing the predicted medical visit data of the transfer prediction and the time series prediction. Based on predicted medical visit data, the capacity occupancy level and the trend of changes in medical visits are calculated to determine the congestion evolution trend of medical sites.

6. The method according to claim 1, characterized in that, Based on the congestion evolution trend and the real-time reception capacity of treatment centers, a path optimization model is established with the objective function of minimizing the average patient waiting time. The optimal diversion path combinations are calculated as follows: Based on the congestion evolution trend, determine the congestion correlation between adjacent road segments and generate the road segment congestion index within the target time window; Obtain data on the status of medical resources and patient reception efficiency at treatment centers, assess the real-time patient reception capacity of treatment centers, and obtain the real-time patient reception volume of treatment centers; The patient's travel time is calculated based on the road congestion index, and the patient's waiting time for treatment is calculated based on the real-time number of patients received at the treatment point. The sum of the travel time and the waiting time for treatment is taken as the average waiting time for the patient. A path optimization model is constructed, using the congestion evolution trend and the real-time patient volume at the treatment point as state parameters, and the patient allocation scheme as decision parameters. An objective function is constructed based on the average patient waiting time. The path optimization model is iteratively calculated until the optimal triage path combination that satisfies the traffic conditions and patient volume limits is obtained.

7. The method according to claim 1, characterized in that, Specific triage and scheduling instructions are generated based on the optimal triage path combination, and personalized navigation paths are planned for patients based on personnel movement trajectory characteristics, including: The regional patient flow allocation ratio is calculated based on the optimal diversion path combination, and the regional transfer threshold is determined based on road capacity and the number of patients received at treatment points, and diversion scheduling instructions are generated. Collect patients' historical movement trajectories, extract movement behavior features, and classify patients' mobility capabilities; The diversion scheduling instructions are broken down according to mobility capability classification, candidate path combinations are calculated in combination with the road network structure, and differentiated weights are assigned to the path combinations. Based on the path combination with the differentiated weights and real-time traffic conditions, intersection turning instructions are generated, and the travel time is calculated by combining signal timing to form a personalized navigation path.

8. The method according to claim 7, characterized in that, The method further includes: Obtain the patient's real-time location, and when the location deviates from the personalized navigation path, select a new path from the candidate path combination and replan the route; The patient's actual travel data is updated to the historical travel trajectory, mobility capability is reclassified, differential weights are adjusted, and personalized navigation paths are updated.

9. An intelligent scheduling and resource optimization system for outpatient processes, used to implement the method described in any one of claims 1-7, characterized in that, The system includes: The data acquisition unit is used to collect images of the waiting areas of each clinic, extract the personnel density characteristics of the waiting areas, and combine them with a multi-target tracking algorithm to extract the personnel movement trajectory. Topology building units are used to construct a topology map of the patient flow network at the clinic based on population density characteristics and population movement trajectory characteristics. The probability calculation unit is used to analyze the convergence and dispersion points of people in the topology of the patient flow network at each treatment point, and to calculate the probability of patient flow transfer between each treatment point. The trend prediction unit is used to predict the congestion evolution trend of each clinic within a future time window based on the probability of patient flow transfer and historical medical data of clinics. The path optimization unit is used to establish a path optimization model with the objective function of minimizing the average patient waiting time, based on the congestion evolution trend and the real-time reception capacity of the treatment points, and to calculate the optimal combination of diversion paths. The scheduling execution unit is used to generate specific diversion scheduling instructions based on the optimal diversion path combination, and at the same time plan personalized navigation paths for patients based on the characteristics of personnel movement trajectories.

10. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method as described in any one of claims 1 to 7.

Images