Regional medical data analysis processing method and system based on big data technology

Abstract

The application provides a regional medical data analysis processing method and system based on big data technology, which comprises the following steps: collecting operation data of each medical institution in the region to construct a service pressure prediction model and calculating a real-time load index. The load index is mapped to a geographic space to form a dynamic distribution map. When a user requests a specific medical institution service, it is automatically judged whether the institution exceeds the congestion critical value. If the target institution is overloaded, the user demand is converted into a standardized demand description, and the most suitable alternative institution is selected through an intelligent matching algorithm. Finally, based on the load comparison between the target institution and the alternative institution, personalized shunt guidance suggestions are generated and pushed to the user. The application can effectively improve the utilization efficiency of regional medical resources.

Description

Technical Field

[0001] This invention belongs to the field of big data processing technology, specifically relating to a method and system for regional medical data analysis and processing based on big data technology. Background Technology

[0002] With an aging population and increased health awareness, the demand for medical services is growing rapidly. However, high-quality medical resources are relatively scarce and unevenly distributed, leading to large tertiary hospitals operating beyond capacity for extended periods, while primary healthcare institutions are underutilized. Traditional healthcare systems primarily rely on simple appointment and queuing mechanisms, lacking intelligent resource allocation capabilities. While online appointments and remote consultations have become more widespread with the rapid development of internet-based healthcare, they still cannot solve the fundamental problem of uneven resource allocation. This imbalance not only reduces the overall efficiency of healthcare services but also increases patients' waiting times and medical costs, impacting the quality of medical care. Summary of the Invention

[0003] This invention provides a regional medical data analysis and processing method and system based on big data technology to solve the above-mentioned technical problems.

[0004] In a first aspect, the present invention provides a method for regional medical data analysis and processing based on big data technology, the method comprising the following steps:

[0005] Multi-dimensional dynamic data of medical institutions in the region are collected. Based on the multi-dimensional dynamic data, a composite stress index model is constructed to predict the service pressure trend of each medical institution, and the real-time composite stress index of each medical institution is calculated.

[0006] The real-time composite pressure index of each medical institution is mapped to geospatial space to construct a dynamic pressure potential energy field that characterizes the degree of congestion of medical services in the region.

[0007] When a user requests services from a target medical institution, it is determined whether the real-time composite pressure index of the target medical institution exceeds the preset congestion threshold.

[0008] If the congestion threshold is exceeded, the service request is transformed into a patient demand vector. Based on the spatiotemporal distribution characteristics of the dynamic pressure potential energy field, a hospital service vector containing a dynamic attraction dimension is constructed for other medical institutions in the region. By calculating the similarity between the patient demand vector and each hospital service vector, one or more candidate medical institutions are selected.

[0009] Based on the pressure characteristics of the target medical institution and the alternative medical institution in the dynamic pressure potential energy field, a diversion and guidance strategy is generated and pushed to the user's user terminal.

[0010] Optionally, the method further comprises the following steps of:

[0011] fusing the static resource data, the quasi-real-time service data and the real-time queuing data in the multi-dimensional dynamic data into a current pressure snapshot value through a weighted fusion function;

[0012] training the pre-deployed time series prediction model using a sequence of historical current pressure snapshot values as training data;

[0013] outputting a future pressure prediction value based on the current pressure snapshot value using the time series prediction model;

[0014] dynamically weighting and combining the current pressure snapshot value and the future pressure prediction value to generate a real-time composite pressure index.

[0015] Optionally, the time series prediction model is a long short-term memory network model.

[0016] Optionally, the method further comprises the following steps of:

[0017] abstracting each medical institution as a virtual particle in a geographic coordinate system, and taking the real-time composite pressure index of the medical institution as a particle value of the virtual particle;

[0018] calculating a pressure potential value of each preset grid cell in the geographic coordinate system based on the geographic positions and the particle values of all virtual particles using a spatial interpolation algorithm;

[0019] collecting the pressure potential values of all grid cells to generate a dynamic pressure potential field covering the entire region and continuously changing.

[0020] Optionally, the method further comprises the following steps of:

[0021] extracting a spatiotemporal distribution feature of the dynamic pressure potential field, the spatiotemporal distribution feature including a pressure potential value of a location where each medical institution is located and a pressure potential gradient, the pressure potential gradient representing a rate and a direction of pressure change at the corresponding location in the dynamic pressure potential field, and being used to quantify whether a service pressure of the medical institution is in a relieving trend or an exacerbating trend;

[0022] generating a patient demand vector using a word embedding model based on keywords and department information in the service request;

[0023] For medical institutions in the region, based on the service ability characteristics of the medical institutions, the reciprocal of the stress potential value and the stress potential gradient, a hospital service vector containing static service dimension, dynamic attraction dimension and stress trend dimension is constructed;

[0024] The cosine similarity algorithm is used to calculate the similarity score between the patient demand vector and each hospital service vector;

[0025] According to the similarity score in descending order, and select one or more medical institutions with a score higher than the preset matching threshold as the alternative medical institutions.

[0026] Optionally, the method of generating a diversion guide strategy based on the stress characteristics of the target medical institution and the alternative medical institution in the dynamic stress potential field and pushing the diversion guide strategy to the user terminal of the user comprises the following steps:

[0027] The stress potential values of the target medical institution and the alternative medical institution are extracted from the dynamic stress potential field;

[0028] Based on the stress potential values, the estimated queuing time of the user at the target medical institution and the alternative medical institution is calculated respectively;

[0029] The structured diversion guide strategy is generated by combining the stress potential value and the estimated queuing time;

[0030] The interface for reservation confirmation operation of the user terminal of the user presents the diversion guide strategy in the form of a pop-up window.

[0031] Optionally, the method further comprises the following steps:

[0032] When the user refuses the diversion guide strategy, the user's refusal behavior is defined as a refusal event;

[0033] Capture and associate the context parameters at the time of the refusal event, including at least the patient demand vector, the service ability characteristics of the target medical institution and the alternative medical institution, the stress potential value difference and the geographical distance difference between the target medical institution and the alternative medical institution;

[0034] A plurality of refusal events and associated context parameters are collected to form a decision friction data set;

[0035] A decision friction map is constructed based on the decision friction data set, and the decision friction map is used to dynamically adjust the diversion guide strategy.

[0036] Optionally, the method of constructing a decision friction map based on a decision friction data set comprises the following steps:

[0037] The clustering algorithm of unsupervised learning is used to cluster the context parameters in the decision friction data set, and multiple rejection modes are identified;

[0038] Each identified rejection mode is defined as a decision friction node;

[0039] The node association edge is constructed by analyzing the association strength between the patient demand vector and the decision friction node;

[0040] All decision friction nodes and node association edges are integrated to form a decision friction map representing the internal reasons for the rejection of the user group.

[0041] Optionally, the step of dynamically adjusting the diversion guide strategy using the decision friction map comprises the following steps:

[0042] When screening the candidate medical institutions for the user's new service request, the target decision friction node with the strongest association with the user's patient demand vector is queried in the decision friction map;

[0043] The rejection mode represented by the target decision friction node is determined;

[0044] Based on the rejection mode, a dynamic adjustment rule is generated, and the dynamic adjustment rule is applied to the step of screening the candidate medical institutions to dynamically adjust the weight parameter when calculating the similarity between the patient demand vector and the hospital service vector.

[0045] In a second aspect, the present application also provides a regional medical data analysis processing system based on big data technology, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to realize the regional medical data analysis processing method based on big data technology as described in the first aspect.

[0046] The present application has the following advantages:

[0047] The application can master the service conditions of each medical institution in the region in real time by establishing an intelligent dynamic monitoring mechanism, completely changing the traditional static management mode, and realizing the fundamental change from passive response to active allocation. The application has strong prediction and analysis capabilities, can accurately identify the congestion trend and distribution characteristics of medical services, provide reliable basis for scientific decision-making, and greatly improve the accuracy and foresight of resource allocation. The application realizes the accurate docking of patient demand and medical resources by building an intelligent matching mechanism, avoids resource waste caused by blind medical treatment, and significantly improves the overall service efficiency. The application also has self-learning and optimization capabilities, can continuously adjust and improve service strategies according to user feedback, and forms a virtuous cycle. The application will significantly alleviate the uneven distribution of medical resources, improve the accessibility of medical services, reduce the cost of patients seeking medical treatment, and provide strong technical support for building a more fair and efficient medical service system. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 It is a flowchart of the regional medical data analysis and processing method based on big data technology in one of the embodiments of the application.

[0049] Figure 2 It is a flowchart of the regional medical data analysis and processing method based on big data technology in one of the embodiments of the application.

[0050] Figure 3 It is a flowchart of the regional medical data analysis and processing method based on big data technology in one of the embodiments of the application. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the application will be described clearly in the following with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the application.

[0052] The terms "first", "second", etc. in the specification and claims of the application are used to distinguish similar objects, not to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are usually a class, not limited to the number of objects, for example, the first object can be one or more. In addition, "and / or" in the specification and claims means at least one of the connected objects, and the character " / ", generally represents a "or" relationship between the associated objects before and after.

[0053] Figure 1This is a flowchart illustrating a regional medical data analysis and processing method based on big data technology in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps. For example Figure 1 As shown, the regional medical data analysis and processing method based on big data technology disclosed in this invention specifically includes the following steps:

[0054] S101. Collect multi-dimensional dynamic data of medical institutions in the region, construct a composite stress index model based on the multi-dimensional dynamic data to predict the service pressure trend of each medical institution, and calculate the real-time composite stress index of each medical institution.

[0055] To accurately quantify and predict the service pressure on each hospital, it is necessary to comprehensively collect multi-dimensional dynamic data from all medical institutions within the region. This data includes relatively fixed static resource data such as the number of beds and doctors, as well as near-real-time business data such as registration volume and patient volume, and real-time queuing numbers in each department. A weighted fusion function can integrate these different types of data into a "current pressure snapshot value" that reflects the current capacity of the hospital. Subsequently, a pre-built time series prediction model, such as a Long Short-Term Memory (LSTM) network, is trained using a historical series of pressure snapshot values ​​to learn the intrinsic patterns of pressure changes. Finally, by dynamically weighting and combining the "future pressure prediction value" output by the model based on the current snapshot value with the "current pressure snapshot value," a more forward-looking real-time composite pressure index can be calculated. The index (in This is a snapshot of the current pressure value. This is a forecast of future stress levels. (As a dynamic weighting coefficient), its effect is to generate an accurate pressure indicator that reflects both reality and predicts the future, laying a solid data foundation for subsequent regional analysis and scheduling.

[0056] S102. Map the real-time composite pressure index of each medical institution to the geographic space to construct a dynamic pressure potential energy field that characterizes the degree of congestion of medical services in the region.

[0057] In order to convert the isolated pressure index of each medical institution into a regional, visual congestion situation, it is necessary to map these indexes onto the geographical space and construct a dynamic pressure potential field. In this process, each medical institution is abstracted as a virtual particle with a specific value in the geographic information system (GIS), and its geographic coordinates are its actual location, and its particle value is the real-time composite pressure index calculated previously . Then, a spatial interpolation algorithm such as inverse distance weighting (IDW) is used to calculate the pressure potential value of each preset grid cell in the entire regional coordinate system. The core idea of this algorithm is that the potential value of any unknown point is the weighted average of the values of all known particles (i.e. medical institutions) around it, and the weight is inversely proportional to the distance. Specifically, it can be calculated by the formula , where V is the pressure potential value of the grid cell, is the distance from the cell to the i-th medical institution, and k is the power index of the distance. By calculating the potential values of all grid cells and collecting them, a continuous and dynamic heat map covering the entire region is finally generated, which intuitively shows the congestion degree and distribution of medical service resources.

[0058] S103. When receiving a service request of a target medical institution from a user, it is determined whether the real-time composite pressure index of the target medical institution exceeds a preset congestion threshold.

[0059] When the system receives a service request (such as online registration) from a user for a target medical institution, a key judgment step is triggered. The system will immediately obtain the real-time composite pressure index of the target medical institution and compare it with a pre-set congestion threshold . This threshold is scientifically set according to historical data and management experience, representing the critical point at which the hospital service capacity is close to saturation. The implementation principle of this judgment step is very direct, which is a simple conditional judgment: if , the subsequent diversion guidance process is started. This step can actively identify service requests that may cause users to wait for a long time and have a poor medical experience, ensuring that the diversion guidance mechanism is only activated when it is really necessary. This avoids unnecessary disturbance to normal medical users, improving the efficiency of the system while also ensuring the smoothness of the user experience.

[0060] S104. If the congestion threshold is exceeded, the service request is converted into a patient demand vector, and based on the spatio-temporal distribution characteristics of the dynamic pressure potential field, a hospital service vector containing a dynamic attraction dimension is constructed for other medical institutions in the region, and by calculating the similarity between the patient demand vector and the hospital service vector of each hospital, one or more alternative medical institutions are selected.

[0061] If the pressure index of the target medical institution exceeds the congestion threshold, the system initiates an intelligent matching process to find alternative institutions. First, the user's service request (such as "child fever, see internal medicine") is converted into a patient demand vector containing specific medical needs using word embedding technology (such as Word2Vec) . At the same time, the system constructs a multi-dimensional hospital service vector for other medical institutions in the region , which not only includes static service dimensions such as specialty capabilities, but also innovatively incorporates dynamic attractiveness dimensions (e.g., inversely proportional to the pressure potential value V, the less congested the higher the attractiveness) and pressure trend dimensions (i.e., the gradient of the pressure potential field, indicating whether the pressure is easing or intensifying). Subsequently, by calculating the cosine similarity between the patient demand vector and each hospital service vector , the matching degree of each alternative hospital to the patient's demand can be quantified. Finally, the hospital with the highest similarity score and exceeding the preset matching threshold is selected as the alternative. This step achieves precise matching beyond geographical location, taking into account medical expertise, real-time congestion, and future trends.

[0062] S105. Based on the pressure characteristics of the target medical institution and the alternative medical institution in the dynamic pressure potential field, a diversion guidance strategy is generated and pushed to the user's user terminal.

[0063] After selecting suitable alternative medical institutions, the final step is to generate and push a persuasive diversion guidance strategy. The core of this strategy is to convert abstract "pressure" into specific information that users can intuitively perceive based on the pressure characteristics in the dynamic pressure potential field. The system extracts the pressure potential value V of the target medical institution and all alternative medical institutions where the user plans to go, and uses a preset function to convert it into an estimated waiting time . Then, these information, together with the distance, expertise of the hospital, etc., are integrated into a structured comparison list. For example, on the interface where the user is about to confirm the appointment, a pop-up window clearly presents: "The A hospital you selected is currently extremely busy, with an estimated queue time of over 90 minutes. We recommend B hospital, which has a high degree of specialty matching, with an estimated queue time of only 20 minutes, and is only a 10-minute drive away from you." This approach effectively improves the user's acceptance of diversion suggestions by providing quantitative, transparent, and easy-to-understand comparison information at key decision-making nodes, thereby effectively guiding passenger flow without forcing the user, achieving balanced allocation of regional medical resources, and ultimately improving the overall patient experience.

[0064] In one embodiment, a composite pressure index model for predicting the service pressure trend of each medical institution is constructed based on multi-dimensional dynamic data, and the real-time composite pressure index of each medical institution is calculated, including the following steps:

[0065] The static resource data, quasi-real-time business data and real-time queuing data in the multi-dimensional dynamic data are fused into a current pressure snapshot value by a weighted fusion function;

[0066] The historical sequence of current pressure snapshot values is used as training data to train a pre-deployed time series prediction model;

[0067] The time series prediction model is used to output future pressure prediction values based on the current pressure snapshot value;

[0068] The current pressure snapshot value and the future pressure prediction value are dynamically weighted and combined to generate a real-time composite pressure index.

[0069] In this embodiment, in order to accurately reflect the instant service load of a medical institution, the multi-dimensional dynamic data collected is first integrated into a single current pressure snapshot value by a weighted fusion function. These data include static resource data (such as the number of doctors and beds) representing service potential, quasi-real-time business data (such as the number of registered patients and patients in the clinic) reflecting current business volume, and real-time queuing data reflecting congestion bottlenecks. Since the units and dimensions of these data are different, they must be normalized first to map them to the interval of 0 to 1. Then, by assigning different weights to different data items to reflect the differences in their impact on service pressure, the current pressure snapshot value is finally calculated. The calculation method can be represented as wherein is the normalized data value of the i-th item, is the corresponding weight. The effect of this step is to refine complex and scattered multi-source information into a standardized and horizontally comparable instant pressure indicator, providing high-quality input for subsequent trend prediction.

[0070] After obtaining a continuous sequence of current pressure snapshot values, the next step is to use these historical data to train a pre-deployed time series prediction model, so that it has the ability to predict future pressure. The principle of this process is to feed a large number of historical pressure snapshot time series (for example, every ten minutes of the past months) into the model, so that it can learn the underlying rules and patterns of pressure changes, and then use these learned rules and patterns to predict future pressure. The output of this model is a sequence of future pressure prediction values, which can be represented as allow the model to autonomously learn the complex patterns embedded within, such as periodicity (daily morning rush, weekend lull) and trend (gradual increase in pressure due to seasonal flu outbreaks). Long Short-Term Memory (LSTM) networks are particularly suitable for this task due to their unique gating structure, which effectively captures long-term dependencies in data sequences, avoiding the gradient vanishing or exploding problems that traditional models may face when dealing with long sequences. This training process is essentially an optimization process, where the model continuously adjusts its internal parameters to minimize the difference between its predictions and the actual historical data.

[0071] When the time series prediction model is trained, it can be applied to real-time prediction. In practice, the system will input the latest calculated into the trained model as input. The model will analyze this current state using the internal laws it has learned and calculate a pressure prediction value for a specific time window in the future, such as the next hour. This output future pressure prediction value can be represented as , where M represents the trained time series prediction model. The core value of this step is to introduce a forward-looking perspective into pressure analysis, no longer just reflecting how crowded it is now, but being able to warn how crowded it may be in the future. This predictive ability allows medical resources to be allocated and patients to be guided more proactively and in advance, effectively preventing or exacerbating congestion, significantly improving the intelligence and proactive service level of the entire system.

[0072] Finally, in order to generate a final indicator that is both stable and sensitive to trends, the current snapshot value representing the current pressure needs to be dynamically weighted and combined with the future pressure prediction value indicating the future, generating a real-time composite pressure index. This combination is not a simple addition or average, but a dynamic weighting, because the importance of the current value and the prediction value is different in different situations. For example, when the prediction value shows that the pressure will soon rise sharply, the weight of the prediction value should be adjusted higher to give an early warning. This dynamic weighting combination process can be achieved through the formula , where is a dynamic weight coefficient between 0 and 1, which can be adjusted in real time according to the rate of change of the prediction value, the current time, etc. The final effect of this step is to produce a highly reliable composite index , which not only smooths out possible transient noise and glitches in the current data, but also fully incorporates scientific predictions of future trends, making it a more comprehensive, intelligent, and decision-supporting core indicator for measuring the service pressure of medical institutions.

[0073] In one embodiment, the real-time composite pressure index of each medical institution is mapped to the geographical space, and a dynamic pressure potential field representing the congestion level of medical services in the region is constructed by the following steps:

[0074] Each medical institution is abstracted as a virtual particle in the geographical coordinate system, and the real-time composite pressure index of the medical institution is taken as the particle value of the virtual particle;

[0075] Based on the geographical position and particle value of all virtual particles, the pressure potential value of each preset grid cell in the geographical coordinate system is calculated by using a spatial interpolation algorithm;

[0076] The pressure potential values of all grid cells are collected to generate a dynamic pressure potential field that covers the entire region and changes continuously.

[0077] In this embodiment, in order to convert the discrete hospital pressure data into a regional macro situation, the first step is to perform mathematical abstraction on each medical institution. Specifically, in the two-dimensional coordinate system of the geographic information system, each medical institution is no longer regarded as a complex entity occupying an area, but is simplified to a virtual particle without volume, and the position of the virtual particle is the precise geographical coordinates of the institution. The key attribute of this particle is its value, which is set to the real-time composite pressure index calculated earlier, which can dynamically reflect the service load of the institution. The essence of this step is to convert the complex real-world problem into a standardized spatial data problem, which lays the foundation for subsequent spatial analysis and calculation, enabling the pressure levels of different medical institutions to be quantified and compared in the same spatial framework, and completes the key conversion from business data to spatial geometric data.

[0078] After all medical institutions are abstracted as particles with values, the next step is to estimate the pressure conditions at any location in the region based on the information of these known points, which is achieved by using a spatial interpolation algorithm. The basic principle of this algorithm is the first law of geography, which states that similar things are more closely related. Therefore, the pressure potential value of an unknown location can be estimated by weighting and averaging the pressure values of all known particles (i.e., medical institutions) around it according to their distance. The closer the medical institution, the greater its influence (weight). One common implementation is the inverse distance weighting method (IDW), whose core formula is where is the pressure potential value of the grid cell to be solved, is the distance from the grid cell to the i-th medical institution, and k is a power index used to control the decay rate of distance influence. The effect of this step is to fill in the blank areas between medical institutions, extending discrete point data to continuous surface data, and providing algorithmic support for constructing a complete potential field.

[0079] Finally, by systematically applying the spatial interpolation algorithm, all the calculated results can be collected to generate a dynamic pressure potential field that covers the entire target area and changes continuously. The implementation is to first divide the entire geographic area into a fine grid composed of a large number of small units (e.g. squares). Then, for each unit in the grid, the pressure potential value of the center point is calculated using the interpolation formula of the previous step. When the potential values of all grid units are calculated, these values are collected to form a large data matrix. This matrix can be intuitively rendered into a heat map with different colors, where the darker the color (the higher the potential value), the higher the "pressure area" of medical services, and the lighter the color (the lower the potential value), the "low pressure area". The effect of this final result is to create a visual and dynamic decision support tool that not only shows the current congestion points, but also reveals the distribution, spread trend and mutual influence between regions of pressure, providing a macro and global perspective for intelligent diversion and resource optimization.

[0080] In one embodiment, based on the spatio-temporal distribution characteristics of the dynamic pressure potential field, a hospital service vector including a dynamic attraction dimension is constructed for other medical institutions in the region, and one or more candidate medical institutions are selected by calculating the similarity between the patient demand vector and each hospital service vector, including the following steps:

[0081] Extracting the spatio-temporal distribution characteristics of the dynamic pressure potential field, the spatio-temporal distribution characteristics including the pressure potential value of the location of each medical institution and the pressure potential gradient, the pressure potential gradient representing the rate and direction of pressure change at the corresponding location in the dynamic pressure potential field, used to quantify whether the service pressure of the medical institution is in a relief trend or an aggravation trend;

[0082] Based on the keywords and department information in the service request, a patient demand vector is generated using a word embedding model;

[0083] For medical institutions in the region, based on the service capability characteristics of the medical institutions, the reciprocal of the pressure potential value and the pressure potential gradient, a hospital service vector including a static service dimension, a dynamic attraction dimension and a pressure trend dimension is constructed;

[0084] Cosine similarity algorithm is used to calculate the similarity score between the patient demand vector and each hospital service vector;

[0085] According to the similarity score, arrange in descending order, and select one or more medical institutions with a score higher than the preset matching threshold as candidate medical institutions.

[0086] In this embodiment, in order to achieve accurate hospital recommendation, first of all, the key spatiotemporal distribution characteristics of each medical institution need to be extracted from the dynamic pressure potential field. This not only includes directly reading the pressure potential value of the geographical position of each institution, which represents its current congestion degree, but further needs to calculate the pressure potential gradient of the position. The gradient is a vector, which is calculated by analyzing the difference in potential values of the surrounding grid, and its direction points to the direction of the fastest pressure growth, and its size represents the severity of pressure change. This gradient can therefore quantify the future trend of service pressure, clearly indicating whether the congestion of the institution is easing or continuing to intensify. By extracting these two characteristics, the potential value of the current state and the potential gradient of the future trend, a dynamic and insightful pressure portrait of each medical institution is constructed, providing a core basis for subsequent intelligent matching beyond static information.

[0087] Next, the user's vague natural language service request, such as seeing what disease and hanging what department, needs to be converted into a standardized data structure that machines can understand and calculate. This process utilizes a word embedding model. This model is pre-trained on a large amount of medical text data, learning the semantic relationships between words. When receiving the key words and department information in the service request, the model will map each word (such as fever, pediatric department) to a high-dimensional numerical vector. By combining these word vectors, such as taking the average, a unique patient demand vector will be generated. The position and direction of this vector in the mathematical space accurately encodes the core appeal of the user's current medical visit. The effect of this step is to successfully convert unstructured text requests into structured numerical vectors, laying the foundation for subsequent quantitative similarity calculations.

[0088] Corresponding to the construction of the patient demand vector, a comprehensive, multi-dimensional hospital service vector also needs to be constructed for each medical institution in the region. This vector is designed to contain information in three key dimensions. The first is the static service dimension, which represents the inherent service capabilities of the hospital, such as its key specialties, equipment level, etc. The second is the dynamic attraction dimension, which is quantified by the reciprocal of the pressure potential value, i.e. the smaller the pressure, the higher the attraction, which intuitively reflects the convenience of medical treatment. The third is the pressure trend dimension, which is constructed using the pressure potential gradient calculated in the previous step, to represent whether the hospital's pressure is easing or intensifying, thus reflecting its future service potential. By integrating the information of these three dimensions into a vector, a dynamically updated mathematical portrait of each hospital is created, which can fully reflect its service status.

[0089] With the vector representing the patient's needs and the vector representing the service capacity of each hospital, the cosine similarity algorithm can be used to accurately calculate the matching degree between the two. The core principle of this algorithm is to calculate the cosine value of the angle between the two vectors in the multi-dimensional space. The value is independent of the length of the vector itself, and only cares about whether the direction of the two vectors is consistent. The calculation formula is The result Sim is a score between -1 and 1. The closer the score is to 1, the more consistent the direction of the hospital service vector and the direction of the patient demand vector, i.e. the more matching the service profile of the hospital and the medical needs of the patient. By performing this calculation for each candidate hospital in the region, the effect is to generate an objective and quantitative matching score for each hospital, thus simplifying the complex and multi-dimensional matching problem into a clear score sorting problem.

[0090] After calculating the similarity scores of all candidate medical institutions and patient needs, the last step is to filter according to these scores to determine the final recommendation list. This process first arranges all medical institutions in descending order of their similarity scores, with the highest score at the top. However, not all institutions with scores will be recommended, and a preset matching threshold must be set. Only those medical institutions with a similarity score higher than the threshold are considered as qualified candidates. The setting of this threshold plays a quality control role, filtering out options that may be ranked high but have low actual matching degree. Finally, from this list that has passed the threshold test, one or more medical institutions ranked at the top are selected as the final candidate. The effect of this step is to ensure that the recommendation suggestions pushed to the user are of high quality and highly relevant, thus completing the entire intelligent screening and matching process.

[0091] In one embodiment, generating a diversion guide strategy based on the pressure characteristics of the target medical institution and the candidate medical institution in the dynamic pressure potential field and pushing the diversion guide strategy to the user terminal of the user includes the following steps:

[0092] Extracting the pressure potential values of the target medical institution and the candidate medical institution from the dynamic pressure potential field;

[0093] Based on the pressure potential values, the estimated waiting time of the user at the target medical institution and the candidate medical institution is calculated respectively;

[0094] Combining the pressure potential values and the estimated waiting time to generate a structured diversion guide strategy;

[0095] The interface for reservation confirmation operation on the user terminal of the user presents the diversion guide strategy in the form of a pop-up window.

[0096] In this embodiment, to generate targeted diversion suggestions, first, the pressure potential values of the target medical institution currently selected by the user and the alternative medical institutions filtered by the system at their respective geographic locations need to be accurately extracted from the dynamic pressure potential field covering the entire region. This process is technically a direct data query operation. By obtaining the geographic coordinates of each medical institution, the pressure potential values at the corresponding locations are directly indexed and read from the already generated and stored pressure potential field data grid. The implementation effect of this step is to convert the macroscopic, visual regional congestion situation into precise quantitative values directly related to the current user's decision. These values are the basis for subsequent calculation and comparison, providing raw data input for converting the abstract pressure concept into specific user perception indicators.

[0097] After obtaining the pressure potential values, the next step is to convert them into indicators that users can most intuitively understand and perceive, namely estimated queuing time. This conversion is not linear, but is achieved through a function based on historical data modeling. This function needs to be able to reflect the non-linear growth of queuing time with pressure, i.e., when the pressure is low, the queuing time grows slowly, and when the pressure approaches the saturation point, the queuing time increases sharply. This function can be expressed as where is the estimated queuing time. By substituting the pressure potential values of the target medical institution and the alternative medical institutions into this function, the corresponding estimated queuing times can be calculated. The core effect of this step is to successfully build a bridge, translating a professional backend data indicator into frontend information that has a decisive impact on user decision-making, greatly improving the persuasiveness of the diversion suggestions.

[0098] After calculating the estimated queuing times, these core information needs to be combined with other auxiliary information to generate a clear and contrasting diversion guidance strategy. This is not just a list of numbers, but a strategic organization of information. A structured data object is created, which not only contains the pressure potential values and the estimated queuing times calculated therefrom, but also integrates other dimensions of comparison information such as hospital specialization matching degree, geographic distance, and transportation convenience. By juxtaposing the structured information of the target hospital and the alternative hospitals, a clear comparison list is formed. The implementation effect of this step is to integrate scattered data points into a logical and focused decision support solution, helping users quickly weigh the pros and cons through multi-dimensional information presentation, and preparing the content for the final interface presentation.

[0099] The last step is to present the carefully designed structured diversion guidance strategy to the user at the most critical decision node. This node is chosen at the interface where the user is about to complete the appointment and click the confirmation button. The strategy is presented through an automatically triggered pop-up window (or modal dialog box) to ensure that the information is fully noticed before the user makes the final decision. The pop-up interface clearly shows the huge difference in estimated queuing time between the original option and the recommended option in a highly visualized manner, such as using different colors or icons to mark the duration and congestion level, and provides clear operation buttons. The effect of this step is to complete the closed loop of the entire guidance process, by intervening at the right time in a friendly way, maximizing the user's acceptance of the diversion suggestion, and successfully converting the complex data analysis results in the background into actual effectiveness of regional medical resource optimization.

[0100] In one embodiment, the method further comprises the steps of:

[0101] When the user rejects the diversion guidance strategy, the user's rejection behavior is defined as a rejection event;

[0102] Capture and associate the context parameters at the time of the rejection event, including at least the patient demand vector, the service capacity characteristics of the target medical institution and the alternative medical institution, the pressure potential value difference between the target medical institution and the alternative medical institution, and the geographical distance difference;

[0103] Collect multiple rejection events and associated context parameters to form a decision friction data set;

[0104] Based on the decision friction data set, a decision friction map is constructed, and the decision friction map is used to dynamically adjust the diversion guidance strategy.

[0105] In this embodiment, in order to enable the guidance strategy to be self-optimized, the user's rejection behavior needs to be first converted into analyzable data. When the user faces the pop-up recommendation and finally chooses to ignore the suggestion and stick to the original option, this behavior is defined as a clear rejection event. This definition process technically means that the front-end interface of the system needs to be able to capture the user's final confirmation operation and determine whether the operation accepts or rejects the diversion. If it is determined to be a rejection, a specific event signal will be triggered. The implementation effect of this step is to convert a silent user choice into a landmark event with clear business meaning that can be recorded and processed by the back-end system. This lays the foundation for the event source for subsequent in-depth analysis of the underlying reasons behind user decision-making and the establishment of a feedback loop that can learn from failure.

[0106] After the rejection event is defined and captured, the next critical step is to immediately correlate and record the full context parameters at the time of the event. Because the user's rejection is not an isolated act, but a comprehensive judgment based on all available information at the time. Therefore, the system must capture and package a series of key data, which at least includes the patient demand vector representing the user's specific needs; the service capability characteristics of the target medical institution and the recommended alternative medical institution respectively; the difference between the two in the stress potential value (wherein and are the potential values of the target and alternative hospitals respectively); and the difference between the two in the geographical distance This series of context parameters fully reproduces all the information faced by the user when making decisions. The effect of this step is to attach a detailed snapshot to each rejection event, so that subsequent analysis is no longer blind guesswork, but data mining with evidence.

[0107] With time, a large number of rejection events and their associated context parameters will be captured continuously. In order to conduct pattern analysis, these isolated event records need to be aggregated to form a dedicated decision friction dataset. In terms of technical implementation, this is a data engineering task, that is, the complete context parameters of each rejection event are stored as a record in a dedicated database or data warehouse. Each row of this dataset represents the friction between the system suggestion and the user preference, and each column is a potential factor that may cause this friction, such as pressure difference, distance difference, department matching degree, etc. The implementation effect of this step is to transform countless individual and instantaneous decision behaviors into macroscopic and collectively intelligent assets that can be statistically analyzed. Based on the aggregated decision friction dataset, a decision friction atlas that reveals the internal correlation of rejection reasons can be finally constructed, and it can be used to dynamically optimize future diversion strategies. First, a clustering algorithm based on unsupervised learning is used to analyze the context parameters in the dataset, aggregate similar rejection scenarios, and identify multiple typical rejection patterns, such as extremely sensitive to distance mode or highly loyal to specific hospital brand mode. Each pattern is defined as a decision friction node in the atlas. Then, by analyzing the association strength between different types of patient demand vectors and these nodes, the association edges between nodes are constructed. When a new user's service request arrives, the system can query the most matched friction node in the atlas according to the user's demand, so as to predict the most possible rejection reason of the user. Based on this prediction, the system can dynamically adjust the weight parameters when calculating the similarity of the alternative hospitals for the user, for example, if it is identified that the user belongs to the distance sensitive type, the negative weight of the distance factor in the similarity calculation will be increased. The effect of this step is to make the diversion strategy have the ability to learn and evolve, from a universal rule engine to an intelligent and personalized guidance system that can understand the user's implicit preferences and avoid potential decision friction in advance.

[0108] In one embodiment, constructing a decision friction atlas based on the decision friction dataset includes the following steps:

[0109] Using a clustering algorithm based on unsupervised learning to analyze the context parameters in the decision friction dataset and identify multiple rejection patterns;

[0110] Defining each identified rejection pattern as a decision friction node;

[0111] Constructing node association edges by analyzing the association strength between patient demand vectors and decision friction nodes;

[0112] Integrating all decision friction nodes and node association edges to form a decision friction atlas representing the internal reasons for user group rejection and diversion.

[0113] In this embodiment, to reveal the underlying reasons behind users' rejection of intelligent shunting suggestions, unsupervised learning cluster analysis is first performed on the collected decision friction dataset. This process aims to automatically discover natural groupings present in the data without predefining any labels. Cluster algorithms such as K-Means or DBSCAN are used to process each rejection event as a multi-dimensional data point, with dimensions composed of context parameters such as pressure difference and distance difference. The algorithm groups similar rejection events into a class based on their distance in the feature space. For example, all events that reject due to excessive distance naturally cluster together, forming a cluster. The implementation of this step refines and summarizes thousands of seemingly chaotic rejection records into a limited number of typical rejection patterns with clear business implications, completing the first transition from raw data to abstract patterns. After identifying multiple rejection patterns, the next step is to assign a clear identity to each pattern under the framework of graph construction.

[0114] Specifically, each cluster obtained through cluster analysis is defined as an independent decision friction node. This node conceptually represents a specific user decision preference or obstacle. For example, a cluster with a central point characterized by a large distance difference and a small pressure difference can be understood as a distance highly sensitive node. This step is a concept mapping process in technology, which converts statistical clustering results into nodes, the basic elements in graph theory. The effect is to establish a clear and discrete operating object for subsequent analysis, solidify vague patterns into basic units that can be connected and analyzed in the graph, and lay a structural foundation for building a knowledge network.

[0115] After obtaining nodes representing rejection patterns, it is necessary to construct node association edges to connect users with these patterns to reflect the tendency of different user groups to produce specific rejection behaviors. The core of this step is to analyze the association strength between different types of patient demand vectors and each decision friction node in the original dataset. For each decision friction node, the frequency of each type of patient demand vector in all rejection events attributed to the node can be counted. Through calculation, an association strength score can be obtained, for example, using conditional probability to quantify, where represents the association strength of the demand vector with the node . The higher this score, the more likely it is that users with this demand feature will exhibit the rejection pattern represented by the node. The effect of this step is to establish a probability link from user demand to rejection behavior, enabling the graph to have predictive ability.

[0116] Finally, all the created decision friction nodes and node association edges are integrated to form a complete decision friction graph that represents the internal reasons for the user group's rejection diversion. In terms of technical implementation, this is a process of combining a node set and an edge set into a unified graph structure, which can be represented as where N is the set of all decision friction nodes, E is the set of all node association edges, and the weight of each edge is its association strength. This graph is not a static picture, but a computable and dynamic knowledge base. The final effect is to create an analysis tool that can comprehensively and deeply reveal the user's decision logic. By querying and traversing this graph, we can understand the decision preferences behind different medical needs, so that the diversion guidance strategy can fundamentally understand and adapt to the user's real ideas, achieving true intelligence and personalization.

[0117] In one embodiment, dynamically adjusting the diversion guidance strategy using the decision friction graph includes the following steps:

[0118] When screening the candidate medical institutions for the user's new service request, query the target decision friction node associated with the user's patient demand vector in the decision friction graph;

[0119] Determine the rejection pattern represented by the target decision friction node;

[0120] Generate a dynamic adjustment rule based on the rejection pattern, and apply the dynamic adjustment rule to the step of screening candidate medical institutions to dynamically adjust the weight parameter when calculating the similarity between the patient demand vector and the service vector of each hospital.

[0121] In this embodiment, when a new service request is generated, the system will first use its corresponding patient demand vector to query the pre-constructed decision friction graph. The purpose of this query is to find the decision friction node that is most strongly associated with the current user demand. In terms of specific implementation, the pre-set association strength score between the patient demand vector and each node in the graph is calculated, and this score reflects the probability that users with such demand have historically exhibited a specific rejection pattern. By comparing all scores, the node with the highest score, i.e. the target decision friction node, can be locked. The selection process can be represented as where is the edge weight representing the association strength. The effect of this step is to accurately predict the decision bias or concerns that the user is most likely to have before the recommendation is generated.

[0122] After the target decision friction node is locked, the next step is to determine the specific rejection mode represented by the node. Since each node is built by clustering rejection events with similar context parameters, each node naturally carries the semantic information of the mode it represents. This process is an information extraction operation, that is, reading the core characteristics from the attributes of the target node. For example, if the characteristics of the target node are a large geographical distance difference and a small pressure potential value difference, the rejection mode it represents is determined to be extremely sensitive to distance. The effect of this step is to convert the abstract node query result in the previous step into a specific and understandable business rule, such as user prioritizing distance over queuing time, providing clear guidance for subsequent dynamic adjustment strategies.

[0123] Finally, based on the determined rejection mode, a dynamic adjustment rule is generated and applied to the screening process of the alternative medical institutions. This rule directly acts on the step of calculating the similarity between the patient demand vector and the hospital service vector, and is realized by dynamically adjusting the weight of each parameter. For example, if the user's rejection mode is determined to be distance sensitive, the negative weight related to geographical distance in the hospital service vector will be temporarily and significantly increased when calculating the similarity. The adjusted similarity calculation can be regarded as wherein is the dynamic weight vector generated according to the rejection mode. The final effect of this step is to enable the recommendation algorithm to actively avoid the user's potential rejection point, generate a highly personalized alternative list for the user that is more in line with his implicit preferences, and thus greatly improve the success rate of the diversion guidance strategy and user satisfaction.

[0124] The application also discloses a regional medical data analysis processing system based on big data technology, which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, and the processor implements the regional medical data analysis processing method based on big data technology as described in any one of the above embodiments when executing the computer program.

[0125] The processor can be a central processing unit (CPU), of course, according to actual use, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), ready-to-program gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. can also be used, and the general-purpose processor can be a microprocessor or any conventional processor, etc. The present application does not limit this.

[0126] The memory can be an internal storage unit of the computer device, for example, a hard disk or a memory of the computer device, or an external storage device of the computer device, for example, a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD), a flash card (FC) or the like, or a combination of the internal storage unit and the external storage device. The memory is configured to store a computer program and other programs and data required by the computer device, and is also configured to temporarily store data that has been output or is to be output. The present application does not limit the memory.

[0127] It should be understood by those of ordinary skill in the art that the above discussion of any of the embodiments is merely exemplary and is not intended to suggest any limitation as to the scope of the present application. The above embodiments or technical features in different embodiments can be combined, and the steps can be implemented in any order, and there are many other changes and variations of the different aspects of one or more embodiments of the present application as described above. For the sake of brevity, they are not provided in detail.

[0128] One or more embodiments of the present application are intended to cover all such alternatives, modifications and variations as fall within the broad scope of the present application. Therefore, any omission, modification, equivalent replacement, improvement and the like made within the spirit and principle of one or more embodiments of the present application should be included in the scope of the present application.

Claims

1.A regional medical data analysis processing method based on big data technology, characterized in that, The method comprises the following steps: Collecting multi-dimensional dynamic data of each medical institution in the region, and fusing static resource data, quasi-real-time business data and real-time queuing data in the multi-dimensional dynamic data into a current pressure snapshot value through a weighted fusion function; Training a pre-deployed time series prediction model using a sequence of historical current pressure snapshot values as training data; Outputting a future pressure prediction value based on the current pressure snapshot value using the time series prediction model; Generating a real-time composite pressure index by dynamically weighting and combining the current pressure snapshot value and the future pressure prediction value; Mapping the real-time composite pressure index of each medical institution to a geographic space to construct a dynamic pressure potential field representing the congestion level of medical services in the region; When receiving a service request from a user for a target medical institution, determining whether the real-time composite pressure index of the target medical institution exceeds a preset congestion threshold; If the congestion threshold is exceeded, the service request is converted into a patient demand vector, and the spatio-temporal distribution characteristics of the dynamic pressure potential field are extracted, including the pressure potential value of the location of each medical institution and the pressure potential gradient, which represents the rate and direction of pressure change in the dynamic pressure potential field, and is used to quantify the service pressure of the medical institution in a relieving or intensifying trend; Generating a patient demand vector using a word embedding model based on the keywords and department information in the service request; For medical institutions in the region, based on the service capacity characteristics of the medical institutions, the reciprocal of the pressure potential value and the pressure potential gradient, a hospital service vector is constructed, which includes static service dimensions, dynamic attraction dimensions and pressure trend dimensions; Using a cosine similarity algorithm, the similarity score between the patient demand vector and each hospital service vector is calculated; According to the similarity score, the medical institutions with a score higher than a preset matching threshold are selected as candidate medical institutions in descending order; Based on the pressure characteristics of the target medical institution and the candidate medical institutions in the dynamic pressure potential field, a diversion guidance strategy is generated and pushed to the user terminal of the user. 2.The regional medical data analysis processing method based on big data technology according to claim 1, wherein, The time series prediction model is a long short-term memory network model. 3.The regional medical data analysis processing method based on big data technology according to claim 1, characterized in that, The mapping of the real-time composite pressure index of each medical institution to the geographic space to construct a dynamic pressure potential field representing the congestion level of medical services in the region comprises the following steps: Each medical institution is abstracted as a virtual particle in a geographic coordinate system, and the real-time composite pressure index of the medical institution is used as the particle value of the virtual particle; Based on the geographic location and particle value of all virtual particles, a spatial interpolation algorithm is used to calculate the pressure potential value of each preset grid cell in the geographic coordinate system; Collecting the pressure potential values of all grid cells to generate a dynamic pressure potential field that covers the entire region and changes continuously. 4.The regional medical data analysis processing method based on big data technology according to claim 1, characterized in that, The steps of generating a diversion guidance strategy based on the pressure characteristics of the target medical institution and the candidate medical institutions in the dynamic pressure potential field, and pushing the diversion guidance strategy to the user terminal of the user include the following steps: Extracting the pressure potential values of the target medical institution and the candidate medical institutions from the dynamic pressure potential field; Based on the pressure potential values, the estimated queuing time of the user at the target medical institution and the candidate medical institutions is calculated respectively; The structured diversion guidance strategy is generated in combination with the pressure potential value and the estimated queuing time; The interface for the reservation confirmation operation of the user terminal of the user presents the diversion guidance strategy in the form of a pop-up window. 5.The regional medical data analysis processing method based on big data technology according to claim 4, characterized in that, The method further includes the following steps: When the user rejects the diversion guidance strategy, the rejection behavior of the user is defined as a rejection event; Contextual parameters when the rejection event occurs are captured and associated, including at least a patient demand vector, service capability characteristics of the target medical institution and the alternative medical institution, a difference in pressure potential value between the target medical institution and the alternative medical institution, and a difference in geographical distance; A plurality of rejection events and associated contextual parameters are collected to form a decision friction data set; A decision friction graph is constructed based on the decision friction data set, and the diversion guidance strategy is dynamically adjusted using the decision friction graph. 6.The regional medical data analysis processing method based on big data technology according to claim 5, characterized in that, The construction of the decision friction graph based on the decision friction data set includes the following steps: A clustering algorithm of unsupervised learning is used to perform clustering analysis on the contextual parameters in the decision friction data set, and a plurality of rejection modes are identified; Each identified rejection mode is defined as a decision friction node; A node association edge is constructed by analyzing the association strength between the patient demand vector and the decision friction node; All decision friction nodes and node association edges are integrated to construct a decision friction graph representing the internal reasons for the user group to reject diversion. 7.The regional medical data analysis processing method based on big data technology according to claim 6, characterized in that, The dynamic adjustment of the diversion guidance strategy using the decision friction graph includes the following steps: When screening alternative medical institutions for a new service request of a user, the target decision friction node with the strongest association with the patient demand vector of the user is queried in the decision friction graph; The rejection mode represented by the target decision friction node is determined; A dynamic adjustment rule is generated based on the rejection mode, and the dynamic adjustment rule is applied to the step of screening alternative medical institutions to dynamically adjust the weight parameter when calculating the similarity between the patient demand vector and the service vector of each hospital. 8.A regional medical data analysis processing system based on big data technology, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, The processor executes the computer program to implement the regional medical data analysis processing method based on big data technology according to any one of claims 1 to 7.

Images