Medical resource intelligent scheduling method, system and chip based on multi-modal data

Abstract

The application provides a kind of medical resource intelligent scheduling method, system and chip based on multi-modal data, belongs to collaborative scheduling technical field, specifically includes: database construction module is responsible for collecting the multi-modal data related to medical resources, the semantic alignment of the multi-modal data is carried out, redundancy is removed and associated fusion, and a unified multi-modal medical resource supply and demand dynamic database is formed;Predictive management module with the prediction strategy of the medical resources of different medical institutions, and combining the multi-modal fusion data, a dynamic supply and demand prediction model of a deep learning framework is used to output the prediction result of medical resource demand;Scheduling execution module is responsible for taking the prediction result of the dynamic supply and demand prediction model library as input, generating the scheduling scheme under the scene type, issuing the scheduling scheme to the corresponding execution subject, and tracking the scheduling execution effect, to ensure the reliability of the supply of medical resources.

Description

Technical Field

[0001] This invention belongs to the field of collaborative scheduling technology, and in particular relates to a method, system and chip for intelligent scheduling of medical resources based on multimodal data. Background Technology

[0002] Currently, my country's healthcare sector is at a critical stage of digital transformation. The traditional model of medical resource allocation has long relied on manual planning, which presents numerous deep-seated problems. On the one hand, multi-source medical data is scattered and fragmented across different departments and institutions, lacking a unified integration mechanism, making it difficult to fully realize the value of the data. On the other hand, resource allocation is mainly based on static planning, failing to dynamically adapt to changes in demand brought about by population movement, disease outbreaks, and emergencies. Simultaneously, the lack of sound cross-departmental and cross-regional collaboration mechanisms, along with information barriers and ambiguous division of responsibilities, leads to slow emergency response and difficulty in handling complex scenarios such as public health emergencies.

[0003] Existing technical solutions often rely on historical medical data, demographic data, and other statistical analysis methods to predict future medical resource needs and formulate annual or quarterly resource allocation plans. However, these solutions suffer from the following technical problems: Only by combining multimodal data related to medical resources, such as satellite remote sensing data and ground sensing data, can the accuracy of medical resource demand predictions be guaranteed. However, using multimodal data for prediction processing on different medical institutions may lead to high difficulty in prediction processing and complexity in data processing. Therefore, it is an urgent technical problem to solve how to combine the matching degree of the model's prediction results to determine the medical institutions that need secondary correction of the prediction results of medical resource demand, that is, the medical institutions with reliable supply guarantee, and to determine the prediction optimization strategy for different medical institutions based on the distribution data of medical institutions with reliable supply guarantee, so as to reduce the difficulty of prediction processing.

[0004] In summary, to address the aforementioned technical issues, this application provides a method, system, and chip for intelligent scheduling of medical resources based on multimodal data. Summary of the Invention

[0005] To achieve the objectives of this invention, the following technical solution is adopted: Specifically, in the first aspect, this application provides a medical resource intelligent scheduling system based on multimodal data, specifically including: The database construction module is responsible for collecting multimodal data related to medical resources, performing semantic alignment, redundancy removal, and correlation fusion on the multimodal data to form a unified dynamic database of supply and demand for multimodal medical resources. The prediction management module determines the prediction strategy for medical resources of the analysis target based on the changes in the multimodal data of the analysis target and the updated results of the selected medical institutions. Based on the prediction strategies for medical resources of different medical institutions and combined with the multimodal fusion data, a dynamic supply and demand prediction model using a deep learning framework is adopted to output the prediction results of medical resource demand. The scheduling and execution module is responsible for generating a scheduling scheme for the scenario type by taking the prediction results of the dynamic supply and demand prediction model library as input, distributing the scheduling scheme to the corresponding execution entity, and tracking the scheduling execution effect.

[0006] Furthermore, the multimodal data includes satellite remote sensing data, ground sensing data, and medical data.

[0007] Furthermore, the satellite remote sensing data and ground remote sensing data are remote sensing data of traffic flow on the roads between the medical resource allocation center and medical institutions.

[0008] Secondly, this application provides a method for intelligent scheduling of medical resources based on multimodal data, applied to the aforementioned intelligent scheduling system for medical resources based on multimodal data, specifically including: S1 uses the model's prediction results to determine the degree of matching between the predicted medical resource demand in different medical institutions and the user's demand. Using the degree of matching and the distance between medical institutions whose matching degree does not meet the requirements and other medical institutions, when it is determined that the prediction model can be simplified, proceed to the next step. S2 determines the matching degree of the demand matching deviation institutions in the medical institutions based on the matching degree, and determines the identification strategy of the medical institutions that perform medical resource scheduling after correcting the prediction results according to a preset ratio based on the overlap of the predicted deviation dates of the demand matching deviation institutions, and uses it as the identification strategy for screening medical institutions. S3 determines the screened medical institutions among the medical institutions based on the identification strategy, takes the medical institutions other than the screened medical institutions as the analysis targets, and determines the identification and analysis strategy of the multimodal data of the analysis targets based on the distribution data of the screened medical institutions and the interval distance between the screened medical institutions and different analysis targets. S4 uses the identification and analysis strategy to determine the changes in the multimodal data of the analysis target, and based on the changes, determines the prediction strategy for the medical resources of the analysis target according to the updated results of the screened medical institutions in the medical institutions.

[0009] Furthermore, the degree of matching between the predicted medical resource demand in the medical institution and the user demand is determined based on the deviation rate between the predicted short-term medical resource demand and the user demand.

[0010] Furthermore, it was determined that the prediction model could be simplified, specifically including: S11 determines the date on which the predicted medical resource demand is less than the user demand and the deviation rate from the user demand is greater than a preset deviation rate threshold based on the degree of matching between the predicted medical resource demand in the medical institution and the user demand, and uses this date as the predicted deviation date. S12 determines the prediction deviation weight value of the medical institution based on the proportion of prediction deviation dates in the medical institution; S13 determines whether a simplification of the prediction model can be performed by using the prediction bias weights of different medical institutions and the distance between medical institutions that do not meet the matching requirements and other medical institutions.

[0011] Thirdly, this application provides a chip device, which includes: a data acquisition module, an analysis module, and a control module, wherein the data acquisition module is responsible for acquiring data, the analysis module is responsible for the above-mentioned intelligent medical resource scheduling method based on multimodal data, and the control module is responsible for outputting control commands based on the analysis results.

[0012] The beneficial effects of this invention are as follows: Based on the degree of overlap between demand-matching deviation institutions and their predicted deviation dates, the risk level of scheduling medical resources among different medical institutions on different dates is determined. Based on the risk level of scheduling medical resources among different medical institutions on different dates, an identification strategy for screening medical institutions is determined. This reduces the over-allocation of medical resources while ensuring the reliability of scheduling medical resources among medical institutions.

[0013] Based on the updated results of the screened medical institutions, the reliability of cross-institutional scheduling for different analytical objectives is determined. By combining the reliability of cross-institutional scheduling for different analytical objectives with the changes in multimodal data of the analytical objectives, a prediction strategy for medical resources of the analytical objectives is developed. This not only reduces the difficulty of predicting and processing medical resources, but also ensures the reliability of the supply of medical resources.

[0014] Other features and advantages will be set forth in the following description, and the objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] The above and other features and advantages of the present invention will become more apparent from a detailed description of exemplary embodiments thereof with reference to the accompanying drawings.

[0017] Figure 1 This is a framework diagram of a medical resource intelligent scheduling system based on multimodal data; Figure 2 This is a flowchart of a method for intelligent scheduling of medical resources based on multimodal data; Figure 3 This is a flowchart illustrating methods for simplifying predictive models. Figure 4 This is a flowchart illustrating the method for determining the identification and analysis strategy for the multimodal data of the analysis target. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0019] Example 1 Specifically, such as Figure 1 As shown, this application provides a medical resource intelligent scheduling system based on multimodal data, specifically including: The database construction module is responsible for collecting multimodal data related to medical resources, performing semantic alignment, redundancy removal, and correlation fusion on the multimodal data to form a unified dynamic database of supply and demand for multimodal medical resources. The prediction management module is responsible for determining the prediction strategy for medical resources of the analysis target based on the changes in the multimodal data of the analysis target and the updated results of the selected medical institutions. Based on the changes in the multimodal fusion data, the module uses a dynamic supply and demand prediction model with a deep learning framework to output the prediction results of medical resource demand. The scheduling and execution module is responsible for generating a scheduling scheme for the scenario type using the prediction results from the dynamic supply and demand prediction model library as input, distributing the scheduling scheme to the corresponding execution entity, and tracking the scheduling execution effect.

[0020] Specifically, the implementation of the database construction module is as follows: In the multimodal data acquisition phase, remote sensing imagery data of traffic flow on roads between the medical resource allocation center and various medical institutions is acquired via a satellite remote sensing platform. This traffic flow imagery data includes road congestion index, vehicle density distribution, and traffic speed distribution. Ground sensing data is collected through ground-based sensing devices deployed at key road nodes. This ground-based sensing data includes real-time traffic flow counts, road weather data, and traffic light status data. Medical data is collected through the information systems of medical institutions. This medical data includes bed occupancy rates, departmental visit volumes, drug inventory levels, and medical device usage status data for each medical institution. These three types of data are collected synchronously using a unified timestamp and spatial coordinates to ensure temporal and spatial consistency of the multimodal data.

[0021] In the semantic alignment stage, using the spatial location of medical institutions as the baseline coordinates, the road congestion index in satellite remote sensing data is mapped according to road segments between medical institutions, and the real-time traffic flow count in ground sensing data is matched according to the corresponding road segments. Medical data such as bed occupancy rate and patient visits are aggregated according to medical institutions, forming a unified data structure with medical institutions as nodes in the spatial dimension. In the redundancy removal stage, for cases where both satellite remote sensing data and ground sensing data exist for the same road segment, the real-time traffic flow count in the ground sensing data is used as a benchmark to verify the corresponding data in the satellite remote sensing data. Redundant data with a deviation rate exceeding a preset deviation rate threshold is removed, along with duplicate index data recorded in different data sources. In the association and fusion stage, using medical institutions as nodes, the medical data of each medical institution is associated with the traffic flow remote sensing data and ground sensing data of the surrounding roads, forming a multi-dimensional feature vector containing medical resource supply capacity and transportation channel traffic conditions. A unified multimodal dynamic database of medical resource supply and demand is constructed, and this database is dynamically updated according to a preset update cycle.

[0022] The specific implementation method of the prediction management module is as follows: In the prediction strategy determination stage, based on the medical resource prediction strategies for each analysis target determined in S4 of Example 2, the data input method adopted by each medical institution is determined. Specifically, for analysis targets using the basic prediction strategy, daily dynamically identified satellite remote sensing data, ground sensing data, and medical data are used as inputs; for analysis targets using the screening prediction strategy, multimodal data from the available date group replaces the daily dynamically identified data as inputs; for analysis targets using the secondary screening prediction strategy, the inputs are determined based on the scheduling matching weight value and the proportion of dates. In the dynamic supply and demand prediction model operation stage, the determined inputs are input into the dynamic supply and demand prediction model, and the model outputs the medical resource demand prediction results for each medical institution. These prediction results include the bed demand, drug demand, and equipment demand for each medical institution within a preset future timeframe.

[0023] The dynamic supply and demand forecasting model is constructed as follows: It uses medical data from various medical institutions, satellite remote sensing data of surrounding roads, and ground-based sensor data as input features, and the actual medical resource consumption of each medical institution after a preset future time period as the output label. A deep learning framework fusing Long Short-Term Memory (LSTM) networks and Multilayer Perceptrons (MLPs) is employed for model construction. The LSTM part handles temporal feature inputs, including historical multi-day sequences of bed occupancy rates, patient visit volumes, and road traffic flow, utilizing the LSTM's temporal memory capability to capture the dynamic trend of medical resource demand over time. The MLP part handles static feature inputs, including the total number of beds, department configuration, geographical location, and surrounding road network topology of medical institutions, utilizing the MLP's nonlinear mapping capability to capture the impact of inherent attributes of medical institutions on resource demand. The temporal feature vector output from the LSTM and the static feature vector output from the MLP are concatenated and fused, then input into a fully connected layer for regression prediction, outputting the predicted medical resource demand values ​​for each medical institution. The training data is constructed based on historical operation records. Multimodal data from each historical day is used as input samples, and the actual medical resource consumption for a preset duration after the corresponding day is used as label samples. The backpropagation algorithm and Adam optimizer are used to iteratively optimize the model parameters. The mean squared error loss function is used to ensure that the deviation between the model prediction results and the actual needs is minimized.

[0024] The update mechanism of the dynamic supply and demand forecasting model is as follows: After each scheduling cycle, the actual medical resource consumption data and the corresponding input multimodal data within that cycle are added to the training dataset as new training samples to incrementally train the model and update its parameters. During the update process, if the deviation rate between the predicted result and the actual demand of a certain medical institution continuously exceeds a preset deviation rate threshold for multiple consecutive cycles, the sampling weight of the relevant features of that medical institution in the training dataset is increased. This allows the model to pay more attention to the feature patterns of that medical institution in subsequent training, thereby achieving dynamic adaptation of the model to the evolution of demand features of different medical institutions and ensuring that the prediction results accurately reflect the changing trends of medical resource demand.

[0025] The specific implementation of the scheduling execution module is as follows: In the scheduling plan generation stage, the predicted medical resource demand of each medical institution, output by the dynamic supply and demand forecasting model, is used as input. This is combined with the screening status of each medical institution, the multimodal data changes of the target, and the distance between medical institutions to generate a scheduling plan for the specified scenario. The scheduling plan includes the source medical institution, target medical institution, type of dispatched materials, quantity, dispatch route, and estimated arrival time. For screened medical institutions, the dispatch quantity is determined based on the revised forecast results; for non-screened medical institutions, the dispatch quantity is determined based on the original forecast results. In the dispatch route determination stage, satellite remote sensing data and ground sensor data of the roads between the source and target medical institutions are used as the basis, selecting routes with lower traffic volume and higher travel speeds as the dispatch routes. In the scheduling plan distribution stage, the scheduling plan is distributed to the corresponding execution entities via a communication network. These execution entities include medical resource delivery vehicles, delivery personnel, and receiving personnel at the medical institutions. Each execution entity executes the dispatch and transportation tasks of medical resources according to the scheduling plan. In the execution effect tracking stage, the location and traffic status of dispatched transport vehicles are monitored in real time through satellite remote sensing data and ground sensor data. The inventory changes and usage of medical resources after receiving them are monitored in real time through the information system of medical institutions. The actual effect of dispatch execution is compared with the prediction results to identify deviations in the dispatch plan. The deviation data is fed back to the prediction management module for model iterative optimization, thus forming a closed-loop management from prediction to dispatch to feedback optimization.

[0026] The system in this embodiment achieves a complete technical closed loop from multimodal data acquisition and fusion to prediction strategy determination, supply and demand prediction model operation, scheduling scheme generation and execution, and execution effect feedback optimization through the collaborative work of the database construction module, prediction management module, and scheduling execution module. Its working principle corresponds to the method steps described in Embodiment 2. The database construction module corresponds to the acquisition and fusion processing of multimodal data, the prediction management module corresponds to the prediction strategy determination and model prediction in S1 to S4, and the scheduling execution module corresponds to the application of prediction results and scheduling closed-loop management.

[0027] Furthermore, the multimodal data includes satellite remote sensing data, ground sensing data, and medical data.

[0028] Furthermore, the satellite remote sensing data and ground remote sensing data are remote sensing data of traffic flow on the roads between the medical resource allocation center and medical institutions.

[0029] Example 2 Secondly, such as Figure 2 As shown, this application provides a medical resource intelligent scheduling method based on multimodal data, applied to the aforementioned medical resource intelligent scheduling system based on multimodal data, specifically including: S1 uses the model's prediction results to determine the degree of matching between the predicted medical resource demand in different medical institutions and the user's demand. Using the degree of matching and the distance between medical institutions whose matching degree does not meet the requirements and other medical institutions, when it is determined that the prediction model can be simplified, proceed to the next step. The core objective of this embodiment is to determine whether the prediction model can be simplified by comparing the degree of matching between the predicted medical resources and user needs. It then determines the strategy for identifying medical institutions by comparing the degree of overlap between the matched deviation institutions and the predicted deviation dates. The distribution and distance between the selected medical institutions determine the identification and analysis strategy. Finally, it determines the prediction strategy for the analysis target by considering the changes in multimodal data and the updated results of the selected medical institutions. The core logic uses the degree of matching of the prediction results as input. First, it determines whether the model can be simplified by using the prediction deviation weight value and the distance. Then, it determines the selected medical institutions by comparing the proportion of matched deviation institutions and the degree of overlap in the deviation dates. Next, it determines the identification and analysis strategy by considering the distribution and distance of the selected institutions. Finally, it determines the prediction strategy by considering the changes in multimodal data and the scheduling matching weight value. The overall logic follows the process of "predictive model simplification determination → medical institution identification → identification and analysis strategy determination → prediction strategy determination".

[0030] Furthermore, the degree of matching between the predicted medical resource demand in the medical institution and the user demand is determined based on the deviation rate between the predicted short-term medical resource demand and the user demand.

[0031] The degree of matching refers to the degree of agreement reflected by the deviation rate between the short-term medical resource demand forecast and the actual user demand. The smaller the deviation rate, the higher the degree of matching. The short term refers to the recent period within a preset time frame. The user demand refers to the actual demand for medical resources within the service area of ​​the medical institution.

[0032] Suppose that a medical institution's predicted medical resource demand for a certain day is 100 units, and the actual user demand is 85 units. Then the deviation rate is (100-85) / 85≈0.176, and the degree of matching is reflected by the deviation rate.

[0033] The method for determining the degree of matching was clarified, and the deviation rate between the prediction results and the actual needs was used as the objective basis for evaluating the degree of matching, thereby providing a quantitative basis for subsequent identification of prediction deviation institutions and simplified processing judgment.

[0034] It is understandable that, such as Figure 3 As shown, the simplification of the prediction model can be determined, specifically including: In this embodiment, based on the degree of matching between the predicted medical resource demand in medical institutions and user demand, as well as the distance between medical institutions whose matching degree does not meet the requirements and other medical institutions, it is determined whether the prediction model can be simplified. This enables the assessment of the reliability of dynamic scheduling of medical resources among medical institutions based on the distribution data of medical institutions whose matching degree does not meet the requirements. Based on the reliability of dynamic scheduling of medical resources among medical institutions, it is determined whether the prediction model can be simplified. This lays the foundation for further reducing the processing difficulty of predicting medical resources while ensuring the reliability of the supply of medical resources in medical institutions.

[0035] The simplified processing of the prediction model refers to replacing the daily dynamically identified multimodal data with data from historically available date groups as model input to reduce the amount of data processing; the "matching degree not meeting the requirements" refers to medical institutions whose prediction deviation rate exceeds a preset deviation rate threshold. The meaning of "not meeting the requirements" needs to be clearly defined in the pre-verification confirmation, as indicated here.

[0036] If the medical institutions that do not meet the matching requirements are close to other medical institutions, it means that the prediction bias can be made up by cross-institutional scheduling. Therefore, the prediction model can be simplified to reduce the processing difficulty.

[0037] By determining whether the process can be simplified based on the degree of matching and the distance between intervals, the prediction deviation information and the spatial distribution information of medical institutions are transformed into a simplified decision-making basis, thereby reducing the difficulty of prediction processing while ensuring the reliability of supply.

[0038] S11, based on the degree of matching between the predicted results of medical resource demand in the medical institution and user demand, determine the date on which the predicted results of medical resource demand are less than user demand and the deviation rate from user demand is greater than a preset deviation rate threshold, and use this date as the predicted deviation date. The prediction deviation date refers to the date on which the prediction result is lower than the user's needs and the deviation rate exceeds the preset deviation rate threshold, that is, the insufficient prediction on that day led to resource shortage.

[0039] Suppose a medical institution's prediction result on a certain day is 80 units, the user demand is 100 units, and the deviation rate is (100-80) / 80=0.25. If it exceeds the preset deviation rate threshold of 0.15, then that day is the prediction deviation day.

[0040] By filtering predicted deviation dates using the deviation rate, dates with insufficient prediction leading to shortages can be identified from historical operational data, providing a basis for the subsequent calculation of prediction deviation weight values.

[0041] S12 determines the prediction deviation weight value of the medical institution based on the proportion of prediction deviation dates in the medical institution; The prediction deviation weight value refers to the proportion of the number of prediction deviation dates of a medical institution to the total number of historical dates. The value ranges from 0 to 1. The larger the value, the more frequent the prediction deviation of the institution.

[0042] If a medical institution has 6 prediction bias dates in the past 30 days, then the prediction bias weight value = 6 / 30 = 0.20.

[0043] The predicted deviation weight value is calculated by the proportion of the deviation date, and the frequency of the predicted deviation is quantified into a comparable weight metric, providing a quantitative basis for judging whether subsequent processing can be simplified.

[0044] S13 determines whether a simplification of the prediction model can be performed by using the prediction bias weights of different medical institutions and the distance between medical institutions that do not meet the matching requirements and other medical institutions.

[0045] Assuming that the prediction bias weights and intervals for each medical institution are already determined, determine whether the processing can be simplified based on the judgment results of S131 and S132.

[0046] It is understandable that, based on the prediction bias weights across different medical institutions and the distance between medical institutions that do not meet the matching requirements and other medical institutions, it is determined whether a simplification of the prediction model can be performed. Specifically, this includes: S131. Based on the prediction deviation weight values ​​of different medical institutions, determine whether the average value of the prediction deviation weight values ​​of different medical institutions is greater than the preset deviation weight threshold. If so, then once the prediction model is simplified, it may cause the prediction deviation of different medical institutions to become more serious, and there are a large number of medical institutions with serious prediction deviations, making it impossible to achieve the coordinated scheduling of medical resources between medical institutions. Therefore, it is determined that the prediction model cannot be simplified. If not, proceed to step S132. The preset deviation weight threshold refers to the critical value used to determine whether the average value of the prediction deviation weight is too high.

[0047] If the average prediction bias weight is greater than the preset bias weight threshold, it indicates that the overall prediction bias is serious and cannot be compensated by cross-agency scheduling. In this case, simplification is not allowed. Simplification is prohibited when the average bias weight is high. When the overall prediction bias is serious, the prediction accuracy is ensured by maintaining the complete prediction model, thereby avoiding the exacerbation of bias due to simplification.

[0048] S132, medical institutions whose prediction deviation weight value is greater than a preset deviation weight threshold are designated as demand matching deviation institutions, and demand matching deviation institutions whose interval distance with other non-demand matching deviation institutions is greater than a preset distance threshold are designated as screening deviation institutions. It is determined whether the proportion of the number of the screening deviation institutions in the medical institutions is greater than a preset deviation institution proportion threshold. If so, then once the prediction model is simplified, it may lead to more serious prediction deviations among different medical institutions, and the coordinated scheduling of medical resources between demand matching deviation institutions and other medical institutions will be difficult. Therefore, it is determined that the prediction model cannot be simplified. If not, it is determined that the prediction model can be simplified.

[0049] The "demand matching deviation institution" refers to a medical institution whose prediction deviation weight value exceeds a preset deviation weight threshold, indicating that the institution's prediction deviation is relatively serious. The "screening deviation institution" refers to a demand matching deviation institution that is far from both non-deviation institutions, indicating that the institution cannot compensate for the deviation through cross-institution scheduling. The preset distance threshold is a critical value for judging whether the distance between medical institutions is too far. The preset deviation institution proportion threshold is a critical value for judging whether the proportion of screening deviation institutions is high.

[0050] If the proportion of institutions with screening deviations is high, it indicates that there are many medical institutions whose deviations are difficult to compensate for through cross-institutional scheduling, and therefore simplification is not possible; if the proportion is low, it indicates that most institutions with deviations can compensate for through scheduling, and therefore simplification is possible.

[0051] By screening the proportion of deviation mechanisms, it is determined whether the processing can be simplified. When the average deviation weight is controllable, the simplification decision is dynamically adjusted according to the spatial distribution characteristics of the deviation mechanisms, thereby reducing the difficulty of prediction processing while ensuring the reliability of scheduling.

[0052] This embodiment realizes a complete process from prediction deviation date identification to prediction deviation weight value calculation and simplified processing judgment through steps S11 to S13. The core value is reflected in two aspects: First, by judging the average prediction deviation weight value, simplification is prohibited when the overall deviation is serious in order to ensure prediction accuracy; Second, by judging the proportion of deviation institutions, the simplification decision is dynamically adjusted according to the spatial distribution of deviation institutions when the deviation is controllable, thereby achieving an adaptive balance between supply reliability and processing efficiency.

[0053] A medical resource scheduling system covers 10 medical institutions, H1 to H10. The deviation rate between the short-term medical resource demand forecast results and user demand of the dynamic supply and demand forecast model for each medical institution has been calculated. Taking H1 as an example, the prediction deviation dates (prediction < demand and deviation rate > 0.15) in the historical 30 days are 4 days, and the prediction deviation weight value = 4 / 30 ≈ 0.133.

[0054] Step S11: The prediction deviation dates and prediction deviation weights for each medical institution are as follows: H1: Deviation date 4 days, weight value 0.133; Step S12: The prediction bias weight values ​​for each medical institution are as described above.

[0055] Step S13: Step S131: Mean of prediction deviation weight value = (0.133+0.233+0.067+0.200+0.100+0.267+0.033+0.167+0.067+0.133) / 10 = 1.40 / 10 = 0.140.

[0056] Determine whether the mean 0.140 is greater than the preset deviation weight threshold 0.20: 0.140 is not greater than 0.20, no, proceed to S132.

[0057] Step S132: The demand matching deviation mechanisms (weight value > 0.20) are H2 (0.233) and H6 (0.267), a total of 2.

[0058] Determine whether H2 and H6 are screening deviation mechanisms (the distance between them and non-deviation mechanisms is >5 km): The distance between H2 and H1 is 3km, and the distance between H2 and H3 is 4km, neither of which is greater than 5km. Therefore, H2 is not a screening deviation mechanism.

[0059] H6 is 6km away from H1, 7km away from H3, 8km away from H4, 6km away from H7, 7km away from H8, 8km away from H9, and 6km away from H10. It is also more than 5km away from all non-deviation mechanisms. H6 is the screening deviation mechanism.

[0060] The screening bias mechanism is H6, with a total of 1 unit. Its percentage is 1 / 10 = 0.10.

[0061] Determine if the proportion of 0.10 is greater than the preset deviation mechanism proportion threshold of 0.30: If 0.10 is not greater than 0.30, then the simplification of the prediction model can be performed and proceed to S2.

[0062] S2 determines the matching degree of the demand matching deviation institutions in the medical institutions based on the matching degree, and determines the identification strategy of the medical institutions that perform medical resource scheduling after correcting the prediction results according to a preset ratio based on the overlap of the predicted deviation dates of the demand matching deviation institutions, and uses it as the identification strategy for screening medical institutions. Specifically, the method for determining the identification strategy for screening medical institutions is as follows: In this embodiment, the risk level of scheduling medical resources between different medical institutions on different dates is determined based on the overlap between the demand matching deviation institutions and the predicted deviation dates of the demand matching deviation institutions. Based on the risk level of scheduling medical resources between different medical institutions on different dates, an identification strategy for screening medical institutions is determined. This reduces the over-allocation of medical resources while ensuring the reliability of scheduling medical resources between medical institutions.

[0063] Specifically, the method for determining the identification strategy for screening medical institutions is as follows: In this embodiment, the risk level of scheduling medical resources between different medical institutions on different dates is determined based on the overlap between the demand matching deviation institutions and the predicted deviation dates of the demand matching deviation institutions. Based on the risk level of scheduling medical resources between different medical institutions on different dates, an identification strategy for screening medical institutions is determined. This reduces the over-allocation of medical resources while ensuring the reliability of scheduling medical resources between medical institutions.

[0064] The selected medical institutions refer to those whose medical resource allocation is processed after the prediction results have been corrected according to a preset ratio. This means that the demand prediction results for these institutions are amplified and corrected to compensate for prediction bias. The preset ratio refers to the amplification factor used to correct the prediction results, which is greater than 1. The pre-screening confirms that the deviation weight threshold is less than the preset deviation weight threshold; this threshold is used for a more lenient screening process.

[0065] If the predicted deviation date of a certain demand matching deviation institution highly overlaps with that of other deviation institutions, it indicates that multiple institutions will have deviations on the same day, and the risk of cross-institutional scheduling is high. In this case, the institution should be included in the screening of medical institutions to correct the prediction results and reduce the scheduling risk.

[0066] The screening strategy is determined by the proportion of deviation institutions and the degree of overlap of deviation dates. The information on the distribution of deviation institutions and the overlap of dates is transformed into the basis for screening decisions, thereby reducing over-allocation while ensuring scheduling reliability.

[0067] S21 Based on the demand matching deviation institutions in the medical institution, determine the proportion of demand matching deviation institutions in the medical institution, and use the proportion of demand matching deviation institutions in the medical institution as the proportion of matching deviation institutions; The "matching deviation ratio" refers to the proportion of institutions with demand mismatches to the total number of all medical institutions.

[0068] If we assume that 2 out of 10 medical institutions are institutions with demand mismatch, then the proportion of institutions with mismatch is 2 / 10 = 0.20.

[0069] By calculating the proportion of matching deviation mechanisms, a basis for determining the magnitude of the difference in screening strategies based on the proportion is provided.

[0070] It is understandable that if the proportion of the matched deviation institutions is greater than the preset deviation institution proportion threshold, the prediction deviation is more serious. In order to ensure the reliability of cross-hospital scheduling, medical institutions whose prediction deviation weight value is greater than the deviation weight threshold (less than the preset deviation weight threshold) are selected as screening medical institutions.

[0071] The preset deviation mechanism proportion threshold refers to the critical value for determining whether the proportion of matched deviation mechanisms is high; the deviation weight threshold refers to a more lenient screening threshold that is less than the preset deviation weight threshold, used to include more mechanisms with moderate deviations in the screening range when the proportion of deviation mechanisms is high. Pre-screening confirms that the deviation weight threshold is less than the preset deviation weight threshold.

[0072] If the proportion of institutions with mismatch is high, it indicates that the prediction error is serious. In this case, a more lenient error weight threshold is used to screen more institutions as medical institutions to ensure the reliability of cross-hospital scheduling. When the proportion of institutions with mismatch is high, a lenient screening strategy is adopted. When the error is serious, more institutions are included to correct the prediction results and ensure the reliability of scheduling.

[0073] It is also understood that if the proportion of the matching deviation mechanism is not greater than the preset deviation mechanism proportion threshold, the process proceeds to step S22.

[0074] For example, if the proportion of matching deviation mechanisms is not greater than the preset deviation mechanism proportion threshold, it indicates that there are not many deviation mechanisms, and then proceed to S22 for more detailed analysis.

[0075] S22 determines the demand matching deviation mechanism that belongs to the predicted deviation date based on the degree of overlap of the predicted deviation date of the demand matching deviation mechanism, and uses the proportion of the demand matching deviation mechanism that belongs to the predicted deviation date as the demand matching deviation proportion of the date. The demand matching deviation ratio refers to the proportion of demand matching deviation institutions that belong to the predicted deviation date on a certain date to the total number of demand matching deviation institutions, reflecting the degree of overlap of deviation institutions on that date.

[0076] If two demand matching deviation agencies both predict deviation dates on a certain day, then the demand matching deviation ratio for that day = 2 / 2 = 1.0, indicating that the deviations completely overlap.

[0077] The demand matching deviation ratio is calculated by measuring the degree of overlap between deviation dates, which quantifies the overlap between deviation institutions from the date dimension and provides a basis for the refined determination of subsequent screening strategies.

[0078] It should be noted that the above steps include the following: Based on the demand matching deviation ratio on different dates, determine whether the average demand matching deviation ratio on different dates is less than the preset matching deviation ratio threshold. If so, the overlap of the predicted deviation dates of different demand matching deviation institutions is not high. Therefore, medical institutions with predicted deviation weight values ​​greater than the preset deviation weight threshold are selected as screening medical institutions. If not, proceed to step S23. The preset matching deviation ratio threshold refers to the critical value used to determine whether the average ratio of demand matching deviation is too low.

[0079] If the average demand matching deviation ratio is less than the preset matching deviation ratio threshold, it indicates that the overlap of predicted deviation dates between the institutions is not high and the risk of cross-institutional scheduling is low. In this case, institutions with deviation weight values ​​exceeding the preset deviation weight threshold will be selected as medical institutions. If the average value is not less than the threshold, it indicates that the overlap is high and the scheduling risk is high. In this case, proceed to S23 for further analysis.

[0080] Judging the degree of overlap of deviation dates by the average deviation ratio is significant because it allows for the use of simple screening strategies to reduce resource waste when the degree of overlap is low, and for determining the screening strategy through more refined analysis when the degree of overlap is high.

[0081] S23, determine the identification strategy for screening medical institutions based on the proportion of matching deviation institutions and the proportion of demand matching deviation for different dates.

[0082] Assuming that the proportion of matching deviation institutions and the proportion of demand matching deviation on different dates have been determined, the screening strategy is determined based on the judgment results of S231 and S232 respectively.

[0083] It is understood that the identification strategy for screening medical institutions is determined based on the proportion of institutions with matching deviations and the proportion of demand matching deviations on different dates, specifically including: S231, dates with a demand matching deviation ratio not less than a preset matching deviation ratio threshold are taken as overlapping dates. It is determined whether the ratio of overlapping dates meets the requirements. If so, it means that the number of overlapping dates is small, so proceed to step S232. If not, medical institutions with a prediction deviation weight value greater than the deviation weight threshold (less than the preset deviation weight threshold) are selected as screening medical institutions.

[0084] The overlapping date refers to the date on which the demand matching deviation ratio reaches or exceeds the preset matching deviation ratio threshold, that is, the degree of overlap of the deviation institutions on that day is relatively high; the "meeting the requirements" refers to the case where the overlapping date ratio does not exceed the preset overlapping date ratio threshold, that is, there are fewer overlapping dates.

[0085] If the proportion of overlapping dates meets the requirement (i.e., is low), it indicates that the deviation overlap is not serious, and then proceed to S232 for further analysis; if the proportion of overlapping dates does not meet the requirement (i.e., is high), it indicates that the deviation overlap is serious and the scheduling risk is high, and then the deviation weight threshold is used to screen more institutions.

[0086] Judging whether more detailed analysis is needed by the proportion of overlapping dates is significant because it allows for the determination of a screening strategy when there are few overlapping dates, while a more lenient screening strategy can be adopted when there are many overlapping dates to ensure scheduling reliability.

[0087] S232, based on the proportion of matching deviation institutions and the proportion of demand matching deviation on different dates, determine the scheduling demand value, and determine whether the scheduling demand value is greater than a preset scheduling demand threshold. If so, medical institutions whose predicted deviation weight value is greater than the deviation weight threshold (less than the preset deviation weight threshold) are selected as screening medical institutions. If not, medical institutions whose predicted deviation weight value is greater than the deviation weight threshold and have demand matching deviation institutions within a preset distance interval from the medical institution, as well as medical institutions whose predicted deviation weight value is greater than the preset deviation weight threshold, are selected as screening medical institutions. The preliminary verification confirms that the scheduling demand value calculation formula needs to be supplemented.

[0088] The scheduling demand value refers to a quantitative indicator calculated by combining the ratio of the comprehensive matching deviation mechanism and the average ratio of the demand matching deviation; the preset scheduling demand threshold refers to a critical value for judging whether the scheduling demand value is too high.

[0089] The scheduling demand value is calculated as follows: Let the matching deviation ratio be P_deviation, and the average of the demand matching deviation ratios for different dates be P_overlap. Then: Scheduling demand value = P_deviation + P_overlap.

[0090] If the calculated scheduling demand value is greater than the preset scheduling demand threshold, it indicates that the combined effect of the proportion of deviation institutions and the degree of deviation overlap is significant. In this case, the deviation weight threshold is used to screen more institutions. If it is not greater than the threshold, a more refined screening strategy is adopted to include institutions whose deviation weight value exceeds the deviation weight threshold and which have deviation institutions nearby, as well as institutions whose deviation weight value exceeds the preset deviation weight threshold, in the screening scope.

[0091] By further differentiating the screening strategy by scheduling demand values, the screening range can be dynamically adjusted according to the degree of joint impact when there are few overlapping dates, thereby achieving a fine balance between screening coverage and screening accuracy.

[0092] It should be noted that if the medical institution is among the screened medical institutions, the prediction result is determined by multiplying the preset ratio and the prediction result, and this result is used as the prediction result of the medical institution's medical resource demand.

[0093] The preset ratio refers to the amplification factor that corrects the prediction results. It is greater than 1, which means that the prediction results of the screened medical institutions are amplified to make up for the insufficiency of the prediction.

[0094] Assuming the predicted result for a certain screening medical institution is 80 units, and the preset ratio is 1.2, then the corrected predicted result = 80 × 1.2 = 96 units.

[0095] This embodiment realizes a complete process from calculating the proportion of institutions with matching deviations to analyzing the degree of overlap of deviation dates and determining the screening strategy for medical institutions. Its core value is reflected in three aspects: First, by judging the proportion of institutions with matching deviations, a lenient screening strategy is adopted when the deviation is serious to ensure scheduling reliability; second, by judging the average deviation ratio, a simple screening strategy is adopted when the degree of overlap of deviation dates is not high to improve efficiency; and third, by judging the grading of the overlap date ratio and scheduling demand value, the screening range is dynamically adjusted when the degree of overlap is different, thereby achieving an adaptive balance between screening coverage and screening accuracy.

[0096] Continuing with the calculation results of S1, the demand matching deviation mechanisms are H2 (weight value 0.233) and H6 (weight value 0.267), a total of 2.

[0097] Step S21: Matching deviation mechanism ratio = 2 / 10 = 0.20.

[0098] Determine whether the matching deviation mechanism ratio of 0.20 is greater than the preset deviation mechanism ratio threshold of 0.25: 0.20 is not greater than 0.25, no, proceed to S22.

[0099] Step S22: Calculate the overlap of predicted deviation dates for demand matching deviation agencies H2 and H6 on each date.

[0100] Assume the prediction deviation dates for H2 are D3, D5, D7, D9, D11, D13, D15 (7 days), and the prediction deviation dates for H6 are D1, D5, D7, D11, D13, D15, D17, D19 (8 days). The overlapping dates for the two institutions are D5, D7, D11, D13, D15 (5 days).

[0101] Demand matching deviation percentage for each date: D5, D7, D11, D13, D15: 2 / 2 = 1.0 (Both institutions have deviation dates) Other deviation dates: 1 / 2 = 0.5 The average proportion of demand matching deviation = (1.0×5+0.5×5) / 10 = 7.5 / 10 = 0.75.

[0102] Determine if the mean value of 0.75 is less than the preset matching deviation ratio threshold of 0.50: 0.75 is not less than 0.50; otherwise, proceed to S23.

[0103] Step S23: Step S231: Overlapping dates (demand matching deviation ratio ≥ 0.50) are D5, D7, D11, D13, and D15, totaling 5 days. Overlapping date ratio = 5 / 30 = 0.167. Determine if the overlapping date ratio of 0.167 meets the requirement (i.e., does not exceed the preset overlapping date ratio threshold of 0.30): 0.167 is not greater than 0.30, meeting the requirement (less than 0.30), proceed to S232.

[0104] Step S232: Calculate the scheduling demand value. Scheduling demand value = matching deviation ratio + average demand matching deviation ratio = 0.95. Determine whether the scheduling demand value of 0.95 is greater than the preset scheduling demand threshold of 0.8. If yes.

[0105] Based on the path that meets the conditions in S232, medical institutions whose prediction deviation weight value is greater than the deviation weight threshold of 0.15 are selected as the screening medical institutions.

[0106] The selected medical institutions are H2, H4, H6, and H8, a total of 4 institutions. The non-selected medical institutions (the target of analysis) are H1, H3, H5, H7, H9, and H10, a total of 6 institutions.

[0107] Correction of prediction results for screening medical institutions: Taking H2 as an example, if the prediction result is 80 units, the corrected prediction result = 80 × 1.2 = 96 units.

[0108] S3 determines the screened medical institutions among the medical institutions based on the identification strategy, takes the medical institutions other than the screened medical institutions as the analysis targets, and determines the identification and analysis strategy of the multimodal data of the analysis targets based on the distribution data of the screened medical institutions and the interval distance between the screened medical institutions and different analysis targets. It should be noted that, as Figure 4 As shown, the method for determining the identification and analysis strategy for the multimodal data of the analysis target is as follows: In this embodiment, the reliability of using the screening of medical institutions' distribution data and the distance between the screening medical institutions and different analysis targets is determined. Based on the reliability of using the screening of medical institutions to schedule medical resources among medical institutions, the identification and analysis strategy of the multimodal data of the analysis targets is determined. That is, which analysis targets are reliably identified and processed for changes in multimodal data. This reduces the difficulty of analysis and processing, and also lays the foundation for reducing the difficulty of predicting some analysis targets.

[0109] The identification and analysis strategy refers to the decision-making scheme for determining which analytical targets to analyze for multimodal data changes; the multimodal data includes satellite remote sensing data and ground sensing data, wherein the satellite remote sensing data and ground sensing data are remote sensing data of traffic flow on the road between the medical resource allocation center and medical institutions.

[0110] If there are screening medical institutions near a certain analysis target, it indicates that the bias of the analysis target can be compensated by cross-institution scheduling. Therefore, multimodal data variation analysis is performed on the analysis target to determine its data variation.

[0111] By screening the distribution and spacing of medical institutions to determine the identification and analysis strategy, the spatial distribution information of the screened institutions is transformed into the basis for identification and analysis decision-making, thereby reducing the difficulty of analysis and processing while laying the foundation for reducing the difficulty of prediction and processing.

[0112] S31, using the distribution data of the screened medical institutions, determine the proportion of the number of screened medical institutions among the medical institutions; The percentage of screened medical institutions refers to the proportion of screened medical institutions to the total number of all medical institutions.

[0113] If 4 out of 10 medical institutions are selected, then the percentage is 4 / 10 = 0.40.

[0114] By calculating the proportion of screened institutions, a basis for determining the magnitude of the identification and analysis strategy based on the differences in proportion is provided.

[0115] It is understandable that if the proportion of the selected medical institutions in the medical institutions is greater than the preset institution proportion threshold, the selected medical institutions can be used to schedule medical resources among different medical institutions. Therefore, the identification and analysis strategy for the multimodal data of the analysis target is to compare and analyze the satellite remote sensing data and ground sensing data of all analysis targets on different dates to determine the changes in the satellite remote sensing data and ground sensing data.

[0116] The preset institution proportion threshold refers to the critical value for judging whether the proportion of screened medical institutions is high. If the proportion of screened institutions is greater than the preset institution proportion threshold, it indicates that the coverage of screened institutions is good and cross-institution scheduling can be carried out using screened institutions. Then, multimodal data change analysis is performed on all analysis targets.

[0117] When the proportion of selected institutions is high, a comprehensive analysis strategy is adopted. When the coverage of selected institutions is good, multimodal data variation analysis of all analysis targets is used to ensure data accuracy, thereby providing a sufficient data foundation for determining subsequent prediction strategies.

[0118] It should also be noted that if the proportion of the number of screened medical institutions in the medical institutions is not greater than the preset institution proportion threshold, proceed to step S32. S32, by filtering the distance between medical institutions and different analysis targets, determine the medical institutions within the preset distance range of the analysis targets; The screening medical institutions within the preset distance range refer to screening medical institutions whose distance from a certain analysis target is within the preset distance range, that is, there are schedulable screening institutions near the analysis target.

[0119] If there are two screening medical institutions within a 5-kilometer radius of a certain analysis target, then the analysis target has screening medical institutions within a preset distance interval.

[0120] By filtering medical institutions within a preset distance range through intervals, schedulable screening institutions near each analysis target are identified from a spatial perspective, providing a foundation for the refined determination of subsequent identification and analysis strategies.

[0121] S33 determines the identification and analysis strategy for the multimodal data of the analysis target based on the proportion of the number of screened medical institutions in the medical institutions and the screened medical institutions within the preset distance range of different analysis targets.

[0122] Assuming that the proportion of screened institutions and the data of screened institutions near each analysis target have been determined, the identification and analysis strategies are determined based on the judgment results of S331 and S332.

[0123] It is understood that, based on the proportion of screened medical institutions among the medical institutions and the screened medical institutions within the preset distance range of different analysis targets, the multimodal data identification and analysis strategy for the analysis targets is determined, specifically including: S331, using different analysis targets within a preset distance range to screen medical institutions, identify analysis targets that have medical institutions within the preset distance range, and determine whether the proportion of analysis targets that have medical institutions within the preset distance range among all analysis targets is greater than a preset target proportion threshold. If yes, proceed to step S332; otherwise, compare and analyze the satellite remote sensing data and ground sensing data of the analysis targets that have medical institutions within the preset distance range on different dates to determine the changes in the satellite remote sensing data and ground sensing data. The preset target proportion threshold refers to the critical value used to determine whether the proportion of analytical targets with screening mechanisms is high.

[0124] If the proportion of analysis targets with screening agencies is greater than the preset target proportion threshold, it indicates that most analysis targets are near screening agencies, and then proceed to S332 for further analysis; if the proportion is not greater than the threshold, then only the analysis targets with screening agencies are subjected to change analysis.

[0125] The need for more granular analysis can be determined by the percentage of analytical targets with screening institutions. When the percentage is low, only the analytical targets with screening institutions are analyzed for changes to save resources. When the percentage is high, more granular analysis is used to determine the strategy.

[0126] S332: Determine whether the proportion of screened medical institutions among the medical institutions is less than a preset proportion threshold. If so, compare and analyze the satellite remote sensing data and ground sensing data of the analysis targets that exist within the preset distance range on different dates to determine the changes in the satellite remote sensing data and ground sensing data. If not, determine that the multimodal data identification and analysis strategy for the analysis targets is to compare and analyze the satellite remote sensing data and ground sensing data of all analysis targets on different dates to determine the changes in the satellite remote sensing data and ground sensing data. Preliminary verification confirms that the "preset proportion threshold" and the "preset institution proportion threshold" in S31 are different thresholds. If the preset proportion threshold is less than the preset institution proportion threshold, S332 is used to determine whether the proportion is too low to adopt a more stringent identification strategy.

[0127] The preset percentage threshold refers to a more stringent screening threshold that is less than the preset institution percentage threshold, used to determine whether the percentage of screened institutions is too low.

[0128] If the proportion of screened institutions is less than the preset proportion threshold, it indicates that there are few screened institutions and the cross-institutional scheduling capability is limited. In this case, only the analysis targets with screened institutions will be subjected to change analysis. If the proportion is not less than the preset proportion threshold, it indicates that the number of screened institutions is acceptable. In this case, all analysis targets will be subjected to change analysis.

[0129] The significance of further differentiating the strategy by screening the proportion of institutions lies in the ability to dynamically adjust the scope of analysis when the proportions are different, thereby achieving a fine balance between analytical coverage and resource efficiency.

[0130] Furthermore, the changes are determined based on the degree of similarity between satellite remote sensing data and ground sensing data on different dates.

[0131] The aforementioned variation refers to the consistency characteristics of satellite remote sensing data and ground sensing data across different dates; the lower the similarity, the greater the data variation.

[0132] If the similarity between the satellite remote sensing traffic flow data of a certain analysis target on a certain day and the data of the previous day is low, then the variation on that day is relatively large.

[0133] This embodiment realizes a complete process from calculating the proportion of screening institutions to interval distance analysis and then to determining the identification and analysis strategy. Its core value is reflected in three aspects: First, by judging the proportion of screening institutions, a comprehensive analysis strategy is adopted when the coverage is good to ensure data sufficiency; second, by judging the proportion of analysis targets with screening institutions, only the analysis targets with screening institutions are analyzed for change when the proportion is not high, so as to save resources; third, by further judging the proportion of screening institutions, the scope of change analysis is dynamically adjusted when the proportion is different, thereby achieving an adaptive balance between analysis coverage and resource efficiency.

[0134] Continuing with the calculation results of S2, the selected medical institutions are H2, H4, H6, and H8, a total of 4 institutions. The analysis targets are H1, H3, H5, H7, H9, and H10, a total of 6 institutions.

[0135] Step S31: The percentage of medical institutions screened = 4 / 10 = 0.40.

[0136] Determine if the proportion of 0.40 is greater than the preset institution proportion threshold of 0.35: If 0.40 is greater than 0.35, then yes.

[0137] Based on the path that meets the conditions in S31, the identification and analysis strategy is to compare and analyze the satellite remote sensing data and ground sensing data of all analysis targets on different dates to determine the changes, that is, to perform multimodal data change analysis on all six analysis targets H1, H3, H5, H7, H9, and H10.

[0138] S4 uses the identification and analysis strategy to determine the changes in the multimodal data of the analysis target, and based on the changes, determines the prediction strategy for the medical resources of the analysis target according to the updated results of the screened medical institutions in the medical institutions.

[0139] Furthermore, the changes are determined based on the degree of similarity between satellite remote sensing data and ground sensing data on different dates.

[0140] Specifically, the method for determining the prediction strategy for the medical resources of the analysis target is as follows: In this embodiment, based on the updated results of the screened medical institutions, the reliability of cross-medical institution scheduling for different analysis targets is determined. The reliability of cross-medical institution scheduling for different analysis targets is combined with the changes in the multimodal data of the analysis targets to formulate a prediction strategy for the medical resources of the analysis targets. This not only reduces the difficulty of predicting and processing medical resources, but also ensures the reliability of the supply of medical resources.

[0141] The prediction strategy refers to the decision-making scheme for determining how to predict the demand for medical resources for the analysis target, including the basic prediction strategy, the screening prediction strategy, and the secondary screening prediction strategy; the available scheduling analysis target refers to the analysis target that has screening medical institutions within a preset distance range, that is, the analysis target can be scheduled across institutions through the screening institutions.

[0142] If multiple screening medical institutions exist near a certain analysis target, the cross-institutional scheduling of the analysis target is highly reliable, and a simpler prediction strategy can be adopted to reduce the processing difficulty.

[0143] By determining the forecasting strategy based on the reliability of scheduling and the changes in multimodal data, the information on scheduling reliability and data changes is transformed into the basis for selecting the forecasting strategy, thereby reducing the difficulty of forecasting while ensuring supply reliability.

[0144] S41, based on the updated results of the screened medical institutions in the medical institutions, determine the analysis targets that exist within the preset distance interval and use them as available scheduling analysis targets; The available scheduling analysis target refers to the analysis target that has at least one screening medical institution within a preset distance range, that is, there is a scheduling screening institution near the analysis target.

[0145] If there are two screening medical institutions, H2 and H4, within a 5-kilometer radius of a certain analysis target, then this analysis target is a usable scheduling analysis target.

[0146] By screening the updated results of the agencies, available scheduling analysis targets are determined, and those with cross-agency scheduling capabilities are identified from the analysis targets, providing a basis for the calculation of subsequent scheduling matching weight values.

[0147] S42, determine the scheduling matching weight value of the available scheduling analysis target based on the number of medical institutions within a preset distance range that are available for scheduling analysis; The scheduling matching weight value refers to a quantitative indicator determined based on the preset weight value corresponding to the number of medical institutions selected within a preset distance range of the available scheduling analysis target. The more institutions selected, the greater the weight value.

[0148] The scheduling matching weight value is calculated as follows: it is determined based on the preset weight value corresponding to the number of available scheduling analysis targets within a preset distance range. The more screening institutions there are, the greater the scheduling matching weight value.

[0149] Suppose that there are two screened medical institutions within a preset distance range for a certain available scheduling analysis target, and the corresponding preset weight value is 0.6, then the scheduling matching weight value is 0.6.

[0150] By determining the scheduling matching weight value by screening the number of institutions, the scheduling resources of the analysis target are quantified into comparable weight measures, providing a quantitative basis for determining the subsequent prediction strategy.

[0151] S43 determines the prediction strategy for the medical resources of the analysis target based on the scheduling matching weight values ​​of different available scheduling analysis targets.

[0152] Assuming that the scheduling matching weight values ​​of each available scheduling analysis target have been determined, the prediction strategy is determined based on the judgment results of Case 1, Case 2 and Case 3 respectively.

[0153] It is understood that the scheduling matching weight value of the available scheduling analysis target is determined based on the preset weight value corresponding to the number of medical institutions screened by the available scheduling analysis target within the preset distance interval. The more medical institutions screened by the available scheduling analysis target within the preset distance interval, the greater the scheduling matching weight value of the available scheduling analysis target.

[0154] The positive correlation between the scheduling matching weight value and the number of screening institutions was clarified, ensuring that the scheduling matching weight value can accurately reflect the cross-institutional scheduling capability of the analysis target.

[0155] It is understood that, based on the scheduling matching weight values ​​of different available scheduling analysis targets, the prediction strategy for the medical resources of the analysis target is determined, specifically including: It should be noted that the determination of the forecasting strategy includes the following three situations.

[0156] Case 1: If the proportion of the available scheduling analysis target in the medical institution is less than the preset available target proportion threshold, then the prediction strategy for the medical resources of the analysis target is determined as the basic prediction strategy.

[0157] The preset available target ratio threshold refers to the critical value for judging whether the proportion of available scheduling analysis targets is sufficient; the basic prediction strategy refers to the strategy of replacing dynamic data with available group data when the proportion of available date group dates reaches the threshold, otherwise using dynamic data for prediction.

[0158] If the proportion of available scheduling analysis targets is less than the preset threshold for the proportion of available targets, it indicates that the cross-agency scheduling capability is insufficient. In this case, a basic prediction strategy is adopted, and group data is used to replace dynamic data only when the proportion of available date groups is high enough. When the available scheduling analysis targets are insufficient, the basic prediction strategy is adopted, and when the cross-agency scheduling capability is insufficient, more conservative data replacement conditions are used to ensure the accuracy of the prediction.

[0159] Specifically, the basic prediction strategy includes: Based on the similarity between satellite remote sensing data and ground sensing data across different dates, dates meeting the similarity requirement are grouped into the same date group. The date group with the highest percentage of dates is selected as the available date group. If the percentage of dates in the available date group exceeds a preset threshold, the satellite remote sensing data and ground sensing data in the available date group replace the dynamically identified satellite remote sensing data and ground sensing data as input to the model for predicting the medical resource demand of the analysis target, thereby reducing the amount of data processing. Otherwise, the daily dynamically identified satellite remote sensing data and ground remote sensing data are still used as input to the model for predicting the medical resource demand of the analysis target.

[0160] The date group refers to the set of dates clustered according to the similarity between different dates of satellite remote sensing data and ground sensing data; the available date group refers to the date group with the largest proportion of dates; the preset date proportion threshold refers to the critical value for judging whether the proportion of available group dates is sufficient.

[0161] If, under the basic forecasting strategy, the proportion of available date groups reaches or exceeds the preset date quantity proportion threshold, then available date group data is used instead of dynamic data to reduce processing volume; otherwise, daily dynamic data is still used.

[0162] The execution method of the basic forecasting strategy was clarified. When the cross-institutional scheduling capacity is insufficient, the amount of data processing is reduced by replacing date group data. At the same time, when the proportion of group data is insufficient, dynamic data is maintained to ensure the accuracy of forecasting.

[0163] Case 2: If the proportion of the available scheduling analysis target in the medical institution is not less than the preset available target proportion threshold, determine whether the average value of the scheduling matching weight value of different available analysis targets is above the preset matching weight threshold. If yes, determine that the prediction strategy for the medical resources of the analysis target is the screening prediction strategy. If no, proceed to the judgment of Case 3.

[0164] The preset matching weight threshold refers to the critical value for judging whether the average value of the scheduling matching weight is high; the filtering prediction strategy refers to the strategy of directly using available date group data instead of dynamic data for prediction, without needing to judge the group proportion condition.

[0165] Assuming that the proportion of available scheduling analysis targets is sufficient and the average scheduling matching weight value is high, indicating that the cross-organizational scheduling capability is sufficient and the scheduling reliability is high, a screening and prediction strategy is adopted to directly replace the dynamic data with the available group data to minimize the processing volume.

[0166] When scheduling capacity is sufficient and the average weight is high, a screening prediction strategy is adopted. When the reliability of cross-agency scheduling is high, direct data replacement is used to minimize the amount of data processing, thereby improving prediction efficiency.

[0167] Specifically, the screening and prediction strategy includes: Based on the similarity between satellite remote sensing data and ground sensing data across different dates, dates that meet the similarity requirements are grouped into the same date group. The date group with the largest proportion of dates is selected as the available date group. Satellite remote sensing data and ground sensing data from the available date group are used instead of dynamically identified satellite remote sensing data and ground remote sensing data as input to the model for predicting the medical resource demand of the analysis target, thereby reducing the amount of data processing.

[0168] This step clarifies how the screening and prediction strategy is executed. Its significance lies in unconditionally replacing dynamic data with available group data when scheduling reliability is high, thereby minimizing the amount of data processing.

[0169] Case 3: Based on the proportion of the available scheduling analysis targets in medical institutions and the scheduling matching weight values ​​of different available scheduling analysis targets, determine the scheduling processing matching value, and determine whether the scheduling processing matching value is greater than the preset matching value. If yes, then determine the prediction strategy for the medical resources of the analysis target as a screening prediction strategy; if no, then determine the prediction strategy for the medical resources of the analysis target as a secondary screening prediction strategy.

[0170] The scheduling processing matching value refers to a quantitative indicator calculated by combining the proportion of available scheduling analysis targets and the average of scheduling matching weight values; the matching preset value refers to a critical value for judging whether the scheduling processing matching value is too high.

[0171] The method for calculating the matching value in the scheduling process is as follows: Let P_available represent the proportion of available scheduling analysis targets in the medical institution, and let Wavg be the mean of the scheduling matching weight values ​​for different available scheduling analysis targets. Then, the scheduling processing matching value is the mean of the two.

[0172] If the calculated matching value is greater than the preset matching value, it indicates that the overall matching degree is high, and a filtering prediction strategy is adopted; if it is not greater than the preset matching value, a secondary filtering prediction strategy is adopted.

[0173] By further differentiating prediction strategies through scheduling and processing of matching values, when scheduling capacity is sufficient but the average weight is not high, the strategy is dynamically selected based on the comprehensive matching degree, thereby achieving a fine balance between prediction efficiency and prediction accuracy.

[0174] Specifically, the secondary screening and prediction strategy includes: Based on the similarity between satellite remote sensing data and ground sensing data on different dates, dates that meet the similarity requirements are grouped into the same date group. The date group with the largest number of dates is selected as the available date group. If the scheduling matching weight value of the scheduling analysis target is above the preset matching weight threshold, the satellite remote sensing data and ground sensing data in the available date group are used instead of the dynamically identified satellite remote sensing data and ground remote sensing data as input to the model to predict the medical resource demand of the analysis target, thereby reducing the amount of data processing. If the scheduling matching weight value of the scheduling analysis target is not above the preset matching weight threshold, and if the proportion of the number of dates in the available date group is above the preset date proportion threshold, the satellite remote sensing data and ground sensing data in the available date group will replace the dynamically identified satellite remote sensing data and ground remote sensing data as the input of the model to predict the medical resource demand of the analysis target, thereby reducing the amount of data processing. In other cases, the daily dynamically identified satellite remote sensing data and ground remote sensing data will still be used as the input of the model to predict the medical resource demand of the analysis target.

[0175] The secondary screening and prediction strategy refers to a strategy that uses different data replacement conditions based on whether the scheduling matching weight value meets the standard. That is, the analysis target with the weight value meets the standard directly uses group data, while the analysis target with the weight value does not meet the standard only uses group data when the group proportion is sufficient, otherwise dynamic data is still used.

[0176] If, under the secondary screening prediction strategy, the scheduling matching weight value of a certain analysis target reaches or exceeds the preset matching weight threshold, then the available group data is directly used to replace the dynamic data; if the weight value does not meet the threshold, then the group data is used only when the proportion of available groups is sufficient, otherwise the dynamic data is still used.

[0177] The execution method of the secondary screening prediction strategy was clarified. When the scheduling matching degree is moderate, the data replacement conditions are determined according to the scheduling weight value of each analysis target and the group proportion, so as to achieve a fine balance between prediction efficiency and prediction accuracy.

[0178] This embodiment realizes a complete process from determining the available scheduling analysis targets to calculating the scheduling matching weight values ​​and then determining the prediction strategy. Its core value is reflected in three aspects: First, by judging the proportion of available scheduling analysis targets, a basic prediction strategy is adopted to ensure prediction accuracy when cross-agency scheduling capacity is insufficient. Second, by judging the average of scheduling matching weight values, a screening prediction strategy is adopted to minimize the processing volume when scheduling capacity is sufficient and the weights are high. Third, by judging the hierarchical nature of scheduling processing matching values, data replacement conditions are dynamically selected based on the weight values ​​of each analysis target and the group proportion when scheduling capacity is moderate, thereby achieving an adaptive balance between prediction efficiency and prediction accuracy.

[0179] Continuing with the calculation results of S3, multimodal data variation analysis was performed on all six analysis targets H1, H3, H5, H7, H9, and H10. Medical institutions were selected as H2, H4, H6, and H8.

[0180] The preset distance range is set at 5 kilometers. The number of medical institutions within a 5-kilometer radius for each analysis objective is as follows: H1: H2 (3km), H4 (4km), a total of 2 locations; H3:H4 (4km), 1 location in total; H5: H6 (5km), 1 store in total; H7: None; H9: H8 (4km), 1 store in total; H10: H8 (5km), 1 location in total; Step S41: The available scheduling analysis targets (with filtering mechanisms within the preset distance interval) are H1, H3, H5, H9, and H10, a total of 5. H7 is not an available scheduling analysis target.

[0181] Step S42: Calculate the scheduling matching weight value for each available scheduling analysis target.

[0182] H1: 2 companies → 0.6; H3: 1 company → 0.4; H5: 1 store → 0.4; H9: 1 company → 0.4; H10: 1 company → 0.4; Step S43: The proportion of available scheduling analysis targets in medical institutions = 5 / 10 = 0.50.

[0183] Case 1 Judgment: Is the ratio 0.50 less than the preset available target ratio threshold of 0.50? If 0.50 is not less than 0.50, then it does not belong to Case 1.

[0184] Case 2 Judgment: If the ratio of 0.50 is not less than the preset available target ratio threshold of 0.50, determine whether the average scheduling matching weight value is above the preset matching weight threshold of 0.50. Average scheduling matching weight value = (0.6 + 0.4 + 0.4 + 0.4 + 0.4) / 5 = 2.2 / 5 = 0.44.

[0185] Determine if the mean of 0.44 is above the preset matching weight threshold of 0.50: 0.44 is not greater than 0.50, no, proceed to case 3.

[0186] Case 3 Judgment: Calculate the scheduling processing matching value. Scheduling processing matching value = available scheduling analysis target proportion × average scheduling matching weight value = 0.47. Determine whether the scheduling processing matching value of 0.47 is greater than the preset matching value of 0.60. If not, determine that the prediction strategy for the medical resources of the analysis target is a secondary screening prediction strategy.

[0187] That is, under the secondary screening prediction strategy: For analysis targets with scheduling matching weight values ​​above the preset matching weight threshold of 0.50 (H1 is 0.6>0.50), the available date group data is directly used instead of dynamic data as the model input for prediction processing.

[0188] For analysis targets whose scheduling matching weight values ​​are not above 0.50 (H3, H5, H9, and H10 are all 0.4 < 0.50), group data will be used instead of dynamic data only if the proportion of the number of dates in the available date group is above the preset date proportion threshold of 0.50; otherwise, daily dynamic data will still be used.

[0189] For H7 (non-available scheduling analysis target), no data replacement is performed; the daily dynamic data is still used.

[0190] This embodiment implements an intelligent medical resource scheduling method based on multimodal data. S1, through hierarchical analysis of the mean of predicted deviation weight values ​​and the proportion of institutions with deviations, determines whether a simplified prediction model can be performed, achieving an adaptive balance between supply reliability and processing efficiency. S2, through hierarchical analysis of the proportion of institutions with deviations and the degree of overlap in deviation dates, determines a strategy for identifying medical institutions, achieving an adaptive balance between screening coverage and screening accuracy. S3, through hierarchical analysis of the proportion of institutions with deviations and interval distances, determines an identification analysis strategy, achieving an adaptive balance between analysis coverage and resource efficiency. S4, through hierarchical analysis of scheduling matching weight values ​​and scheduling processing matching values, determines a prediction strategy, achieving an adaptive balance between prediction efficiency and prediction accuracy. Overall, this method constructs a complete technical closed loop from prediction deviation analysis to prediction strategy determination through the coordinated linkage of prediction deviation analysis, simplified processing judgment, screening and identification of medical institutions, hierarchical determination of identification and analysis strategies, and dynamic management of prediction strategies. Its core value is reflected in four aspects: First, the objectivity of simplified processing judgment is ensured through quantitative analysis of prediction deviation weight values ​​and interval distances; second, the adaptive selection of medical institution screening identification strategies is achieved through hierarchical judgment of the proportion of deviating institutions and the degree of overlap of deviating dates; third, the adaptive determination of identification and analysis strategies is achieved through hierarchical analysis of the distribution and interval distance of screened institutions; and fourth, the dynamic determination of prediction strategies is achieved through feedback analysis of scheduling matching weight values ​​and scheduling processing matching values, thereby ensuring the reliability of medical resource supply while rationally controlling the consumption of prediction processing resources.

[0191] Example 3 Thirdly, this application provides a chip device, which includes: a data acquisition module, an analysis module, and a control module, wherein the data acquisition module is responsible for acquiring data, the analysis module is responsible for the above-mentioned intelligent medical resource scheduling method based on multimodal data, and the control module is responsible for outputting control commands based on the analysis results.

[0192] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0193] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0194] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. A medical resource intelligent scheduling system based on multimodal data, characterized in that, Specifically, it includes: The database construction module is responsible for collecting multimodal data related to medical resources, performing semantic alignment, redundancy removal, and correlation fusion on the multimodal data to form a unified dynamic database of supply and demand for multimodal medical resources. The prediction management module determines the prediction strategy for medical resources of the analysis target based on the changes in the multimodal data of the analysis target and the updated results of the selected medical institutions. Based on the prediction strategies for medical resources of different medical institutions and combined with the multimodal fusion data, a dynamic supply and demand prediction model using a deep learning framework is adopted to output the prediction results of medical resource demand. The scheduling and execution module is responsible for generating a scheduling scheme for the scenario type by taking the prediction results of the dynamic supply and demand prediction model library as input, distributing the scheduling scheme to the corresponding execution entity, and tracking the scheduling execution effect.

2. The intelligent medical resource scheduling system based on multimodal data as described in claim 1, characterized in that, The multimodal data includes satellite remote sensing data, ground sensing data, and medical data.

3. The intelligent medical resource scheduling system based on multimodal data as described in claim 2, characterized in that, The satellite remote sensing data and ground remote sensing data are remote sensing data of traffic flow on the roads between the medical resource allocation center and medical institutions.

4. A method for intelligent scheduling of medical resources based on multimodal data, applied to the intelligent scheduling system for medical resources based on multimodal data as described in any one of claims 1-3, characterized in that, Specifically, it includes: The model's prediction results determine the degree of matching between the predicted medical resource demand in different medical institutions and the user's demand. Using the degree of matching and the distance between medical institutions whose matching degree does not meet the requirements and other medical institutions, when it is determined that the prediction model can be simplified, proceed to the next step. Based on the matching degree, the matching degree of the demand matching deviation institutions in the medical institutions is determined, and combined with the degree of overlap of the predicted deviation dates of the demand matching deviation institutions, the identification strategy of the medical institutions that perform medical resource scheduling after correcting the prediction results according to a preset ratio is determined, and this strategy is used as the identification strategy for screening medical institutions. Based on the identification strategy, screened medical institutions are identified among the medical institutions. Medical institutions other than the screened medical institutions are taken as analysis targets. Based on the distribution data of the screened medical institutions and the interval distance between the screened medical institutions and different analysis targets, the identification and analysis strategy of the multimodal data of the analysis targets is determined. The identification and analysis strategy is used to determine the changes in the multimodal data of the analysis target. Based on the changes, and according to the updated results of the screened medical institutions in the medical institutions, a prediction strategy for the medical resources of the analysis target is determined.

5. The intelligent medical resource scheduling method based on multimodal data as described in claim 4, characterized in that, The degree of matching between the predicted medical resource demand in the medical institution and the user demand is determined based on the deviation rate between the predicted short-term medical resource demand and the user demand.

6. The intelligent medical resource scheduling method based on multimodal data as described in claim 4, characterized in that, The simplification of the prediction model can be determined, specifically including: Based on the degree of matching between the predicted medical resource demand in the medical institution and the user demand, the date on which the predicted medical resource demand is less than the user demand and the deviation rate from the user demand is greater than a preset deviation rate threshold is determined and is taken as the prediction deviation date. The prediction deviation weight value of the medical institution is determined based on the proportion of prediction deviation dates in the medical institution. By using the prediction bias weights of different medical institutions and the distance between medical institutions that do not meet the matching requirements and other medical institutions, it can be determined whether the prediction model can be simplified.

7. The intelligent medical resource scheduling method based on multimodal data as described in claim 6, characterized in that, Based on the prediction bias weights across different medical institutions and the distance between medical institutions that do not meet the matching requirements and other medical institutions, it is determined whether a simplification of the prediction model can be performed. Specifically, this includes: Based on the prediction bias weight values ​​of different medical institutions, if the average prediction bias weight values ​​of different medical institutions are greater than the preset bias weight threshold, it is determined that the prediction model cannot be simplified.

8. The intelligent medical resource scheduling method based on multimodal data as described in claim 4, characterized in that, The method for determining the identification strategy for screening medical institutions is as follows: Based on the demand-matching deviation institutions in the medical institutions, determine the proportion of demand-matching deviation institutions in the medical institutions, and use the proportion of demand-matching deviation institutions in the medical institutions as the proportion of deviation institutions. Based on the degree of overlap between the predicted deviation dates of the demand matching deviation agencies, the demand matching deviation agencies that belong to the predicted deviation dates on the stated dates are identified, and the proportion of demand matching deviation agencies that belong to the predicted deviation dates on the stated dates is taken as the demand matching deviation proportion for the stated dates. The identification strategy for screening medical institutions is determined based on the proportion of institutions with matching deviations and the proportion of demand matching deviations on different dates.

9. The intelligent medical resource scheduling method based on multimodal data as described in claim 4, characterized in that, The method for determining the prediction strategy for the medical resources of the analysis target is as follows: Based on the updated results of the screened medical institutions in the medical institutions, the analysis targets that exist within the preset distance range are determined and used as available scheduling analysis targets; The scheduling matching weight value of the available scheduling analysis target is determined by the number of medical institutions within a preset distance range that are available for scheduling analysis; Based on the scheduling matching weight values ​​of different available scheduling analysis targets, a prediction strategy for the medical resources of the analysis target is determined.

10. A chip device, the chip device comprising: The system comprises a data acquisition module, an analysis module, and a control module, wherein the data acquisition module is responsible for acquiring data, and the analysis module is responsible for the intelligent medical resource scheduling method based on multimodal data as described in any one of claims 4-9, and the control module is responsible for outputting control commands based on the analysis results.

Images