First aid resource scheduling method and system based on multi-dimensional positioning and intelligent correction

Abstract

The application provides an emergency resource scheduling method and system based on multi-dimensional positioning and intelligent correction, relates to the technical field of resource scheduling, and comprises the following steps: obtaining initial positioning by fusing multi-source position data, correcting the deviation in combination with the event type and historical distribution characteristics when the deviation is detected, and obtaining the corrected patient position. Meanwhile, the emergency degree is dynamically evaluated based on the time sequence characteristics of the patient state information. Finally, a multi-objective optimization model is constructed to generate a scheduling scheme according to the corrected position, the real-time resource state and the adjusted emergency degree. The application improves the positioning accuracy and the real-time performance of state evaluation, and realizes efficient and optimized scheduling of emergency resources.

Description

Technical Field

[0001] This invention relates to resource scheduling technology, and more particularly to an emergency medical resource scheduling method and system based on multi-dimensional positioning and intelligent deviation correction. Background Technology

[0002] In the field of emergency medical services, quickly and accurately locating patients and efficiently dispatching emergency resources are crucial for improving the success rate of treatment. Current technologies rely on location information from single or limited sources, making their accuracy susceptible to various factors. For example, in complex urban environments, indoor settings, or areas with poor signal, base station positioning errors can be significant; and the address verbally described by the caller may be vague, incorrect, or impossible to pinpoint. Inaccurate patient location directly leads to ambulance arrival delays, difficulty in locating patients, and a significant waste of valuable emergency response time. On the other hand, assessments of patient urgency are often based on instantaneous information at the time of the call, employing static and fixed classification standards, lacking consideration of the dynamic evolution of the patient's condition. Different diseases have different development trajectories, and the severity of the same symptom can change significantly over different time intervals. Static assessment models cannot capture this temporal evolution, potentially leading to a mismatch between resource dispatch priorities and actual medical needs, such as misjudging a rapidly deteriorating patient as having low urgency, thus delaying treatment. These shortcomings collectively constrain the overall response efficiency and treatment effectiveness of the emergency medical system. Summary of the Invention

[0003] The embodiments of the present invention provide an emergency medical resource scheduling method and system based on multi-dimensional positioning and intelligent deviation correction, which can solve the problems in the prior art.

[0004] A first aspect of this invention provides an emergency medical resource scheduling method based on multi-dimensional positioning and intelligent deviation correction, comprising:

[0005] In response to emergency call requests, collect multi-source location data from the calling terminal and obtain patient status information;

[0006] The multi-source location data is fused to obtain an initial positioning result;

[0007] The initial positioning result is matched and verified. When the positioning deviation is detected to exceed the dynamic deviation threshold, the deviation of the initial positioning result is estimated based on the event type characteristics in the patient status information and the spatial distribution characteristics of similar historical events. The deviation is then cross-verified in combination with the reference location information to generate the corrected patient location.

[0008] Based on the characteristic parameters and temporal change characteristics in the patient status information, combined with the temporal evolution pattern of historical similar characteristic data and the time interval between the current call time and the initial time, the urgency of the patient status information is dynamically evaluated and adjusted through a temporal prediction network.

[0009] Based on the corrected patient location, the real-time status of dispatchable ambulances and target medical institutions is obtained. According to the adjusted urgency level, the weight coefficients of multiple dispatch objectives are determined, a multi-objective optimization function is constructed and solved, and a dispatch scheme is generated.

[0010] Based on the event type characteristics and spatial distribution characteristics of similar historical events in the patient status information, the initial positioning result is deviated and cross-validated using reference location information to generate a corrected patient location, including:

[0011] The occurrence location data of historical similar events associated with the event type feature are retrieved from a pre-built associated knowledge base. The occurrence location data includes a set of offset vectors of the actual occurrence location of the historical event relative to the initial location and the corresponding environment type feature.

[0012] Based on the environmental type characteristics of the location corresponding to the initial positioning result, a subset of offset vectors matching the environmental type characteristics is selected from the offset vector set, and the dominant offset direction and average offset magnitude of the offset vector subset are calculated.

[0013] Using the initial positioning result as a reference point, the spatial position is offset along the dominant offset direction according to the average offset amplitude to generate the estimated position;

[0014] Obtain reference location information within a spatiotemporal range, calculate the spatial distance distribution between the estimated location and the reference location information; compare the spatial distance distribution with preset cross-validation conditions, and verify and correct the estimated location based on the comparison results to generate the corrected patient location.

[0015] Based on the characteristic parameters and temporal change features in the patient status information, combined with the temporal evolution patterns of historical similar characteristic data and the time interval between the current call time and the initial time, a temporal prediction network dynamically assesses and adjusts the urgency of the patient status information, including:

[0016] Feature parameters and temporal change features are extracted from the patient status information. The feature parameters include symptom intensity indicators and physiological status indicators. The temporal change features include the rate of change of each feature parameter during the call process.

[0017] Retrieve historical similar feature data that matches the feature parameters from the historical database, and extract the temporal evolution pattern of the urgency over time from the historical similar feature data. The temporal evolution pattern includes the correlation mapping relationship between the feature parameter change trajectory and the urgency change trajectory.

[0018] Calculate the time interval between the current call time and the initial time, and use the time interval as a time decay factor;

[0019] The feature parameters, temporal change features, temporal evolution patterns, and time decay factors are input into a temporal prediction network for encoding. The temporal dependencies of the feature parameters are captured through the hidden layer units of the temporal prediction network. The temporal prediction network outputs an adjusted urgency assessment value, which is then updated in the patient status information.

[0020] The training process of the time-series prediction network includes:

[0021] A training sample set is constructed, which includes the feature parameter sequence, temporal evolution pattern and actual urgency label of historical emergency cases. Statistical analysis is performed on samples of different urgency categories in the training sample set to identify sparse urgency categories with a sample number lower than a preset threshold.

[0022] For the samples of the sparse urgency category, synthetic samples are generated based on the feature parameter distribution features of samples of the same category, and the synthetic samples are added to the training sample set to form an expanded training sample set;

[0023] A time decay attention mechanism is set in the encoding layer of the time series prediction network. The time decay attention mechanism calculates the attention weight based on the time interval between the feature parameter acquisition time and the current prediction time. The feature parameter with a smaller time interval is assigned a larger attention weight.

[0024] A loss function is constructed, which sets a corresponding penalty coefficient based on the urgency label value of the sample. A time sensitivity penalty term is introduced into the loss function, which calculates an additional penalty weight based on the product of the time interval and the rate of increase of urgency in the sample.

[0025] The time-series prediction network is trained and pruned using the expanded training sample set and the loss function.

[0026] Based on the corrected patient location, the real-time status of dispatchable ambulances and target medical institutions is obtained. Weight coefficients for multiple dispatch objectives are determined according to the adjusted urgency level. A multi-objective optimization function is constructed and solved to generate a dispatch scheme, including:

[0027] Based on the real-time status of dispatchable ambulances and the real-time resource status of the target medical institution, a set of candidate dispatchable objects is constructed.

[0028] The system obtains the ambulance load distribution status and medical institution reception capacity within the current area. The ambulance load distribution status reflects the task saturation level of each ambulance, and the medical institution reception capacity reflects the remaining reception capacity of each medical institution.

[0029] The adjusted urgency level is input into a preset weight adjustment strategy space to select a target weight adjustment strategy. The weight adjustment strategy space contains multiple sets of candidate weight adjustment strategies for different urgency levels.

[0030] The ambulance load distribution status and the medical institution's reception capacity are input into the target weight adjustment strategy to dynamically calculate the weight coefficients of the response time target, medical capacity matching target, and resource utilization target.

[0031] A multi-objective optimization function is constructed, which combines the response time objective, the medical capacity matching objective, and the resource utilization objective according to their respective weight coefficients.

[0032] The multi-objective optimization function is solved in the set of candidate scheduling objects to obtain the ambulance scheduling scheme and medical institution recommendation scheme corresponding to the optimal solution.

[0033] The adjusted urgency level is input into a preset weight adjustment strategy space to select a target weight adjustment strategy, including:

[0034] Event type features and time window features are extracted from the patient status information and combined with the adjusted urgency level to form a query feature vector; historical cases similar to the query feature vector are retrieved from the historical scheduling database, the historical cases containing historical usage strategy identifiers and corresponding scheduling effect evaluation indicators, the scheduling effect evaluation indicators including response time target completion rate and resource utilization efficiency; the retrieved historical cases are sorted according to the scheduling effect evaluation indicators;

[0035] Statistically analyze the frequency of occurrence of each historical strategy usage identifier in historical cases that exceed a preset scoring threshold, and construct a strategy effectiveness evaluation matrix.

[0036] Extract candidate weight adjustment strategies that match the query feature vector from the weight adjustment strategy space, calculate the effect score of each candidate weight adjustment strategy in the strategy effect evaluation matrix, and select the candidate weight adjustment strategy with the highest effect score as the target weight adjustment strategy.

[0037] The ambulance load distribution and the medical institution's capacity are input into the target weight adjustment strategy, and the weight coefficients are dynamically calculated, including:

[0038] The task saturation of each ambulance in the ambulance load distribution state is quantitatively calculated to obtain the regional ambulance load index; the remaining reception capacity of each medical institution in the medical institution reception capacity is quantitatively calculated to obtain the regional medical institution capacity index.

[0039] The regional ambulance load index is compared with a preset load threshold, and the current load status type is determined based on the comparison result.

[0040] The baseline weight coefficient corresponding to the current load state type is extracted from the target weight adjustment strategy. The baseline weight coefficient includes the weight allocation of response time target, medical capacity matching target and resource utilization target.

[0041] The weight adjustment factor is calculated based on the regional ambulance load index and the regional medical institution capacity index. The weight adjustment factor is then used to correct the baseline weight coefficient, resulting in a dynamically adjusted weight coefficient.

[0042] A second aspect of the present invention provides an emergency medical resource dispatching system based on multi-dimensional positioning and intelligent deviation correction, comprising:

[0043] The call response unit is used to respond to emergency call requests, collect multi-source location data from the call terminal, and obtain patient status information.

[0044] A data fusion unit is used to fuse the multi-source location data to obtain an initial positioning result;

[0045] The positioning verification unit is used to match and verify the initial positioning result. When the positioning deviation is detected to exceed the dynamic deviation threshold, the initial positioning result is deduced based on the event type characteristics and spatial distribution characteristics of similar historical events in the patient status information, and cross-verification is performed in combination with the reference position information to generate the corrected patient position.

[0046] The status assessment unit is used to dynamically assess and adjust the urgency of the patient's status information based on the characteristic parameters and temporal change characteristics in the patient's status information, combined with the temporal evolution pattern of historical similar characteristic data and the time interval between the current call time and the initial time, through a temporal prediction network.

[0047] The scheduling optimization unit is used to obtain the real-time status of dispatchable ambulances and target medical institutions based on the corrected patient location, determine the weight coefficients of multiple scheduling objectives according to the adjusted urgency level, construct and solve a multi-objective optimization function, and generate a scheduling scheme.

[0048] A third aspect of the present invention provides an electronic device, comprising:

[0049] processor;

[0050] Memory used to store processor-executable instructions;

[0051] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0052] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0053] This method effectively overcomes the limitations of single signal sources in complex urban environments by fusing multi-source location data, improving the reliability and stability of initial positioning. By analyzing the temporal changes of characteristic parameters, combining the evolution patterns of historical similar data with the current time interval, and using a temporal prediction network for dynamic prediction, the assessment of urgency is no longer a one-time static judgment, but a dynamic process that can be updated in real time according to changes in the patient's condition. This provides a more accurate and realistic priority basis for subsequent scheduling decisions.

[0054] Based on the corrected and accurate location and dynamically adjusted urgency level, a more realistic real-time status of ambulances and medical institutions can be obtained. The scheduling objective weights are determined according to the urgency level, and a multi-objective optimization function is constructed and solved. The generated scheduling scheme can comprehensively consider various factors such as emergency response time, resource utilization efficiency, and patient transport routes. This achieves intelligent optimization of the entire process from patient location and status assessment to resource matching and route planning. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the emergency resource scheduling method based on multi-dimensional positioning and intelligent deviation correction according to an embodiment of the present invention.

[0056] Figure 2 Generate a flowchart for multi-objective optimization scheduling schemes. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0059] Figure 1 This is a flowchart illustrating the emergency resource scheduling method based on multi-dimensional positioning and intelligent deviation correction according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0060] In response to emergency call requests, collect multi-source location data from the calling terminal and obtain patient status information;

[0061] The multi-source location data is fused to obtain an initial positioning result;

[0062] The initial positioning result is matched and verified. When the positioning deviation is detected to exceed the dynamic deviation threshold, the deviation of the initial positioning result is estimated based on the event type characteristics in the patient status information and the spatial distribution characteristics of similar historical events. The deviation is then cross-verified in combination with the reference location information to generate the corrected patient location.

[0063] Based on the characteristic parameters and temporal change characteristics in the patient status information, combined with the temporal evolution pattern of historical similar characteristic data and the time interval between the current call time and the initial time, the urgency of the patient status information is dynamically evaluated and adjusted through a temporal prediction network.

[0064] Based on the corrected patient location, the real-time status of dispatchable ambulances and target medical institutions is obtained. According to the adjusted urgency level, the weight coefficients of multiple dispatch objectives are determined, a multi-objective optimization function is constructed and solved, and a dispatch scheme is generated.

[0065] Based on the event type characteristics and spatial distribution characteristics of similar historical events in the patient status information, the initial positioning result is deviated and cross-validated using reference location information to generate a corrected patient location, including:

[0066] The occurrence location data of historical similar events associated with the event type feature are retrieved from a pre-built associated knowledge base. The occurrence location data includes a set of offset vectors of the actual occurrence location of the historical event relative to the initial location and the corresponding environment type feature.

[0067] Based on the environmental type characteristics of the location corresponding to the initial positioning result, a subset of offset vectors matching the environmental type characteristics is selected from the offset vector set, and the dominant offset direction and average offset magnitude of the offset vector subset are calculated.

[0068] Using the initial positioning result as a reference point, the spatial position is offset along the dominant offset direction according to the average offset amplitude to generate the estimated position;

[0069] Obtain reference location information within a spatiotemporal range, and calculate the spatial distance distribution between the estimated location and the reference location information;

[0070] The spatial distance distribution is compared with preset cross-validation conditions, and the estimated position is verified and corrected based on the comparison results to generate the corrected patient position.

[0071] For example, when an emergency dispatch center receives an emergency call request, it automatically initiates a multi-source location data acquisition process. The multi-source location data includes GPS positioning data from the calling terminal, base station positioning data, and IP address positioning data. GPS positioning data is provided by the calling terminal's satellite positioning module and includes latitude and longitude coordinates and positioning accuracy parameters; base station positioning data is obtained through the mobile communication network, calculating location coordinates based on the signal strength between the terminal and the base station and a multi-base station triangulation algorithm; IP address positioning data is obtained by querying the corresponding geographical location information from the IP address database when the call is initiated via the network. Simultaneously, patient status information is collected through voice recognition or text input, including subjective descriptions such as symptom descriptions, pain levels, and level of consciousness, as well as physiological indicators such as heart rate and blood pressure collected through the smart terminal.

[0072] When fusing the collected multi-source location data, the positioning accuracy of each data source is first evaluated. GPS positioning accuracy is affected by signal strength and obstruction; in open outdoor environments, accuracy can reach within 10 meters, but it drops significantly indoors or in densely populated areas with tall buildings. Base station positioning accuracy is related to base station density; in urban areas, accuracy is typically between 50 and 200 meters. IP address positioning accuracy is lower, usually only able to locate at the block or area level. Based on the positioning accuracy parameters of each data source, a weighted average fusion algorithm is used, assigning greater weight to data sources with higher positioning accuracy, and calculating the weighted average location coordinates as the initial positioning result.

[0073] When verifying the initial positioning results, the initial positioning coordinates are compared with reachable locations in the geographic information database. Reachable location information such as roads, building entrances, and public places within a 100-meter radius of the initial positioning point is extracted, and the distance between the initial positioning point and the nearest reachable location is calculated. If this distance exceeds a dynamic deviation threshold, a positioning deviation is determined. The dynamic deviation threshold is dynamically adjusted based on the positioning data source type and environmental characteristics; it is set to 50 meters in outdoor environments with good GPS signals and 150 meters in indoor or signal-constrained environments to ensure the accuracy of deviation detection.

[0074] After obtaining the initial location results, the historical dataset corresponding to the event type features is retrieved from the associated knowledge base. The associated knowledge base stores spatial data of historical emergency events categorized by event type. Each historical record includes the initial location coordinates at the time of the call, the actual patient location coordinates confirmed after the ambulance's arrival, the offset vector between the two, and the environmental type label of the event location. Environmental type features cover spatial attributes such as building type, regional functional attributes, and road network density. For cardiac arrest events, a set of historical offset vectors for this type of event is extracted, with each vector represented as a two-dimensional coordinate offset in the form of (Δx, Δy).

[0075] Based on the initial positioning result, the environmental type characteristics of the location are identified. Attribute information such as building density, whether it is located within a commercial complex, and floor structure of the area where the positioning point is located is obtained through a geographic information service interface. These attributes are then compared with the environmental type characteristics in the offset vector set to calculate the matching degree. Specifically, for building types, hierarchical coding matching is used, classifying building types into three levels: "major category - intermediate category - minor category." The building type codes of the current location and historical locations are compared level by level, starting from the highest level: 1 point for a complete match, 0.6 points for an intermediate category match, 0.3 points for a major category match only, and 0 points for no match. For regional functional attributes, attribute vector cosine similarity calculation is used. Functional attributes such as residential, commercial, office, medical, and transportation are quantified into multi-dimensional vectors, and the cosine similarity between the functional attribute vectors of the current location and historical locations is calculated as the score for that dimension. For road network density, numerical difference normalization scoring is used, calculating the absolute difference between the road density values ​​of the current location and historical locations, and applying Formula 1. |Density Difference| / Maximum Density is converted into a 0-1 range score. The three-dimensional scores are weighted and summed (0.4, 0.3, 0.3) to obtain the overall matching degree. Offset vectors with environmental type similarity exceeding a set threshold (e.g., 0.7) are selected to form a subset. Statistical analysis is performed on the selected subset of offset vectors, decomposing all vectors into two dimensions: orientation angle and magnitude. Kernel density estimation is used to identify clusters of offset directions, and the direction corresponding to the density peak is determined as the dominant offset direction. The mean magnitude of all offset vectors in the subset is calculated as the average offset amplitude, and the standard deviation is recorded for subsequent verification and judgment.

[0076] Based on the dominant offset direction and average offset amplitude, the spatial location of the initial positioning result is calculated. Starting from the initial positioning coordinates, the distance corresponding to the average offset amplitude is extended along the dominant offset direction to calculate the new coordinates of the calculated position. When the average offset amplitude is 120 meters and the dominant offset direction is 35 degrees east of north, the specific changes in latitude and longitude of the calculated position relative to the initial position are calculated through coordinate transformation.

[0077] Obtain reference location information within the spatiotemporal range surrounding the estimated location. This information includes nearby landmarks, public facility entrances, road intersections, and other spatially defined reference points. Set a search area with a radius of 200 meters centered on the estimated location, and extract the coordinates of all reference points within this area from the geographic information database. Calculate the Euclidean distance between the estimated location and each reference point, forming a spatial distance distribution sequence.

[0078] The cross-validation criteria are set as follows: the estimated location should be within the area formed by connecting at least two different types of reference points, and the distance to the nearest reference point should not exceed the 75th percentile of the distance distribution sequence. The calculated spatial distance distribution is compared with these criteria for verification. When the estimated location meets the verification criteria, it is directly output as the corrected patient location. When the verification fails, a weighted interpolation is performed between the estimated location and the nearest reference point. The interpolation weights are determined based on the standard deviation of historical offset data; if the standard deviation is large, the weight of the reference point is increased to generate the final corrected location coordinates.

[0079] This method effectively reduces the impact of positioning deviation on dispatching decisions and improves the accuracy of ambulances reaching the target location through a dual mechanism of environment type matching and reference location cross-validation.

[0080] Based on the characteristic parameters and temporal change features in the patient status information, combined with the temporal evolution patterns of historical similar characteristic data and the time interval between the current call time and the initial time, a temporal prediction network dynamically assesses and adjusts the urgency of the patient status information, including:

[0081] Feature parameters and temporal change features are extracted from the patient status information. The feature parameters include symptom intensity indicators and physiological status indicators. The temporal change features include the rate of change of each feature parameter during the call process.

[0082] Retrieve historical similar feature data that matches the feature parameters from the historical database, and extract the temporal evolution pattern of the urgency over time from the historical similar feature data. The temporal evolution pattern includes the correlation mapping relationship between the feature parameter change trajectory and the urgency change trajectory.

[0083] Calculate the time interval between the current call time and the initial time, and use the time interval as a time decay factor;

[0084] The feature parameters, temporal variation features, temporal evolution patterns, and time decay factors are input into a temporal prediction network for encoding. The temporal dependencies of the feature parameters are captured through the hidden layer units of the temporal prediction network.

[0085] The time-series prediction network outputs an adjusted urgency assessment value, which is then updated into the patient status information.

[0086] For example, feature parameters are extracted from patient status information, encompassing symptom intensity indicators and physiological status indicators. Symptom intensity indicators are quantified through the caller's voice description or input text information; for instance, pain level can be coded using a rating scale of 0 to 10, and consciousness status can be divided into discrete state values ​​such as awake, drowsy, and comatose. Physiological status indicators include measurable parameters such as heart rate, blood pressure, and respiratory rate. When the call terminal has physiological monitoring capabilities, the values ​​are directly collected; otherwise, they are converted into estimated values ​​based on the caller's description.

[0087] Simultaneously, temporal variation features are extracted, reflecting the dynamic trends of each feature parameter during the call duration. The rate of change is obtained by calculating the difference between feature parameters at adjacent time points and dividing by the time interval. For example, if the patient's heart rate is 110 beats per minute and 125 beats per minute in two separate data collections, with a time interval of 2 minutes, then the rate of change in heart rate is an increase of 7.5 beats per minute. The rate of change in pain intensity, from a score of 6 to a score of 9, is an increase of 1.5 points per minute.

[0088] Historical case records with similar feature parameters to the current case are retrieved from a historical database. This database stores historical emergency cases categorized by event type and symptom category. Each record contains a complete time series of feature parameters, the corresponding urgency level at each moment, the final diagnosis, and the timeliness of treatment. Euclidean distance is used to calculate the matching degree between feature vectors. Specifically, the current patient's symptom intensity index and physiological state index are used to construct an n-dimensional feature vector. The Euclidean distance between this vector and the feature vectors of historical cases is calculated, and historical cases with a distance less than a set threshold (e.g., 3.5 based on the feature space dimension) are selected. The evolutionary pattern of urgency over time in these historical cases is extracted to construct a temporal evolution model. This model establishes a regression mapping relationship by statistically analyzing the correlation between the rate of change of feature parameters and the rate of change of urgency in historical cases. This model represents the magnitude of the increase in urgency and the time window when the symptom intensity index increases at a specific rate. For example, historical data shows that when the pain score of patients with chest pain increases by 1 point per minute, the emergency level rises from moderate to high within an average of 30 minutes; when the respiratory rate of patients with dyspnea increases by 3 breaths per minute, the emergency level rises from mild to moderate within 15 minutes.

[0089] Calculate the time interval between the current call time and the initial call access time, and convert this time interval into a time decay factor. The time decay factor is implemented using an exponential decay function, and the calculation formula is as follows: Where t is the time interval, β is the attenuation intensity coefficient ranging from 0.5 to 2.0, and λ is the attenuation rate parameter ranging from 0.01 to 0.1, reflecting the increasing weight of the impact on the urgency assessment as the waiting time lengthens. Specifically, a baseline time of 10 minutes is set. When the actual interval exceeds the baseline time, the attenuation factor value increases significantly, prompting the urgency assessment value to be adjusted to a higher level.

[0090] The extracted feature parameters, temporal variation features, temporal evolution patterns, and time decay factors are combined into an input vector, which is then fed into the temporal prediction network. This network employs a long short-term memory (LSTM) network structure, comprising an input layer, two LSTM hidden layers, and an output layer.

[0091] The input layer receives a multi-dimensional feature vector containing feature parameters, temporal variation features, temporal evolution patterns, and time decay factors, and performs normalization processing, mapping the values ​​of each dimension to the interval between 0 and 1. The feature vector is organized according to a time step sequence, with each time step containing the symptom intensity index, physiological state index, and their rate of change at the current moment, while incorporating the temporal evolution patterns of similar historical cases as auxiliary feature dimensions.

[0092] The LSTM hidden layers employ a two-layer stacked structure, with each layer containing 256 hidden units. The first LSTM layer receives the temporal feature sequence from the input layer and selectively retains key state information from historical moments through gating mechanisms such as forget gates, input gates, and output gates, capturing short-term trends in feature parameters. The second LSTM layer receives the output sequence from the first layer as input, further extracting long-term dependencies across time steps and identifying deep patterns in the evolution of urgency. A dropout layer with a dropout rate of 0.3 is placed between the two LSTM layers to prevent overfitting.

[0093] After the output of the second LSTM layer at the last time step, an attention mechanism layer is connected. This layer weights the hidden states at each time step according to a time decay factor, assigning higher attention weights to hidden states whose time intervals are closer to the current time. The weighted comprehensive feature representation is then calculated. The output layer adopts a fully connected structure, containing a 128-dimensional fully connected hidden layer and an output layer. The fully connected hidden layer uses the ReLU activation function, and the output layer uses a softmax function to map the feature representation to a probability distribution of three urgency levels: mild, moderate, and severe. The level with the highest probability is selected as the adjusted urgency assessment result and updated in the patient status information, replacing the initial assessment result. This updated assessment value is used for subsequent scheduling decisions, influencing the allocation of weight coefficients and the priority ranking of optimization objectives.

[0094] This method integrates temporal evolution patterns and time decay mechanisms to achieve dynamic tracking and prospective early warning of the urgency of patients' conditions, avoiding the risk of delayed treatment due to static assessment.

[0095] The training process of the time-series prediction network includes:

[0096] A training sample set is constructed, which includes the feature parameter sequence, temporal evolution pattern and actual urgency label of historical emergency cases. Statistical analysis is performed on samples of different urgency categories in the training sample set to identify sparse urgency categories with a sample number lower than a preset threshold.

[0097] For the samples of the sparse urgency category, synthetic samples are generated based on the feature parameter distribution features of samples of the same category, and the synthetic samples are added to the training sample set to form an expanded training sample set;

[0098] A time decay attention mechanism is set in the encoding layer of the time series prediction network. The time decay attention mechanism calculates the attention weight based on the time interval between the feature parameter acquisition time and the current prediction time. The feature parameter with a smaller time interval is assigned a larger attention weight.

[0099] A loss function is constructed, which sets a corresponding penalty coefficient based on the urgency label value of the sample. A time sensitivity penalty term is introduced into the loss function, which calculates an additional penalty weight based on the product of the time interval and the rate of increase of urgency in the sample.

[0100] The time-series prediction network is trained and pruned using the expanded training sample set and the loss function.

[0101] For example, during the training of the time-series prediction network, historical emergency case data is first extracted from the emergency medical services database. Each case includes a sequence of characteristic parameters such as the caller's described symptoms, physiological indicators, and symptom duration. Simultaneously, the evolution of the patient's condition from the initial call to the arrival of the ambulance is recorded, and the actual urgency level confirmed by medical personnel is labeled. When performing category distribution statistics on the training sample set, the proportion of samples corresponding to each urgency level is calculated. When the number of samples in a certain category is less than 5% of the total sample size or the absolute number is less than 100 cases, that category is identified as a sparse urgency category.

[0102] To expand the sparse class samples, a kernel density estimation-based method is used to model the distribution of these samples in the feature space. Specifically, for continuous features such as heart rate and blood pressure values, a Gaussian kernel function is used to calculate the probability density distribution, with the kernel bandwidth automatically determined using the Silverman rule. For discrete features such as symptom description keywords, class frequency statistics are used. While maintaining the statistical characteristics of the original samples, synthetic samples are generated by sampling from the estimated distribution. A random perturbation parameter is introduced during the generation process, with the perturbation amplitude set to 0.1 to 0.3 times the standard deviation of the original samples. This ensures that the feature parameter values ​​of the synthetic samples fluctuate within a reasonable range near the mean of the original samples, guaranteeing that the synthetic samples both conform to the statistical regularity of real data and possess a certain degree of diversity. The number of synthetic samples generated for each sparse class is such that the total number of samples in that class reaches 80% of the average number of samples in the training set, forming an expanded training sample set.

[0103] The time-decay attention mechanism employs an exponential decay function to calculate attention weights. For the feature at time step i in the feature parameter sequence, its attention weight is calculated as follows: , where Δt i This represents the time interval between the current prediction time and the current time step, where λ is the decay coefficient ranging from 0.05 to 0.15. In the encoding layer, the attention weights for all time steps are first normalized, and the normalized weights are then calculated. Where T is the total number of time steps, and j represents the index variable of the time step. The normalized attention weights are multiplied element-wise with the LSTM hidden state vector hih_ihi at the corresponding time step to obtain the weighted feature representation. Hi represents the LSTM hidden state vector at the i-th time step; then the weighted feature representations of all time steps are summed. This generates a comprehensive feature vector that incorporates time decay information, enabling feature parameters closer to the current moment to play a greater role in state assessment, while retaining historical feature information to capture long-term evolution trends.

[0104] The loss function is constructed using a weighted cross-entropy approach, with differentiated penalty coefficients applied to different urgency levels. Specifically, urgency is categorized into mild, moderate, and severe levels, with corresponding penalty coefficients set to 1.0, 2.0, and 4.0, respectively, ensuring the model has higher sensitivity to high-risk cases. The time-sensitivity penalty term is calculated by quantifying the relationship between the rate of patient deterioration and time intervals in the sample. The rate of escalation of urgency is defined as the change in the urgency level value divided by the time interval. For example, if a patient deteriorates from mild (level 1) to severe (level 3) within 10 minutes, the rate of escalation is... The urgency level is calculated per minute. The additional penalty weight is calculated as β = k·Δt·r, where the adjustment coefficient k is 2.0, Δt is the time interval from the initial call to the current assessment (in minutes), and r is the urgency level escalation rate as defined above. The final loss function is expressed as... L CE For weighted cross-entropy loss, y pred and y true These are numerical representations of the predicted urgency and the actual urgency, respectively. When the sample shows that the patient's condition deteriorates rapidly in a short period of time, a larger β value will impose a stronger penalty on the prediction bias, prompting the model to learn the dynamic change pattern of urgency that is sensitive to time.

[0105] The training process employed mini-batch gradient descent with a batch size of 32. The initial learning rate was 0.001, dynamically adjusted using cosine annealing over 100 training epochs. After training, the network was pruned and optimized using a weight magnitude-based pruning strategy. The absolute value of each connection weight was calculated as an importance score, and redundant connections with scores below 20% of the median global weight magnitude were removed. This reduced the number of model parameters by approximately 30% to 40% while maintaining a prediction accuracy decrease of no more than 2%, thereby shortening the model inference latency to within 50 milliseconds, meeting the stringent response speed requirements of emergency rescue scenarios.

[0106] This invention effectively improves the model's ability to identify rare critical cases and rapidly deteriorating conditions by expanding sparse class samples and using a time-sensitive penalty mechanism, thereby enhancing the timeliness and accuracy of urgency assessment.

[0107] Based on the corrected patient location, the real-time status of dispatchable ambulances and target medical institutions is obtained. Weight coefficients for multiple dispatch objectives are determined according to the adjusted urgency level. A multi-objective optimization function is constructed and solved to generate a dispatch scheme, including:

[0108] Based on the real-time status of dispatchable ambulances and the real-time resource status of the target medical institution, a set of candidate dispatchable objects is constructed.

[0109] The system obtains the ambulance load distribution status and medical institution reception capacity within the current area. The ambulance load distribution status reflects the task saturation level of each ambulance, and the medical institution reception capacity reflects the remaining reception capacity of each medical institution.

[0110] The adjusted urgency level is input into a preset weight adjustment strategy space to select a target weight adjustment strategy. The weight adjustment strategy space contains multiple sets of candidate weight adjustment strategies for different urgency levels.

[0111] The ambulance load distribution status and the medical institution's reception capacity are input into the target weight adjustment strategy to dynamically calculate the weight coefficients of the response time target, medical capacity matching target, and resource utilization target.

[0112] A multi-objective optimization function is constructed, which combines the response time objective, the medical capacity matching objective, and the resource utilization objective according to their respective weight coefficients.

[0113] The multi-objective optimization function is solved in the set of candidate scheduling objects to obtain the ambulance scheduling scheme and medical institution recommendation scheme corresponding to the optimal solution.

[0114] Combination Figure 2 Flowchart for generating a multi-objective optimized scheduling scheme: Based on the corrected patient location, the scheduling platform initiates a query request to the geographic information service interface, with the query range set to an area with a radius of 8 kilometers centered on the patient's location. For ambulances, it obtains their current location coordinates, onboard equipment status, medical staff configuration level, remaining fuel, and task execution status; for medical institutions, it obtains their resource status, such as department opening status, bed occupancy rate, number of patients waiting in the emergency department, and specialist doctor on-duty status. Ambulances with a status of "idle" or "about to complete the task," as well as medical institutions with unsaturated capacity, are selected to construct a candidate scheduling object set.

[0115] Ambulance load distribution is quantified by statistically analyzing the number of missions performed by each ambulance within the current region, cumulative mileage, and mission intervals. A mission saturation index is calculated for each ambulance, comparing the number of missions performed in the most recent two hours with a standard mission threshold. The standard threshold is determined based on the average mission allocation for the same historical period in the region. Specifically, it is calculated by extracting the number of missions completed by each ambulance within the same time period over the past 30 days, calculating their arithmetic mean, and then increasing it by 20% to obtain the standard threshold. This ensures that the threshold reflects both normal working load and emergency capacity. Ambulances with mission counts exceeding 80% of the threshold are marked as high-load. Medical institution capacity is calculated as the ratio of the current number of patients waiting in the emergency department to the standard capacity. The standard capacity is determined by each medical institution based on the total number of emergency department beds, the standard number of medical staff, and the average patient handling time. The calculation formula is the total number of beds multiplied by the unit-time turnover coefficient. The turnover coefficient is obtained by statistically analyzing historical data to determine the ratio of the average number of patients treated per hour to the number of beds at the medical institution. A ratio below 0.6 indicates sufficient patient capacity, a ratio between 0.6 and 0.85 indicates moderate patient capacity, and a ratio above 0.85 indicates strained patient capacity.

[0116] Five candidate strategies are pre-set in the weight adjustment strategy space, corresponding to urgency levels from 1 to 5. For low-urgency events (levels 1 and 2), the strategy focuses on resource utilization, with baseline weight coefficients set at 0.25 for response time, 0.25 for medical capacity matching, and 0.5 for resource utilization. For medium-urgency events (level 3), the weight coefficients for the three objectives are set at 0.4 for response time, 0.35 for medical capacity matching, and 0.25 for resource utilization. For high-urgency events (levels 4 and 5), the strategy significantly increases the weight of the response time objective, with weight coefficients set at 0.65 for response time, 0.25 for medical capacity matching, and 0.1 for resource utilization. Based on the adjusted urgency level, the corresponding target weight adjustment strategy is selected from the strategy space.

[0117] The ambulance load distribution and medical institution capacity are input into the selected strategy for dynamic fine-tuning. When the proportion of high-load ambulances in the region exceeds 60%, the strategy automatically shifts 0.1 from the resource utilization target weight to the response time target weight to avoid further burdening high-load vehicles. When the target medical institution's capacity is strained, the strategy shifts 0.15 from the response time target weight to the medical capacity matching target weight, prioritizing medical institutions with sufficient capacity. The final output is the weight coefficients of the three objectives: w1 represents the response time target weight, w2 represents the medical capacity matching target weight, and w3 represents the resource utilization target weight, all of which satisfy w1 + w2 + w3 = 1.

[0118] When constructing the multi-objective optimization function, the response time objective is quantified as the sum of the estimated travel time for the ambulance to reach the patient's location from its current location and the estimated travel time for the patient to be transferred from the current location to the medical institution. The travel time is calculated using real-time road network traffic data and distance. The medical capacity matching objective is quantified as the weighted sum of the medical institution's specialty matching degree and equipment configuration matching degree. The specialty matching degree is scored based on the correspondence between the patient's event type and the medical institution's specialty departments, with a perfect match being scored as 1 point, a partial match as 0.6 points, and a mismatch as 0 points. The equipment configuration matching degree is calculated based on the overlap between the types of medical equipment the patient estimates need and the medical institution's existing equipment. The overlap degree is equal to the number of types of equipment already equipped by the medical institution among the patient's required equipment divided by the total number of types of equipment required by the patient. The two matching degrees are weighted and summed with weights of 0.6 and 0.4 to obtain the quantified value of the medical capacity matching objective. The resource utilization rate objective is quantified as the weighted average of the ambulance task saturation and the medical institution's bed occupancy rate, with weights of 0.5 and 0.5, respectively. The lower the objective value, the more balanced the resource utilization. The three objective functions are linearly weighted and combined according to their weight coefficients to construct a comprehensive objective function.

[0119] Under the constraint of the candidate scheduling object set, a genetic algorithm is used to solve the multi-objective optimization function, ensuring that the solution search is only performed on the selected schedulable ambulances and unsaturated medical institutions. The population size is set to 50, and the maximum number of iterations is set to 100 generations. Integer encoding is used, with each chromosome containing the ambulance number and the medical institution number, and the encoding value range is limited to the index range of the candidate scheduling object set. The crossover probability is set to 0.8, and the mutation probability is set to 0.05. The fitness function is the negative value of the comprehensive objective function; the higher the fitness value, the better the solution. After iterative optimization, the chromosome with the highest fitness is obtained. After decoding, the ambulance scheduling scheme and the recommended medical institution scheme corresponding to the optimal solution are obtained and output to the scheduling instruction generation module.

[0120] This method uses a dynamic weight adjustment mechanism and multi-objective collaborative optimization to adaptively balance response speed, medical resource matching, and system load balancing under different urgency scenarios, thereby improving the scientific nature and effectiveness of scheduling decisions.

[0121] The adjusted urgency level is input into a preset weight adjustment strategy space to select a target weight adjustment strategy, including:

[0122] The event type features and time window features are extracted from the patient status information and combined with the adjusted urgency level to form a query feature vector;

[0123] Retrieve historical cases similar to the query feature vector from the historical scheduling database. The historical cases include historical usage strategy identifiers and corresponding scheduling effect evaluation indicators, including response time target completion rate and resource utilization efficiency.

[0124] The retrieved historical cases are sorted according to the aforementioned scheduling effect evaluation indicators;

[0125] Statistically analyze the frequency of occurrence of each historical strategy usage identifier in historical cases that exceed a preset scoring threshold, and construct a strategy effectiveness evaluation matrix.

[0126] Extract candidate weight adjustment strategies that match the query feature vector from the weight adjustment strategy space, and calculate the effect score of each candidate weight adjustment strategy in the strategy effect evaluation matrix;

[0127] The candidate weight adjustment strategy with the highest performance score is selected as the target weight adjustment strategy.

[0128] For example, after obtaining the adjusted urgency level value, this value needs to be mapped to a specific weight adjustment strategy to guide the subsequent construction of the multi-objective optimization function. Event type features are extracted from patient status information, represented by enumerated values, such as cardiac arrest, stroke, and trauma, each corresponding to different type codes. Simultaneously, time window features are extracted, reflecting the time period attribute of the call. This is achieved by dividing the day into several time periods, such as morning peak 7:00-9:00, noon peak 11:00-13:00, and evening peak 17:00-19:00, assigning a specific identifier to each time period. The event type features, time window features, and adjusted urgency level value are concatenated to form a three-dimensional query feature vector.

[0129] The historical dispatch database stores a large number of historical dispatch cases. Each case record includes the patient's event type, the time period of occurrence, the assessed urgency level at the time, the actual weight adjustment strategy used, and the dispatch effectiveness evaluation indicators. Dispatch effectiveness evaluation indicators include response time compliance rate and resource utilization efficiency. The response time compliance rate is calculated as the ratio of actual arrival time to standard response time. Resource utilization efficiency is quantified by comprehensively considering the rationality of the ambulance's travel route and the matching degree of the medical institution's receiving capacity. Specifically, the rationality of the ambulance's travel route is first assessed, using the ratio of the actual travel distance to the shortest path distance as the route efficiency indicator; the closer the ratio is to 1, the more rational the route. Next, the matching degree of the medical institution's receiving capacity is assessed, using the ratio of the remaining capacity of the medical institution to which the patient is sent to to the institution's standard capacity as the receiving matching indicator; the higher the ratio, the more sufficient the institution has after receiving the patient. The route efficiency indicator and the receiving matching indicator are weighted and summed with weights of 0.4 and 0.6 respectively to obtain the quantified value of resource utilization efficiency. Using vector similarity calculation methods, the similarity between the query feature vector and the feature vectors in historical cases is calculated. A weighted Euclidean distance metric is used, with the urgency dimension having a weight coefficient of 0.5, and event type and time window each accounting for 0.25. The top 50 historical cases with the highest similarity are retrieved as a reference set.

[0130] Historical cases in the reference set are ranked and scored based on a comprehensive evaluation index of scheduling effectiveness. The comprehensive score is obtained by weighted summation of response time target achievement rate and resource utilization efficiency after normalization, with weight coefficients set to 0.6 and 0.4, respectively. A preset scoring threshold of 0.75 is set to select high-quality cases with comprehensive scores higher than this threshold. The frequency of occurrence of each historical usage strategy identifier in these high-quality cases is counted to construct a strategy effectiveness evaluation matrix. This matrix uses the strategy identifier as the row index and the frequency of occurrence and average effectiveness score as the column data, forming a two-dimensional data structure.

[0131] The weight adjustment strategy space pre-configured five strategy schemes corresponding to weight 5, each strategy corresponding to an urgency level. The baseline weight coefficients for level 1 and 2 strategies are: response time target 0.25, medical capacity matching target 0.25, and resource utilization target 0.5; for level 3 strategies, the baseline weight coefficients are: response time target 0.4, medical capacity matching target 0.35, and resource utilization target 0.25; and for levels 4 and 5 strategies, the baseline weight coefficients are: response time target 0.65, medical capacity matching target 0.25, and resource utilization target 0.1. Based on the event type and urgency level range in the query feature vector, suitable candidate weight adjustment strategies are selected from the strategy space, typically 3 to 5 candidate strategies.

[0132] For each candidate strategy, its corresponding historical usage record is searched in the strategy effectiveness evaluation matrix. If a candidate strategy has a corresponding row in the matrix, the average effectiveness score of that row is directly extracted as its effectiveness score; if no direct corresponding record exists, the nearest neighboring strategy record with the closest strategy parameters is searched, and its effectiveness score is estimated using a linear interpolation method. The effectiveness scores of all candidate strategies are compared, and the strategy with the highest score is selected as the target weight adjustment strategy. The baseline weight coefficients of each scheduling target included in this strategy will be used for subsequent dynamic fine-tuning based on the ambulance load distribution and the medical institution's capacity to ensure that the scheduling scheme achieves optimal results in the current scenario.

[0133] This method utilizes historical case feedback and a strategy evaluation matrix to achieve data-driven selection of weight adjustment strategies, thereby enhancing the adaptability of scheduling decisions to different scenarios.

[0134] The ambulance load distribution and the medical institution's capacity are input into the target weight adjustment strategy, and the weight coefficients are dynamically calculated, including:

[0135] The task saturation of each ambulance in the ambulance load distribution state is quantitatively calculated to obtain the regional ambulance load index.

[0136] The remaining capacity of each medical institution within the total capacity of medical institutions is quantitatively calculated to obtain the regional medical institution capacity index.

[0137] The regional ambulance load index is compared with a preset load threshold, and the current load status type is determined based on the comparison result.

[0138] The baseline weight coefficient corresponding to the current load state type is extracted from the target weight adjustment strategy. The baseline weight coefficient includes the weight allocation of response time target, medical capacity matching target and resource utilization target.

[0139] The weight adjustment factor is calculated based on the regional ambulance load index and the regional medical institution capacity index. The weight adjustment factor is then used to correct the baseline weight coefficient, resulting in a dynamically adjusted weight coefficient.

[0140] For example, after obtaining the ambulance load distribution and the medical institution's reception capacity, the task saturation of each ambulance is quantified. Specifically, for each ambulance, the number of tasks currently being performed, the standby time, and the frequency of recent task completion are recorded. The task occupancy rate is calculated by dividing the number of currently performed tasks by the ambulance's standard task capacity, which is set to 1, meaning each ambulance can only perform one task at a time. Therefore, the task occupancy rate of an ambulance currently performing a task is 1, and the task occupancy rate of an idle ambulance is 0. The availability index is calculated based on the standby time, converting the standby time into a value between 0 and 1. Ambulances with standby times exceeding 30 minutes have an availability index of 1, while standby times between 0 and 30 minutes are mapped linearly, calculated by dividing the standby time in minutes by 30. The fatigue parameter is determined based on the frequency of task completion within the past 2 hours. The initial fatigue value is obtained by dividing the number of task completions by the standard task threshold. Values ​​exceeding 1 are truncated to 1 to ensure the fatigue parameter falls within the range of 0 to 1. The single-vehicle saturation value is calculated using a weighted summation, with task occupancy weighted at 0.5, availability index at 0.3, and fatigue parameter at 0.2. The saturation values ​​of all ambulances within the target area are arithmetically averaged to obtain the regional ambulance load index. This index ranges from 0 to 1; a higher value indicates a heavier overall load on the regional ambulances.

[0141] To assess the capacity of each medical institution, three indicators were extracted: the current number of patients in the emergency department, the number of available beds, and the number of medical staff on duty. The ratio of patients in the emergency department to the total number of beds was calculated as the bed occupancy rate. The difference between the number of available beds and the total number of beds was divided by the total number of beds to calculate the bed vacancy rate. The ratio of on-duty medical staff to the standard staffing level was used to determine staff adequacy; a ratio greater than or equal to 1 was recorded as 1, and a ratio less than 1 was used directly as staff adequacy. The smaller value between the bed vacancy rate and staff adequacy was taken as the immediate capacity of the medical institution. A weighted average of the immediate capacity of all medical institutions in the region was then calculated, with the weights determined by the historical proportion of each institution's patient volume, to obtain the regional medical institution capacity index. This index also ranged from 0 to 1, with a higher value indicating a stronger remaining capacity of the regional medical institutions.

[0142] The calculated regional ambulance load index is compared with three preset load thresholds: 0.3, 0.6, and 0.8. When the load index is less than 0.3, the current load status is determined to be "low load"; when the load index is between 0.3 and 0.6, it is determined to be "medium load"; when the load index is between 0.6 and 0.8, it is determined to be "high load"; and when the load index exceeds 0.8, it is determined to be "overloaded".

[0143] Based on the identified load state type, the corresponding baseline weight coefficient set is extracted from the target weight adjustment strategy library. For low load states, the baseline weight coefficients are set as follows: response time target 0.3, medical capacity matching target 0.3, and resource utilization target 0.4; for medium load states, the adjustment is to response time target 0.4, medical capacity matching target 0.35, and resource utilization target 0.25; for high load states, the adjustment is to response time target 0.5, medical capacity matching target 0.3, and resource utilization target 0.2; and for overload states, the adjustment is to response time target 0.6, medical capacity matching target 0.25, and resource utilization target 0.15. This tiered setting ensures a reasonable allocation of optimization objectives at different load levels, consistent with the baseline weight coefficients determined according to the urgency level in weight 5. The final weight coefficients are obtained by adjusting the baseline values ​​determined by the urgency level for load state.

[0144] Weight adjustment factors are calculated based on the regional ambulance load index and the regional medical institution capacity index. The difference between the load index and the capacity index is calculated, reflecting the degree of supply-demand imbalance. When the difference is positive and large, it indicates that ambulances are under heavy pressure while medical institutions have sufficient capacity, and the response time weight should be increased to accelerate turnover. When the difference is negative, it indicates that medical institutions are under heavy pressure, and the medical capacity matching weight should be increased for precise allocation. Specifically, the difference is normalized using a hyperbolic tangent function to obtain an adjustment amount ranging from -1 to +1. This adjustment amount is multiplied by 0.1 to obtain the weight adjustment factor for the response time objective, multiplied by -0.08 to obtain the weight adjustment factor for the medical capacity matching objective, and multiplied by -0.02 to obtain the weight adjustment factor for the resource utilization objective. Each adjustment factor is added to the corresponding baseline weight coefficient. After correction, the three weight coefficients are normalized proportionally by dividing each corrected weight coefficient by the sum of the three values ​​to ensure that the sum is 1, generating the final dynamically adjusted weight coefficients for subsequent multi-objective optimization solutions.

[0145] This method achieves a sensitive response of the weighting coefficient to the real-time status of regional resources by classifying the load status and dynamically adjusting the supply-demand difference, thus ensuring that the urgency level dominates the scheduling priority while taking into account the overall load balance of the system.

[0146] A second aspect of the present invention provides an emergency medical resource dispatching system based on multi-dimensional positioning and intelligent deviation correction, comprising:

[0147] The call response unit is used to respond to emergency call requests, collect multi-source location data from the call terminal, and obtain patient status information.

[0148] A data fusion unit is used to fuse the multi-source location data to obtain an initial positioning result;

[0149] The positioning verification unit is used to match and verify the initial positioning result. When the positioning deviation is detected to exceed the dynamic deviation threshold, the initial positioning result is deduced based on the event type characteristics and spatial distribution characteristics of similar historical events in the patient status information, and cross-verification is performed in combination with the reference position information to generate the corrected patient position.

[0150] The status assessment unit is used to dynamically assess and adjust the urgency of the patient's status information based on the characteristic parameters and temporal change characteristics in the patient's status information, combined with the temporal evolution pattern of historical similar characteristic data and the time interval between the current call time and the initial time, through a temporal prediction network.

[0151] The scheduling optimization unit is used to obtain the real-time status of dispatchable ambulances and target medical institutions based on the corrected patient location, determine the weight coefficients of multiple scheduling objectives according to the adjusted urgency level, construct and solve a multi-objective optimization function, and generate a scheduling scheme.

[0152] A third aspect of the present invention provides an electronic device, comprising:

[0153] processor;

[0154] Memory used to store processor-executable instructions;

[0155] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0156] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0157] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for emergency medical resource scheduling based on multi-dimensional positioning and intelligent deviation correction, characterized in that, include: In response to emergency call requests, collect multi-source location data from the calling terminal and obtain patient status information; The multi-source location data is fused to obtain an initial positioning result; The initial positioning result is matched and verified. When the positioning deviation is detected to exceed the dynamic deviation threshold, the deviation of the initial positioning result is estimated based on the event type characteristics in the patient status information and the spatial distribution characteristics of similar historical events. The deviation is then cross-verified in combination with the reference location information to generate the corrected patient location. Based on the characteristic parameters and temporal change characteristics in the patient status information, combined with the temporal evolution pattern of historical similar characteristic data and the time interval between the current call time and the initial time, the urgency of the patient status information is dynamically evaluated and adjusted through a temporal prediction network. Based on the corrected patient location, the real-time status of dispatchable ambulances and target medical institutions is obtained. According to the adjusted urgency level, the weight coefficients of multiple dispatch objectives are determined, a multi-objective optimization function is constructed and solved, and a dispatch scheme is generated.

2. The method according to claim 1, characterized in that, Based on the event type characteristics and spatial distribution characteristics of similar historical events in the patient status information, the initial positioning result is deviated and cross-validated using reference location information to generate a corrected patient location, including: The occurrence location data of historical similar events associated with the event type feature are retrieved from a pre-built associated knowledge base. The occurrence location data includes a set of offset vectors of the actual occurrence location of the historical event relative to the initial location and the corresponding environment type feature. Based on the environmental type characteristics of the location corresponding to the initial positioning result, a matching subset of offset vectors is selected from the offset vector set, and the dominant offset direction and average offset magnitude of the offset vector subset are calculated. Using the initial positioning result as a reference point, the spatial position is offset along the dominant offset direction according to the average offset amplitude to generate the estimated position; Obtain reference location information within a spatiotemporal range, calculate the spatial distance distribution between the estimated location and the reference location information; compare the spatial distance distribution with preset cross-validation conditions, and verify and correct the estimated location based on the comparison results to generate the corrected patient location.

3. The method according to claim 1, characterized in that, Based on the characteristic parameters and temporal change features in the patient status information, combined with the temporal evolution patterns of historical similar characteristic data and the time interval between the current call time and the initial time, a temporal prediction network dynamically assesses and adjusts the urgency of the patient status information, including: Feature parameters and temporal change features are extracted from the patient status information. The feature parameters include symptom intensity indicators and physiological status indicators. The temporal change features include the rate of change of each feature parameter during the call process. Retrieve historical similar feature data that matches the feature parameters from the historical database, and extract the temporal evolution pattern of the urgency over time from the historical similar feature data. The temporal evolution pattern includes the correlation mapping relationship between the feature parameter change trajectory and the urgency change trajectory. Calculate the time interval between the current call time and the initial time, and use the time interval as a time decay factor; The feature parameters, temporal change features, temporal evolution patterns, and time decay factors are input into a temporal prediction network for encoding. The temporal dependencies of the feature parameters are captured through the hidden layer units of the temporal prediction network. The temporal prediction network outputs an adjusted urgency assessment value, which is then updated in the patient status information.

4. The method according to claim 3, characterized in that, The training process of the time-series prediction network includes: A training sample set is constructed, which includes the feature parameter sequence, temporal evolution pattern and actual urgency label of historical emergency cases. Statistical analysis is performed on samples of different urgency categories in the training sample set to identify sparse urgency categories with a sample number lower than a preset threshold. For the samples of the sparse urgency category, synthetic samples are generated based on the feature parameter distribution features of samples of the same category, and the synthetic samples are added to the training sample set to form an expanded training sample set; A time decay attention mechanism is set in the encoding layer of the time series prediction network. The time decay attention mechanism calculates the attention weight based on the time interval between the feature parameter acquisition time and the current prediction time. The feature parameter with a smaller time interval is assigned a larger attention weight. A loss function is constructed, which sets a corresponding penalty coefficient based on the urgency label value of the sample. A time sensitivity penalty term is introduced into the loss function, which calculates an additional penalty weight based on the product of the time interval and the rate of increase of urgency in the sample. The time-series prediction network is trained and pruned using the expanded training sample set and the loss function.

5. The method according to claim 1, characterized in that, Based on the corrected patient location, the real-time status of dispatchable ambulances and target medical institutions is obtained. Weight coefficients for multiple dispatch objectives are determined according to the adjusted urgency level. A multi-objective optimization function is constructed and solved to generate a dispatch scheme, including: Based on the real-time status of dispatchable ambulances and the real-time resource status of the target medical institution, a set of candidate dispatchable objects is constructed. The system obtains the ambulance load distribution status and medical institution reception capacity within the current area. The ambulance load distribution status reflects the task saturation level of each ambulance, and the medical institution reception capacity reflects the remaining reception capacity of each medical institution. The adjusted urgency level is input into a preset weight adjustment strategy space to select a target weight adjustment strategy. The weight adjustment strategy space contains multiple sets of candidate weight adjustment strategies for different urgency levels. The ambulance load distribution status and the medical institution's reception capacity are input into the target weight adjustment strategy to dynamically calculate the weight coefficients of the response time target, medical capacity matching target, and resource utilization target. A multi-objective optimization function is constructed, which combines the response time objective, the medical capacity matching objective, and the resource utilization objective according to their respective weight coefficients. The multi-objective optimization function is solved in the set of candidate scheduling objects to obtain the ambulance scheduling scheme and medical institution recommendation scheme corresponding to the optimal solution.

6. The method according to claim 5, characterized in that, The adjusted urgency level is input into a preset weight adjustment strategy space to select a target weight adjustment strategy, including: Event type features and time window features are extracted from the patient status information and combined with the adjusted urgency level to form a query feature vector; historical cases similar to the query feature vector are retrieved from the historical scheduling database, the historical cases containing historical usage strategy identifiers and corresponding scheduling effect evaluation indicators, the scheduling effect evaluation indicators including response time target completion rate and resource utilization efficiency; the retrieved historical cases are sorted according to the scheduling effect evaluation indicators; Statistically analyze the frequency of occurrence of each historical strategy usage identifier in historical cases that exceed a preset scoring threshold, and construct a strategy effectiveness evaluation matrix. Extract candidate weight adjustment strategies that match the query feature vector from the weight adjustment strategy space, calculate the effect score of each candidate weight adjustment strategy in the strategy effect evaluation matrix, and select the candidate weight adjustment strategy with the highest effect score as the target weight adjustment strategy.

7. The method according to claim 5, characterized in that, The ambulance load distribution and the medical institution's capacity are input into the target weight adjustment strategy, and the weight coefficients are dynamically calculated, including: The task saturation of each ambulance in the ambulance load distribution state is quantitatively calculated to obtain the regional ambulance load index; the remaining reception capacity of each medical institution in the medical institution reception capacity is quantitatively calculated to obtain the regional medical institution capacity index. The regional ambulance load index is compared with a preset load threshold, and the current load status type is determined based on the comparison result. The baseline weight coefficient corresponding to the current load state type is extracted from the target weight adjustment strategy. The baseline weight coefficient includes the weight allocation of response time target, medical capacity matching target and resource utilization target. The weight adjustment factor is calculated based on the regional ambulance load index and the regional medical institution capacity index. The weight adjustment factor is then used to correct the baseline weight coefficient, resulting in a dynamically adjusted weight coefficient.

8. An emergency medical resource dispatch system based on multi-dimensional positioning and intelligent deviation correction, used to implement the method of any one of claims 1-7, characterized in that, include: The call response unit is used to respond to emergency call requests, collect multi-source location data from the call terminal, and obtain patient status information. A data fusion unit is used to fuse the multi-source location data to obtain an initial positioning result; The positioning verification unit is used to match and verify the initial positioning result. When the positioning deviation is detected to exceed the dynamic deviation threshold, the initial positioning result is deduced based on the event type characteristics and spatial distribution characteristics of similar historical events in the patient status information, and cross-verification is performed in combination with the reference position information to generate the corrected patient position. The status assessment unit is used to dynamically assess and adjust the urgency of the patient's status information based on the characteristic parameters and temporal change characteristics in the patient's status information, combined with the temporal evolution pattern of historical similar characteristic data and the time interval between the current call time and the initial time, through a temporal prediction network. The scheduling optimization unit is used to obtain the real-time status of dispatchable ambulances and target medical institutions based on the corrected patient location, determine the weight coefficients of multiple scheduling objectives according to the adjusted urgency level, construct and solve a multi-objective optimization function, and generate a scheduling scheme.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images