Medical emergency system based on dynamic graph flow prediction

Abstract

The invention discloses a medical emergency system based on dynamic graph flow prediction, and the method comprises the steps: S1, collecting traffic data recorded by all sensors through a medical emergency center system, and carrying out the preprocessing of the traffic data; s2, enabling the preprocessed traffic data to pass through a spatial-temporal feature embedding layer, and enabling the low-dimensional features of the traffic data to be converted into embedded vectors; s3, constructing a local adaptive graph and constructing a global dynamic graph generated by dynamic graph nodes through an embedded vector; s4, extracting spatial correlation between the local adaptive graph and the global dynamic graph through an adaptive gating fusion graph convolution module; s5, capturing time dependence of different scales through a multi-scale time gating convolution module; s6, forming a dynamic graph traffic prediction model in the steps S1 to S5, and pre-training the model; s7, the medical emergency center system loads the trained dynamic graph traffic prediction model to an emergency dispatching cloud platform for regulating and controlling a medical emergency channel in real time; the efficiency, the timeliness and the first-aid success rate of the medical first-aid channel are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of constructing an emergency system in intelligent healthcare, and more particularly to a medical emergency system based on dynamic graph traffic prediction. Background Art

[0002] Traffic flow prediction is an important technology in the smart city system, which plays an important role in urban traffic management and emergency passage, etc., and also has a relatively important impact on the intelligent healthcare system. As a participant in the smart city traffic system and the intelligent healthcare system, the passing efficiency of medical emergency vehicles directly affects the life and health of patients. However, there are still many deficiencies in existing traffic flow prediction methods, and there are very few prediction methods that can be effectively applied to optimize medical emergency channels.

[0003] In recent years, many traffic flow prediction algorithms based on deep learning have greatly improved the accuracy of traffic flow prediction. For example, convolutional neural networks (CNNs) and recurrent neural networks (RNNs) are used to learn the spatio-temporal dependencies of road networks. However, the CNN algorithm is only suitable for learning standard regular grid data graphs and is difficult to effectively handle the traffic flow prediction of road networks with non-regular graph topologies. In addition, the RNN algorithm is prone to problems of gradient vanishing or explosion. Therefore, by using Graph Convolutional Neural Network (GCN) and Gated Tanh Unit (GTU) to extract the spatio-temporal dependencies of non-Euclidean data of road networks, the model will have a significant performance improvement.

[0004] However, most existing models use a predefined static adjacency matrix to represent the topological graph structure of roads, which is difficult to reflect the real dynamic changes of the node dependencies in road networks in real time. Secondly, although some scholars have studied the mining of spatial dynamic correlations, they only stay at single dynamic spatial correlations, without simultaneously extracting local and global dynamic spatial correlations, and rarely capture multi-scale dynamic time dependencies.

[0005] In addition, most of the existing traffic flow prediction methods lack effective implementation connection with the medical emergency system. The connection mechanism has not been studied in detail in the existing technology, resulting in the medical emergency system still being in a relatively inefficient state, and some patients not receiving timely first aid. (For example, the traditional medical emergency system fails to classify based on the severity of the patient's condition and implement an adaptive decision-making mechanism, nor does it consider configuring a vital sign detection interface and a real-time path planning strategy during the emergency transportation process, making it difficult to adjust the transportation method in a timely manner when the patient's condition suddenly changes. Moreover, for medical emergency transportation in harsh environments, there is a lack of multi-mode combined transportation methods and road passage evaluation methods, which greatly limits the application effect of prediction technology in complex emergency environment scenarios. Additionally, there are such problems and so on) (Of course, the above content in parentheses is only an example explanation and does not mean that all these problems need to be solved one by one in the core technical solution of this time)

[0006] Therefore, in the medical emergency channel transportation system, there is an urgent need for a suitable traffic flow prediction method and a novel emergency transportation strategy for connection to improve the efficiency and timeliness of the medical emergency system. Summary of the Invention

[0007] The purpose of the present invention is to overcome the above-mentioned defects in the prior art and provide a medical emergency system based on dynamic graph traffic prediction for improving the efficiency and timeliness of the medical emergency channel of the medical emergency system.

[0008] To achieve the above purpose, the present invention provides a medical emergency system based on dynamic graph traffic prediction, which is characterized by including the following steps:

[0009] S1: The medical emergency center system collects traffic data recorded by sensors everywhere and preprocesses the traffic data;

[0010] S2: The preprocessed traffic data passes through a spatio-temporal feature embedding layer to convert its low-dimensional features into embedding vectors;

[0011] S3: Construct a local adaptive graph and a global dynamic graph for generating dynamic graph nodes through the embedding vectors;

[0012] S4: Extract the spatial correlation between the local adaptive graph and the global dynamic graph through an adaptive gated fusion graph convolution module;

[0013] S5: Capture time dependencies at different scales through a multi-scale time gated convolution module;

[0014] S6: Steps S1 - S5 form a dynamic graph traffic prediction model, and pre-train this model;

[0015] S7: The medical emergency center system loads a trained dynamic graph traffic prediction model into the emergency dispatch cloud platform for real-time regulation of the medical emergency channels.

[0016] Preferably, in step S1, the traffic data collected by the medical emergency command center includes GPS positioning data on emergency transport vehicles everywhere, urban traffic monitoring data everywhere, and data from the crowd flow monitoring devices at the emergency event sites everywhere.

[0017] Preferably, in step S1, it also includes the medical emergency command center collecting the patient condition grading information and the real-time vital sign values of patients everywhere, and / or collecting the parked and available states of multi-mode transportation means everywhere, and / or collecting weather forecast data everywhere.

[0018] Preferably, in step S2, the preprocessed traffic data passes through a spatio-temporal feature embedding layer and is combined with hyperedge embedding based on spatial features and periodic feature embedding based on the time dimension to convert its low-dimensional features into embedding vectors. The processing process is as follows:

[0019] Assume the preprocessed traffic data is X, which is converted to X through a convolutional layer data ∈R t×N×D . The structural embedding based on spatial features uses hyperedge embedding data represented as Z. First, the traditional graph is converted into a hypergraph, the traditional edges are converted into hyperedges, the nodes and hyperedges are combined to construct a hypergraph structure. Then, a feature vector is initialized for each node to map these features generated from the traffic data X into a low-dimensional space. Then, the hypergraph convolutional network can process the information of multiple nodes in the hyperedge through convolutional operations and learn the embedding representation of the nodes through this operation. Next, the periodic feature embedding based on the time dimension divides the data by time granularity. One day is divided into 1440 time steps, and one week is divided into seven days. Therefore, the time feature embedding outputs the periodic time feature representation X w ∈R t×D and the trend time feature representation X d ∈R t×D through fully connected layers, and uses the basic architecture of the Transformer self-attention mechanism to design X data as the relative position encoding X tq in the encapsulated sequence;

[0020] The hyperedge embedding method for the road network structure based on spatial features is:

[0021]

[0022] Among them, is the graph representing the road network in the traffic data, is a set of nodes, representing intersections in the road network, is a set of hyperedges. Each hyperedge can connect multiple nodes, representing a group of road nodes with similar functions or specific relationships, is a hyperedge weight matrix, representing the importance of each hyperedge.

[0023]

[0024] Among them, Z is the embedded road network node feature matrix, and each row represents the low-dimensional embedding vector of a node, F ME-HGCN is a super-resolution convolution function, Xe is the original road network node matrix, is the road network hypergraph structure, is the prior information related to medical first aid, including the current position of the first aid vehicle, the location of the target hospital, and the event impact range,

[0025]

[0026] Among them, Z (l) is the node feature embedding matrix of the l-th layer, σ is a non-linear activation function, is the diagonal matrix of node degrees, is the diagonal matrix of hyperedge degrees, H is the incidence matrix of the hypergraph, representing the subordination relationship between nodes and hyperedges. If a node belongs to a hyperedge, then H(v,e)=1, otherwise it is 0. α, β, γ are learnable exponential parameters that control the degree of normalization, Θ (l) is the learnable parameter matrix of the l-th layer, δ is the parameter that controls the influence degree of the first aid path selection item, Ω is a multi-factor fusion function used to integrate the weight matrix, and Ψ is a path-aware structure embedding enhancement function,

[0027] Finally, the embedding vector X embed of the embedding layer is output as:

[0028] X embed =X data +Z+X w +X d +X tq .

[0029] Preferably, in step S3, a local adaptive graph based on the local range is constructed, and an improved multi-head attention is used to process the masking matrix. Each attention head focuses on different traffic flow patterns, and the calculated attention scores are combined with the dynamic masking matrix to achieve multi-dimensional traffic relationship modeling,

[0030] Construct a global dynamic graph for generating dynamic graph nodes, that is, use the graph neural network sampling strategy to perform priority sampling on the neighbor information of the nodes after processing by the Flow-Node2vec algorithm for the medical emergency scenario, update the dynamic feature representation of the nodes, and model the long-term propagation impact of traffic flow.

[0031] 6. A medical emergency system based on dynamic graph traffic prediction according to claim 5, wherein in step S3, first, the query, key, and value matrices are represented using the self-attention formula. To extract the spatial correlation of the adjacent area, a masking matrix M is constructed in the stacked spatio-temporal blocks local , that is, if the distance between two adjacent nodes is too small and reaches the threshold p, the edge weight is set to 1, otherwise it is 0. Then, the masking matrix is processed using multi-head attention. By dividing the attention into multiple traffic flow characteristic heads, different levels of spatio-temporal dependence relationships are adaptively learned. The processing process is as follows:

[0032]

[0033] Among them, are the query transformation weight matrix, key transformation weight matrix, and value transformation weight matrix of the i-th head respectively, and ψ i (F t ) is the traffic flow pattern adjustment function corresponding to the i-th head, F t represents the traffic flow characteristics at time t, Q t is the query matrix at time t, K t is the key matrix at time t, V t is the value matrix at time t, d k is the dimension of the key vector, used to scale the dot product attention, ⊙ is the Hadamard product (element-wise multiplication), and softmax() is the activation function used to normalize the attention weights.

[0034] Then, the output of the attention heads is combined using the concatenation strategy, and the finally output local adaptive graph is:

[0035]

[0036] Among them, σ is the non-linear activation function, and α i (t) is the weight coefficient of the time-varying attention head;

[0037] To effectively capture the complex and changing characteristics contained in the road network, a dynamic graph node generation module is proposed in the stacked spatio-temporal blocks to obtain a node dynamic graph with timestamp identification. First, the embedding vector and hidden state of the traffic data at the current moment are input into the dynamic graph node generation module. The embedding vector of this traffic data assigns trainable weights to the nodes through multi-head attention, and its output is aggregated with the hidden state. Then, spatial dependency relationship S is extracted through a convolution operation. eff It is then input into the Node2vec algorithm, which controls the random walk strategy through the "bias" parameter, captures the dynamic structural relationship between nodes, and then continues to optimize through iteration to generate a learnable node embedding matrix X. embed Next, element-wise multiplication is used to generate node dynamic features, and its formula is as follows:

[0038] Z flow = σ λ (S eff ⊙ X embed )

[0039] where Z flow is the node dynamic feature, S eff is the learned traffic space dependency relationship, X embed is the learnable node embedding matrix, and λ is the saturation rate hyperparameter of the dynamically adjusted activation function σ.

[0040] Furthermore, an Emergency-GraphSAGE sampling strategy is proposed. For emergency medical situations, priority sampling is performed on the neighbor information of the nodes processed by the Flow-Node2vec algorithm to update the dynamic feature representation of the nodes. Then, using Z flow as the initial node feature representation, several nodes are sampled based on medical emergency relevance from the neighbors of all nodes, and the information of the neighbor nodes is extracted through an aggregation function. The embeddings of the neighbor nodes are weighted and averaged according to the relevance of different nodes to the emergency path, then concatenated with the embedding of its own node, and then a non-linear transformation is performed. Finally, the self-multiplication method is used to update the dynamic representation of the nodes, thereby generating a global dynamic graph. Using Z flow as the initial value, its generation formula is as follows:

[0041]

[0042] where, represents the embedding feature of neighbor node u in the previous layer, N flow represents the neighbor set of node v after sampling, W k represents the learnable weight matrix at the k-th layer, ρ(f vu ) represents the weight function of the correlation between nodes u and v, and σ is the non-linear activation function, τ(Pdensity ) is the traffic density parameter P density which is a dynamic adjustment function of the generated global dynamic traffic flow graph.

[0043] Preferably, in step S4, an adaptive gated fusion graph convolutional module with spectral domain enhancement is introduced. Through the decomposition and reconstruction in the spectral domain, the characteristics of these two graphs are optimized respectively at different frequency levels, and the spatial correlation between the local adaptive graph and the global dynamic graph is extracted. The processing process is as follows:

[0044] First, a multi-level feature frequency division mechanism is adopted to decompose the local adaptive graph and the global dynamic graph in the spectral domain:

[0045]

[0046] Among them, represents the spectral transformation function, Ψ loc and Ψ glo are respectively the sets of spectral basis functions of the local and global graphs,

[0047] Then, an adaptive spectral gating unit is introduced to dynamically adjust the importance of different frequencies based on traffic changes. The formula is:

[0048]

[0049] Among them, B represents the number of frequency bands, γ b (S t ) is the importance coefficient of the b-th frequency band adaptively generated based on the current traffic state S t , is the gating weight matrix of the b-th frequency band, and are respectively the components of the local graph and the global graph in the b-th frequency band, Φ b (X,F t ) is the enhancement function for the b-th frequency band, is the corresponding bias term,

[0050] Subsequently, an inverse transformation mechanism from the spectral domain to the spatial domain is used to restore the spectral representation after fusing the learnable reconstruction basis functions to the spatial domain. The formula is:

[0051]

[0052] Among them represents the inverse spectral transformation, Ω is the set of learnable reconstruction basis functions,

[0053] Finally, the spatial dependence characteristics of the fused graph are extracted:

[0054]

[0055] where α j (F t ) is the weight coefficient of the j-th graph neural network that dynamically changes according to the current traffic state, GCN j represents the j-th graph neural network, Θ j is the corresponding set of learnable parameters, ResConnect() is a function that fuses the local adaptive graph and the global dynamic graph through residual connection, and β is a balance parameter that controls the size of the residual connection.

[0056] Preferably, in step S5, the multi-scale time-gated convolution module includes three GTU modules with different scale receptive fields. The GTU module is a gated time unit module. Stacking the gated time unit modules can expand the receptive field of the time dimension. Its final output is as follows:

[0057] T out = ReLU(Concat(Pooling(ε1 * T), Pooling(ε2 * T), Pooling(ε3 * T)) + T)

[0058] where ε1, ε2, and ε3 are convolutional kernels of 1×3, 1×5, and 1×7, T is the time information after being processed by tanh and sigmoid, Pooling is a pooling operation, Concat is a concatenation operation, and ReLU(·) is a non-linear activation function.

[0059] Preferably, in step S7, the medical emergency center system loads a trained dynamic graph traffic prediction model into the emergency dispatch cloud platform to predict the medical emergency channel conditions for real-time regulation of medical vehicles,

[0060] and adopts different objective functions according to the patient's condition grading to adaptively adjust the medical emergency strategy,

[0061] and configures a vital sign detection interface to facilitate re-planning the transportation channel when the patient's condition suddenly changes,

[0062] and implements a three-level traffic control system according to the traffic congestion degree and the patient's condition. The three-level traffic control system is respectively signal optimization, area control, and city-wide emergency response.

[0063] Preferably, the medical emergency center system loads a trained dynamic graph traffic prediction model into the emergency dispatch cloud platform to generate real-time traffic prediction results for several hours in the future. The medical emergency center system obtains multiple medical emergency channels based on the prediction results and adjusts the best emergency channel based on the locations of multiple hospitals, the urgency of the patient's condition, and the road congestion situation.

[0064] Meanwhile, the prediction results are presented to the medical emergency command center through a dedicated visualization interface, which includes a traffic flow prediction map, a congested road section display map, and key information on the estimated round-trip time.

[0065] In addition, to consider the characteristics of medical emergencies, an adaptive decision-making mechanism based on patient condition grading is used. In the patient condition grading strategy, the emergency situation of the patient is converted into an urgency parameter, and different objective functions are set according to different levels of patients. The processing process is as follows:

[0066] min(T transport ) Red (critically ill)

[0067] α·min(T transport )+(1-α)·max(M resource ) Yellow (urgent)

[0068] β·min(T transport )+(1-β)·max(M distribution ) Green (general)

[0069] Among them, T transport is the transfer time, M resource is the matching degree of medical resources, M distribution represents the balance degree of medical resource distribution. α and β are balance parameters. Based on the objective function of the patient's condition, the medical emergency center system can adaptively adjust the priority of decision-making according to the urgency of the patient: for extremely critically ill patients in red, the system will transfer the patient in the fastest way; for urgent patients in yellow, while ensuring a short transfer time, the system considers an appropriate matching degree of medical resources to ensure that the patient can receive more appropriate treatment; for green patients, the system focuses more on the allocation of medical resources.

[0070] In addition, considering that the patient's condition may mutate during the transfer process, a vital sign monitoring interface and a real-time path planning mechanism for the patient are added:

[0071] ifΔV vital >V threshold then replanning()

[0072] Among them, ΔV vital is the change degree of the patient's vital signs, V threshold is the threshold matrix for measuring whether to re-plan. Different thresholds are set for diseases in different emergency situations. When the vital signs mutate, the medical emergency transfer strategy can be adjusted in time to improve the safety of the first aid process.

[0073] When re-planning the emergency rescue channel, to ensure the stability and safety of the transfer process, a path transition algorithm is used:

[0074] P transition = argmin(p ∈ P current ∩ P new )(D(p, P current ) + D(p, P new ) + λ

[0075] ·R(p))

[0076] where P current is the current path, P new is the re - planned path, D(p, P) is the distance metric from node p to path P, R(p) is the risk score of node p, and λ is the risk weight coefficient. Considering the optimality of medical channel transformation here, road risk factors are taken into account to ensure the safety of the transfer location and avoid the transformation of the first - aid channel on dangerous sections.

[0077] In addition, for medical first - aid situations in special harsh environments, a multi - mode joint transportation strategy is added. For the selection algorithm of transfer points, methods of multi - mode joint transportation such as vehicles, ships, and helicopters are considered:

[0078] J optimal = argmax j∈J (Q access (j)·Q safety (j)·Q resource (j)·Q weather (j))

[0079] where J is the set of all possible transfer locations, Q access is the accessibility score, Q safety is the safety score, Q resource is the remaining resource score, and Q weather is the weather adaptability score.

[0080] In addition, considering the impact of harsh weather on first - aid transportation, a road passage evaluation index under harsh weather conditions is set:

[0081] C weather (r, w) = C normal (r)·φ(r, w)

[0082] where r is the road section and w is the weather condition. C normal (r) is the traffic efficiency of road section r under normal weather, and φ(r, w) is the harsh weather impact coefficient. The formula is as follows:

[0083]

[0084] where η i is the weight coefficient of the i - th type of harsh weather, WIi is the intensity index of the i-th kind of severe weather, S i (r) is the sensitivity of road section r to the i-th kind of severe weather, which combines the characteristics of different road sections with severe weather factors to optimize the channels affected by severe weather.

[0085] In addition, a three-level traffic control system is set up in the medical control center. Combining the predicted traffic conditions and the emergency situations of patients' conditions, corresponding-level traffic control systems are adopted. For green patients or low-traffic-jam road sections, the optimized passage of medical vehicles is achieved by optimizing traffic lights and in-vehicle GPS systems. For yellow patients or medium-traffic-jam road sections, regional traffic control is carried out, including temporary vehicle avoidance, adjustment of bus stops, and adjustment of bicycle parking areas. For red patients or high-traffic-jam road sections, a city-wide emergency response is used, including traffic police guidance, road section closure, and reverse traffic.

[0086] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0087] 1. A medical first-aid system based on dynamic graph flow prediction according to the present invention

[0088] Because an embedded data modeling local adaptive graph and a global dynamic graph are used, and an adaptive gated fusion graph convolution is used to capture time-varying dynamic spatial correlations, the sensitivity to emergencies is improved;

[0089] Moreover, a multi-scale time gated convolution is used to capture multi-time-scale dynamic time features through GTUs with different receptive fields, enhancing the robustness of the model to complex time patterns. It mainly solves the obvious deficiencies in combining local and global time-varying dynamic spatial correlations and multi-scale dynamic time dependencies in the existing methods, resulting in large errors in prediction, low computational efficiency, and weak robustness.

[0090] 2. Through the above improvements, the requirements for a certain efficiency and timeliness of the medical first-aid channel can be met. Thus, further, grading is carried out based on the degree of the patient's condition, an adaptive decision-making mechanism is implemented, a vital sign detection interface and a real-time path planning strategy are configured, and the problem of being difficult to adjust the transportation method in a timely manner when the patient's condition suddenly changes is solved. In addition, for medical first-aid transportation in harsh environments, a multi-mode combined transportation method and a road passage evaluation method are adopted to improve the application effect of prediction technology in complex first-aid environment scenarios. Therefore, first aid is carried out according to different dynamically changing traffic conditions and different first-aid situations, thus greatly improving the first-aid success rate.

[0091] 3. Overall, a medical emergency system based on dynamic graph traffic prediction according to the present invention aims, firstly, to improve the efficiency and timeliness of the medical emergency channels in the medical emergency system. On this basis, secondly, it adapts to different dynamic traffic conditions, makes decisions on different emergency situations, and improves the success rate of first aid in the medical emergency channels of the medical emergency system. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0093] Figure 1 is the flow schematic diagram of a medical emergency system based on dynamic graph traffic prediction provided by an embodiment of the present invention;

[0094] Figure 2 is the system framework diagram of a medical emergency system based on dynamic graph traffic prediction provided by an embodiment of the present invention;

[0095] Figure 3 is the structural schematic diagram for constructing a global dynamic graph provided by an embodiment of the present invention;

[0096] Figure 4 is the model schematic diagram of a multi-scale time-gated convolution module provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0097] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present embodiments. Obviously, the described embodiments are one embodiment of the present invention, rather than all embodiments of the present invention. Based on the present embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0098] Please refer to Figures 1 to 4 , an embodiment of the present invention provides a medical emergency system based on dynamic graph traffic prediction, including a medical emergency system based on dynamic graph traffic prediction, characterized in that it includes the following steps:

[0099] S1: The medical emergency center system collects traffic data recorded by sensors everywhere and preprocesses the traffic data;

[0100] Among them, in the step S1, the traffic data collected by the medical emergency command center includes GPS positioning data on emergency transport vehicles everywhere, urban traffic monitoring data everywhere, and data of the crowd flow monitoring devices at the emergency event sites everywhere;

[0101] Among them, it also includes the medical emergency command center collecting the patient condition classification information everywhere (such as the red, yellow, and green classifications of the three-level classification method, which will be mentioned in the following content), the real-time vital sign values of the patients, and / or collecting the parked status and available status of multi-modal means of transportation everywhere, and / or collecting weather forecast data everywhere;

[0102] Specifically, the specific processing process of step S1 is as follows: In the step S1, the medical emergency center system, as the management center system, is responsible for collecting in real time the data from the medical emergency center system and the urban traffic network. These data usually come from the GPS positioning on emergency transport vehicles, the hospital emergency resource management system, urban traffic monitoring data, and the crowd flow monitoring devices at the emergency event sites everywhere. These data include the real-time position, speed, and destination information of the emergency vehicles, the availability of hospital beds, specialist doctors, and emergency equipment, the real-time traffic flow, vehicle speed, occupancy rate, etc. of the roads around the incident location, and the density and flow trend of on-site personnel in special events, etc. It is updated every minute. At the same time, the system also integrates the traffic emergency response data during historical special events as training samples to improve the model's recognition and prediction ability for emergencies;

[0103] In addition, the medical emergency center system also collects the patient condition classification information (such as the red, yellow, and green classifications of the three-level classification method), the real-time vital sign values of the patients (such as blood pressure, heart rate, etc.). For the rescue in special harsh environments, it also collects the parked positions and available status of multi-modal means of transportation (such as the position and available status of helicopters, the status of ships, etc.). For the conditions in extreme weather, the system also integrates real-time weather forecast data and considers the rainfall, wind speed, visibility, road surface conditions, etc. on the rescue roads.

[0104] S2: The preprocessed traffic data passes through the spatio-temporal feature embedding layer to convert its low-dimensional features into embedding vectors;

[0105] Specifically, in the step S2, the preprocessed traffic data passes through the spatio-temporal feature embedding layer and is combined with the hyper-edge embedding based on spatial features and the periodic feature embedding based on the time dimension to convert its low-dimensional features into embedding vectors. The processing process is as follows:

[0106] In the complex structure of the road network of traffic data, significant periodic characteristics are shown in the time dimension. Therefore, in the spatio-temporal feature embedding layer, hyperedge embedding based on the road network structure of spatial features and periodic feature embedding based on the time dimension are simultaneously adopted for the input traffic data, enabling subsequent modules to learn deeper spatio-temporal information. The specific processing process is as follows:

[0107] Assume that the preprocessed traffic data is X, which is converted to X through the convolutional layer data ∈R t×N×D . The structure embedding based on spatial features uses hyperedge embedding to represent the data as Z. First, the traditional graph is transformed into a hypergraph, the traditional edges are transformed into hyperedges, and the nodes and hyperedges are combined to construct a hypergraph structure. Then, a feature vector is initialized for each node, and these features generated from the traffic data X are mapped into a low-dimensional space. Then, the Hypergraph Convolutional Networks can process the information of multiple nodes in the hyperedge through convolutional operations and learn the embedding representation of the nodes through this operation. This convolutional operation updates the embedding vector of each node by propagating information in the hypergraph structure. The traditional graph is composed of nodes and adjacent edges, while each hyperedge of the hypergraph can connect multiple nodes, which can more naturally represent complex multi-relational data. Next, the periodic feature embedding based on the time dimension divides the data by time granularity. One day is divided into 1440 time steps, and one week is divided into seven days. Therefore, the time feature embedding outputs the periodic time feature representation X w ∈R t×D and the trend time feature representation X d ∈R t×D through the fully connected layer, and uses the basic architecture of the Transformer self-attention mechanism to design X data as the relative position encoding X tq encapsulated in the sequence; compared with the fixed absolute position encoding, the relative position encoding has higher flexibility and can automatically adjust the relationship between elements according to different contexts. Therefore, it can better adapt to different data structures, especially when the length of the data sequence changes;

[0108] Data usually has the problem of data scarcity, especially in some remote roads or areas. Through hypergraph modeling of the road network structure embedding, the similarity and adjacent relationship between nodes are effectively utilized, and the model can obtain more information from the traffic data in neighboring regions or roads to make up for the problem of incomplete data. Through periodic feature embedding, the model can use known time patterns to infer data for incomplete time periods, which helps to process missing or abnormal traffic data during some time periods;

[0109] The hyperedge embedding method for the road network structure based on spatial features is as follows:

[0110]

[0111] Among them, is a graph representing the road network in traffic data, is a set of nodes, representing intersections in the road network, is a set of hyperedges. Each hyperedge can connect multiple nodes and represents a group of road nodes with similar functions or specific relationships, is a hyperedge weight matrix, representing the importance of each hyperedge;

[0112]

[0113] Among them, Z is the embedded road network node feature matrix, and each row represents the low-dimensional embedding vector of a node, F ME-HGCN is a super-resolution convolution function, Xe is the original road network node matrix, containing features such as traffic flow and vehicle speed, is the road network hypergraph structure, is the prior information related to medical first aid, including the current position of the emergency vehicle, the location of the target hospital, the scope of the event impact, etc.;

[0114] Furthermore,

[0115]

[0116] Among them, Z (l) is the node feature embedding matrix of the l-th layer, σ is a non-linear activation function, is the diagonal matrix of the node degree, is the diagonal matrix of the hyperedge degree, H is the incidence matrix of the hypergraph, representing the subordination relationship between nodes and hyperedges. If node belongs to the hyperedge, then H(v,e) = 1, otherwise it is 0. α, β, γ are learnable exponential parameters, controlling the normalization degree, Θ (l) is the learnable parameter matrix of the l-th layer, δ is the parameter controlling the influence degree of the emergency path selection item, Ω is a multi-factor fusion function, used to integrate the weight matrix, Ψ is a path-aware structure embedding enhancement function,

[0117] Finally, the embedding vector X embed of the embedding layer is output as:

[0118] X embed = X data + Z + X w + X d + X tq .

[0119] The advantages of the above processing are as follows: In the subsequent local adaptive graph construction process, the spatial relationship information in the super-resolution embedded features enables it to determine the spatial proximity between nodes, thereby generating a more accurate masking matrix and solving the problem of data scarcity. In the subsequent global dynamic graph construction, the periodic feature embedding based on the time dimension helps the Flow-Node2vec algorithm capture the traffic trends in different time periods more accurately and generate traffic-aware node dynamic features more precisely;

[0120] S3: Construct a local adaptive graph through the embedding vector and a global dynamic graph generated by constructing dynamic graph nodes;

[0121] Among them, in step S3, a local adaptive graph based on a local range is constructed, and the masking matrix is processed using improved multi-head attention. Each attention head focuses on different traffic flow patterns (such as commuting flow, commercial flow, occasional flow, etc.), and the calculated attention scores are combined with the dynamic masking matrix to achieve multi-dimensional traffic relationship modeling.

[0122] The global dynamic graph generated by constructing dynamic graph nodes is to use the graph neural network sampling strategy to perform priority sampling on the neighbor information of the nodes processed by the Flow-Node2vec algorithm for the medical emergency scenario, update the dynamic feature representation of the nodes, and model the long-term propagation impact of traffic flow. The dual construction mechanism of the proposed local adaptive graph and global dynamic graph, the former can accurately capture the local fluctuations of emergencies, and the latter models the long-term propagation impact of traffic flow through the Flow-Node2vec algorithm;

[0123] Among them, in the step S3, first, the query, key, and value matrices are represented using the self-attention formula. In order to extract the spatial correlation of the adjacent region, a masking matrix M is constructed in the stacked spatio-temporal blocks. local , that is, if the distance between two adjacent nodes is too small to reach the threshold p, then its edge weight is set to 1, otherwise it is 0. In this way, the influence of distant nodes can be blocked, and the masking matrix makes the connection between distant nodes ignored, thereby reducing the calculation amount and paying more attention to the dependence relationship between adjacent nodes. Then, the masking matrix is processed using multi-head attention. By dividing the attention into multiple traffic flow characteristic heads, different levels of spatio-temporal dependence relationships are adaptively learned. Different attention heads focus on different traffic flow patterns (such as commuting flow, commercial area flow, etc.), and the calculated attention is combined with the dynamic masking matrix to achieve multi-dimensional traffic flow pattern modeling. The processing process is as follows:

[0124]

[0125] Among them, are the query transformation weight matrix, key transformation weight matrix, and value transformation weight matrix of the i-th head, ψ i (Ft ) is the traffic pattern adjustment function corresponding to the i-th head, F t represents the traffic flow characteristics at time t, Q t is the query matrix at time t, K t is the key matrix at time t, V t is the value matrix at time t, d k is the dimension of the key vector, used to scale the dot product attention, ⊙ is the Hadamard product (element-wise multiplication), and softmax() is the activation function used to normalize the attention weights.

[0126] Then, use the concatenation strategy to combine the outputs of the attention heads. The finally output local adaptive graph is:

[0127]

[0128] where σ is the non-linear activation function, and α i (t) is the weight coefficient of the time-varying attention head; the local adaptive graph construction method adaptively focuses on the main node relationships under different traffic patterns through the attention mechanism, significantly improving the model prediction performance. During the local adaptive graph construction process, the spatial relationship information in the super-resolution embedded features can determine the spatial similarity between nodes, thus forming a more accurate masking matrix.

[0129] To effectively capture the complex and changing characteristics contained in the road network, a dynamic graph node generation module is proposed in the stacked spatio-temporal blocks to obtain a node dynamic graph with timestamp identification. First, input the embedded vector and hidden state of the traffic data at the current moment into the dynamic graph node generation module. The embedded vector of this traffic data assigns trainable weights to the nodes through multi-head attention, enabling the network to automatically focus relatively more on valuable information, further enhancing the representation of dynamic spatio-temporal dependence, and aggregating its output with the hidden state. Then, extract the spatial dependence relationship S eff , and input it into the Node2vec algorithm. It controls the random walk strategy through the "bias" parameter, captures the dynamic structural relationship between nodes, and then continues to optimize through iteration to generate a learnable node embedding matrix X embed , and then use element-wise multiplication to generate node dynamic features. The formula is as follows:

[0130] Z flow = σ λ (S eff ⊙ X embed )

[0131] where Z flow is the node dynamic feature, S eff is the learned traffic space dependence relationship, and X embedis the learnable node embedding matrix, and λ is the saturation rate hyperparameter of the dynamically adjusted activation function σ.

[0132] Further, the Emergency-GraphSAGE sampling strategy is proposed. For emergency medical situations, priority sampling is performed on the neighbor information of the nodes processed by the Flow-Node2vec algorithm to update the dynamic feature representation of the nodes. Then, taking Z flow as the initial node feature representation, by sampling several nodes from the neighbors of all nodes based on medical emergency relevance and extracting the information of the neighbor nodes through an aggregation function, a more comprehensive global dynamic graph is generated. This combination can better capture the complex dynamic spatio-temporal dependence relationships between nodes, making the final node embedding not only contain the dynamic information of medical emergencies but also consider the allocation characteristics of global nodes. The neighbor node embeddings are weighted and averaged according to the relevance of different nodes to the emergency path, then concatenated with the embedding of its own node, and then non-linearly transformed. Finally, the self-multiplication method is used to update the dynamic representation of the nodes, thereby generating a global dynamic graph. Taking Z flow as the initial value, its generation formula is as follows:

[0133]

[0134]

[0135] where, represents the embedding feature of neighbor node u in the previous layer, N flow represents the neighbor set of node v after sampling, W k represents the learnable weight matrix at the k-th layer, ρ(f vu ) represents the weight function of the correlation between nodes u and v, σ is the non-linear activation function, τ(P density ) is the dynamic adjustment function of the traffic density parameter P density , is the generated global dynamic graph of traffic flow. In the construction of the global dynamic graph, the periodic feature embedding based on the time dimension helps the Flow-Node2vec algorithm to more accurately capture the traffic flow relationship in different time periods, thereby generating more accurate node dynamic features of traffic flow sensitivity.

[0136] S4: Extract the spatial correlation between the local adaptive graph and the global dynamic graph through the adaptive gating fusion graph convolution module;

[0137] Among them, in step S4, an adaptive gating fusion graph convolutional module with spectral domain enhancement is introduced. Through the decomposition and reconstruction in the spectral domain, the heterogeneity problem between the local adaptive graph and the global dynamic graph is effectively solved. Through the innovative spectral decomposition and reconstruction technology, the characteristics of these two graphs are optimized at different frequency levels, retaining both the sensitivity of the local graph to emergencies and the modeling ability of the global graph for long-term dependencies, and extracting the local and global dynamic spatial correlations. The specific processing process is as follows:

[0138] First, a multi-level feature frequency division mechanism is adopted to decompose the local adaptive graph and the global dynamic graph in the spectral domain:

[0139]

[0140] Among them, represents the spectral transformation function, Ψ loc and Ψ glo are the sets of spectral basis functions of the local and global graphs respectively,

[0141] Next, an adaptive spectral gating unit (ASGU) is introduced to dynamically adjust the importance of different frequencies based on traffic changes. The formula is:

[0142]

[0143] Among them, B represents the number of frequency bands, γ b (S t ) is the importance coefficient of the b-th frequency band adaptively generated based on the current traffic state S t , is the gating weight matrix of the b-th frequency band, and are the components of the local graph and the global graph in the b-th frequency band respectively, Φ b (X,F t ) is the enhancement function for the b-th frequency band, is the corresponding bias term,

[0144] Subsequently, an inverse transformation mechanism from the spectral domain to the spatial domain is used to restore the spectral representation after fusing the learnable reconstruction basis functions to the spatial domain. The formula is:

[0145]

[0146] Among them represents the inverse spectral transformation, Ω is the set of learnable reconstruction basis functions,

[0147] Finally, extract the spatial dependence characteristics of the fused graph:

[0148]

[0149] where α j (F t ) is the weight coefficient of the j-th graph neural network that dynamically changes according to the traffic state at the current moment, GCN j represents the j-th graph neural network, Θ j is the corresponding set of learnable parameters, ResConnect() is a function that fuses the local adaptive graph and the global dynamic graph through residual connection, and β is a balance parameter that controls the size of the residual connection;

[0150] The proposed module for modulus adaptive gated fusion of graph convolution with spectral domain enhancement effectively solves the problem of heterogeneous fusion between the local adaptive graph and the global dynamic graph through spectral domain decomposition and reconstruction. The local adaptive graph focuses on capturing the associations between adjacent nodes in the road network and can effectively model the local impact of emergencies; while the global dynamic graph focuses on describing the long-term impact of traffic and is good at capturing the diffusion effect of distant nodes. Through spectral domain decomposition and fusion, this module can optimize the characteristics of these two graphs at different frequencies, retaining both the feature extraction ability of the local graph for emergencies and the modeling ability of the global graph for long-term dependencies;

[0151] S5: Capturing time dependencies at different scales through a multi-scale time gated convolution module, thereby learning short-term time dynamics and long-term stability and capturing time dependencies at different scales;

[0152] Among them, in step S5, a new multi-scale time gated convolution module is used to extract short-term time dynamics and long-term stability in the data. Traditional GTUs use convolutional kernels to double the number of channels, and the kernel size is generally 1×1, making it difficult to capture long-term time dependencies in the data. Therefore, on this basis, a multi-scale time gated convolution module is improved. The multi-scale time gated convolution module includes three GTU modules with different scale receptive fields. The GTU module is a gated time unit module. Stacking gated time unit modules can expand the receptive field in the time dimension and solve the problem of missing long-term time information. Its final output is as follows:

[0153] T out = ReLU(Concat(Pooling(ε1*T),Pooling(ε2*T),Pooling(ε3*T))+T)

[0154] where ε1, ε2, and ε3 are convolutional kernels of 1×3, 1×5, and 1×7, T is the time information processed by tanh and sigmoid, Pooling is a pooling operation, Concat is a concatenation operation, and ReLU(·) is a non-linear activation function;

[0155] GTUs with different convolutional kernels have completely different receptive fields and can obtain temporal information from different ranges of traffic data. Small convolutional kernels can capture dependencies within short time intervals, which is very effective for capturing frequently changing short-term patterns, emergencies, or the dynamics of local moments. Larger convolutional kernels can focus on temporal dependencies over longer time spans and are suitable for capturing long-term trend information. By fusing different-sized convolutional kernels, GTUs can obtain feature information at different time spans in the time dimension, thus better modeling the short-term dynamics and long-term stability of traffic data. The concatenation operation can combine multi-scale temporal features extracted by different GTUs to obtain fused features containing short-term dynamic and long-term stability information. After dimensionality reduction by the pooling layer, the computational cost can be reduced, the feature representation can be generalized, and overfitting to the training data can be avoided. In this process, skip connections are used to more efficiently utilize the learned features and enable faster convergence during training. Finally, GTU combined with residual connections can well reduce the risk of gradient vanishing or explosion and improve the training effect of the model.

[0156] S6: Steps S1 - S5 form a dynamic graph traffic prediction model and pre-train this model;

[0157] S7: The medical emergency center system loads the trained dynamic graph traffic prediction model into the emergency dispatch cloud platform for real-time regulation of medical emergency channels;

[0158] Among them, in step S7, the medical emergency center system loads the trained dynamic graph traffic prediction model into the emergency dispatch cloud platform to predict the status of medical emergency channels (ambulance channels), for real-time regulation of medical vehicles, and adopts different objective functions according to the patient's condition grading (green / yellow / red), adaptively adjusts the medical emergency strategy, and configures a vital sign detection interface to facilitate re-planning the transportation route when the patient's condition suddenly changes. And according to the traffic congestion degree and the patient's condition, a three-level traffic control system is implemented. The three-level traffic control system is respectively signal optimization, area control, and city-wide emergency response. Based on the traffic flow prediction, the patient's destination is adjusted in real time to prevent the over-concentration of medical emergency resources;

[0159] Furthermore, a multi-mode joint transportation mechanism can be introduced. Under comprehensive consideration, various joint transportation methods such as vehicles, ships, and helicopters are used to enable different patients to quickly reach the hospital. An evaluation index for road traffic under bad weather can also be introduced, and the channel can be optimized and selected based on road characteristics and weather conditions. An all-weather emergency route library is established, and multiple alternative routes are preset for different bad weather conditions. Record the difference between the actual travel time and the predicted time, and continuously optimize the medical transportation channel;

[0160] More specifically, the processing process of step S7 is as follows:

[0161] The medical emergency center system loads a dynamic graph traffic prediction model after training into the emergency dispatch cloud platform. The processed traffic data, along with information such as emergency vehicles and weather, is used as the input to the model to generate real-time traffic prediction results for the next few hours. The medical emergency center system determines multiple medical emergency channels based on the prediction results and adjusts the optimal emergency channels based on the locations of multiple hospitals, the urgency of the patients' conditions, and the road congestion situation.

[0162] Meanwhile, the prediction results are presented to the medical emergency command center through a dedicated visualization interface, which includes key information such as traffic prediction graphs, congestion section display graphs, and estimated round-trip times.

[0163] To consider the characteristics of medical emergencies, an adaptive decision-making mechanism based on patient condition grading is used. In the patient condition grading strategy, the emergency situation of the patient is converted into an urgency parameter, and different objective functions are set according to different levels of patients. The processing process is as follows:

[0164] min(T transport ) Red (critically ill)

[0165] α·min(T transport )+(1-α)·max(M resource ) Yellow (urgent)

[0166] β·min(T transport )+(1-β)·max(M distribution ) Green (general)

[0167] Where T transport is the transfer time, M resource is the matching degree of medical resources, M distribution represents the balance degree of medical resource distribution, and α and β are balance parameters. Based on the objective function of the patient's condition, the medical emergency center system can adaptively adjust the decision-making priority according to the urgency of the patient: for extremely critically ill patients in red, the system will transfer the patient in the fastest way; for urgent patients in yellow, while ensuring a short transfer time, the system considers an appropriate matching degree of medical resources to ensure that the patient can receive appropriate treatment; for green patients, the system focuses more on the allocation of medical resources; avoiding over-concentration of resources in a certain hospital, thus effectively improving the efficiency of medical treatment.

[0168] Considering that the patient's condition may mutate during the transfer process, a vital sign monitoring interface for the patient and a real-time path planning mechanism are added:

[0169] ifΔV vital >V threshold then replanning()

[0170] Among them, ΔV vital is the change degree of the patient's vital signs, and V threshold is the threshold matrix for measuring whether to re-plan. Different thresholds are set for diseases in different emergency situations. When there is a sudden change in vital signs, the medical emergency transport strategy can be adjusted in a timely manner to improve the safety of the first aid process.

[0171] When re-planning the emergency channel, to ensure the stability and safety of the transport process, the path transition algorithm is used:

[0172] P transition = argmin(p∈P current ∩P new )(D(p, P current ) + D(p, P new ) + λ

[0173] ·R(p))

[0174] Among them, P current is the current path, P new is the re-planned path, D(p, P) is the distance metric from node p to path P, R(p) is the risk score of node p, and λ is the risk weight coefficient. Here, considering the optimality of the medical channel transformation, the road risk factor is taken into account to ensure the safety of the transfer location and avoid the transformation of the emergency channel in dangerous sections (such as sharp turns and steep slopes);

[0175] In addition, for medical emergency situations in special harsh environments, a multi-mode combined transport strategy is added. For the selection algorithm of transfer points, methods of multi-mode combined transport such as vehicles, ships, and helicopters will be considered:

[0176] J optimal = argmax j∈J (Q access (j)·Q safety (j)·Q resource (j)·Q weather (j))

[0177] Among them, J is the set of all possible transfer locations, Q access is the reachability score, Q safety is the safety score, Q resource is the resource remaining score, Q weather is the weather adaptability score. Facing the complex and changeable transfer environment (such as mountainous areas, flood or snow disaster areas), the medical emergency center system can adjust the means of transportation for multi-mode combined transport in real time according to the specific calculated scores and consider the selectivity of transfer locations;

[0178] In addition, considering the impact of bad weather on emergency transportation, road passage evaluation indicators under bad weather conditions are set:

[0179] C weather (r, w) = C normal (r)·φ(r, w)

[0180] where r is the road section, w is the weather condition, and C normal (r) is the traffic efficiency of road section r under normal weather, and φ(r, w) is the bad weather impact coefficient. The formula is as follows:

[0181]

[0182] where η i is the weight coefficient of the i-th type of bad weather, WI i is the intensity index of the i-th type of bad weather, S i (r) is the sensitivity of road section r to the i-th type of bad weather, which combines the characteristics of different road sections with bad weather factors. For example, low-lying road sections are more affected in rainy conditions, and elevated road sections are more affected in strong wind conditions, thus realizing the optimization of channels affected by bad weather;

[0183] Therefore, the medical emergency center system has established an all-weather emergency route library in the city, and multiple sets of alternative routes are pre-set for different bad weathers (such as heavy rain, hail, etc.). Different alternative routes are used for different bad weathers. For example, in heavy snow weather, the medical channels of ambulances preferentially select road sections with less snowfall or snow melting equipment;

[0184] In addition, in order to achieve the purpose of optimizing medical emergency channels, a three-level traffic control system is set up in the medical emergency center system. Combining the predicted traffic conditions and the emergency conditions of patients' illnesses, corresponding levels of traffic control systems are adopted.

[0185] For green patients or low-traffic congestion road sections, the optimized passage of medical vehicles is achieved by optimizing traffic lights and in-vehicle GPS systems. For yellow patients or medium-traffic congestion road sections, regional traffic control is carried out, including temporary vehicle avoidance, adjustment of bus stops, and adjustment of bicycle parking areas. For red patients or high-traffic congestion road sections, a city-wide emergency response is used, including traffic police guidance, road section closure, and reverse passage.

[0186] Based on the embedding of medical resource information, the destination of the patient is adaptively adjusted. Judgments are made according to the mid-course changes in the patient's condition and the real-time changes in road conditions. When the judgment index reaches a certain level, a better hospital can be selected for the patient. In addition, the medical emergency center system will also predict the remaining medical resources of the hospital to avoid emergency congestion caused by too many ambulances going to the same hospital.

[0187] The medical emergency center system can also deploy the positions of emergency vehicles in advance according to the severity of the patient's condition, weather conditions and road congestion, adjust the vehicles to the area with the least impact of bad weather, or preferentially select other emergency transportation means to shorten the time to reach the patient's destination. For the emergency vehicles on the way of transporting patients, the center will detect the road conditions in real time and check the meteorological data, dynamically adjust the speed and route, and order them to take temporary evasive actions in case of emergencies.

[0188] In addition, the medical emergency center system will also record the difference between the actual passing time and the predicted time of each emergency task, establish an analysis chart of the passing time curve of the exclusive transportation means, and continuously optimize the medical channel. Select the best departure point and target hospital according to the predicted traffic conditions, start remote medical guidance in advance in case of congestion, and optimize the multi-vehicle collaborative rescue strategy.

[0189] In summary, the embodiments of the present invention provide a medical emergency system based on dynamic graph flow prediction, which can be seamlessly integrated into the medical emergency control center of the smart city system. Its advantages are as follows:

[0190] Use embedded data to model local adaptive graphs and global dynamic graphs, use spectrum-domain enhanced adaptive gated fusion graph convolution to capture time-varying dynamic spatial correlations, and improve the sensitivity to emergencies; secondly, use multi-scale time gated convolution to capture multi-time scale dynamic time features through GTUs with different receptive fields, and enhance the robustness of the model to complex time patterns. To mainly solve the obvious deficiencies in combining local and global time-varying dynamic spatial correlations and multi-scale dynamic time dependencies in existing methods, resulting in large errors in prediction, low computational efficiency and weak robustness. In addition, classify according to the severity of the patient's condition, implement an adaptive decision-making mechanism, configure a vital sign detection interface and a real-time path planning strategy to solve the problem of difficult timely adjustment of the transportation method in case of sudden changes in the patient's condition.

[0191] Secondly, for medical emergency transportation in harsh environments, adopt a multi-mode combined transportation method and a road passage evaluation method to improve the application effect of prediction technology in complex emergency environment scenarios.

[0192] Finally, in the case of large-scale event events, disaster events or major accidents, etc., emergency personnel can effectively reach the incident site quickly in a complex and changeable traffic environment and send the patient to the hospital, while the traditional fixed green channel strategy often has difficulty coping with this dynamically changing traffic condition.

[0193] Therefore, the medical emergency system based on dynamic graph traffic prediction not only improves the efficiency and timeliness of the medical emergency channels of the medical emergency system, but also adapts to different dynamic traffic conditions and makes decisions on different emergency situations, thereby increasing the success rate of first aid in the medical emergency channels of the medical emergency system.

[0194] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A medical first aid system based on dynamic graph traffic prediction, characterized in that, It includes the following steps: S1: The medical emergency center system collects traffic data recorded by sensors everywhere and preprocesses the traffic data. S2: The preprocessed traffic data passes through a spatio-temporal feature embedding layer to convert its low-dimensional features into embedding vectors. S3: Construct a local adaptive graph through the embedding vectors and a global dynamic graph generated by constructing dynamic graph nodes. S4: Extract the spatial correlation between the local adaptive graph and the global dynamic graph through an adaptive gating fusion graph convolution module. S5: Capture time dependencies at different scales through a multi-scale time gating convolution module. S6: Steps S1 - S5 form a dynamic graph traffic prediction model, and pre-train this model. S7: The medical emergency center system loads the trained dynamic graph traffic prediction model into the emergency dispatch cloud platform for real-time regulation of the medical emergency channel.

2. The medical first aid system based on dynamic graph traffic prediction according to claim 1, wherein In step S1, the traffic data collected by the medical emergency command center from sensors everywhere includes GPS positioning data on emergency transport vehicles everywhere, urban traffic monitoring data everywhere, and data from the crowd flow monitoring devices at the emergency event sites everywhere.

3. The medical first aid system based on dynamic graph traffic prediction according to claim 1, wherein In step S1, it also includes the medical emergency command center collecting the patient condition grading information and the real-time vital sign values of patients everywhere, and / or collecting the parked and available states of multi-mode transportation means everywhere, and / or collecting weather forecast data everywhere.

4. The medical first aid system based on dynamic graph traffic prediction according to claim 1, wherein In step S2, the preprocessed traffic data passes through a spatio-temporal feature embedding layer and is combined with hyperedge embedding based on spatial features and periodic feature embedding based on the time dimension to convert its low-dimensional features into embedding vectors. The processing process is as follows: Assume that the preprocessed traffic data is X, which is converted to X through the convolutional layer data ∈R t×N×D . The structural embedding based on spatial features uses hyperedge embedding data, denoted as Z. First, the traditional graph is transformed into a hypergraph, the traditional edges are transformed into hyperedges, and the nodes and hyperedges are combined to construct a hypergraph structure. Then, a feature vector is initialized for each node, and the features generated from the traffic data X are mapped into a low-dimensional space. Subsequently, the hypergraph convolutional network can process the information of multiple nodes in the hyperedge through convolutional operations and learn the embedded representation of the nodes through this operation. Next, the periodic feature embedding based on the time dimension divides the data by time granularity. One day is divided into 1440 time steps, and one week is divided into seven days. Therefore, the time feature embedding outputs the periodic time feature representation X w ∈R t×D and the trend time feature representation X d ∈R t×D respectively, and uses the basic architecture of the Transformer self-attention mechanism to design X data as the relative position encoding X tq in the encapsulated sequence; The hyperedge embedding method for the road network structure based on spatial features is: Among them, is a graph representing the road network in traffic data, is a set of nodes, representing intersections in the road network, is a set of hyperedges. Each hyperedge can connect multiple nodes, representing a group of road nodes with similar functions or specific relationships, is a hyperedge weight matrix, representing the importance of each hyperedge. Among them, Z is the feature matrix of the road network nodes after embedding, and each row represents the low-dimensional embedding vector of a node, F ME-HGCN is the super-resolution convolution function, and Xe is the original road network node matrix, is the hypergraph structure of the road network, is the prior information related to medical first aid, including the current position of the emergency vehicle, the location of the target hospital, and the scope of the event impact, Among them, Z (l) is the node feature embedding matrix of the l-th layer, σ is a non-linear activation function, is the diagonal matrix of node degrees, is the diagonal matrix of hyperedge degrees, H is the incidence matrix of the hypergraph, representing the subordination relationship between nodes and hyperedges. If node belongs to hyperedge, then H(v, e)=1, otherwise it is 0. α, β, γ are learnable exponential parameters that control the degree of normalization, Θ (l) is the learnable parameter matrix of the l-th layer, δ is the parameter that controls the influence degree of the first-aid path selection item, Ω is the multi-factor fusion function used to integrate the weight matrix, and Ψ is the path-aware structure embedding enhancement function. Finally, the embedding vector X of the embedding layer embed The output representation is: X embed = X data + Z + X w + X d + X tq 。 5. The medical emergency system based on dynamic graph traffic prediction according to claim 1, wherein, In step S3, construct a local adaptive graph based on a local range, use an improved multi-head attention to process the masking matrix, each attention head focuses on different traffic flow patterns, and the calculated attention scores are combined with the dynamic masking matrix to achieve multi-dimensional traffic relationship modeling. Construct a global dynamic graph generated by dynamic graph nodes, that is, use the graph neural network sampling strategy to perform priority sampling on the neighbor information of the nodes after processing by the Flow-Node2vec algorithm for the medical emergency scenario, update the dynamic feature representation of the nodes, and model the long-term propagation influence of the traffic flow.

6. The medical first-aid system based on dynamic graph traffic prediction according to claim 5, wherein, In the step S3, first, the query, key, and value matrices are represented using the self-attention formula. To extract the spatial correlation of adjacent regions, a masking matrix M is constructed in the stacked spatio-temporal blocks local , that is, if the distance between two adjacent nodes is too small and reaches the threshold p, the edge weight is set to 1, otherwise it is 0. Then, the masked matrix is processed using multi-head attention. By dividing the attention into multiple traffic flow characteristic heads, different levels of spatio-temporal dependence relationships are adaptively learned. The processing process is as follows: Among them, are respectively the query transformation weight matrix, key transformation weight matrix, and value transformation weight matrix of the $i$-th head, $\psi$ i (F t ) is the traffic pattern adjustment function corresponding to the $i$-th head, $F$ t represents the traffic flow characteristics at time $t$, $Q$ t is the query matrix at time $t$, $K$ t is the key matrix at time $t$, $V$ t is the value matrix at time $t$, $d$ k is the dimension of the key vector, used to scale the dot product attention, $\odot$ is the Hadamard product (element-wise multiplication), and $\text{softmax}()$ is the activation function used to normalize the attention weights. Then use the concatenation strategy to combine the outputs of the attention heads, and the finally output local adaptive graph is: where σ is a non-linear activation function, and α i (t) is the weight coefficient of the time-varying attention head; To effectively capture the complex and changing characteristics contained in the road network, a dynamic graph node generation module is proposed in the stacked spatio-temporal blocks to obtain a node dynamic graph with timestamp identification. First, the embedding vector and hidden state of the traffic data at the current moment are input into the dynamic graph node generation module. The embedding vector of this traffic data assigns trainable weights to the nodes through multi-head attention, and its output is aggregated with the hidden state. Then, spatial dependence relationship S is extracted through convolution operations eff , and then input into the Node2vec algorithm, which controls the random walk strategy through the "bias" parameter to capture the dynamic structural relationship between nodes, and then continues to optimize through iteration to generate a learnable node embedding matrix X embed , and then element-wise multiplication is used to generate node dynamic features, and its formula is as follows: Z flow = σ λ (S eff ⊙ X embed ) Among them, Z flow is the dynamic feature of the node, S eff is the learned passage space dependence relationship, X embed is the learnable node embedding matrix, and λ is the saturation rate hyperparameter of the activation function σ that is dynamically adjusted Furthermore, an Emergency-GraphSAGE sampling strategy is proposed. For emergency medical situations, priority sampling is performed on the neighbor information of the nodes processed by the Flow-Node2vec algorithm to update the dynamic feature representation of the nodes. Then, taking Z flow as the initial node feature representation, several nodes are sampled based on medical emergency relevance from the neighbors of all nodes, and the information of the neighbor nodes is extracted through an aggregation function. The embeddings of the neighbor nodes are weighted and averaged according to the relevance of different nodes to the emergency path, and then concatenated with the embedding of its own node, followed by a non-linear transformation. Finally, the self-multiplication method is used to update the dynamic representation of the nodes, thereby generating a global dynamic graph. Taking Z flow as the initial value, the formula for its generation is as follows: Among them, represents the embedding feature of neighbor node u in the previous layer, N flow represents the neighbor set of node v after sampling, W k represents the learnable weight matrix in the k-th layer, ρ(f vu ) represents the weight function of the correlation between nodes u and v, σ is the non-linear activation function, τ(P density ) is the dynamic adjustment function of the traffic density parameter P density , is the generated global dynamic graph of traffic flow.

7. The medical first aid system based on dynamic graph traffic prediction according to claim 1, characterized in that, In step S4, introduce an adaptive gating fusion graph convolution module with spectral domain enhancement, and through the decomposition and reconstruction in the spectral domain, optimize the characteristics of these two graphs at different frequency levels respectively, and extract the spatial correlation between the local adaptive graph and the global dynamic graph. The processing process is as follows: First, adopt a multi-level feature frequency division mechanism to decompose the local adaptive graph and the global dynamic graph in the spectral domain. Among them, represents the spectrum transformation function, Ψ loc and Ψ glo are respectively the sets of spectrum basis functions of the local and global graphs, Then, introduce an adaptive spectral gating unit to dynamically adjust the importance of different frequencies based on traffic changes. The formula is: Among them, B represents the number of frequency bands, and γ b (S t ) is the b-th frequency band importance coefficient adaptively generated based on the current traffic state S t , is the gating weight matrix of the b-th frequency band, and are the components of the local graph and the global graph in the b-th frequency band respectively, and Φ b (X, F t ) is the enhancement function for the b-th frequency band, is the corresponding bias term. Subsequently, use the inverse transformation mechanism from the spectral domain to the spatial domain to restore the spectral representation fused with the learnable reconstruction basis function to the spatial domain. The formula is: where denotes the inverse spectral transform, and Ω is a set of learnable reconstruction basis functions Finally, extract the spatial dependence characteristics of the fused graph. Among them, α j (F t ) is the weight coefficient of the j-th graph neural network that dynamically changes according to the current moment traffic state, GCN j represents the j-th graph neural network, Θ j is the corresponding set of learnable parameters, ResConnect() is a function for fusing the local adaptive graph and the global dynamic graph through residual connection, and β is a balance parameter that controls the size of the residual connection.

8. A medical first aid system based on dynamic graph traffic prediction according to claim 1, characterized in that, In step S5, the multi-scale temporal gated convolution module includes three GTU modules with different scale receptive fields. The GTU module is a gated time unit module. Stacking the gated time unit modules can expand the receptive field in the time dimension, and its final output is as follows: T out = ReLU(Concat(Pooling(ε1*T), Pooling(ε2*T), Pooling(ε3*T)) + T) Among them, ε1, ε2, and ε3 are convolutional kernels of 1×3, 1×5, and 1×7. T is the time information after being processed by tanh and sigmoid. Pooling is the pooling operation, Concat is the concatenation operation, and ReLU(·) is the non-linear activation function.

9. The medical first aid system based on dynamic graph traffic prediction according to claim 1, wherein, In step S7, the medical emergency center system loads a trained dynamic graph traffic prediction model into the emergency dispatch cloud platform to predict the status of the medical emergency channels for real-time regulation of ambulances. And different objective functions are adopted according to the patient's condition grading to adaptively adjust the medical emergency strategy. And a vital sign detection interface is configured to facilitate re-planning the transfer route when the patient's condition suddenly changes. And according to the traffic congestion level and the patient's condition, a three-level traffic control system is implemented. The three-level traffic control system is respectively signal light optimization, area control, and city-wide emergency response.

10. The medical first-aid system based on dynamic graph traffic prediction according to claim 9, wherein, The medical emergency center system loads a trained dynamic graph traffic prediction model into the emergency dispatch cloud platform to generate real-time traffic prediction results for several hours in the future. The medical emergency center system obtains multiple medical emergency channels based on the prediction results and adjusts the best emergency channels based on the locations of multiple hospitals, the urgency of the patient's condition, and the road congestion situation. Meanwhile, the prediction results are presented to the medical emergency command center through a dedicated visualization interface, which includes a traffic prediction map, a congested road section display map, and key information on the estimated round-trip time. In addition, to consider the characteristics of medical emergencies, an adaptive decision-making mechanism based on patient condition grading is used. In the patient condition grading strategy, the emergency situation of the patient is converted into an urgency parameter, and different objective functions are set according to different levels of patients. The processing process is as follows: min(T transport ) Red (critical) α·min(T transport )+(1-α)·max(M resource ) Yellow (Emergency) β·min(T transport )+(1-β)·max(M distribution ) Green (general) Among them, T transport is the transfer time, M resource is the matching degree of medical resources, M distribution represents the balance degree of medical resource distribution. α and β are balance parameters. Based on the objective function of the patient's condition, the medical emergency center system can adaptively adjust the decision-making priority according to the urgency of the patient: for extremely critical patients in red, the system will transfer the patient in the fastest way; for urgent patients in yellow, while ensuring a short transfer time, the system considers an appropriate matching degree of medical resources to ensure that the patient can receive more appropriate treatment; for green patients, the system focuses more on the allocation of medical resources In addition, considering that the patient's condition may suddenly change during the transfer process, a vital sign monitoring interface and a real-time path planning mechanism for patients are added: if ΔV vital > V threshold then replanning() Among them, ΔV vital is the degree of change in the patient's vital signs, and V threshold is a threshold matrix for measuring whether to re-plan. Different thresholds are set for diseases in different emergency situations. When sudden changes occur in vital signs, the medical emergency transfer strategy can be adjusted in a timely manner to improve the safety of the first aid process. When re-planning the emergency channel, to ensure the stability and safety of the transfer process, a path transition algorithm is used: P transition = argmin(p ∈ P current ∩ P new )(D(p, P current ) + D(p, P new ) + λ ·R(p)) Among them, P current is the current path, and P new is the re-planned path. D(p, P) is the distance metric from node p to path P, R(p) is the risk score of node p, and λ is the risk weight coefficient. Here, considering the optimality of medical channel transformation, road risk factors are taken into account to ensure the safety of the conversion location and avoid the conversion of the first-aid channel on dangerous sections. In addition, for medical emergency situations in special harsh environments, a multi-mode joint transfer strategy is added. For the selection algorithm of transfer points, methods of multi-mode joint transfer such as vehicles, ships, and helicopters will be considered: J optimal = argmax j∈J (Q access (j)·Q safety (j)·Q resource (j)·Q weather (j)) Among them, J is the set of all possible transfer locations, Q access is the accessibility score, Q safety is the safety score, Q resource is the resource remaining score, Q weather is the weather adaptability score, In addition, considering the impact of harsh weather on emergency transfer, road passage evaluation indicators under harsh weather conditions are set. C weather (r, w) = C normal (r)·φ(r, w) Among them, r is the road section, w is the weather condition, and C normal (r) is the traffic efficiency of road section r under normal weather, and φ(r, w) is the adverse weather impact coefficient. The formula is as follows: Among them, η i is the weight coefficient of the i-th kind of bad weather, WI i is the intensity index of the i-th kind of bad weather, S i (r) is the sensitivity of section r to the i-th kind of bad weather, which combines the characteristics of different sections with bad weather factors, so as to realize the channel optimization of the impact of bad weather. In addition, a three-level traffic control system is set up in the medical control center. Combining the predicted traffic conditions and the emergency conditions of patients' illnesses, corresponding-level traffic control systems are adopted. For green patients or low-traffic-jam sections, the optimized passage of medical vehicles is achieved by optimizing traffic lights and in-vehicle GPS systems. For yellow patients or medium-traffic-jam sections, regional traffic control is carried out, which includes temporary vehicle avoidance, adjustment of bus stops, and adjustment of bicycle parking areas. For red patients or high-traffic-jam sections, a city-wide emergency response is used, which includes traffic police guidance, road section closure, and reverse passage.

Images