Limited open space building explosive power field machine learning and escape planning method

Abstract

The invention discloses a limited open space building explosive power field machine learning and escape planning method, and the method comprises the steps: carrying out the training and optimization of a power field interaction cooperation graph neural network through a limited open space building power field database under the explosion impact, wherein the database comprises a training data set and a verification data set; and a finite open space building power field prediction model under explosion impact is obtained, and machine learning of a finite open space building three-dimensional explosive power field is realized. According to limited open space building power field prediction under explosion impact and historical test data comparison, a personnel damage field of a corresponding underground space building is obtained, a starting point and an ending point of personnel escape are obtained, and a path optimization method is utilized to determine the safety of the underground space building according to a mode meeting the shortest escape time and the maximum survival rate. The method provided by the invention is high in accuracy, high in efficiency and strong in generalization ability.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular to a machine learning and escape planning method for a limited open space building explosion power field with high prediction efficiency, strong generalization ability and high precision. Background Art

[0002] Limited open space refers to a building space that is neither completely open nor completely closed, such as tunnels, cabins, underground spaces, etc.

[0003] The explosion impact power field refers to the explosion effect field that is sufficient to cause damage or kill the target, such as peak overpressure field, peak impulse field, etc.

[0004] Force field prediction refers to predicting the important parameters of the force field under explosion impact, including impact load, overpressure, impulse, and arrival time.

[0005] 1. Impact load: Impact load refers to the change of force acting on an object over time, describing the short-term effect of force on the object. In the case of explosion shock waves, impact load usually refers to the short-term force exerted by the shock wave generated by the explosion on the structure or object. This load is characterized by high energy and short-term action, which can cause serious damage to the structure.

[0006] 2. Overpressure: Overpressure refers to the difference between the pressure on the front of the explosion shock wave and the surrounding atmospheric pressure. When an explosion occurs, the front peak of the shock wave will bring about a sharp increase in local pressure. This part that is higher than the atmospheric pressure is the overpressure. Overpressure is an important parameter to measure the intensity of the shock wave, which directly affects the degree of damage to buildings and biological targets caused by the explosion shock wave.

[0007] 3. Impulse: Impulse is the cumulative effect of force over time and is defined as the integral of force over time. In the context of explosion shock waves, impulse is usually used to describe the total effect of the force exerted by the explosion shock wave on the structure or object over time, and is used to evaluate the destructive potential of the explosion shock wave.

[0008] 4. Arrival time: Arrival time refers to the time when the explosion shock wave reaches a specific area or target. After the explosion, the shock wave needs a certain amount of time to propagate to the target area, and this time is the arrival time. The arrival time is very important for evaluating the impact of the explosion shock wave and formulating protective measures, because it determines when the target begins to be affected by the shock wave after the explosion.

[0009] The prediction of the force field in a confined open space under the impact of an explosion is the basis for safety protection design and an important prerequisite for evaluating the consequences of an explosion and formulating an emergency response plan.

[0010] At present, the prediction of the force field of a building in a confined open space under explosion impact is described in the paper "Current Status and Prospects of Research on Leakage and Explosion Characteristics of Hydrogen Energy Vehicles in Confined Spaces" (Chen Jiayan, Yang Juntao, He Qize, et al. Journal of Wuhan University of Technology, Information and Management Engineering Edition, 2024, 46(3): 382-386, 409. DOI: 10.3963 / j.issn.2095-3852.2024.03.006.), which discloses a method for determining the jet fire and leakage ignition explosion formed by the leakage of a hydrogen energy vehicle in a confined space, including: using the computational fluid dynamics numerical simulation method to study the jet fire behavior of hydrogen under different ventilation conditions, leakage angles, and leakage rates; using the results of numerical simulation to further predict the explosion impact force field on the inner surface of the confined space.

[0011] However, the above methods rely on complex simulation systems. Complex model calculations will greatly increase the calculation time and consume a lot of computing resources, which is very time-consuming and labor-intensive and not efficient enough. On the other hand, the many assumptions made in numerical simulations mean that they can only predict the explosion impact response under specific working conditions, and their generalization ability is weak.

[0012] For example, the paper "Real-time simulation method of explosion flow field based on fusion of physical model analysis and deep neural network" (Zhou Shennan, Wang Zhongqi, Li Zhongqi. Journal of Safety and Environment, 2024, 24(5): 1681-1690. DOI: 10.13637 / j.issn.1009-6094.2023.0758.) discloses a machine learning method for the three-dimensional explosion force field of limited open space buildings based on deep learning, including: using computational fluid dynamics numerical simulation methods to simulate and analyze a variety of typical explosion scenes to realize the construction of data sets, using convolutional variational autoencoder models to map high-dimensional features to latent space variables and reduce the dimensionality of data sets; through a multi-layer feedforward neural network model, a mapping relationship between explosion-related parameters, flow field observation time and latent variables is established to realize numerical simulation. The model is used for regression; the latent variables output by the multi-layer feedforward neural network model are fed back to the convolutional variational autoencoder model, which reconstructs or confirms the corresponding regional explosion pressure field to achieve data generation; the trial and error method is used to tune the hyperparameters related to the model architecture one by one, and a basic deep neural network prediction model is formed through a progressive training method; the prediction results are analyzed by the relative L2 norm error and compared with the numerical simulation results to improve the prediction accuracy and real-time performance of the prediction model; combined with the real-time data provided by the monitoring system, the established model is used to efficiently obtain the regional explosion risk analysis results and realize real-time prediction.

[0013] However, the neural network prediction model used in the above method is limited to specific types of explosion scenarios and conditions, lacks diverse and representative data sets, and the deep learning neural network used in the above method only relies on a multi-layer feedforward neural network, and the prediction accuracy is not high.

[0014] In summary, the problems existing in the existing technology are: the efficiency of predicting the power field of a building in a limited open space under an explosion impact is not high enough, the generalization ability is not strong enough, and the accuracy is not high enough. Summary of the invention

[0015] The purpose of the present invention is to provide a machine learning method for three-dimensional explosion power field of limited open space buildings, which has high efficiency, strong generalization ability and high precision.

[0016] The technical solution to achieve the purpose of the present invention is: In a first aspect, a machine learning method for a three-dimensional explosion force field of a limited open space building is provided, comprising: The power field database of limited open space buildings under explosion impact, including training data sets and verification data sets, is used to train and optimize the power field interactive collaborative graph neural network, and a prediction model of the power field of limited open space buildings under explosion impact is obtained, so as to realize machine learning of the three-dimensional explosion power field of limited open space buildings; wherein the power field includes peak overpressure, peak impulse, and shock wave arrival time; The Power Field Interactive Collaborative Graph Neural Network includes an encoder, a processor, and a decoder, among which: The encoder includes a cascaded multi-head self-attention layer, a first feedforward neural network layer, a residual connection and a layer normalization component, wherein the first feedforward neural network layer includes a first fully connected layer and N identical first sub-layers, each first sub-layer includes a second fully connected layer with a front-end cascaded ReLU activation function, the input of each first sub-layer is the output of the multi-head self-attention layer, and the output of each sub-layer is aggregated to the first fully connected layer through a ReLU activation function; The processor includes a TransGNN processor, a first convolutional layer, a multi-path network structure layer, a third fully connected layer, a pooling layer, and a readout layer, wherein the multi-path network structure layer includes a cascaded graph attention layer, a second feedforward neural network layer, and a RefineNet multi-path network structure, and the readout layer includes a cascaded global pooling layer and a fourth fully connected layer; wherein the second feedforward neural network layer and the first feedforward neural network layer have the same structure; The decoder includes multiple layers of masked self-attention layers, N second identical second sub-layers, stacking and normalization components, and a generating network layer, wherein each second sub-layer includes a second convolutional layer and a fifth fully connected layer connected by a ReLU activation function, and the input of each second convolutional layer is the output of the multiple layers of masked self-attention layers. The stacking and normalization components include cascaded stacking layers, softmax functions, normalization layers, and residual connection layers, and the output of the residual connection layer and the output of all fifth fully connected layers are used as inputs of the stacking layer.

[0017] As a further optimization scheme of the present invention, physical information constraints are embedded in the force field interactive collaborative graph neural network, including: Using the random number method, random sample data were extracted for the three-dimensional explosion of limited open space buildings in the near field, mid-field, and far field according to the proportional distance, and the empirical relationship between peak overpressure, peak impulse, shock wave arrival time and proportional distance was fitted: , , , Where, I + represents the true value of the peak impulse, P represents the true value of the peak overpressure, t + represents the true value of the shock wave arrival time, A1, B1, C1, D1 represent the fitting parameters, Z represents the proportional distance, A2 represents the parameter of the peak impulse, A3 represents the parameter of the shock wave arrival time, and W represents the TNT equivalent; According to the empirical relationship between the near field, midfield and far field at different proportional distances, the empirical relationship is embedded into the loss function of the graph neural network as a constraint condition to realize the physical information constraint embedding of the graph neural network.

[0018] As a further optimization scheme of the present invention, after embedding the physical information constraint, the loss function of the force field interactive collaborative graph neural network obtained adds the physical information constraint loss term L physics : , , , , Where M represents the number of nodes, , and represent the peak overpressure constraint loss term, the peak impulse constraint loss term and the shock wave arrival time constraint loss term, respectively. Indicates the predicted peak overpressure of node j and the actual value of the peak overpressure of node j , represents the proportional distance of node j, A1, B1, C1, D1 represent the fitting parameters, represents the predicted peak impulse of node j, W represents the TNT equivalent, It indicates the predicted arrival time of the shock wave at node j, A3 indicates the parameter of the shock wave arrival time, and dt indicates the infinitesimal element.

[0019] As a further optimization scheme of the present invention, the steps of establishing a database of the force field of a limited open space building under explosion impact include: Through the digital twin of real limited open space buildings and environments, the explosion equivalent and explosion center are randomly generated to construct explosion events; According to the explosion event, a numerical simulation method is used to establish a numerical model of the explosion event power field and obtain the explosion event power field data; The force field data of explosion events are preprocessed to establish a sample set. Combined with the selected explosion features, a force field database of limited open space buildings under explosion impact is established, including a training data set and a verification data set.

[0020] As a further optimization scheme of the present invention, the explosion event construction step includes: Through data crawling, we can obtain the data source of the real limited open space building and the surrounding environment map, and build a 3D model of the limited open space building and the surrounding environment through the building 3D digital twin method; Preset the range of explosion equivalent and explosion center, generate the explosion equivalent and explosion center by random number method, combine the data according to the explosion event number, and obtain the explosion equivalent and explosion center under the explosion event number; The explosion event is constructed by matching the explosion center and explosion equivalent under the same explosion event number with the three-dimensional model of the limited open space building and the surrounding environment.

[0021] As a further optimization scheme of the present invention, the steps of establishing a database of the force field of a limited open space building under explosion impact, including a training data set and a verification data set, include: By completing missing values, adjusting amplitudes, and transforming spatial correlation, the power field data of explosion events are converted into a two-dimensional / three-dimensional matrix form, thus realizing the preprocessing of the power field data of explosion events. Through the reconstruction simulation of the explosion impact pressure field distribution based on the limited measurement point data, the explosion impact power field parameters are digitally reconstructed to build a sample set. The explosion impact power field parameters include peak overpressure, peak impulse, and shock wave arrival time. Through flow field analysis and numerical simulation, the characteristic parameters of the explosion impact force field parameters are obtained; The characteristic parameters of the explosion impact power field parameters are combined with explosion events to construct a power field database of limited open space buildings under explosion impact, and the power field database of limited open space buildings under explosion impact is divided into a training data set and a verification data set.

[0022] In a second aspect, a method for planning explosion escape in a limited open space building is also provided, comprising: Using the machine learning method described above, the prediction results of the force field of a building in a limited open space under the impact of an explosion are obtained; Calculate the percentage of personnel casualties based on the force field prediction results; A weighted directed graph is used to represent the three-dimensional explosion scene of a limited open space building, and the percentage of personnel damage is used as the attribute of the node in the weighted directed graph, that is, the degree of injury of the personnel at the node; Based on the Dijkstra algorithm, the optimal escape path is found and the escape planning is completed with the constraints that the attributes of any node on the planned path do not exceed the preset moderate damage level and the survival rate is maximized and the escape time is shortest.

[0023] As a further optimization scheme of the present invention, the steps for calculating the percentage of personnel damage include: Using the nonlinear relationship data fitting method, the empirical relationship between the percentage of personnel damage and the peak overpressure, peak impulse and shock wave arrival time is obtained: , in, is the percentage of personnel casualties, and E1, F1, and G1 are fitting parameters.

[0024] In a third aspect, a computer-readable storage medium storing one or more programs is also provided, wherein the one or more programs include instructions, which, when executed by a computing device, enable the computing device to perform the machine learning method as described above.

[0025] In a fourth aspect, an electronic device is also provided, comprising one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the machine learning method as described above.

[0026] Compared with the prior art, the present invention has the following significant advantages: 1. Higher analysis and prediction efficiency: The present invention constructs a physical information embedded graph neural network to establish a prediction model for the power field of a limited open space building under explosion impact, directly processes the three-dimensional physical field calculation of the power field of a limited open space building under explosion impact, avoids the traditional numerical format derivation and solution process to improve the calculation efficiency, and realizes efficient response prediction of the power field under explosion impact in a limited open space; 2. Stronger generalization ability: The present invention establishes a database of the power field of limited open space buildings under explosion impact with random regional characteristics, random facility characteristics, and random explosion characteristics, thereby ensuring the diversity of the database and the robustness of the neural network, and realizing efficient generalization prediction of the power field of limited open space buildings under explosion impact (such as peak overpressure, peak impulse, shock wave arrival time, etc.); 3. Higher analysis accuracy: By combining graph neural network and physical information embedding methods, the ability of three-dimensional force field information transmission is enhanced, and interactive collaborative graph neural network processing of explosion power field prediction is realized to achieve high-precision prediction of the physical problems of power field of limited open space buildings under explosion impact.

[0027] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is the main flow chart of the limited open space building explosion power field machine learning and escape planning method of the present invention.

[0029] Figure 2 It is a schematic diagram of the encoder structure.

[0030] Figure 3 It is a schematic diagram of the processor structure.

[0031] Figure 4 It is a schematic diagram of the decoder structure.

[0032] Figure 5 It is a flowchart of escape planning. DETAILED DESCRIPTION

[0033] like Figure 1 As shown, the limited open space building explosion force field machine learning and escape planning method of the present invention includes the following steps: S10. Establishment of explosion impact force field database: construct explosion events, obtain explosion event force field data, establish sample sets, select explosion features, and construct a limited open space building force field database under explosion impact including a training data set and a verification data set.

[0034] The steps of establishing the explosion impact force field database include: S11. Explosion event construction: Through the digital twin of real limited open space buildings and environments, the explosion equivalent and explosion center are randomly generated to construct explosion events.

[0035] The explosion event construction step comprises: S111. Digital twin of real limited open space building and environment: obtain the data source of real limited open space building and surrounding environment map through data crawling, and construct a three-dimensional model of limited open space building and surrounding environment through the building three-dimensional digital twin method.

[0036] S112, generation of explosion equivalent and explosion center: preset the range of explosion equivalent and explosion center, generate the explosion equivalent and explosion center by random number method, combine data according to the explosion event number, and obtain the explosion equivalent and explosion center under the explosion event number.

[0037] S113. Explosion event construction: by matching the explosion center and explosion equivalent under the same explosion event number with the limited open space building and the three-dimensional model of the surrounding environment, the explosion event is constructed; the limited open space building node, explosion equivalent and explosion center are defined as input data.

[0038] S12. Acquisition of explosion power field data: Based on the constructed explosion event, a numerical model of the explosion power field is established using a numerical simulation method, and grid nodes are obtained through simulation division to obtain explosion power field data; the peak overpressure, peak impulse and shock wave arrival time of a limited open space are defined as output data.

[0039] S13. Establishment of force field database: preprocess the force field data of the explosion event, splice the input data and the output data, establish a sample set, and combine the selected explosion characteristics to establish a limited open space building force field database under explosion impact including a training data set and a verification data set.

[0040] The power field database establishment step includes: S131. Data preprocessing: by completing missing values, adjusting amplitudes, and converting spatial correlations, for example, the negative overpressure values ​​are uniformly set to zero, and the explosion event power field data are converted into a two-dimensional / three-dimensional matrix form, thereby implementing the explosion event power field data preprocessing.

[0041] S132, sample set construction: by reconstructing the explosion impact pressure field distribution simulation based on the limited measurement point data, the explosion impact force field parameters are digitally reconstructed to construct a sample set, wherein the explosion impact force field parameters include peak overpressure, peak impulse, and shock wave arrival time.

[0042] S133. Explosion feature selection: Through flow field analysis and numerical simulation, characteristic parameters of explosion impact force field parameters are obtained.

[0043] S134, database construction: combining the characteristic parameters of the explosion impact power field parameters with the explosion event, constructing a database of the power field of limited open space buildings under explosion impact, and dividing the database of the power field of limited open space buildings under explosion impact into a training data set and a verification data set.

[0044] S20. Construction of a power field model of limited open space buildings under explosion impact: extract small sample data from the database, fit the empirical relationship between the peak overpressure in limited open space, the shock wave arrival time and the proportional distance, construct a physical information embedding equation based on the empirical relationship between the proportional distance, and combine it with the interactive collaborative graph neural network to construct an interactive collaborative graph neural network of the power field of limited open space buildings under explosion impact, use the database of the power field of limited open space buildings under explosion impact to train, evaluate and optimize it, and obtain a maturely trained power field model of limited open space buildings under explosion impact.

[0045] The steps of constructing the force field model of a limited open space building under explosion impact include: S21. Fitting of empirical relationship: using the random number method to extract random sample data of explosions in near field, mid field and far field according to proportional distance from the database, and obtain the empirical relationship between peak overpressure, peak impulse and shock wave arrival time and proportional distance in finite open space after fitting.

[0046] 1. Proportional distance: The proportional distance is obtained by dividing the explosion center distance and the TNT equivalent.

[0047] The explosion center distance refers to the distance between the explosion center and the node. It is the spatial coordinates of the known explosion center and the node. The explosion center distance can be obtained through calculation, and further comparison can be used to obtain the proportional distance.

[0048] 2. The empirical relationship between peak overpressure and proportional distance is given by the following formula (1): (1) Where P represents the peak overpressure, Z represents the proportional distance, and A1, B1, C1, and D1 represent the fitting parameters.

[0049] 3. The empirical relationship between peak impulse and proportional distance is given by the following formula (2): (2) In the formula, I + represents the peak impulse, W represents the TNT equivalent, Z represents the proportional distance, and A2 represents the parameter of the peak impulse.

[0050] 4. The empirical relationship between the shock wave arrival time and the proportional distance is given by the following formula (3): (3) Where, t + It represents the arrival time of the shock wave, W represents the TNT equivalent, Z represents the proportional distance, and A3 represents the parameter of the shock wave arrival time.

[0051] The relationship between the peak overpressure and peak impulse in a limited open space and the proportional distance of the shock wave arrival time is fitted according to the empirical relationships of formulas (1) to (3). The parameters of the above empirical relationships are obtained through data fitting, and the empirical relationships in the near field, mid field and far field are obtained.

[0052] S22. Construction of a graph neural network for power field interaction and collaboration: Build a graph neural network composed of an encoder, a processor, and a decoder, construct physical information constraints for the prediction of the explosion power field, and embed the empirical relationships as constraints into the loss function of the graph neural network according to the empirical relationships at different proportional distances corresponding to the near field, midfield, and far field, to achieve dynamic physical information embedding of the graph neural network, realize physical information integration, and build a graph neural network for power field interaction and collaboration.

[0053] The power field interactive collaborative graph neural network construction steps include: S221, encoder like Figure 2 As shown in the figure, the encoder part consists of a multi-head self-attention layer, a feedforward neural network layer, a residual connection and a layer normalization component. The multi-head self-attention layer connects each graph data feature; the feedforward neural network layer consists of a fully connected layer and N identical sub-layers. Each identical sub-layer interacts with information on a fully connected layer through the ReLU activation function, and then uses the ReLU activation function to aggregate to a fully connected layer to achieve information aggregation; residual connections and layer normalization are used to complete the processing of the feedforward neural network, helping the encoder capture the dependencies of all positions in the input sequence, obtain the potential interactive collaborative graph data, and output the interactive collaborative neural network code.

[0054] The input information is preprocessed by defining the structure of the input information, assigning features, and initializing the multi-head self-attention mechanism; through the multi-head self-attention layer, each "head" is given a set of linearly transformed weight matrices, and the attention weights between the self-attention "head" and the neighboring nodes are further calculated. Each head independently calculates the attention weight, updates the data features, and merges the results of multiple heads into one by splicing or averaging to complete the combination of multi-head attention; the ReLU activation function is used to complete the feature information processing on N fully connected layers; the ReLU activation function is used to combine the output information of N fully connected layers with the data processed by interactive collaborative convolution to achieve the aggregation of neighbor information on the new fully connected layer; by including a residual connection that prevents the value of the feature data of each layer of nodes from changing drastically, and a sublayer of a layer normalization component that realizes the integration of node feature data, the data input and output are completed on a sublayer, thereby improving the training speed of the model.

[0055] The above process gradually extracts and integrates the information in the graph structure through multi-head self-attention layers and feedforward network layers until the required encoding depth is reached. Through the coupling of multi-layer multi-head self-attention layers and feedforward neural networks, the latent graph data is obtained, and the encoding representation of each node is output to achieve interactive collaborative neural network encoding.

[0056] S222, Processor like Figure 3 As shown in the figure, the processor part consists of a TransGNN processor, a convolutional layer, a multi-path network structure layer, a fully connected layer, a pooling layer, and a readout layer. Using the TransGNN processor, feature aggregation is completed in the convolutional layer through the ReLU activation function. The multi-path network structure layer consists of a graph attention layer, a feedforward neural network layer, and a RefineNet multi-path network structure including a refinement module and a feature extraction module. The feedforward neural network layer of the multi-path network structure layer in the processor is consistent with the feedforward neural network layer structure of the encoder. The readout layer consists of a global pooling operation and a fully connected layer.

[0057] The encoding of the encoder output uses the TransGNN processor, and uses the Transformer to aggregate more relevant data feature information, expand the receptive field, and disentangle the edge information aggregation to improve the message passing of the GNN. Through position encoding, the GNN layer is used to fuse the structure into node attributes, capture the graph structure information, and realize the improvement of the Transformer in the graph data. The interactive collaborative graph neural network form is obtained by using the ReLU activation function; by performing feature aggregation in the convolution layer, the key feature information of the graph is retained; through the multi-path network structure layer, high-resolution semantic segmentation is performed to achieve graph neural network acceleration and enhance the ability of three-dimensional force field information transmission; through the fully connected layer and the pooling layer, the information exchange and feature conversion between nodes are completed; the explosion power field prediction is realized in the readout layer, and the embedding vectors of all nodes are aggregated by using global pooling operations (such as sum, average or maximum pooling), and then classified through a fully connected layer to obtain graph-level predictions. Through multiple iterations of potential graph data, the interactive collaborative graph neural network processing is completed.

[0058] 1) TransGNN Processor: TransGNN is a new type of graph neural network model that integrates Transformer and Graph Neural Network (GNN) layers alternately to enhance each other's capabilities. In the alternating integration, the GNN layer first processes the features, aggregates the information of neighboring nodes through the message passing mechanism, updates the data representation, and obtains the local graph structure information; using the data representation output by the GNN layer, the Transformer layer aggregates information from other parts of the graph through the self-attention mechanism to achieve global dependency capture and expand the receptive field of each node; using the GNN layer, the global information output from the Transformer layer back to the GNN layer is further updated, and the global context information provided by the Transformer layer is used to further achieve local structure information fusion; through the continuous iteration of the alternating integration process, in each iteration, the GNN layer and the Transformer layer will work alternately, provide information to each other, and continuously update the data representation to ensure that the model can focus on the most important nodes. The alternating integration method allows the model to switch between different layers, using the Transformer layer to broaden the receptive field and decouple information aggregation from the edge, thereby aggregating information from more related nodes and enhancing the message transmission ability of GNN; the GNN layer helps the Transformer layer perceive the graph structure information and obtain more relevant information of neighboring nodes. Through the TransGNN processor, the ability to learn three-dimensional physical information is enhanced.

[0059] 2) Multi-path network structure layer: Graph structure data processing is realized through the graph attention layer; feature conversion is realized through the feedforward neural network layer; RefineNet multi-path network structure technology including feature extraction module and refinement module is used to combine feature extraction and refinement processes in a parallel manner, retain the information in the original image, improve segmentation performance, perform high-resolution semantic segmentation, and realize graph neural network acceleration.

[0060] 3) Readout layer: By using global pooling operations (such as sum, average or maximum pooling), the embedding vectors of all nodes are aggregated, and then classified through a fully connected layer to complete graph-level prediction of complex graph data (such as graph classification) and realize interactive collaborative graph neural network processing.

[0061] S223, decoder like Figure 4 As shown in the figure, the decoder consists of multiple layers of masked self-attention layers, N identical layers, stacking and normalization components, and a generation network layer, and each identical layer contains a convolutional layer, an activation function, and a fully connected layer. Processing through these layers ensures that the decoder can take into account previous outputs when generating sequences and avoid the influence of future information. The final predicted part is obtained through the decoder.

[0062] Through multiple layers of masked self-attention layers, N identical sub-layers are used to aggregate neighbor node information features through convolutional layers, which are processed by activation functions and transmitted to the fully connected layer. The processed fully connected layer information is processed using superposition and normalization components to achieve decoding and complete node updates. The potential graph data is restored to the explosion impact force field data structure and recombined in the generation network layer to achieve data decoding.

[0063] 1) Multi-layer masked attention layer: The graph autoencoder represents the vertices of the graph as low-dimensional vectors, uses the encoder and decoder structure to learn the low-dimensional representation of the graph data, migrates the variational autoencoder to the graph field, learns the distribution (mean and variance) of the low-dimensional vector representation of the node through encoding (graph convolution), samples the vector representation of the node in the distribution, implements multi-layer masked self-attention layers, decodes the attention layer through link prediction, and reconstructs the graph.

[0064] 2) Convolutional layer: Aggregate neighbor node information features through convolution operations, and use convolution based on spectral decomposition and convolution based on node space transformation to update node representation. Node representation is extracted by stacking N graph convolution layers to achieve node classification and graph classification tasks.

[0065] 3) Generative network layer: Use the graph generation network as a building block, recombine the nodes and edges of the graph data according to certain rules, generate a target graph with specific properties and requirements, and realize the generation of graphs based on a set of observable graphs.

[0066] S224. Embedding physical information: constructing physical information constraints, fitting them through the empirical relationship, dynamically adding the corresponding physical information constraints after fitting to the loss of the interactive collaborative graph neural network of the explosion power field prediction, and obtaining the interactive collaborative graph neural network of the power field of limited open space buildings under the explosion impact embedded with dynamic physical information.

[0067] The following formula (4) gives the loss term Loss of the model. The data loss term L between the predicted value and the true value is obtained by MSE method: data and the physical information loss term L physics constitute: (4) In the formula, λ data , physics represent constraint weights respectively.

[0068] The step of embedding physical information comprises: S2241, Construction of physical information constraint loss term: By obtaining the force field parameters of the limited open space building under the explosion impact of the peak overpressure, peak impulse, and shock wave arrival time, the physical information constraint loss term is constructed to ensure that the model output conforms to the laws of physics; As shown in formulas (5)-(8), the physical information constraint loss term L physics The loss term L is constrained by the peak overpressure p , peak impulse constraint loss term L I , the shock wave arrival time constraint loss term L t The corresponding value is obtained by calculation: (5) (6) (7) (8) Where M represents the number of grid nodes for force field numerical simulation, represents the predicted peak overpressure, represents the proportional distance, represents the predicted peak impulse, Indicates the end time of the explosion, W indicates the TNT equivalent, Represents the predicted shock wave arrival time.

[0069] S2242, physical information constraint embedding: The proportional distance is obtained by the explosion equivalent and the distance from the explosion center, and the data is classified according to the near-field and far-field explosion types according to the size of the proportional distance. When embedding the physical information constraint, the empirical relationship formulas of the corresponding near-field, mid-field and far-field explosions are respectively embedded, and the data loss term is constructed. By adding dynamic physical information constraints in the loss term of the model, dynamic physical information constraint embedding classified by proportional distance is realized; S2243, physical information embedding: By dividing the explosion into near-field, mid-field, and far-field according to the proportional distance, dynamic physical information constraints are added to the loss term of the model to achieve dynamic physical information embedding. The loss term of the model consists of data loss term and physical information constraint loss term. Since the final prediction focuses on the power field of limited open space buildings under the impact of the explosion, the corresponding empirical relationship formulas for near-field, mid-field, and far-field explosions are embedded in the physical information constraint embedding to obtain the dynamic physical information constraint loss term and achieve dynamic embedding of physical information.

[0070] S23. Construction of a power field model: using the training data set to train the interactive collaborative graph neural network of the power field of the limited open space building under the explosion impact, and using the verification data set to optimize the interactive collaborative graph neural network of the power field of the limited open space building under the explosion impact, to obtain a power field model of the limited open space building under the explosion impact.

[0071] S231, model training and optimization: Adopt Adam algorithm, update model parameters, train the model in a supervised way, and optimize the model by adjusting hyperparameters.

[0072] The model parameters are randomly generated at the beginning of training, and the back propagation algorithm is needed to back propagate the value of the loss function and calculate the gradient of the model parameters relative to the loss value to update the model parameters. The selected back propagation algorithm is the Adam algorithm, which converts the pressure field in the three-dimensional explosion field into a gradient field, combines the characteristics of gradient descent and momentum optimization, and sets an appropriate transfer function to highlight the shock wave with good performance and convergence speed.

[0073] When performing hyperparameter tuning, you need to pay attention to a series of key hyperparameters, including the number of iterations, batch size, learning rate, and adjustment of the model structure.

[0074] S232. Construction of the power field model of limited open space buildings under explosion impact: Taking the power field parameters peak overpressure, peak impulse, and shock wave arrival time as evaluation targets, combined with relative error, the model is evaluated by bringing in the verification data set to obtain the power field model of limited open space buildings under explosion impact.

[0075] The present invention embeds physical information of the force field of limited open space buildings under explosion impact into an interactive collaborative graph neural network, alternately uses Transformer and GNN layers, enhances the ability of three-dimensional force field information transmission, and realizes interactive collaborative graph neural network processing for explosion force field prediction. It can not only ensure the model's learning ability for the physical problems of the force field of limited open space buildings under explosion impact, but also ensure the model's learning ability for three-dimensional unstructured data, thereby realizing efficient and universal prediction of the force field of limited open space buildings under explosion impact.

[0076] S30. Prediction of the power field of a building in a limited open space under explosion impact: Acquire real explosion event data through digital twin technology and explosion information collection, and use the real explosion event data as the input of the power field model of a building in a limited open space under explosion impact, and the output is the required power field of a building in a limited open space under explosion impact.

[0077] The step of predicting the force field of a limited open space building under explosion impact comprises: S31. Acquisition of real explosion event data: through data collection, the explosion point and explosion information are obtained, and the three-dimensional model of the on-site building and environment is established by using digital twin technology to obtain the real explosion event. Through 2 data processing and conversion, the real explosion event data is obtained; The step of acquiring real explosion event data comprises: S311, explosion information collection: by acquiring the explosion center and explosion equivalent, limited open space explosion information collection is realized to obtain limited open space explosion impact input parameters; S312, Digital Twin Acquisition: By acquiring the explosion location information and combining GIS technology, a three-dimensional model of the on-site building and environment is constructed to obtain the digital twin of the limited open space building; S313, data processing and conversion: converting and combining the above-mentioned limited open space explosion impact input parameters and the limited open space building digital twin according to data characteristics to achieve data processing; formatting the above-mentioned limited open space explosion impact input parameters and the limited open space building digital twin according to data types, standardizing the formatted data, and combining the standardized data according to the power field parameters of peak overpressure, peak impulse, and shock wave arrival time to obtain explosion input data; S314, obtaining real explosion event data: combining the explosion input data according to the explosion cases to obtain real explosion event data.

[0078] S32. Explosion impact force field prediction: The real explosion event data is input into a limited open space building force field model under explosion impact, and the output is the required limited open space building force field under explosion impact.

[0079] For example, the prediction results of the force field test set of limited open space buildings under explosion impact corresponding to the grid nodes divided by a numerical model simulation are expressed as [explosion case number, grid node number, peak overpressure, peak impulse, shock wave arrival time], which is [1,2500,5e6,1e3,2e-1]. Its content shows that in the explosion event with explosion case number 1, the peak overpressure of the unit with grid unit node number 2500 in the three-dimensional physical field is 5e6Pa, the impulse is 1e3Pa•s, and the shock wave arrival time is 2e-1s.

[0080] It should be noted that the scene in the present invention is saved in three-dimensional data formats such as stl, etc., and can be gridded, and grid nodes are obtained through grid division. The nodes in the graph data represent data nodes, which are based on grid nodes, but are also the same nodes, except that the features attached to the nodes will change.

[0081] The present invention also provides an emergency escape planning method for explosions in limited open space buildings: based on the prediction of the power field of limited open space buildings under the impact of the explosion, based on the comparison of historical test data, the personnel damage field of the corresponding underground space building is obtained, the starting point and end point of the personnel escape are obtained, and the path optimization method is used to achieve emergency escape planning in a way that satisfies the shortest escape time and the maximum survival rate. The emergency escape planning steps include: S41. Acquisition of personnel injury field: Compare the personnel injury levels under the explosion power field parameters of peak overpressure, peak impulse, and shock wave arrival time at different proportional distances in the existing limited open space, perform interpolation processing, and obtain the personnel injury percentage; use the data fitting method to obtain the calculation experience relationship between personnel injury and explosion power field parameters; obtain the explosion power field data based on the prediction of the explosion impact power field, substitute it into the constructed calculation experience relationship and apply it to obtain the personnel injury field.

[0082] At present, the research on the empirical relationship between personnel damage and explosion force field parameters is described in the paper "Research on Simplified Calculation of Structural Overpressure-Impulse Curve under Explosion Conditions" (Chen Junjie, Gao Kanghua, Sun Ao. [J]. Vibration and Shock, 2016, 35(13): 224-232. DOI: 10.13465 / j.cnki.jvs.2016.13.036.), which discloses a dynamic fitting formula of the curve: (9), In the formula, represents the equivalent pressure, represents the equivalent impulse, A and B represent the values ​​of the quasi-static asymptote and impulse asymptote respectively, C is a constant under the influence of load shape, and n is a parameter affected by the damping ratio ξ and the ductility ratio η. It can be seen from the formula that the PI curve can quickly evaluate the damage effect of building structures under blast loads by fitting overpressure and impulse.

[0083] Without considering the damping energy consumption, the formula is simplified. In order to better reflect the damage effect, the peak overpressure (P) and peak impulse (I + ) to obtain the empirical formula, which amplifies the impact on the damage percentage calculation. The shock wave arrival time is also introduced. This variable amplifies the effect of time on the damage percentage calculation, resulting in a new empirical relationship as follows: (10), The personnel injury field acquisition step comprises: S411. Classification of personnel damage field: Classify the damage field according to the peak overpressure, peak impulse and shock wave arrival time when the explosion occurs, and perform interpolation to obtain the percentage of personnel damage. According to the different degrees of personnel injuries, the personnel damage field is divided into the following different levels.

[0084] 1. Minor Injury Grade A (0% - 10%): The lowest level of injury, which may only involve superficial abrasions, minor burns or minor impact injuries.

[0085] 2. Mild Injury Grade B (10% - 30%): The severity of the injury has increased and requires medical treatment such as suturing, bandaging or short-term treatment, but usually does not lead to long-term disability.

[0086] 3. Moderate Injury Grade C (30% - 50%): The degree of injury is high, which may involve fractures, deep burns or internal organ damage. The patient cannot walk for a short period of time and requires a longer period of treatment and rehabilitation.

[0087] 4. Severe Injury Level D (50% - 80%): The degree of injury is severe and may result in serious physical damage, requiring urgent medical intervention and long-term treatment, and there may be a risk of long-term disability.

[0088] 5. Extreme Injury Level E (80% - 100%): The degree of injury is extremely high and may result in life-threatening injuries, such as severe internal organ damage, extensive burns or extreme body trauma, requiring immediate medical assistance and possible long-term care.

[0089] S412. Data fitting of personnel casualties: From historical test data, it can be seen that the influence of peak overpressure, peak impulse and shock wave arrival time on personnel casualties is a nonlinear relationship. The nonlinear relationship data fitting method is used. The peak overpressure P is subtracted from the constant parameter E1, and then multiplied by the peak impulse Subtract the constant parameter F1 and multiply it by the shock wave arrival time Subtract the constant parameter G1, and the empirical formula is used to obtain the personnel injury field . The result is a percentage value, which is used to describe the degree of damage that the shock wave may cause to personnel under specific conditions. Considering the relationship between the percentage of personnel damage (Y) and the peak overpressure (P), peak impulse (I + ) and the shock wave arrival time (t + ), the empirical relationship between personnel damage and explosion power field parameters is obtained as shown in the following formula (11): (11) Among them, Y is in the form of percentage, and the symbol is %. is a function of P, I and t, and can be expressed in the form of the following multiplication formula (12): (12) Among them, E1, F1, and G1 are constants, which need to be obtained by polynomial fitting through historical data of personnel damage and test data. P represents the peak overpressure, and I + represents the peak impulse, Indicates the shock wave arrival time.

[0090] By comparing with historical test data, the parameters of peak overpressure, peak impulse, and shock wave arrival time are input, and the percentage of personnel casualties is output to achieve data fitting of personnel injury situations.

[0091] S413. Obtaining the personnel injury field: By predicting the explosion power field, the parameters of the power field, namely, peak overpressure (P), peak impulse (I) and shock wave arrival time (t), are obtained, and the corresponding parameters are substituted into the calculation empirical relationship to further obtain the personnel injury field.

[0092] S42. Escape path optimization: The problem is defined as a shortest path problem with multiple objective constraints, with the requirements that casualties at the waypoints do not exceed a moderate injury level (which can be set to 50%), the shortest escape time and the maximum survival rate being used as multiple constraints for path optimization design.

[0093] The steps of optimizing the escape route include: S421. Model establishment: Use a weighted directed graph in graph theory to represent the environment. Each node of the graph data represents a location, and each edge of the graph data represents a path from one location to another. The weight of each edge of the graph data represents the cost (such as time, distance, etc.) of going from one location to another. Each node and edge of the graph data has corresponding attributes, such as injury level, escape time, and survival rate. For example, each node has an injury level attribute, which indicates the injury level of the person at that location.

[0094] S422. Constraint setting: By setting the injury level of any node on the path not to exceed the moderate injury level as the overall constraint, using the rule that the total time to complete the path should be as short as possible and the maximum survival rate on the path as the arrangement constraint, the constraint setting is completed.

[0095] Maximum survival rate: Substitute the personnel injury field into the established model to calculate and obtain the survival rate of a single node. If the calculated value Y is greater than the moderate injury level, the node is severely damaged and is not included in the overall survival rate calculation; by calculating the percentage of node damage on the entire path, the corresponding node damage level is obtained, and all nodes that meet the damage level not exceeding the moderate damage level are obtained. By calculating the proportion of all characteristic nodes that meet the requirements on the entire path, the maximum survival rate is obtained through weighted calculation. The obtained survival rate is used as a factor in the final arrangement and selection of the path. For example, the weight of the edge of the graph data can be adjusted to reflect the survival rate, so that the algorithm tends to choose a path with a higher survival rate.

[0096] S423, path optimization algorithm: Use Dijkstra algorithm to process weighted graphs. The flow chart of path optimization is as follows: Figure 5 As shown, by initializing the distance according to satisfying all constraints, the distance corresponding to each node is obtained; by creating a set of unvisited nodes, a check step is added to the algorithm to check whether the injury level of the new node exceeds the moderate injury level. If the injury level exceeds the moderate injury level, the node is not added to the path to implement the constraint of the injury level, ensuring that the injury level of each node on the path does not exceed the moderate injury level; according to the constraints, update until all nodes are visited to obtain all paths that meet the overall constraints; according to the arrangement rules, the paths that meet the arrangement constraints are arranged using the requirements of the shortest total time based on the path and the maximum survival rate. After comprehensive consideration, the best path from the starting point to the end point that meets all constraints is found to achieve the optimization of the escape path.

[0097] S43. Emergency escape planning: With the help of digital twin technology, the underground limited open space building is digitized, and rapid emergency escape planning is achieved based on the path optimization method.

[0098] Based on the same technical solution, the present invention also discloses a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, enable the computing device to execute the above-mentioned limited open space building explosion force field machine learning and escape planning method.

[0099] Based on the same technical solution, the present invention also discloses a computing device, including one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the above-mentioned limited open space building explosion force field machine learning and escape planning method.

[0100] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0101] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0102] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0103] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0104] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A machine learning method for three-dimensional explosion force field of limited open space buildings, characterized in that: include: The power field database of limited open space buildings under explosion impact, including training data sets and verification data sets, is used to train and optimize the power field interactive collaborative graph neural network, and a prediction model of the power field of limited open space buildings under explosion impact is obtained, so as to realize machine learning of the three-dimensional explosion power field of limited open space buildings; wherein the power field includes peak overpressure, peak impulse, and shock wave arrival time; The Power Field Interactive Collaborative Graph Neural Network includes an encoder, a processor, and a decoder, among which: The encoder includes a cascaded multi-head self-attention layer, a first feedforward neural network layer, a residual connection and a layer normalization component, wherein the first feedforward neural network layer includes a first fully connected layer and N identical first sub-layers, each first sub-layer includes a second fully connected layer with a front-end cascaded ReLU activation function, the input of each first sub-layer is the output of the multi-head self-attention layer, and the output of each sub-layer is aggregated to the first fully connected layer through a ReLU activation function; The processor includes a TransGNN processor, a first convolutional layer, a multi-path network structure layer, a third fully connected layer, a pooling layer, and a readout layer, wherein the multi-path network structure layer includes a cascaded graph attention layer, a second feedforward neural network layer, and a RefineNet multi-path network structure, and the readout layer includes a cascaded global pooling layer and a fourth fully connected layer; wherein the second feedforward neural network layer and the first feedforward neural network layer have the same structure; The decoder includes multiple layers of masked self-attention layers, N second identical second sub-layers, stacking and normalization components, and a generating network layer, wherein each second sub-layer includes a second convolutional layer and a fifth fully connected layer connected by a ReLU activation function, and the input of each second convolutional layer is the output of the multiple layers of masked self-attention layers. The stacking and normalization components include cascaded stacking layers, softmax functions, normalization layers, and residual connection layers, and the output of the residual connection layer and the output of all fifth fully connected layers are used as inputs of the stacking layer.

2. The machine learning method according to claim 1, characterized in that Embed physical information constraints in the force field interactive collaborative graph neural network, including: Using the random number method, random sample data were extracted for the three-dimensional explosion of limited open space buildings in the near field, mid-field, and far field according to the proportional distance, and the empirical relationship between peak overpressure, peak impulse, shock wave arrival time and proportional distance was fitted: ， ， ， Where, I + represents the true value of the peak impulse, P represents the true value of the peak overpressure, t + represents the true value of the shock wave arrival time, A1, B1, C1, D1 represent the fitting parameters, Z represents the proportional distance, A2 represents the parameter of the peak impulse, A3 represents the parameter of the shock wave arrival time, and W represents the TNT equivalent; According to the empirical relationship between the near field, midfield and far field at different proportional distances, the empirical relationship is embedded into the loss function of the graph neural network as a constraint condition to realize the physical information constraint embedding of the graph neural network.

3. The machine learning method according to claim 1, characterized in that After embedding the physical information constraint, the loss function of the resulting force field interactive collaborative graph neural network adds the physical information constraint loss term L physics : ， ， ， ， Where M represents the number of nodes, , and represent the peak overpressure constraint loss term, the peak impulse constraint loss term and the shock wave arrival time constraint loss term, respectively. Indicates the predicted peak overpressure of node j and the actual value of the peak overpressure of node j , represents the proportional distance of node j, A1, B1, C1, D1 represent the fitting parameters, represents the predicted peak impulse of node j, W represents the TNT equivalent, It indicates the predicted arrival time of the shock wave at node j, A3 indicates the parameter of the shock wave arrival time, and dt indicates the infinitesimal element.

4. The machine learning method according to claim 1, characterized in that The steps for establishing the force field database of limited open space buildings under explosion impact include: Through the digital twin of real limited open space buildings and environments, the explosion equivalent and explosion center are randomly generated to construct explosion events; According to the explosion event, a numerical simulation method is used to establish a numerical model of the explosion event power field and obtain the explosion event power field data; The force field data of explosion events are preprocessed to establish a sample set. Combined with the selected explosion features, a force field database of limited open space buildings under explosion impact is established, including a training data set and a verification data set.

5. The machine learning method according to claim 4, characterized in that The steps to build an explosion event include: Through data crawling, we can obtain the data source of the real limited open space building and the surrounding environment map, and build a 3D model of the limited open space building and the surrounding environment through the building 3D digital twin method; Preset the range of explosion equivalent and explosion center, generate the explosion equivalent and explosion center by random number method, combine the data according to the explosion event number, and obtain the explosion equivalent and explosion center under the explosion event number; The explosion event is constructed by matching the explosion center and explosion equivalent under the same explosion event number with the three-dimensional model of the limited open space building and the surrounding environment.

6. The machine learning method according to claim 4, characterized in that The steps for establishing the force field database of limited open space buildings under explosion impact, including training data sets and verification data sets, include: By completing missing values, adjusting amplitudes, and transforming spatial correlation, the power field data of explosion events are converted into a two-dimensional / three-dimensional matrix form, thus realizing the preprocessing of the power field data of explosion events. Through the reconstruction simulation of the explosion impact pressure field distribution based on limited measurement point data, the explosion impact power field parameters are digitally reconstructed to build a sample set. The explosion impact power field parameters include peak overpressure, peak impulse, and shock wave arrival time. Through flow field analysis and numerical simulation, the characteristic parameters of the explosion impact force field parameters are obtained; The characteristic parameters of the explosion impact power field parameters are combined with explosion events to construct a power field database of limited open space buildings under explosion impact, and the power field database of limited open space buildings under explosion impact is divided into a training data set and a verification data set.

7. A method for planning escape from explosion in a limited open space building, characterized in that: include: Using the machine learning method as described in any one of claims 1 to 6, a prediction result of the force field of a building in a limited open space under the impact of an explosion is obtained; Calculate the percentage of personnel casualties based on the force field prediction results; A weighted directed graph is used to represent the three-dimensional explosion scene of a limited open space building, and the percentage of personnel damage is used as the attribute of the node in the weighted directed graph, that is, the degree of injury of the personnel at the node; Based on the Dijkstra algorithm, the optimal escape path is found and the escape planning is completed with the constraints that the attributes of any node on the planned path do not exceed the preset moderate damage level and the survival rate is maximized and the escape time is shortest.

8. The emergency escape planning method according to claim 7, characterized in that: The steps for calculating the casualty percentage include: Using the nonlinear relationship data fitting method, the empirical relationship between the percentage of personnel damage and the peak overpressure, peak impulse and shock wave arrival time is obtained: ， in, is the percentage of personnel casualties, E1, F1, G1 are fitting parameters, I + represents the true value of the peak impulse, P represents the true value of the peak overpressure, t + Indicates the true value of the shock wave arrival time.

9. A computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, characterized in that: When the instructions are executed by a computing device, the computing device performs the machine learning method as described in any one of claims 1-6.

10. An electronic device, characterized in that: The method comprises one or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the machine learning method as described in any one of claims 1 to 6.

Images