Emergency situation intelligent decision-making method, device and system based on deep learning

Abstract

The invention discloses an emergency situation intelligent decision-making method, device and system based on deep learning, and the method comprises the steps: firstly collecting multi-modal data of an affected area, carrying out the fusion processing of the multi-modal data into a unified embedded vector, carrying out the time and space dimension feature extraction of the unified embedded vector through a space-time perception neural network, and constructing a state sequence containing a space-time evolution characteristic; constructing a dynamic causal graph based on the sequence to identify a potential causal relationship, and predicting a future emergency situation development trend according to the potential causal relationship; and finally, generating an emergency decision strategy accurately adaptive to the actual condition of the affected area in combination with historical disposal experience and the current resource constraint condition. According to the invention, accurate perception of disaster situations, evolution prediction and intelligent decision generation under resource constraint can be realized.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device and system for intelligent decision-making on emergency situations based on deep learning. Background Art

[0002] In the field of emergency management, a typical approach in existing technologies is to build disaster situation awareness systems based on data from a single source, such as remote sensing imagery, ground sensors, or social media text. For example, some systems characterize disaster situations using manually designed feature extraction rules (such as pixel change rates in disaster areas and public opinion keyword frequency statistics) and generate response instructions based on preset logical judgment conditions. Another approach employs traditional machine learning models (such as support vector machines or shallow neural networks) to predict resource scheduling plans by pattern matching static features of historical disaster cases. Furthermore, recent research has attempted to introduce time series models (such as LSTM) or spatial graph networks to model disaster states. However, due to their lack of deep integration capabilities for multimodal and heterogeneous data, dynamic coupled analysis of the evolution of disaster events is difficult. Furthermore, existing systems typically rely on expert knowledge bases or static causal graphs for causal reasoning, unable to dynamically modify the correlations of disaster chain reactions based on real-time situation data. This results in low confidence in predictions of future emergency scenarios. Furthermore, the generation of final decision-making strategies often ignores the coordinated optimization of historical response experience and resource constraints, making it difficult to meet the dynamic response requirements in complex disaster scenarios.

[0003] Therefore, existing technologies make it difficult to achieve accurate perception of disaster situations, evolution prediction, and intelligent decision-making under resource constraints, which affects the effectiveness of emergency management. Summary of the Invention

[0004] The embodiments of the present invention provide a method, device and system for intelligent decision-making on emergency situations based on deep learning, which can realize accurate perception of disaster situations, evolution prediction and intelligent decision-making under resource constraints, effectively improving the effect of emergency management.

[0005] An embodiment of the present invention provides an intelligent decision-making method for emergency situations based on deep learning, comprising:

[0006] The multimodal data from the disaster-stricken areas are fused to generate a unified representation of the multimodal emergency situation embedding vector;

[0007] Extract features from the multimodal emergency situation embedding vector using a spatiotemporal perception neural network model, model the state of the emergency situation in the time series dimension and the spatial correlation dimension, and output a state sequence of the emergency situation containing spatiotemporal evolution characteristics;

[0008] Using the emergency situation state sequence to construct a dynamic causal diagram to identify potential causal relationships between disaster events and influencing factors;

[0009] Predicting the development trend of the future emergency situation in the disaster-stricken area based on the potential causal relationship, and generating a causal evolution result of the emergency situation in the disaster-stricken area;

[0010] Based on the causal evolution results of the emergency situation, and combined with historical handling experience data and current resource constraints, an emergency decision-making strategy for the disaster-stricken area is generated.

[0011] As an improvement to the above solution, the fusion processing of multimodal data from the disaster-stricken area to generate a unified representation of the multimodal emergency situation embedding vector includes the following sub-steps:

[0012] Perform feature extraction on satellite image data, meteorological data, and public opinion text data from the multimodal data of the disaster-stricken area to obtain spatial features of the satellite image data, change features of the meteorological data, and semantic features of the public opinion text data;

[0013] A cross-modal attention mechanism is used to calculate the association weights between the spatial features of the satellite image data, the change features of the meteorological data, and the semantic features of the public opinion text data to generate an inter-modal association matrix;

[0014] The spatial features, the change features and the semantic features are weightedly fused based on the inter-modal association matrix, and a unified multi-modal emergency situation embedding vector is generated through dimensionality reduction processing.

[0015] As an improvement to the above solution, the feature extraction of the multimodal emergency situation embedding vector is performed by using a spatiotemporal perception neural network model, modeling is performed in the time series dimension and the spatial correlation dimension respectively, and outputting an emergency situation state sequence containing spatiotemporal evolution characteristics, including the following sub-steps:

[0016] The multimodal emergency situation embedding vector is input into a temporal convolutional network to extract the time series characteristics of the disaster diffusion speed and resource consumption rate;

[0017] Based on the geographic grid division of the disaster-stricken area, a spatial topology map is constructed, and the spatial correlation characteristics of the disaster impact between regions are extracted through the graph attention network;

[0018] Performing tensor splicing on the time series features and the spatial correlation features to generate a spatiotemporal joint feature vector;

[0019] The spatiotemporal joint feature vectors are serialized and reorganized to output an emergency situation state sequence representing the spatiotemporal evolution law of the disaster.

[0020] As an improvement to the above solution, the method of constructing a dynamic causal graph using the emergency situation state sequence to identify the potential causal relationship between disaster events and influencing factors includes the following sub-steps:

[0021] Mapping each dimension of the emergency situation state sequence into a node of a causal graph, and initializing causal strength parameters between the nodes;

[0022] Based on the causal strength parameters between nodes, the causal relationship between each node in the causal graph is analyzed to determine the causal edge;

[0023] Generate a dynamic causal graph based on the causal edges between each node;

[0024] Based on the dynamic causal diagram, the potential causal relationship between disaster events and influencing factors is identified.

[0025] As an improvement to the above solution, the method of predicting the development trend of the future emergency situation in the disaster-stricken area based on the potential causal relationship and generating the causal evolution result of the emergency situation in the disaster-stricken area includes the following sub-steps:

[0026] Extracting a node causal association matrix based on the potential causal relationship as prior causal knowledge;

[0027] Inputting the emergency situation state sequence and the causal association matrix into a pre-trained time series prediction model to perform multi-step trend prediction and obtain a multi-step trend prediction result;

[0028] Based on the multi-step trend prediction results, the situation evolution path of the disaster-stricken area under different emergency measures is simulated to generate the causal evolution results of the emergency situation in the disaster-stricken area.

[0029] As an improvement to the above solution, the generation of an emergency decision-making strategy for the disaster-stricken area based on the causal evolution results of the emergency situation and in combination with historical handling experience data and current resource constraints includes the following sub-steps:

[0030] Obtain historical disposal experience data including historical emergency plans, resource consumption data and disposal effects;

[0031] Convert the current number of available rescue resources, team distribution locations, and traffic network status into resource constraints;

[0032] Generate an initial emergency decision-making strategy through a pre-trained deep learning model, taking the causal evolution results of the emergency situation, the historical handling experience data and the resource constraints as input;

[0033] Based on the causal logic of the dynamic causal graph, the secondary disaster risks potentially caused by the initial decision-making strategy are verified, and the final emergency decision-making strategy for the disaster-stricken area is output.

[0034] Another embodiment of the present invention provides an intelligent decision-making device for emergency situations based on deep learning, comprising:

[0035] The data fusion module is used to fuse the multimodal data from the disaster-stricken areas and generate a unified multimodal emergency situation embedding vector;

[0036] A feature extraction module is used to extract features from the multimodal emergency situation embedding vector through a spatiotemporal perception neural network model, modeling in the time series dimension and the spatial correlation dimension respectively, and outputting an emergency situation state sequence containing spatiotemporal evolution characteristics;

[0037] A causal identification module is used to construct a dynamic causal graph using the emergency situation state sequence to identify potential causal relationships between disaster events and influencing factors;

[0038] A prediction module, configured to predict the development trend of the future emergency situation in the disaster-stricken area based on potential causal relationships, and generate causal evolution results of the emergency situation in the disaster-stricken area;

[0039] The strategy generation module is used to generate an emergency decision-making strategy for the disaster-stricken area based on the causal evolution results of the emergency situation and in combination with historical handling experience data and current resource constraints.

[0040] As an improvement to the above solution, the data fusion module is specifically used to:

[0041] Perform feature extraction on satellite image data, meteorological data, and public opinion text data from the multimodal data of the disaster-stricken area to obtain spatial features of the satellite image data, change features of the meteorological data, and semantic features of the public opinion text data;

[0042] A cross-modal attention mechanism is used to calculate the association weights between the spatial features of the satellite image data, the change features of the meteorological data, and the semantic features of the public opinion text data to generate an inter-modal association matrix;

[0043] The spatial features, the change features and the semantic features are weightedly fused based on the inter-modal association matrix, and a unified multi-modal emergency situation embedding vector is generated through dimensionality reduction processing.

[0044] As an improvement to the above solution, the feature extraction module is specifically used to:

[0045] The multimodal emergency situation embedding vector is input into a temporal convolutional network to extract the time series characteristics of the disaster diffusion speed and resource consumption rate;

[0046] Based on the geographic grid division of the disaster-stricken area, a spatial topology map is constructed, and the spatial correlation characteristics of the disaster impact between regions are extracted through the graph attention network;

[0047] Performing tensor splicing on the time series features and the spatial correlation features to generate a spatiotemporal joint feature vector;

[0048] The spatiotemporal joint feature vectors are serialized and reorganized to output an emergency situation state sequence representing the spatiotemporal evolution law of the disaster.

[0049] Another embodiment of the present invention provides an emergency situation intelligent decision-making system based on deep learning, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the emergency situation intelligent decision-making method based on deep learning described in the above-mentioned embodiment of the invention.

[0050] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0051] Multimodal data from disaster-stricken areas is collected through multi-source equipment. Fusion technology is used to integrate image, text, and voice data into a unified multimodal emergency situation embedding vector, eliminating data noise and redundancy, allowing the data to more comprehensively and accurately reflect the emergency situation. Using a spatiotemporal-aware neural network model, feature extraction and modeling are performed on the embedding vector in both the time series dimension and the spatial correlation dimension, deeply exploring the spatiotemporal evolution of the disaster situation and outputting a state sequence of the emergency situation that incorporates spatiotemporal characteristics. This state sequence is used to construct a dynamic causal graph to identify the potential causal relationships between disaster events and influencing factors, providing a basis for prediction. Based on the causal relationships, future emergency situation trends are predicted and causal evolution results are generated. Finally, emergency decision-making strategies are generated by combining historical response experience data with current resource constraints. Multimodal data fusion provides a comprehensive data foundation for subsequent analysis, spatiotemporal feature extraction accurately captures the spatiotemporal changes of disasters, and dynamic causal graph construction enables precise identification of potential causal relationships. Decision-making strategies generated based on a comprehensive multi-factor approach are more in line with actual needs. In summary, the embodiments of the present invention achieve deep integration and analysis of emergency situation data through deep fusion, dynamic modeling and causal reasoning mechanisms, accurately grasp the spatiotemporal characteristics and causal relationships of disasters, and thus achieve accurate perception of disaster situations, evolution prediction and intelligent decision-making under resource constraints. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1This is a flowchart of an intelligent decision-making method for emergency situations based on deep learning provided by one embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram of the structure of an emergency situation intelligent decision-making device based on deep learning provided by one embodiment of the present invention;

[0054] Figure 3 This is a structural diagram of an emergency situation intelligent decision-making system based on deep learning provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0056] See also Figure 1 , is a flow chart of a method for intelligent decision-making in emergency situations based on deep learning provided by one embodiment of the present invention. The method for intelligent decision-making in emergency situations based on deep learning comprises the following steps:

[0057] S10, fusing the multimodal data from the disaster-stricken area to generate a unified representation of the multimodal emergency situation embedding vector;

[0058] S11, extracting features from the multimodal emergency situation embedding vector using a spatiotemporal perception neural network model, modeling the model in the time series dimension and the spatial correlation dimension, and outputting an emergency situation state sequence containing spatiotemporal evolution characteristics;

[0059] S12, constructing a dynamic causal graph using the emergency situation state sequence to identify potential causal relationships between disaster events and influencing factors;

[0060] S13, predicting the development trend of the future emergency situation in the disaster-stricken area based on the potential causal relationship, and generating a causal evolution result of the emergency situation in the disaster-stricken area;

[0061] S14, based on the causal evolution results of the emergency situation, and combined with historical handling experience data and current resource constraints, generate an emergency decision-making strategy for the disaster-stricken area.

[0062] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0063] Multimodal data from disaster-stricken areas is collected through multi-source equipment. Fusion technology is used to integrate image, text, and voice data into a unified multimodal emergency situation embedding vector, eliminating data noise and redundancy, allowing the data to more comprehensively and accurately reflect the emergency situation. Using a spatiotemporal-aware neural network model, feature extraction and modeling are performed on the embedding vector in both the time series dimension and the spatial correlation dimension, deeply exploring the spatiotemporal evolution of the disaster situation and outputting a state sequence of the emergency situation that incorporates spatiotemporal characteristics. This state sequence is used to construct a dynamic causal graph to identify the potential causal relationships between disaster events and influencing factors, providing a basis for prediction. Based on the causal relationships, future emergency situation trends are predicted and causal evolution results are generated. Finally, emergency decision-making strategies are generated by combining historical response experience data with current resource constraints. Multimodal data fusion provides a comprehensive data foundation for subsequent analysis, spatiotemporal feature extraction accurately captures the spatiotemporal changes of disasters, and dynamic causal graph construction enables precise identification of potential causal relationships. Decision-making strategies generated based on a comprehensive multi-factor approach are more in line with actual needs. In summary, the embodiments of the present invention achieve deep integration and analysis of emergency situation data through deep integration, dynamic modeling and causal reasoning mechanisms, accurately grasp the spatiotemporal characteristics and causal relationships of disasters, and thus achieve accurate perception of disaster situations, evolution prediction and intelligent decision-making under resource constraints, effectively improving the effectiveness of emergency management.

[0064] As one example, fusing multimodal data from the disaster-stricken area to generate a unified multimodal emergency situation embedding vector includes the following sub-steps:

[0065] Perform feature extraction on satellite image data, meteorological data, and public opinion text data from the multimodal data of the disaster-stricken area to obtain spatial features of the satellite image data, change features of the meteorological data, and semantic features of the public opinion text data;

[0066] A cross-modal attention mechanism is used to calculate the association weights between the spatial features of the satellite image data, the change features of the meteorological data, and the semantic features of the public opinion text data to generate an inter-modal association matrix;

[0067] The spatial features, the change features and the semantic features are weightedly fused based on the inter-modal association matrix, and a unified multi-modal emergency situation embedding vector is generated through dimensionality reduction processing.

[0068] In this embodiment, for the multimodal data (satellite image data, meteorological data and public opinion text data) with a wide range of sources and diverse types in the disaster-stricken areas, the time-space alignment technology is first used to eliminate the differences in the time and space dimensions of the data, and a time-synchronized multimodal data set is generated to lay the foundation for subsequent analysis and ensure the consistency and comparability of the data. Then, the cross-modal attention mechanism is used to deeply explore the potential connections between the spatial characteristics of satellite image data, the changing characteristics of meteorological data and the semantic characteristics of public opinion text data. By calculating the correlation weights, an inter-modal correlation matrix is ​​generated to achieve a quantitative assessment of the importance of features of different modal data and fully capture the complex correlation relationship between the data. Finally, based on the inter-modal correlation matrix, various features are weighted and fused, and through dimensionality reduction processing, the high-dimensional and complex multimodal data are converted into a unified representation of the multimodal emergency situation embedding vector, which reduces the data dimension while retaining key information, improves data processing efficiency and the accuracy of subsequent analysis. Therefore, through this embodiment, it is possible to effectively solve the problems of data heterogeneity, difficulty in mining feature correlation, and difficulty in processing high data dimensions in the process of multimodal data fusion, achieve efficient integration and deep fusion of multimodal data in disaster-stricken areas, generate more representative and usable emergency situation data representation, and provide high-quality data support for subsequent deep learning-based spatiotemporal feature extraction, causal relationship analysis, and emergency decision-making strategy generation, significantly improve the comprehensiveness and accuracy of emergency situation analysis, and thus optimize the scientificity and effectiveness of emergency decision-making.

[0069] Specifically, the working process of this embodiment is as follows:

[0070] (1) Multimodal data feature extraction:

[0071] For satellite imagery data, convolutional neural network (CNN) technology, such as ResNet, is used. The satellite imagery data is fed into the network and processed through multiple convolutional and pooling layers to extract spatial features that include information such as the geographic outline of the disaster area, damaged buildings, and road interruptions. For example, by sliding the convolution kernel across the image, the characteristic responses of local pixels are calculated, gradually extracting spatial features at different scales.

[0072] For meteorological data, the LSTM (long short-term memory network) in the time series analysis algorithm is used to input meteorological data over a period of time, such as wind speed, rainfall, temperature, etc., into the network in chronological order. The information is memorized and updated through the gating mechanism to extract the changing characteristics of meteorological data in the time dimension, such as the changing trends and mutation points of meteorological elements.

[0073] For public opinion text data, the BERT (Bidirectional Transformer Encoder) model in the field of natural language processing is used. The text data is segmented and vectorized and then input into the model. The semantic relationship between words in the text is learned through the self-attention mechanism, and semantic features are extracted, such as the needs of the affected people, descriptions of the disaster site, and other semantic information.

[0074] (2) Cross-modal attention mechanism calculates association weights:

[0075] Assume that the spatial features extracted from satellite image data are S = [s1, s2, ..., s m ], where m is the feature dimension; the change feature of meteorological data is W = [w1, w2, ..., w n ], n is the feature dimension; the semantic feature of public opinion text data is T = [t1, t2, ..., t p ], p is the feature dimension.

[0076] First, the features of the three modalities are mapped to the same dimension through linear transformation, and we get S′=WS+b1, where W is the learnable weight matrix and b1 is the bias vector; W′=VW+b2, where V is the learnable weight matrix and b2 is the bias vector; and T′=UT+b3, where U is the learnable weight matrix and b3 is the bias vector.

[0077] Then, the attention score between the modalities is calculated. For satellite images and meteorological data, the attention score A SW The calculation formula is: where d k is the feature dimension after mapping, and the softmax function is used to normalize the score into a probability distribution, which represents the degree of correlation between satellite image features and meteorological features. Similarly, the attention score of satellite images and public opinion text data is Attention scores of meteorological data and public opinion text data

[0078] Finally, the three attention scores are combined into the inter-modality correlation matrix M: Among them A WS 、A TS 、A TW A SW 、A ST 、A WT The transpose of , the 1 on the diagonal indicates that the correlation degree of the own mode is 1.

[0079] (3) Weighted fusion and dimensionality reduction processing:

[0080] The spatial features S of satellite images, the change features W of meteorological data, the semantic features T of public opinion text data and the inter-modal correlation matrix M are weightedly fused. The calculation formula of the fused feature F is F = M 11 S+M 12 W+M 13 T+M 21 S+M 22 W+M 23 T+M 31 S+M 32 W+M 33 T, here M ij Represents the element in the i-th row and j-th column of the inter-modal association matrix M.

[0081] The fused feature F is subjected to dimensionality reduction processing using principal component analysis (PCA). By calculating the covariance matrix, eigenvalues, and eigenvectors, the main characteristic components are selected, redundant information is removed, and a unified representation of the multimodal emergency situation embedding vector is generated. This vector contains the key information of the three modal data and is used for subsequent emergency situation analysis.

[0082] As one example, the feature extraction of the multimodal emergency situation embedding vector by the spatiotemporal perception neural network model is performed, modeling is performed in the time series dimension and the spatial correlation dimension respectively, and the emergency situation state sequence containing spatiotemporal evolution characteristics is output, including the following sub-steps:

[0083] The multimodal emergency situation embedding vector is input into a temporal convolutional network to extract the time series characteristics of the disaster diffusion speed and resource consumption rate;

[0084] Based on the geographic grid division of the disaster-stricken area, a spatial topology map is constructed, and the spatial correlation characteristics of the disaster impact between regions are extracted through the graph attention network;

[0085] Performing tensor splicing on the time series features and the spatial correlation features to generate a spatiotemporal joint feature vector;

[0086] The spatiotemporal joint feature vectors are serialized and reorganized to output an emergency situation state sequence representing the spatiotemporal evolution law of the disaster.

[0087] In this embodiment, targeted feature extraction and fusion are performed on the multimodal emergency situation embedding vector based on the spatiotemporal characteristics of the disaster emergency situation. First, a temporal convolutional network is used to process the embedding vector using a causal convolutional structure with a memory gate to capture time series characteristics such as the speed of disaster spread and resource consumption rate. The memory gate retains historical information, enhancing the ability to capture dynamic changes in the time series. A spatial topological map is then constructed based on the geographic grid of the affected area. Using a dual-scale graph attention network, the spatial correlation characteristics of inter-regional disaster impacts are explored, taking into account both local and global node relationships. The time series features and spatial correlation feature tensors are then concatenated, and a weighted fusion is used to generate a spatiotemporal joint feature vector, balancing the importance of spatiotemporal features. Finally, the spatiotemporal joint feature vector is serialized and reorganized, and a fully connected network is used to output the emergency situation state sequence, presenting the spatiotemporal evolution of the disaster. In summary, this embodiment effectively separates and extracts the spatiotemporal characteristics of disasters, overcoming the problem of traditional methods' inadequate handling of spatiotemporal heterogeneity; the dual-scale attention mechanism enhances the capture of disaster propagation paths; weighted fusion and serialized reorganization allow the model to adapt to different disaster scenarios, and the output state sequence provides strong support for disaster causal analysis, situation prediction, and emergency decision-making, thereby improving the scientific nature and timeliness of emergency response.

[0088] Specifically, the working process of this embodiment is as follows:

[0089] 3. Example of feature extraction of spatiotemporal perception neural network model

[0090] 3.1 Time series feature extraction: The multimodal emergency situation embedding vector generated above is used as the input data of this step. The embedding vector is the feature expression obtained by integrating satellite image data, meteorological data and public opinion text data. It exists in the form of a matrix and is recorded as Among them, T represents the time step, that is, the number of time points set during the data collection process, reflecting the observation length of the disaster situation in the time series; D is the embedding vector dimension, which represents the number of features of the fused data. These features contain multi-faceted disaster situation information. This step uses the existing temporal convolutional network (TCN) to extract time series features. TCN is constructed based on the principle of causal convolution. By setting different dilation factors (Dilation Factor), it effectively expands the network receptive field without significantly increasing the number of parameters, thereby capturing long-distance dependencies in time series data. In actual operation, the multimodal emergency situation embedding vector is input into TCN according to the time dimension. The network automatically learns time series features such as the speed of disaster spread and the rate of resource consumption through multiple layers of causal convolution layers and residual connection structures. After training and calculation, TCN stably outputs the time series feature matrix Among them D T It is the characteristic dimension of time series, and the matrix lays the foundation for subsequent processing.

[0091] 3.2 Spatial correlation feature extraction: The disaster-stricken area is evenly divided into N×N grids in the geographic space, and a spatial topology map is constructed based on this. Among them, the node set Each node corresponds to a geographic grid; the edge set ε represents the connection relationship between adjacent grids, that is, if two grids are geographically adjacent, there is an edge between their corresponding nodes. i The initial eigenvector of It is composed of the multimodal emergency situation embedding vector features corresponding to the grid position, D S is the initial dimension of spatial features.

[0092] An improved Graph Attention Network (GAT) is used to extract spatial correlation features of inter-regional disaster impacts. Traditional GATs struggle to capture complex, long-range impacts when processing geospatial data. To address this issue, a multi-level propagation mechanism is designed.

[0093] First, calculate the basic attention weight between nodes. For node i and its neighbor node j, the basic attention weight β ij The calculation is as follows:

[0094]

[0095] Where W is the dimension The weight matrix is ​​used to linearly transform the initial features of the node, D′ S is the feature dimension after transformation; a is the dimension The attention parameter vector [Wf i ||Wf j ] means concatenating the transformed feature vectors of nodes i and j; LeakyReLU is the activation function; is the set of neighbor nodes of node i.

[0096] Then, a multi-level propagation mechanism is introduced, and the number of propagation levels is set to P. At the p-th level of propagation, the feature update formula of node i is:

[0097]

[0098] in, is the feature vector of node i after propagation at level p, is the initial feature; W p is the weight matrix for the p-th level of propagation; σ is the activation function, such as ReLU. Through multi-level propagation, nodes gradually aggregate the influence information of distant nodes.

[0099] Finally, the node features obtained from the last level of propagation are integrated to obtain the spatial correlation feature matrix Complete spatial feature extraction.

[0100] 3.3 Generation of spatiotemporal joint feature vectors: Obtaining time series features and spatial association features Finally, the existing tensor splicing technology is used to perform feature fusion. Due to the difference in temporal and spatial feature dimensions, the time series feature T is first copied and expanded along the spatial dimension, and the spatial correlation feature S is copied and expanded along the temporal dimension to match the two dimensions. Then, standard tensor splicing operations (such as the torch.cat() function in the PyTorch framework) are used to splice on the feature dimension to generate a spatiotemporal joint feature vector. This operation can directly integrate feature information in time and space dimensions, providing a complete feature representation for subsequent analysis.

[0101] 3.4 Emergency Situation State Sequence Output: After obtaining the spatiotemporal joint feature vector After that, it needs to be converted into an emergency situation state sequence. First, the three-dimensional space-time joint feature vector is flattened into a two-dimensional matrix in sequence Then, the flattened matrix is ​​input into the two-layer fully connected network in sequence. The first layer of the fully connected network performs a preliminary transformation on the input features through matrix multiplication and activation function operations to extract basic semantic features; the second layer of the fully connected network further abstracts the features and outputs a feature vector with higher semantic expression capabilities. The operation of the two-layer fully connected network can be expressed as: H1 = ReLU (W1·J flat +b1), H2=ReLU(W2·H1+b2), where W1 and W2 are the weight matrices of the two-layer fully connected network, b1 and b2 are the corresponding bias vectors, and ReLU is the activation function used to introduce nonlinearity and enhance the network's expressiveness. Finally, the feature vector H2 output by the second-layer fully connected network is reshaped and converted into an emergency state sequence. This sequence completely contains the evolution information of disasters in time and space dimensions, and can be directly used for subsequent operations such as constructing dynamic causal diagrams and predicting the development trend of emergency situations, providing core data support for emergency decision-making.

[0102] As one example, the method of constructing a dynamic causal graph using the emergency situation state sequence to identify potential causal relationships between disaster events and influencing factors includes the following sub-steps:

[0103] Mapping each dimension of the emergency situation state sequence into a node of a causal graph, and initializing causal strength parameters between the nodes;

[0104] Based on the causal strength parameters between nodes, the causal relationship between each node in the causal graph is analyzed to determine the causal edge;

[0105] Generate a dynamic causal graph based on the causal edges between each node;

[0106] Based on the dynamic causal diagram, the potential causal relationship between disaster events and influencing factors is identified.

[0107] In this embodiment, the multi-dimensional observation variables of the emergency situation (such as disaster intensity, environmental parameters, emergency resources, etc.) are first mapped to causal graph nodes, and the causal strength parameters between nodes are initialized based on data-driven methods (such as time series correlation, information entropy, or neural networks). Subsequently, significant causal edges are screened and false associations are eliminated through causal inference algorithms (such as Granger causality test, Bayesian structure learning, or reinforcement learning strategy). Then, a dynamically updateable causal graph structure is generated by combining a time series sliding window or state transition model to capture the causal interactions of various factors in disaster evolution in real time. Finally, through reverse reasoning, path search, or intervention analysis of the causal graph, the key driving factors and chain reaction paths of the disaster event are located. From the above analysis, it can be seen that this embodiment realizes dynamic and interpretable modeling of disaster causal relationships. It can not only extract a causal network with time series dependence from complex multidimensional data, but also adapt to the sudden change characteristics of disaster scenarios through a dynamic adjustment mechanism, providing a scientific basis for emergency decision-making with "traceable causal associations, quantifiable key paths, and predictable intervention effects", thereby improving the accuracy of disaster response.

[0108] Specifically, the working process of this embodiment is as follows:

[0109] 1. Map each dimension of the emergency situation state sequence to a node of the causal graph and initialize the causal strength parameters between the nodes: First, obtain the emergency situation state sequence output above, which is a characteristic representation of the spatiotemporal evolution law of the disaster, denoted as Where T represents the time step, reflecting the changing process of the disaster in the time dimension; N is the sequence dimension, corresponding to various disaster-related characteristics, such as disaster intensity, resource status, geographical environment parameters, etc.

[0110] Map each dimension of the sequence Z to a node in the causal graph to construct a node set For example, if a dimension represents rainfall, the corresponding node represents the rainfall factor; if a dimension represents the number of rescue teams, the corresponding node represents the rescue resource factor.

[0111] When initializing the causal strength parameters between nodes, an improved initialization method based on mutual information and time-space distance is adopted. i and v j , initialize the causal strength parameter ω ij The calculation formula is as follows:

[0112]

[0113] Among them, I(v i ;v j ) represents node v i and v j The mutual information between them is used to measure the degree of dependence between the features of two nodes. It is obtained by calculating the relationship between the joint entropy of the probability distribution of the two and the entropy of each node. The higher the mutual information value, the stronger the correlation between the two nodes. When calculating specifically, first count the node v i and v j The value distribution in the emergency state sequence is then calculated according to the mutual information formula I(X;Y)=H(X)+H(Y)-H(X,Y) (where H(X) is the entropy of variable X and H(X,Y) is the joint entropy of variables X and Y). max k，l I(v l ;v l ) is the maximum value of the mutual information of all node pairs, which is used to normalize the mutual information so that ω ij The value range of d is between [0,1]. ij Represents node v i and v j The comprehensive distance metric of the corresponding features in the time and space dimensions. If the node corresponds to a geographic space feature, then d ij Euclidean distance is calculated based on geographic coordinates; if time dimension is involved, weighted calculation is performed based on time step difference. For example, when v i and v j When d represents the disaster intensity of different geographical locations at adjacent time steps, ij The distance between the two places and the time interval will be taken into consideration. max The maximum spatiotemporal distance between all pairs of nodes is used to normalize the spatiotemporal distances, ensuring that the exponential term is between [0, 1]. This calculation process considers both the intrinsic degree of inter-node connections and the spatiotemporal distance, making the initialized causal strength parameter more consistent with the causal relationship characteristics in actual disaster scenarios.

[0114] 2. Based on the causal strength parameters between nodes, analyze the causal relationship between each node in the causal graph and determine the causal edge: After completing the causal strength parameter initialization, analyze the causal relationship between nodes based on these parameters to determine the causal edge. This step uses the improved dynamic time causal verification algorithm. For each pair of nodes v i and v j , calculate its dynamic time causal score S ij , the formula is as follows: Where T is the time step of the emergency state sequence; τ is the set time lag step, which is used to take into account the possible time delay of causal relationships. For example, in a flood disaster, changes in rainfall may not immediately lead to a rise in water levels, but may take some time. τ is used to capture this delay. j,t+τ Represents node v j The change in the eigenvalue at time t+τ; Δx i，t Represents node v i The change in the eigenvalue at time t. ω ij The node v initialized in the above process i and v j In general, by calculating the causal strength parameter between nodes v i The change of v j The degree of influence after the lag τ time is weighted with the initialized causal strength parameter to obtain the dynamic time causal score. Set the threshold ε, when S ij >ε, determine the slave node v i To node v j There is a causal edge with a weight of S ij In this way, not only the static association strength between nodes is considered, but also the causal influence in the time dimension is combined to more accurately determine the causal relationship between nodes.

[0115] 3. Generate a dynamic causal graph based on the causal edges between each node: Based on the node set determined in the first two steps and the causal edge set ε, generating a dynamic causal graph In order to reflect the dynamic change of causal relationship over time, the time decay factor γ∈(0,1) is introduced. For the causal edge (v i ,v j ), the update formula of its weight at the subsequent time t′ (t′>t) is: w ij (t′)=w ij (t)×γ t′-t , where w ij (t) is the causal edge at time t (v i , v j ). Over time, if there is no new evidence that the causal relationship persists or strengthens, its weight will gradually decay, reflecting the dynamic nature of causal relationships in disaster scenarios. By integrating nodes and edges at different times and using existing graph data structures (such as adjacency matrices or adjacency lists) for storage and representation, a dynamic causal graph is finally generated. This dynamic causal diagram can reflect in real time the changes in the causal relationship between disaster events and influencing factors in the time dimension.

[0116] 4. Based on the dynamic causal graph, identify the potential causal relationship between disaster events and influencing factors: Based on the generated dynamic causal graph, an improved causal path search and analysis algorithm is used to identify potential causal relationships. First, determine the target disaster event node (such as the node corresponding to "serious casualties" and "huge property losses") and the set of possible influencing factor nodes. Then, starting from the influencing factor node, use the depth-first search (DFS) or breadth-first search (BFS) algorithm, combined with the weight of the causal edge, to search for all possible paths to the target disaster event node. For each causal path searched Calculate its causal path strength CPS(P) using the following formula: in is the adjacent node in path P and The weight of the causal edge between the two paths. A higher CPS(P) value indicates a stronger causal relationship on that path, and a higher likelihood of it being the key causal path leading to the target disaster event. By ranking and analyzing the strength of all causal paths, we screen for the strongest ones, thereby identifying the potential causal relationship between the disaster event and the influencing factors. Furthermore, based on practical disaster domain knowledge, we verify and correct the identified results to ensure the rationality and reliability of the potential causal relationship, providing a scientific basis for subsequent emergency decision-making.

[0117] As one example, predicting the development trend of the future emergency situation in the disaster-stricken area based on the potential causal relationship to generate the causal evolution result of the emergency situation in the disaster-stricken area includes the following sub-steps:

[0118] Extracting a node causal association matrix based on the potential causal relationship as prior causal knowledge;

[0119] Inputting the emergency situation state sequence and the causal association matrix into a pre-trained time series prediction model to perform multi-step trend prediction and obtain a multi-step trend prediction result;

[0120] Based on the multi-step trend prediction results, the situation evolution path of the disaster-stricken area under different emergency measures is simulated to generate the causal evolution results of the emergency situation in the disaster-stricken area.

[0121] In this embodiment, based on potential causal relationships, a complete chain from causal knowledge extraction to situation prediction and simulation is established. First, a node causal association matrix is ​​extracted from the identified potential causal relationships, and the abstract causal logic is converted into a computable matrix form. This serves as prior knowledge for subsequent predictions, injecting causal constraints into the prediction process. Next, the emergency situation state sequence and the causal association matrix are input into a pre-trained time series prediction model. With the help of the model's ability to process time series data and combined with the causal association matrix, a multi-step trend prediction of the emergency situation in the disaster-stricken area is achieved, so that the prediction results not only contain the change laws of the time dimension but also conform to the causal relationship logic. Finally, based on the multi-step trend prediction results, the situation evolution path of the disaster-stricken area under the intervention of different emergency measures is simulated. By changing the relevant parameters in the causal association matrix, the impact of different emergency measures on the situation is simulated, thereby generating the causal evolution results of the emergency situation. In summary, on the one hand, this embodiment overcomes the problem of the lack of causal interpretability of the traditional time series prediction model by introducing the causal association matrix as prior knowledge, thereby improving the reliability and interpretability of the prediction results; on the other hand, this embodiment can effectively capture the long-term changing trend of the emergency situation through multi-step trend prediction, reserving sufficient time for emergency decision-making; in addition, by simulating the situation evolution path under different emergency measures, it can intuitively present the impact of various measures on the development of the emergency situation, helping decision makers to quickly evaluate the pros and cons of different plans.

[0122] Specifically, the working process of this embodiment is as follows:

[0123] 1. After completing the construction of the dynamic causal graph and the identification of potential causal relationships, extract the causal relationship information between nodes and generate a node causal relationship matrix Where N is the total number of nodes in the causal graph, corresponding to the number of dimensions of the emergency state sequence. ij Represents node v i For node v j The causal relationship strength is based on the weight of the edge in the dynamic causal graph. In the specific extraction process, if the node v i To node v j There is a causal edge with edge weight w ij , then C ij =w ij ; If there is no causal edge, then C ij = 0. To enhance the reliability of causal association, the matrix is ​​normalized: In this way, the sum of the elements in each row of the matrix is ​​made 1, which is convenient for subsequent use as weights in the prediction model, thereby converting the complex causal relationship in the dynamic causal graph into a structured prior causal knowledge matrix.

[0124] 2. Output the emergency status sequence from the above process The node causal association matrix C obtained from the above process is used as input to the improved time series prediction model. This model is based on the Transformer architecture and introduces a causal attention mechanism to fully utilize prior causal knowledge to guide the prediction process.

[0125] Model input processing: The emergency state sequence Z is input sequentially according to the time step length T, and the dimension of the feature vector of each time step is N. The node causal association matrix C is introduced as global context information in the attention calculation process of the model.

[0126] Causal attention mechanism calculation: In the multi-head attention module of Transformer, the attention calculation method is improved. Key Vector Sum value vector (where d k and d v The causal association matrix C is introduced to adjust the attention weight, and the calculation formula is as follows: Here, α is a hyperparameter used to adjust the influence of the causal association matrix on the attention weight. Through this calculation process, the model not only considers the feature correlation of the input sequence when calculating attention, but also enhances the attention weights between nodes with causal relationships based on prior causal knowledge, thereby more accurately capturing the causal dependencies between emergency situation features.

[0127] Multi-step prediction: During the training phase, the model uses the historical emergency situation sequence as input to predict the situation characteristics of multiple time steps in the future. The teacher forcing strategy is used to gradually guide the model to learn long-sequence prediction capabilities during the training process. In the prediction phase, based on the current input emergency situation sequence, the model's forward propagation calculation outputs the multi-step trend prediction results for the next L time steps. Where L is the preset prediction step size.

[0128] 3. Multi-step trend prediction results obtained based on the above process Construct an emergency response simulation system to generate the evolution path of the disaster-stricken areas under different emergency response measures.

[0129] Emergency response modeling involves abstracting various emergency response measures into interventions on the causal matrix C and the sequence of emergency statuses. For example, increasing the input of relief supplies can be seen as strengthening the negative causal influence of the "relief supply quantity" node on the "disaster loss extent" node, simulated by decreasing the value of the corresponding element in C. Opening new rescue routes can be seen as changing the strength of causal relationships between geographical and transportation-related nodes.

[0130] In addition, situation evolution simulation: using simulation algorithms based on causal reasoning, from the prediction results Starting from this, combined with the modified causal association matrix C′, the situation evolution is simulated through iterative calculation. For each time step t, the update formula for calculating the node state based on the causal relationship is as follows: in, Represents node v j The predicted state value at time t, where f(·) is an activation function (such as ReLU) that introduces nonlinearity. This formula indicates that the state change of each node is affected by the state of all its causally related nodes at the previous moment. In this way, the evolution path of the situation under different emergency measures over a period of time in the future is gradually generated.

[0131] The simulated situation evolution path is output as visual charts (such as dynamic curves and heat maps) and structured data (such as node state matrices at different time steps), forming the causal evolution results of the emergency situation in the disaster-stricken area. This result intuitively shows the changing trends and causal transmission processes of various emergency situation indicators in the disaster-stricken area under the intervention of different emergency measures, providing a comprehensive reference for emergency decision-making.

[0132] As one example, generating an emergency decision-making strategy for the disaster-stricken area based on the causal evolution results of the emergency situation and combining historical handling experience data with current resource constraints includes the following sub-steps:

[0133] Obtain historical disposal experience data including historical emergency plans, resource consumption data and disposal effects;

[0134] Convert the current number of available rescue resources, team distribution locations, and traffic network status into resource constraints;

[0135] Generate an initial emergency decision-making strategy through a pre-trained deep learning model, taking the causal evolution results of the emergency situation, the historical handling experience data and the resource constraints as input;

[0136] Based on the causal logic of the dynamic causal graph, the secondary disaster risks potentially caused by the initial decision-making strategy are verified, and the final emergency decision-making strategy for the disaster-stricken area is output.

[0137] In this embodiment, emergency plans, resource consumption and disposal effect data under multiple scenarios are first extracted from the historical database to construct an empirical knowledge base based on case reasoning; at the same time, real-time rescue resources (such as material inventory and team location) and traffic network status are encoded as constrained mathematical programming conditions (such as a group of linear inequalities) to ensure the feasibility of decision-making; then, the causal evolution prediction results (such as disaster diffusion path and resource demand time series) and the above data are input into the pre-trained deep reinforcement learning model, and the initial decision (such as material distribution plan and evacuation route) that meets the resource constraints is generated through the policy network; finally, based on the causal logic of the dynamic causal graph, counterfactual reasoning and risk propagation algorithms are used to simulate the intervention effect of the initial strategy on key nodes (such as transportation hubs and secondary disaster sources), quantitatively evaluate the probability of secondary risks (such as road congestion causing rescue delays), and optimize and adjust the strategy weights through game theory to output the final decision that takes into account timeliness, safety and resource efficiency. Therefore, this embodiment realizes the closed loop of "causal explainability driving decision generation and dynamic risk chain active defense", which not only improves the scientific nature of decision-making through historical experience and causal reasoning, but also reduces the risk of secondary disaster triggering through the causal verification mechanism, thereby enhancing the robustness and feasibility of decision-making in complex disaster scenarios.

[0138] Specifically, the working process of this embodiment is as follows:

[0139] 1. Obtain historical emergency response experience data, including historical emergency plans, resource consumption data, and response results: Build a historical response experience data collection and storage system to obtain relevant data from multiple channels. By establishing data interfaces with emergency management department archives, rescue team work record systems, and disaster statistics databases, complete information on historical disaster events can be collected.

[0140] Historical emergency response plans include detailed rescue action plans, such as rescue team dispatch arrangements, material drop-off points, and emergency evacuation route planning, along with text and charts. Resource consumption data details the consumption and allocation of various resources, such as medical supplies, food, and machinery, during each disaster response. Disaster response effectiveness data is reflected in post-disaster assessment reports for affected areas, including quantitative indicators such as the reduction in casualties, the degree of economic loss control, and the restoration of social order. After structured processing, these data are stored in a historical response experience database, forming a standardized historical response experience dataset H. Each row corresponds to a historical disaster response event, and each column represents different data attributes, such as the emergency plan code, resource category, and response effectiveness score.

[0141] 2. Convert the current number of available rescue resources, team locations, and traffic network status into resource constraints: Leverage IoT devices and geographic information systems (GIS) to collect data in real time. Sensors deployed in material warehouses capture the current number of available rescue resources, such as medicine inventory and tent reserves. Using positioning terminals carried by rescue teams, combined with GIS technology, teams' locations are precisely determined, generating a dataset containing latitude and longitude coordinates, team personnel, and equipment configurations. Using data from traffic cameras, road sensors, and navigation platforms, the current traffic network status is analyzed to capture information such as road capacity, congested road sections, and bridge and tunnel conditions.

[0142] The above data are integrated into a resource constraint matrix R. The rows of the matrix correspond to different types of resource constraints, such as "upper limit of medical supplies", "maximum number of rescue teams that can be deployed", "traffic flow restrictions on key roads", etc.; the columns correspond to different disaster-stricken areas or rescue mission nodes. For example, R ij Indicates the upper limit or limit of the availability of the i-th type of resources in the j-th disaster area. 1j represents the upper limit of the quantity of medical supplies, and its value is the total amount of medical supplies that can be allocated to the j-th disaster-stricken area.

[0143] 3. Using a pre-trained deep learning model, the causal evolution of the emergency situation, the historical response experience data, and the resource constraints are used as input to generate an initial emergency decision-making strategy: This involves building and training an improved deep learning model based on the Transformer architecture. This model, based on the traditional Transformer architecture, incorporates a gating mechanism to better integrate different types of input data.

[0144] Among them, input data processing: the causal evolution result E of the emergency situation is organized into a tensor containing multiple time steps and multiple situation indicators in the future, and each element E t，k represents the predicted value of the kth situation indicator (e.g., affected area, number of trapped people) at the tth time step in the future. Historical disaster response experience data H undergoes feature extraction and encoding, converting it into a fixed-length vector representation to retain key information. The resource constraint matrix R is normalized and then merged with other data.

[0145] Gating mechanism calculation: A gating unit is added between the input layer and the attention layer of the model. For the input emergency situation causal evolution result vector e, historical disposal experience data vector h and resource constraint condition vector r, the gating weights g1, g2, and g3 are calculated: g1 = σ(W g1 e+b g1 ), g2=σ(W g2 h+b g2 )、g3=σ(W g3·r+b g3 ), where σ is the sigmoid activation function, which maps the output to the (0,1) interval; W g1 、W g2 、W g3 is the weight matrix, b g1 、b g2 、b g3 is the bias vector. These parameters are learned through model training. The final fused input vector x is: x = g1⊙e + g2⊙h + g3⊙r, where ⊙ represents element-wise multiplication.

[0146] Model training and prediction: Use historical disaster data and its corresponding effective emergency decision-making strategies as training samples, and use the cross entropy loss function and Adam optimizer for model training. After training, the current emergency situation causal evolution results, historical handling experience data, and resource constraints are input into the model. After feature extraction and decision reasoning of the multi-layer Transformer module, the initial emergency decision-making strategy D is output. init This strategy covers rescue resource allocation, rescue operation schedule, personnel scheduling, and other specific content. For the specific training process, please refer to the existing model training technology.

[0147] 4. Based on the causal logic of the dynamic causal graph, verify the secondary disaster risk potentially caused by the initial decision-making strategy and output the final emergency decision-making strategy for the disaster-stricken area: init This translates into intervention operations on nodes in the dynamic causal graph. For example, increasing the delivery of relief supplies to a certain area corresponds to adjusting the state value of the "supply quantity" node, while changing the rescue team's route corresponds to modifying the causal strength between geographic and transportation-related nodes.

[0148] Based on the causal logic of the dynamic causal graph, the causal reasoning algorithm is used to calculate the secondary disaster risk that may be caused by each intervention operation. i , after implementing the decision strategy, its state changes Δv i The influence of the nodes connected by the relevant causal edges is calculated as follows: Among them, ε is the edge set of the dynamic causal graph, (v j , v i ) represents the slave node v j To node v i The causal edge, w ji is the weight of the causal edge, Δv j For node v j Through iterative calculation, the state changes of all nodes are obtained, and then the possibility and impact of secondary disasters are evaluated.

[0149] Set a risk threshold τ. If the calculated secondary disaster risk value is higher than the threshold, the initial emergency decision-making strategy is adjusted, such as re-planning resource allocation, modifying the rescue action sequence, etc. If the risk value is lower than the threshold, the initial emergency decision-making strategy is determined as the final emergency decision-making strategy D for the disaster-stricken area. final , output complete decision-making documents including detailed rescue plans and risk assessment reports to guide actual emergency rescue work.

[0150] See also Figure 2 , is a schematic diagram of the structure of an emergency situation intelligent decision-making device based on deep learning provided by one embodiment of the present invention. The emergency situation intelligent decision-making device based on deep learning includes:

[0151] The data fusion module 10 is used to fuse the multimodal data from the disaster-stricken area and generate a unified multimodal emergency situation embedding vector;

[0152] A feature extraction module 11 is used to extract features from the multimodal emergency situation embedding vector using a spatiotemporal perception neural network model, modeling the time series dimension and the spatial correlation dimension respectively, and outputting an emergency situation state sequence containing spatiotemporal evolution characteristics;

[0153] A causal identification module 12 is configured to construct a dynamic causal graph using the emergency situation state sequence to identify potential causal relationships between disaster events and influencing factors;

[0154] A prediction module 13 is configured to predict the development trend of the future emergency situation in the disaster-stricken area based on potential causal relationships, and generate a causal evolution result of the emergency situation in the disaster-stricken area;

[0155] The strategy generation module 14 is used to generate an emergency decision-making strategy for the disaster-stricken area based on the causal evolution results of the emergency situation and in combination with historical handling experience data and current resource constraints.

[0156] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0157] Multimodal data from disaster-stricken areas is collected through multi-source equipment. Fusion technology is used to integrate image, text, and voice data into a unified representation of a multimodal emergency situation embedding vector, eliminating data noise and redundancy, allowing the data to more comprehensively and accurately reflect the emergency situation. Using a spatiotemporal-aware neural network model, feature extraction and modeling are performed on the embedding vector in both the time series dimension and the spatial correlation dimension, deeply exploring the spatiotemporal evolution of the disaster situation and outputting a state sequence of the emergency situation that incorporates spatiotemporal characteristics. This state sequence is then used to construct a dynamic causal graph to identify the potential causal relationships between disaster events and influencing factors, providing a basis for prediction. Based on these causal relationships, future emergency situation trends are predicted and causal evolution results are generated. Finally, an emergency decision-making strategy is generated by combining historical response experience with current resource constraints. Multimodal data fusion provides a comprehensive data foundation for subsequent analysis, spatiotemporal feature extraction accurately captures the spatiotemporal changes of disasters, and dynamic causal graph construction enables precise identification of potential causal relationships. Decision-making strategies generated based on a comprehensive multi-factor approach are more in line with actual needs. In summary, the embodiments of the present invention achieve deep integration and analysis of emergency situation data through deep integration, dynamic modeling and causal reasoning mechanisms, accurately grasp the spatiotemporal characteristics and causal relationships of disasters, and thus achieve accurate perception of disaster situations, evolution prediction and intelligent decision-making under resource constraints, effectively improving the effectiveness of emergency management.

[0158] As one example, the data fusion module is specifically configured to:

[0159] Perform feature extraction on satellite image data, meteorological data, and public opinion text data from the multimodal data of the disaster-stricken area to obtain spatial features of the satellite image data, change features of the meteorological data, and semantic features of the public opinion text data;

[0160] A cross-modal attention mechanism is used to calculate the association weights between the spatial features of the satellite image data, the change features of the meteorological data, and the semantic features of the public opinion text data to generate an inter-modal association matrix;

[0161] The spatial features, the change features and the semantic features are weightedly fused based on the inter-modal association matrix, and a unified multi-modal emergency situation embedding vector is generated through dimensionality reduction processing.

[0162] As one example, the feature extraction module is specifically used to:

[0163] The multimodal emergency situation embedding vector is input into a temporal convolutional network to extract the time series characteristics of the disaster diffusion speed and resource consumption rate;

[0164] Based on the geographic grid division of the disaster-stricken area, a spatial topology map is constructed, and the spatial correlation characteristics of the disaster impact between regions are extracted through the graph attention network;

[0165] Performing tensor splicing on the time series features and the spatial correlation features to generate a spatiotemporal joint feature vector;

[0166] The spatiotemporal joint feature vectors are serialized and reorganized to output an emergency situation state sequence representing the spatiotemporal evolution law of the disaster.

[0167] It can be understood that the content of the above-mentioned embodiment of the emergency situation intelligent decision-making device based on deep learning can refer to the content of the embodiment of the above-mentioned emergency situation intelligent decision-making method based on deep learning, and will not be repeated here.

[0168] See also Figure 3 , is a schematic diagram of an emergency situation intelligent decision-making system based on deep learning provided by an embodiment of the present invention. The emergency situation intelligent decision-making system based on deep learning of this embodiment includes: a processor 100, a memory 101, and a computer program stored in the memory 101 and executable on the processor 100, such as an emergency situation intelligent decision-making program based on deep learning. When the processor 100 executes the computer program, the steps in each of the above-mentioned embodiments of the emergency situation intelligent decision-making method based on deep learning are implemented. Alternatively, when the processor 100 executes the computer program, the functions of each module / unit in the above-mentioned device embodiments are implemented.

[0169] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution process of the computer program in the deep learning-based emergency situation intelligent decision-making system.

[0170] The deep learning-based emergency situation intelligent decision-making system can be a computing device such as a desktop computer, laptop, PDA, or cloud server. The deep learning-based emergency situation intelligent decision-making system can include, but is not limited to, a processor and a memory. Those skilled in the art will understand that the schematic diagram is merely an example of a deep learning-based emergency situation intelligent decision-making system and does not constitute a limitation of the deep learning-based emergency situation intelligent decision-making system. The system can include more or fewer components than shown in the diagram, or a combination of certain components, or different components. For example, the deep learning-based emergency situation intelligent decision-making system can also include input and output devices, network access devices, buses, etc.

[0171] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the deep learning-based emergency situation intelligent decision-making system, and utilizes various interfaces and lines to connect the various parts of the deep learning-based emergency situation intelligent decision-making system.

[0172] The memory can be used to store the computer programs and / or modules, and the processor implements the various functions of the deep learning-based emergency situation intelligent decision-making system by running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created based on the use of the mobile phone (such as audio data, a phone book, etc.). In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0173] Among them, if the module / unit integrated in the emergency situation intelligent decision-making system based on deep learning is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0174] It should be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without inventive effort.

[0175] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for intelligent decision-making on emergency situations based on deep learning, characterized in that: The following steps are involved: The multimodal data from the disaster-stricken areas are fused to generate a unified multimodal emergency situation embedding vector. Extract features from the multimodal emergency situation embedding vector using a spatiotemporal perception neural network model, model the state of the emergency situation in the time series dimension and the spatial correlation dimension, and output a state sequence of the emergency situation containing spatiotemporal evolution characteristics; Using the emergency situation state sequence to construct a dynamic causal diagram to identify potential causal relationships between disaster events and influencing factors; Predicting the development trend of the future emergency situation in the disaster-stricken area based on the potential causal relationship, and generating a causal evolution result of the emergency situation in the disaster-stricken area; Based on the causal evolution results of the emergency situation, and combined with historical handling experience data and current resource constraints, an emergency decision-making strategy for the disaster-stricken area is generated.

2. The deep learning-based intelligent decision-making method for emergency situations according to claim 1, characterized in that: The method of fusing the multimodal data from the disaster-stricken area to generate a unified multimodal emergency situation embedding vector includes the following sub-steps: Perform feature extraction on satellite image data, meteorological data, and public opinion text data from the multimodal data of the disaster-stricken area to obtain spatial features of the satellite image data, change features of the meteorological data, and semantic features of the public opinion text data; A cross-modal attention mechanism is used to calculate the association weights between the spatial features of the satellite image data, the change features of the meteorological data, and the semantic features of the public opinion text data to generate an inter-modal association matrix; The spatial features, the change features and the semantic features are weightedly fused based on the inter-modal association matrix, and a unified multi-modal emergency situation embedding vector is generated through dimensionality reduction processing.

3. The deep learning-based intelligent decision-making method for emergency situations according to claim 2, characterized in that: The method of extracting features from the multimodal emergency situation embedding vector using a spatiotemporal perception neural network model, modeling the vectors in the time series dimension and the spatial correlation dimension, and outputting an emergency situation state sequence containing spatiotemporal evolution characteristics comprises the following sub-steps: The multimodal emergency situation embedding vector is input into a temporal convolutional network to extract the time series characteristics of the disaster diffusion speed and resource consumption rate; Based on the geographic grid division of the disaster-stricken area, a spatial topology map is constructed, and the spatial correlation characteristics of the disaster impact between regions are extracted through the graph attention network; Performing tensor splicing on the time series features and the spatial correlation features to generate a spatiotemporal joint feature vector; The spatiotemporal joint feature vectors are serialized and reorganized to output an emergency situation state sequence representing the spatiotemporal evolution law of the disaster.

4. The deep learning-based intelligent decision-making method for emergency situations according to claim 3, characterized in that: The method of constructing a dynamic causal graph using the emergency situation state sequence to identify the potential causal relationship between disaster events and influencing factors includes the following sub-steps: Mapping each dimension of the emergency situation state sequence into a node of a causal graph, and initializing causal strength parameters between the nodes; Based on the causal strength parameters between nodes, the causal relationship between each node in the causal graph is analyzed to determine the causal edge; Generate a dynamic causal graph based on the causal edges between each node; Based on the dynamic causal diagram, the potential causal relationship between disaster events and influencing factors is identified.

5. The deep learning-based intelligent decision-making method for emergency situations according to claim 4, characterized in that: The method of predicting the development trend of the future emergency situation in the disaster-stricken area based on the potential causal relationship to generate the causal evolution result of the emergency situation in the disaster-stricken area includes the following sub-steps: Extracting a node causal association matrix based on the potential causal relationship as prior causal knowledge; Inputting the emergency situation state sequence and the causal association matrix into a pre-trained time series prediction model to perform multi-step trend prediction and obtain a multi-step trend prediction result; Based on the multi-step trend prediction results, the situation evolution path of the disaster-stricken area under different emergency measures is simulated to generate the causal evolution results of the emergency situation in the disaster-stricken area.

6. The deep learning-based intelligent decision-making method for emergency situations according to claim 5, characterized in that: The generation of an emergency decision-making strategy for the disaster-stricken area based on the causal evolution results of the emergency situation and in combination with historical handling experience data and current resource constraints includes the following sub-steps: Obtain historical disposal experience data including historical emergency plans, resource consumption data and disposal effects; Convert the current number of available rescue resources, team distribution locations, and traffic network status into resource constraints; Generate an initial emergency decision-making strategy through a pre-trained deep learning model, taking the causal evolution results of the emergency situation, the historical handling experience data and the resource constraints as input; Based on the causal logic of the dynamic causal graph, the secondary disaster risks potentially caused by the initial decision-making strategy are verified, and the final emergency decision-making strategy for the disaster-stricken area is output.

7. An intelligent decision-making device for emergency situations based on deep learning, characterized in that: include: The data fusion module is used to fuse the multimodal data from the disaster-stricken areas and generate a unified multimodal emergency situation embedding vector; A feature extraction module is used to extract features from the multimodal emergency situation embedding vector through a spatiotemporal perception neural network model, modeling in the time series dimension and the spatial correlation dimension respectively, and outputting an emergency situation state sequence containing spatiotemporal evolution characteristics; A causal identification module is used to construct a dynamic causal graph using the emergency situation state sequence to identify potential causal relationships between disaster events and influencing factors; A prediction module, configured to predict the development trend of the future emergency situation in the disaster-stricken area based on potential causal relationships, and generate causal evolution results of the emergency situation in the disaster-stricken area; The strategy generation module is used to generate an emergency decision-making strategy for the disaster-stricken area based on the causal evolution results of the emergency situation and in combination with historical handling experience data and current resource constraints.

8. The deep learning-based intelligent decision-making device for emergency situations according to claim 7, characterized in that: The data fusion module is specifically used for: Perform feature extraction on satellite image data, meteorological data, and public opinion text data from the multimodal data of the disaster-stricken area to obtain spatial features of the satellite image data, change features of the meteorological data, and semantic features of the public opinion text data; A cross-modal attention mechanism is used to calculate the association weights between the spatial features of the satellite image data, the change features of the meteorological data, and the semantic features of the public opinion text data to generate an inter-modal association matrix; The spatial features, the change features and the semantic features are weightedly fused based on the inter-modal association matrix, and a unified multi-modal emergency situation embedding vector is generated through dimensionality reduction processing.

9. The deep learning-based intelligent decision-making device for emergency situations according to claim 8, characterized in that: The feature extraction module is specifically used for: The multimodal emergency situation embedding vector is input into a temporal convolutional network to extract the time series characteristics of the disaster diffusion speed and resource consumption rate; Based on the geographic grid division of the disaster-stricken area, a spatial topology map is constructed, and the spatial correlation characteristics of the disaster impact between regions are extracted through the graph attention network; Performing tensor splicing on the time series features and the spatial correlation features to generate a spatiotemporal joint feature vector; The spatiotemporal joint feature vectors are serialized and reorganized to output an emergency situation state sequence representing the spatiotemporal evolution law of the disaster.

10. An emergency situation intelligent decision-making system based on deep learning, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for intelligent decision-making of emergency situations based on deep learning as described in any one of claims 1 to 6 is implemented.

Images