Disaster correlation analysis system, server and method

Abstract

The invention relates to a disaster correlation analysis system and a disaster correlation analysis method, which are used for analyzing simultaneous multiple or chain disasters and specific processes of accident damage in order to carry out preventive disaster management on composite disasters by using an artificial intelligence technology. Besides, in order to improve the understanding degree of analysis results through the embodiment, the correlation analysis results of the disasters are distinguished according to the disasters and accident types, and information is provided in the form of scenes and supervision. In addition, according to the embodiment, disaster networking is used as the most important element of composite disaster analysis, and relevance between disasters is analyzed on the basis of the disasters and the relation between accident types and fragile elements.

Description

Technical Field

[0001] The present disclosure relates to a disaster correlation analysis system and method, and more particularly to a system and method for generating a hazard list-related disaster intermediary supervision by analyzing the relationship between disasters based on the relationship between disaster and accident types and vulnerable elements. Background Art

[0002] Unless otherwise mentioned in this specification, the contents described in this section are not prior art related to the claims of this application and cannot be considered as prior art even if included in this section.

[0003] Disaster connectivity analysis is the process of evaluating and analyzing the cooperation and connectivity between multiple agencies, organizations, and resources in a disaster situation. Through this, ways to improve disaster response capabilities and strengthen coordination and cooperation in disaster situations can be identified. Disaster connectivity, as an aspect of disaster management and response, is a concept that expresses the ability to respond effectively in a disaster situation through cooperation and connectivity between multiple agencies and organizations. This includes cooperation and coordination processes between government agencies, non-governmental organizations, local governments, international organizations, and various social groups.

[0004] Disaster relevance analysis is performed for the purpose of agency and organizational collaboration assessment, resource and capacity analysis, information and intent communication analysis, program and training assessment, and outcome measurement and evaluation.

[0005] The assessment of institutional and organizational cooperation refers to the assessment of how multiple institutions and organizations cooperate and connect in disaster situations. This is a process used to understand the cooperation and coordination between various entities such as government agencies, non-governmental organizations, local governments, international organizations, and social groups. Resource and capacity analysis is the process of confirming the resources and capabilities required in disaster situations and understanding the institutions and organizations required to provide and utilize the resources and capabilities, providing methods to effectively coordinate and use medical, rescue supplies, communications, transportation, and human resources.

[0006] Information and communication analysis is used to improve information flow and communication networks by analyzing information sharing and communication procedures in disaster situations, which can strengthen the rapid and accurate information transmission and decision-making process. In addition, plan and training evaluation is to evaluate the effectiveness and appropriateness of disaster response plans and training between institutions and organizations in actual situations, which can provide a basis for improving and enhancing response strategies. In addition, outcome measurement and evaluation refers to evaluating the effects achieved through correlation analysis after disaster response and deriving issues that need to be improved.

[0007] Disaster correlation analysis is an important tool for strategically improving disaster management and response. However, because the data that needs to be analyzed and evaluated is huge and the correlation extraction process and causal inference process of each data are too complicated, there is an urgent need for a system that can accurately analyze disaster correlation and generate intermediary supervision based on this.

[0008] (Prior art literature)

[0009] (Patent Document)

[0010] Patent Document 1: Korean registered patent No. 10-1891267 (August 17, 2018)

[0011] Patent Document 2: Korean Patent Publication No. 10-2019-0080414 (2019.07.08) Summary of the invention

[0012] (Problems to be solved by the invention)

[0013] The present invention relates to a disaster correlation analysis system and method, which analyzes the specific process of simultaneous multiple or chain disasters and accident damage in order to use artificial intelligence technology to carry out preventive disaster management of complex disasters.

[0014] Furthermore, in order to improve the understanding of the analysis results through the embodiment, the disaster correlation analysis results are differentiated according to the disaster and accident types, and the information is provided in the form of scenarios and supervision.

[0015] In addition, the networking of disasters is taken as the most important element of complex disaster analysis through the embodiment, and the correlation between disasters is analyzed based on the relationship between disaster and accident types and vulnerable elements.

[0016] In an embodiment, social network analysis (SNS) is used for disaster correlation analysis, and the phase structure and diffusion process are quantitatively analyzed by modeling the relationships between individuals and groups as nodes and links.

[0017] In addition, through the embodiment, vulnerability factors including disasters, accident types, regions, and facilities are used as nodes, and vulnerability factors of different disasters and accident types responded by experts are used as links to perform structural analysis.

[0018] At the same time, in the embodiment, the relevance of the disaster is explained by measuring the importance of each node considering the network structure.

[0019] Furthermore, based on the definitions of different importance categories including connection, proximity, and betweenness from social network analysis, the characteristics of disasters and objects are derived from the perspective of disaster relevance.

[0020] (Measures taken to solve the problem)

[0021] According to the embodiment, the disaster correlation analysis server includes: a memory for storing at least one instruction; and a processor; by using the processor to execute at least one instruction, the disaster type and the object where the disaster occurs are set as nodes, and the social interaction (Social Network Analysis, SNA) and relationship of each node are modeled into a network and the modeled network is analyzed, so as to analyze the disaster correlation including the centrality (Centrality) representing the importance of each node, the connectivity (Degree) representing the connection between each node and several other nodes, the clustering (Clustering) representing the group formed by the aggregation of nodes with similar attributes or relationships, and the path (path) moving through the links connecting the nodes.

[0022] According to the disaster correlation analysis server of the embodiment, the correlation of disasters is interpreted by measuring the importance of each node considering the network structure, and the characteristics of disasters and objects are derived from the perspective of disaster correlation based on the importance of social network analysis including the connection centrality, closeness centrality and betweenness centrality of each node.

[0023] (Effects of the Invention)

[0024] The disaster correlation analysis system and method described above can analyze how various agencies are connected and respond effectively through disaster correlation analysis, and enhance response capabilities by maximizing the use of resources and capabilities.

[0025] In addition, correlation analysis can be used to determine which institutions own which resources and possess which capabilities, thereby avoiding duplication of resources and achieving optimal utilization.

[0026] In addition, because accurate and rapid information sharing and intention communication must be ensured in disaster situations, information flow and intention communication networks can be analyzed through correlation analysis, thereby improving effective information transmission and intention decision-making processes.

[0027] In addition, the embodiments can be used to evaluate and improve disaster response plans and training between institutions and organizations, thereby enhancing disaster response strategies and effectively responding.

[0028] Furthermore, the embodiments can involve private enterprises and civil society groups and strengthen their cooperation, thereby mobilizing more resources and efforts to strengthen disaster response capabilities.

[0029] In addition, disaster relevance analysis can provide opportunities to simulate scenarios that can help agencies and organizations cope with actual disaster situations, thereby improving their ability to cope with actual situations.

[0030] The effects of the present invention are not limited to the effects described above, but should be understood to include all effects that can be derived from the configurations of the invention described in the detailed description of the present invention or the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a schematic diagram illustrating a disaster correlation analysis system according to an embodiment.

[0032] Figure 2 is a schematic diagram illustrating a block diagram of a disaster relevance analysis server according to an embodiment.

[0033] Figure 3 It is a schematic diagram illustrating the composition of an instruction set stored in a memory for disaster correlation analysis according to an embodiment.

[0034] Figure 4 1 is a schematic diagram illustrating an example of learning a model of data stored in a learning unit according to the embodiment. DETAILED DESCRIPTION

[0035] Next, the embodiments disclosed in this specification will be described in detail with reference to the accompanying drawings, wherein the same reference numbers are assigned to the same or similar constituent elements regardless of the figure numbers, and repeated descriptions related thereto will be omitted. The suffixes "module" and "unit" of the constituent elements used in the following description are only the result of being assigned or mixed in consideration of the convenience of the portable specification, and they themselves do not have different meanings or functions from each other. In addition, in the process of describing the embodiments disclosed in this specification, when it is determined that the specific description of the relevant known technology may cause the gist of the embodiments disclosed in this specification to become unclear, the detailed description related thereto will be omitted. In addition, the drawings are only used to help more easily understand the embodiments disclosed in this specification, and the technical ideas disclosed in this specification are not limited to the drawings, but should be understood to include all changes, equivalents and even substitutes contained in the ideas and technical scope of the present invention.

[0036] In the process of describing a plurality of components, terms including ordinal numbers such as first and second may be used, but the components are not limited by the terms. The terms are only used to distinguish one component from other components.

[0037] When a certain component is recorded as being "connected" or "connected" to other components, it may be directly connected or connected to the other components, but it should be understood that other components may exist between the two. On the contrary, when a certain component is recorded as being "directly connected" or "directly connected" to other components, it should be understood that no other components exist between the two.

[0038] In the present application, terms such as "include" or "have" are only used to indicate the existence of features, numbers, steps, actions, constituent elements, parts or combinations thereof recorded in the specification, and should not be understood as excluding the possibility of the existence or addition of one or more other features, numbers, steps, actions, constituent elements, parts or combinations thereof.

[0039] In this specification, a "unit" includes a unit realized by hardware, a unit realized by software, and a unit realized by both. In addition, one unit may be realized by two or more hardwares, or two or more units may be realized by one hardware.

[0040] In this specification, part of the actions or functions recorded as being performed by a terminal, device, or equipment may be performed by a server connected to the corresponding terminal, device, or equipment. Similarly, part of the actions or functions recorded as being performed by a server may also be performed by a terminal, device, or equipment connected to the corresponding server.

[0041] Next, the present invention will be described in detail with reference to the accompanying drawings.

[0042] Figure 1 is a schematic diagram illustrating a disaster correlation analysis system according to an embodiment.

[0043] See also Figure 1 According to the embodiment, the disaster relevance analysis system may include a disaster relevance analysis server 100 and an administrator terminal 200 .

[0044] The disaster correlation analysis server 100 analyzes the specific process of simultaneous multiple or chain disasters and accident damage in order to use artificial intelligence technology to conduct preventive disaster management of complex disasters. In addition, in an embodiment, in order to improve the understanding of the analysis results, the disaster correlation analysis server 100 distinguishes the disaster correlation analysis results according to the types of disasters and accidents, and provides information in the form of scenes and supervision. In addition, the disaster correlation analysis server 100 regards the networking of disasters as the most important element of complex disaster analysis, and analyzes the correlation between disasters based on the relationship between disaster and accident types and vulnerable elements. In an embodiment, social network analysis (SNS) is used for disaster correlation analysis, and the phase structure and diffusion process are quantitatively analyzed by modeling the relationship between individuals and groups as nodes and links.

[0045] In addition, the disaster relevance analysis server 100 uses vulnerability factors including disasters, accident types, regions, and facilities as nodes and uses vulnerability factors of different disasters and accident types responded by experts as links to perform structural analysis. In the embodiment, the directionality of the analysis process is not considered, and the response frequency is used as a weighted value.

[0046] At the same time, in the embodiment of the disaster relevance analysis server 100, the relevance of the disaster is explained by measuring the importance of each node considering the network structure. In addition, based on the definition of different importance types including connection, proximity and intermediary of social network analysis, the characteristics of the disaster and the object are derived from the perspective of the relevance of the disaster.

[0047] The administrator terminal 200 receives and outputs the analysis results through server-client communication with the disaster relevance analysis server 100, and outputs visualization materials such as matrices, scenes, graphs, and charts to visualize the analysis results.

[0048] Figure 2 is a schematic diagram illustrating a block diagram of a disaster relevance analysis server according to an embodiment.

[0049] In an embodiment, a server refers to a computing system that provides services to other computers or devices or stores and manages data in a computer network. The server 200 can receive requests from other computers or devices called clients and provide responses or data related to the corresponding requests. Figure 2 The configuration of the server 200 shown in FIG. 1 is merely an example for simplified illustration.

[0050] The communication unit 110 may be composed of various communication networks such as a short-distance communication network (PAN: Personal Area Network) and a short-distance communication network (WAN: Wide Area Network). In addition, the communication unit 110 may work based on the well-known World Wide Web (WWW: World Wide Web), and may also use wireless transmission technologies suitable for short-distance communication such as infrared (IrDA: Infrared Data Association) or Bluetooth (Bluetooth). As an example, the communication unit 110 may be responsible for sending and receiving data required to perform the technique according to one embodiment of the present disclosure.

[0051] The memory 120 may refer to any type of storage medium. For example, the memory 120 may include at least one type of storage medium such as a flash memory type, a hard disk type, a multimedia card micro type, a card-type memory (e.g., SD or XD memory, etc.), a random access memory (Random Access Memory), a static random access memory (SRAM: Static Random Access Memory), a read-only memory (ROM: Read-Only Memory), an electrically erasable programmable read-only memory (EEPROM: Electrically Erasable Programmable Read-Only Memory), a programmable read-only memory (PROM: Programmable Read-Only Memory), a magnetic memory, a magnetic disk, and an optical disk. The memory 120 described above may also constitute a storage medium such as Figure 1 The database shown.

[0052] The memory 120 may store at least one instruction executable by the processor 130. In addition, the memory 120 may store any form of information generated or determined by the processor 130 and any form of information received by the server 200. For example, as described later, the memory 120 stores RM data and RM protocols according to the user. In addition, the memory 120 stores various types of modules, instruction sets, and even models.

[0053] The processor 130 may execute the technical features of the embodiments of the present disclosure described below by executing at least one instruction stored in the memory 120. In one embodiment, the processor 130 may be composed of at least one core, and may include a processor such as a central processing unit (CPU) of a computer device, a general purpose graphics processing unit (GPGPU), and a tensor processing unit (TPU) for data analysis and / or processing.

[0054] Figure 3 It is a schematic diagram illustrating the composition of an instruction set stored in a memory for disaster correlation analysis according to an embodiment.

[0055] See also Figure 3 , according to the embodiment, the instruction set may include a collection unit 121, a preprocessing unit 122, a learning unit 123, an analysis unit 124, a process generation unit 125, a data generation unit 126, and a feedback unit 127. The term "unit" used in this specification should be interpreted as including software, hardware, or a combination thereof according to the context in which the term is used. For example, software may be machine language, firmware, embedded code, and application software. As another example, hardware may be a circuit, a processor, a computer, an integrated circuit, an integrated circuit core, a sensor, a micro-electromechanical system (MEMS; Micro-Electro-Mechanical System), a manual device, or a combination thereof.

[0056] The collection unit 121 collects a series of data required for disaster correlation analysis and model learning data. For example, the collection unit 121 collects the disaster history of different regions, disaster types, and attribute information of each disaster as a series of data required for disaster correlation analysis. In addition, the collection unit 121 can collect social network data, disaster-related data, action and intention decision data, geographic data, time data, and social media data as model learning data.

[0057] In an embodiment, the social network data numerical value represents the data of the connection relationship between nodes such as individuals, groups and institutions when a disaster occurs, which can be obtained from surveys or questionnaires, text data of social media and communication records. In an embodiment, the disaster-related data as additional information for the model includes disaster-related data such as the type of disaster, date and time of occurrence, location and scale. The data can be used to analyze the characteristics and impact of the disaster. Action and intention decision data are data that represent people's actions, intention decisions and information sharing patterns in disaster situations. Geographic data refers to geographic information related to the location of the disaster, which belongs to the information required for analyzing the association with the social network, and the scope and characteristics of the disaster can be analyzed from it. Time data is data such as the time and duration of the disaster, and the social network analysis results can be analyzed from the time level. Social media data refers to information, opinions, feelings and other information shared by people through social media in the case of a disaster. In an embodiment, the influence and connectivity in the social network can be analyzed by analyzing the social media data.

[0058] The preprocessing unit 122 preprocesses the collected data in order to remove biased or differential data from the collected artificial intelligence learning data. In an embodiment, the preprocessing unit 120 processes the collected training data set into a form suitable for artificial intelligence model learning by preprocessing it. For example, the preprocessing unit 222 can perform processes such as interference removal, outlier removal, and missing value processing. In addition, the preprocessing unit 222 can standardize the data, remove outliers, or perform data magnification adjustment through data preprocessing, thereby preventing the model from learning unnecessary patterns.

[0059] The learning unit 123 constructs a deep learning model by learning the deep learning neural network using the collected training data set. Figure 4 This is a schematic diagram illustrating an example of learning a model of data stored in a learning unit according to an embodiment. Figure 4 The learning unit 123 of the memory may store the analysis model 1, the generation model 2, and the visualization model 3. Each module or even the model may be in the form of an application program executable by the processor 130.

[0060] Analysis model 1 is an artificial neural network model that analyzes disaster correlation through social network analysis (SNA) method.

[0061] In an embodiment, the analysis model 1 includes a social network analysis model. In an embodiment, the social network analysis model provides statistical mathematical methods and techniques for analyzing and understanding the connections and relationships between nodes (such as individuals, groups, and institutions, etc.) in a social network. The model extracts and analyzes a variety of network characteristics based on social interactions and network structures. In an embodiment, the social network analysis model defines nodes (such as individuals, groups, and institutions, etc.) and edges (connections and relationships), and thereby displays social relationships in a graphical form. In addition, the model provides a variety of centrality indicators for determining which nodes in the network are more important. Centrality is a useful indicator for analyzing information transmission, influence propagation, and leadership in a network. In addition, the network analysis model is involved in detecting clusters and groups. In an embodiment, the model can provide a function to find clusters or groups formed by nodes with similar attributes or relationships. In addition, the model can analyze the information transmission path or influence propagation by analyzing the path moving along the edge connecting between nodes. At the same time, the model can visualize the network, which helps to easily understand its relationship and structure. Furthermore, the model can provide the capability to perform future predictions or scenario simulations based on social interactions and network structure.

[0062] Generative model 2 is an artificial neural network model that generates dangerous scenes based on the results of disaster correlation analysis. In an embodiment, generative model 2 may include models such as generative adversarial neural network (GAN) or variational autoencoder (VAE), but is not limited to this. Generative adversarial neural network (GAN) is composed of two neural networks called generators and discriminators, which compete and train with each other. The generator is used to generate specific data, and the discriminator is used to distinguish between real data and data generated by the generator. In an embodiment, a new dangerous scene can be generated by the generator based on the results of disaster correlation analysis, and the discriminator can be learned in a way that distinguishes between the actual dangerous scene and the generated scene.

[0063] A variational autoencoder (VAE) is a model that reconstructs the original data by re-decoding the data after encoding it into a low-dimensional latent space. In an embodiment, the encoder and decoder may be trained based on the disaster correlation analysis results to generate potential danger scenarios based on the disaster correlation analysis results.

[0064] The generation model 2 can be used to generate a dangerous scenario similar to the actual one based on the disaster correlation analysis results. At this time, the learning data can include social network analysis results and disaster-related data, additional information for generating scenarios, etc.

[0065] The visualization model 3 is an artificial neural network model that generates the disaster correlation analysis results and the scene generation results as visualization data. In an embodiment, the visualization model 3 generates a visualization display of data and generates a variety of visualization data. In an embodiment, the visualization model 3 can generate the disaster correlation analysis results or the scene generation results as visualization data using a generative adversarial neural network (GAN). The generating subnetwork is trained to generate visualization data similar to the actual data, and the discriminating subnetwork is trained to distinguish between the generated visualization data and the actual visualization data. In this way, visualization data similar to the actual data can be generated. The visualization model 3 can learn the latent space of the data through a variational autoencoder (VAE) and generate a variety of visualization data based on this. For example, a variety of visualization data can be generated by mapping the disaster correlation analysis results or the scene generation results to the latent space.

[0066] In this specification, a model may refer to any form of computer program that works based on a network function, a neural network and / or a neural network. In this specification, a model, a neural network and a network function may be used in interchangeable meanings. A neural network is composed of one or more nodes connected to each other by one or more links, thereby forming a relationship between input nodes and output nodes in the neural network. In a neural network, the characteristics of the neural network can be determined based on the number of nodes and links, the association between nodes and links, and the weighted values ​​assigned to each link. A neural network may be composed of a collection of more than one nodes. A portion of the collection of nodes that constitute a neural network may constitute a layer.

[0067] A deep neural network (DNN) may refer to a neural network that includes multiple hidden layers in addition to an input layer and an output layer. A deep neural network may include, for example, a convolutional neural network (CNN), a recurrent neural network (RNN), an autoencoder, a generative adversarial neural network (GAN), a restricted Boltzmann machine (RBM), a deep belief network (DBN), a Q network, a U network, a twin neural network, a generative adversarial network (GAN), and a transformer. The above-described deep neural network is merely an exemplary description, and the present disclosure is not limited thereto.

[0068] A neural network can learn by at least one of supervised learning, unsupervised learning, semi supervised learning, self-supervised learning, or reinforcement learning. Learning of a neural network can be the process of applying knowledge that the neural network uses to perform a specific action to the neural network.

[0069] Neural networks can learn in the direction of minimizing output errors. Learning in neural networks is a process of repeatedly inputting learning data into a neural network and calculating the error between the output of the neural network related to the learning data and the target, and then backpropagating the error of the neural network from the output layer of the neural network to the input layer in the direction used to reduce the error, thereby updating the weighted values ​​of each node of the neural network. In supervised learning, data labeled with the correct answer in each learning data can be used, while in unsupervised learning, data without the correct answer in each learning data can be used. The connection weight value of each node updated can be determined by the learning rate. The calculation of the neural network related to the input data and the backpropagation of the error can constitute a learning cycle. Different learning rates can be applied according to the number of repetitions of the learning cycle of the neural network. In addition, methods such as increasing learning data, regularization, disabling the loss of some nodes, and batch normalization layer can be applied to prevent overfitting.

[0070] In one embodiment, the model may use at least a portion of a converter. The converter may be composed of an encoder that encodes the embedded data and a decoder that decodes the encoded data. The converter may adopt a structure that receives a series of data and outputs a series of data of different types through encoding and decoding steps. In one embodiment, the series of data may be processed into a form that can be calculated by the converter. The process of processing a series of data into a form that can be calculated by the converter may include an embedding process. Expressions such as data tokens, embedding vectors, and embedding tokens may refer to data embedded in a form that can be processed by the converter.

[0071] In order to encode and decode a series of data through the converter, the encoder and decoder in the converter can be processed using an attention algorithm. The attention algorithm can refer to an algorithm that calculates the similarity associated with one or more keys for a given query, and after reflecting the similarity calculated as described above to the value corresponding to each key, calculates the attention value by weighted summing the values ​​reflecting the similarity.

[0072] Depending on how the query, key, and value are set, various types of attention algorithms can be classified. For example, when all the same query, key, and value are set and attention is calculated, it can be referred to as a self-attention algorithm. In order to process a series of input data in parallel, the dimension of the embedding vector is reduced and the attention head is calculated separately for each segmented embedding vector and the attention is calculated thereby, it can be referred to as a multi-head attention algorithm.

[0073] In one embodiment, the converter may be composed of multiple modules that execute a multi-head self-attention algorithm or a multi-head encoder-decoder algorithm. In one embodiment, the converter may also include additional components other than attention algorithms such as embedding, normalization, and Softmax. The method of constructing a converter using an attention algorithm may include the method disclosed in Vaswanietal., Attention Is All You Need, 2017 NIPS, which is incorporated into this specification by reference.

[0074] The converter can be applied to various data domains such as embedded natural language, segmented image data, and audio waveforms, thereby converting a series of input data into a series of output data. In order to convert data with multiple data domains into a series of data that can be input to the converter, the converter can embed the data. The converter can also process additional data that represents the relative position relationship or phase relationship between a series of input data. Alternatively, a vector reflecting the relative position relationship or phase relationship between the input data can be added to a series of input data and the series of input data can be embedded. In one example, the relative position relationship between a series of input data can include, for example, the word order in a natural language sentence, the relative position relationship of each segmented image, and the time sequence of the segmented audio waveform, but is not limited to this. The process of adding information representing the relative position relationship or phase relationship between a series of input data can be called positional encoding.

[0075] In one embodiment, the model may include at least one of a recurrent neural network (RNN), a long short-term memory (LSTM) network, a deep neural network (DNN), a convolutional neural network (CNN), and a bidirectional recurrent deep neural network (BRDNN), but is not limited thereto.

[0076] In one embodiment, the model may be a model learned by transfer learning. Transfer learning refers to a learning method in which a pre-learned model having a first task is obtained by pre-training a large volume of unlabeled learning data in a semi-supervised learning or self-supervised learning manner, and then the pre-learned model is fine-tuned and labeled in a manner consistent with a second task by supervised learning to construct a target model.

[0077] Re-read Figure 3 The analysis unit 124 sets the disaster type and the object where the disaster occurs as nodes, models the social interaction (Social Network Analysis, SNA) and relationship of each node into a network and analyzes the modeled network. In an embodiment, the analysis unit 124 analyzes the disaster relevance including the centrality (Centrality) indicating the importance of each node, the connectivity (Degree) indicating the connection between each node and several other nodes, the clustering (Clustering) indicating the group formed by the aggregation of nodes with similar attributes or relationships, and the path (path) moving through the links connecting the nodes.

[0078] For example, the analysis unit 124 calculates an index representing the importance of each node. Centrality is an index used to measure how important a node is in a network. Centrality analysis can be expressed in a variety of forms, such as connection centrality, proximity centrality, and betweenness centrality. In an embodiment, connection information centrality can be represented by the number of connected nodes, while proximity centrality can be represented by the sum of the shortest distances between nodes. Betweenness centrality can be represented by the degree of connectivity between node groups. To this end, the analysis unit 124 calculates connectivity, an index used to represent the connection of each node with several other nodes. In this way, the degree of active relationship formation of the node can be analyzed. In addition, the analysis unit 124 performs clustering analysis to detect groups formed by aggregation of nodes with similar attributes or relationships. In an embodiment, a collection of nodes associated with each other in a network can be found by clustering. In addition, the analysis unit 124 analyzes the path of movement through the links connected between nodes, and thereby analyzes the propagation path of information or influence.

[0079] In the embodiment, connection centrality is the number of times a node is directly connected to other nodes, and closeness centrality is an indicator of the path that a node affects other nodes. Therefore, the higher the closeness centrality, the analysis unit 124 will analyze that the number of intermediate nodes between two nodes is relatively small. Betweenness centrality refers to the intermediary role played by a node in a network. Thus, when a node is located on more paths between other nodes in the connection network, the analysis unit 124 analyzes that its betweenness centrality is higher.

[0080] The analysis unit 124 interprets the relevance of disasters by measuring the importance of each node in the network structure according to the disaster relevance analysis results, and derives the characteristics of disasters and objects from the perspective of disaster relevance based on the importance of social network analysis including the connection centrality, closeness centrality and betweenness centrality of each node. For example, the analysis unit 124 can derive the characteristics of disasters and objects based on the importance measurement results of social network analysis. In an embodiment, the analysis unit 124 can analyze which nodes are important and through which paths information and influence are transmitted, and apply them to predict and understand the vulnerability or influence of disasters and objects.

[0081] Specifically, the analysis unit 124 can extract the network centrality superordinate groups of disaster types and vulnerability factors based on the analysis results and calculate the object disaster type, object type, and connection, proximity, and intermediary importance values ​​of the superordinate groups. In an embodiment, the analysis unit 124 can extract the importance superordinate groups that consider the connection, proximity, and intermediary importance values ​​separately, or extract the superordinate groups in the case of considering the connection, proximity, and intermediary importance values ​​as a total, or extract the superordinate groups based on the average values ​​of the connection, proximity, and intermediary importance values.

[0082] In addition, the analysis unit 124 uses the generation model to generate a response program that takes into account disasters, accidents, and object characteristics that rank higher according to centrality type, and generates a response program that prioritizes response to areas of concern including areas with dense elderly populations, central general commercial areas, and the metropolitan area.

[0083] To this end, the analysis unit 124 uses centrality indicators (such as connection centrality, proximity centrality, and betweenness centrality, etc.) to evaluate the importance of each node (disaster, accident, and object) and calculate the ranking. Next, the important nodes are identified according to each indicator. In addition, the analysis unit 124 analyzes the characteristics related to disasters and accidents with higher rankings, taking into account the characteristics of disasters and accidents. At this time, the characteristics related to disasters and accidents are derived taking into account the type of disaster, the possibility of occurrence, and the scope of impact. Next, the analysis unit 124 analyzes the characteristics related to the higher-ranked objects. For example, which objects may be affected and the connection and influence between the two are analyzed, thereby deriving the characteristics related to the object. Next, the analysis unit 124 generates a program for responding to disasters and accidents based on the information obtained. That is, a response plan is planned taking into account what measures should be taken in the event of a disaster or accident and what objects need to be protected.

[0084] In addition, the analysis unit 124 will set areas of concern such as areas where the elderly population is densely populated, central general commercial areas, and metropolitan areas. Areas of concern refer to areas with higher risks or greater influence. In an embodiment, the analysis unit 124 will adjust and concretize the response procedures while giving priority to the set areas of concern. For example, the analysis unit 124 generates a program including special response strategies or measures for the areas of concern, and uses it to respond when a disaster or accident occurs.

[0085] The generation unit 125 assumes that disasters occur continuously and simultaneously based on the relationship between disasters, accidents and objects and generates dangerous scenarios. To this end, the generation unit 125 analyzes the relationship between accidents and objects. Understand which accidents will affect which objects and consider the interaction between the two. Next, the generation unit 125 will assume that accidents may occur continuously. This refers to considering the situation that may become the cause of another accident after one accident occurs. For example, the generation unit 125 can consider the scenario that a second accident may occur because of the first accident. In addition, the generation unit 125 will assume that multiple accidents may occur at the same time. For example, the generation unit 125 will consider the situation that multiple accidents occur within an hour or at the same time, and analyze the impact and interaction when multiple accidents occur at the same time. Next, the generation unit 125 will generate multiple dangerous scenarios based on the relationship between continuous and simultaneous accidents and objects. The scenarios generated in the generation unit 125 will represent the continuous or simultaneous impact between accidents and objects, and consider in which order or combination accidents may occur under disaster conditions. In addition, the generation unit 125 will evaluate and analyze the generated dangerous scenarios. For example, the generator 125 may select and decide which scenario to respond to first, taking into account the likelihood, influence, and response strategies of each scenario. In addition, the generator 125 formulates response strategies and plans related to the selected dangerous scenarios. For example, the generator 125 formulates plans related to response plans, measures, and resource allocations related to each scenario, so as to effectively respond to disasters or accidents.

[0086] The data generating unit 126 generates visualization data of the analysis results of the complex disaster occurrence pattern diagram, the correlation network diagram between disasters, and the correlation matrix between disasters.

[0087] In an embodiment, the data generation unit 126 generates a composite disaster occurrence pattern diagram based on the disaster occurrence scene. The pattern diagram will help understand the impact and interaction of each disaster element. In addition, the data generation unit 126 will generate a network diagram showing the relationship between disasters. The network diagram shows whether a disaster may be related to other disasters by visualizing the connection relationship between disasters. In an embodiment, the data generation unit 126 visualizes the network structure by showing the nodes of each disaster and the edges of the connection relationship.

[0088] In addition, the data generation unit 126 generates a correlation matrix between disasters. In an embodiment, the data generation unit 126 can generate a matrix that numerically represents the correlation between disasters. The matrix displays the degree of interaction between disasters in an array form and represents the correlation in numerical values. In this way, the correlation between disasters can be quantitatively evaluated.

[0089] Next, the data generation unit 126 can generate an analysis result visualization data by integrating the generated composite disaster occurrence pattern diagram, the correlation network diagram between disasters, and the correlation matrix between disasters. In this way, multiple visualization elements can be simultaneously displayed, thereby helping to more clearly understand the complexity and correlation of disasters.

[0090] The feedback unit 127 evaluates the learned artificial neural network model and the deep learning model. In an embodiment, the feedback unit 127 can evaluate the artificial neural network model by at least one of accuracy, precision and recall. Accuracy is an indicator for measuring the consistency between the results predicted by the artificial neural network model and the actual results. Precision is an indicator for measuring the proportion of the results predicted to be positive that are actually positive. Recall is an indicator for measuring the proportion of the actual positives predicted by the model. In an embodiment, the feedback unit 127 can evaluate the artificial neural network model based on at least one of the calculated indicators after calculating the accuracy, precision and recall of the artificial neural network model.

[0091] In an embodiment, the feedback unit 127 can use an evaluation data set to measure the accuracy of the artificial neural network model. The evaluation data set is composed of data that is not used in the learning of the model and is used to objectively evaluate the performance of the model. In an embodiment, the feedback unit 127 uses the evaluation data set to execute the artificial neural network model and compares the predicted values ​​of the artificial neural network model related to each input value with the actual correct value of the corresponding data. Next, the accuracy of the model prediction can be measured by the comparison result. For example, the accuracy can be calculated in the feedback unit 127 by the proportion of data correctly predicted by the model in all data.

[0092] In addition, the feedback unit 127 can calculate an index calculated using the coordinated average of precision and reproducibility, that is, an index indicating the balance between precision and reproducibility, namely, an F1 score (F1 Score), and evaluate the artificial neural network model based on the calculated F1 score, and then generate an index that visualizes the performance of the classification model through a graph, namely, an area under the curve-receiver operating characteristic (AUC-ROC) curve, and then evaluate the artificial neural network model based on the generated AUC-ROC curve. In an embodiment, the closer the area under the receiver operating characteristic (ROC) curve (AUC) is to 1, the better the performance of the model can be evaluated by the feedback unit 127.

[0093] In addition, the feedback unit 127 can evaluate the interpretability of the artificial neural network model. In an embodiment, the feedback unit 127 can evaluate the interpretability of the artificial neural network model by Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) methods. Shapley Additive Explanations (SHAP) is a library for explaining the results predicted by the model, and the feedback unit 127 extracts the Shapley Additive Explanations (SHAP) value from the library. In an embodiment, the feedback unit 127 can predict the degree of influence of the characteristic information input into the model on the model prediction by extracting the Shapley Additive Explanations (SHAP) value.

[0094] The Local Interpretable Model-agnostic Explanations (LIME) method is a method for explaining model predictions related to individual samples. In an embodiment, the feedback unit 127 can approximate the sample into an interpretable model and calculate the importance of each characteristic information through the Local Interpretable Model-agnostic Explanations (LIME) method. In addition, the feedback unit 127 can infer the influence of each characteristic variable by analyzing the internal weight value and deviation value of the model.

[0095] The feedback unit 127 performs improvement work when the fairness of the artificial neural network model is low or shows differences. In an embodiment, when the data of a specific set does not reach a certain level or above, the feedback unit 127 collects additional data representing the specific set and performs a data preprocessing process. In an embodiment, the feedback unit 127 prevents the model from learning unnecessary patterns by performing a data preprocessing process including data standardization, outlier removal, and data multiplier adjustment. In addition, in an embodiment, the feedback unit 127 can prevent differences or ensure fairness by adding specific conditions to the model learning algorithm.

[0096] In an embodiment, the feedback unit 127 compares the predicted results of the model with the actual results through a confusion matrix analysis to ensure fairness and thereby evaluates the performance of the model. A confusion matrix is ​​a matrix for evaluating the classification performance of a model in supervised learning. The confusion matrix compares the results predicted by the model with the actual results and represents the classification results. In an embodiment, the feedback unit 127 can calculate the accuracy and misclassification rate associated with each class through a confusion matrix analysis, thereby evaluating the performance of the model.

[0097] In addition, in an embodiment, the feedback unit 127 can confirm the distribution of the data by visually analyzing the learning data. For example, for image data, the image samples related to each class can be visualized to evaluate the diversity and fairness of the data.

[0098] In addition, the feedback unit 127 can verify the engineering and diversity of the learning data through fairness verification and evaluation index calculation, thereby improving the artificial neural network. In an embodiment, fairness verification refers to confirming by the artificial neural network model whether a specific attribute of the data in the learning information shows differences. In an embodiment, the feedback unit 127 can compare the number of samples related to each attribute or evaluate the classification performance related to each attribute, thereby confirming whether there is a difference related to a specific attribute.

[0099] In addition, the feedback unit 127 calculates various indicators for evaluating the performance of the artificial neural network model. For example, the performance of the model can be evaluated by calculating indicators such as accuracy, precision, reproducibility, and F1-score. At this time, the fairness and diversity of the model can be evaluated by calculating indicators related to each class.

[0100] In addition, the feedback unit 127 can collect feedback information related to problems that occur when the artificial neural network is used in an actual environment, and reflect the collected feedback information in the artificial neural network model, thereby continuously improving the artificial neural network model.

[0101] The disaster correlation analysis system and method described above can analyze how various agencies are connected and respond effectively through disaster correlation analysis, and enhance response capabilities by maximizing the use of resources and capabilities.

[0102] In addition, correlation analysis can be used to determine which institutions own which resources and possess which capabilities, thereby avoiding duplication of resources and achieving optimal utilization.

[0103] In addition, because accurate and rapid information sharing and intention communication must be ensured in disaster situations, information flow and intention communication networks can be analyzed through correlation analysis, thereby improving effective information transmission and intention decision-making processes.

[0104] In addition, the embodiments can be used to evaluate and improve disaster response plans and training between institutions and organizations, thereby enhancing disaster response strategies and effectively responding.

[0105] Furthermore, the embodiments can involve private enterprises and civil society groups and strengthen their cooperation, thereby mobilizing more resources and efforts to strengthen disaster response capabilities.

[0106] In addition, disaster relevance analysis can provide opportunities to simulate scenarios that can help agencies and organizations cope with actual disaster situations, thereby improving their ability to cope with actual situations.

[0107] The disclosed contents are merely illustrative and can be implemented in various ways by a person having ordinary knowledge in the relevant technical field without departing from the gist of the claims. Therefore, the protection scope of the disclosed contents is not limited to the specific embodiments described above.

Claims

1. A disaster correlation analysis server, comprising: A memory for storing at least one instruction; as well as, processor; By executing the at least one instruction with the processor, The disaster types and the objects that have been affected by the disaster are set as nodes, and the social interactions (Social Network Analysis, SNA) and relationships of each node are modeled into a network and the modeled network is analyzed. The disaster correlation is analyzed, including the centrality (Centrality) that indicates the importance of each node, the connectivity (Degree) that indicates the connection between each node and several other nodes, the clustering (Clustering) that indicates the grouping of nodes with similar attributes or relationships, and the path (Path) that moves through the links connecting the nodes.

2. The disaster correlation analysis server according to claim 1, characterized in that: The disaster correlation analysis server, The relevance of disasters is explained by measuring the importance of each node considering the network structure, and the characteristics of disasters and objects are derived from the perspective of disaster relevance based on the importance of social network analysis including the connection centrality, closeness centrality and betweenness centrality of each node.

3. The disaster correlation analysis server according to claim 2, characterized in that: The connection centrality is the number of times a node is directly connected to another node, the closeness centrality represents the path of influence of one node on another node, the higher the closeness centrality, the smaller the number of intermediate nodes between the two nodes, and the betweenness centrality refers to the intermediary role played by a node in the network. Regarding the betweenness centrality, when a node is on more paths between other nodes in the connected network, its betweenness centrality is higher.

4. The disaster correlation analysis server according to claim 1, characterized in that: The disaster correlation analysis server, A social interaction analysis model is constructed by using a training data set to learn a deep learning neural network including at least one of a deep neural network (DNN: Deep Neural Network), a convolutional neural network (CNN: Convolutional Neural Network), a recurrent neural network (RNN: Recurrent Neural Network) and a bidirectional recurrent deep neural network (BRDNN: Bidirectional Recurrent Deep Neural Network).

5. The disaster correlation analysis server according to claim 1, characterized in that: The disaster correlation analysis server, Generate a response program that takes into account disasters, accidents, and target entity characteristics that rank high by centrality type, and generate a response program that prioritizes response to areas of concern, including areas with densely populated elderly populations, central general commercial areas, and the metropolitan area.

6. The disaster correlation analysis server according to claim 1, characterized in that: The disaster correlation analysis server, A danger scenario is generated based on the relationship between disasters, accidents, and objects, assuming that disasters occur continuously and simultaneously.

7. The disaster correlation analysis server according to claim 1, characterized in that: The disaster correlation analysis server, Generate visualization data of analysis results including complex disaster occurrence pattern diagram, correlation network diagram between disasters and correlation matrix between disasters.

Images