Dynamic digital plan simulation and analysis system for multiple scenes

Abstract

The invention provides a dynamic digital plan simulation and analysis system for multiple scenes, relates to the field of data intelligent analysis, and adopts a data processing technology based on machine learning to carry out semantic embedding coding on a digital plan to be applied and accident information. And automatically generating a disaster level evolution trend label according to the priori modal cross-domain fusion representation between the to-be-applied digital plan semantic embedding feature and each accident information semantic embedding feature. Therefore, by combining the digital plan information and the real-time accident information, the dynamic state of accident development can be predicted more accurately, so that a coping strategy can be flexibly adjusted to adapt to the actual development of the accident, and more effective emergency management can be realized.

Description

Technical Field

[0001] The present application relates to the field of intelligent data analysis, and more specifically, to a dynamic digital emergency plan simulation and analysis system for multiple scenarios. Background Art

[0002] In today's complex and ever-changing social and natural environment, various accidents occur frequently, covering natural disasters, industrial accidents, public health incidents, etc. These accidents are often characterized by suddenness, complexity and dynamic evolution, bringing huge challenges to social security and economic development.

[0003] The traditional emergency management system has significant limitations when analyzing the evolution of accidents and cannot provide sufficient real-time data and scientific basis to support decision-making. Specifically, the traditional decision-making process often relies on limited historical experience and outdated information, which not only hinders the accurate prediction of the development dynamics of accidents, but also weakens the timeliness and effectiveness of response measures. This information lag and lack of analytical capabilities make it difficult for emergency response to achieve optimal results. In addition, traditional emergency management mainly relies on static plans based on fixed scenarios and historical experience. Due to its inherent limitations, it is difficult to respond immediately to these rapidly changing situations and flexibly adjust response strategies to adapt to the actual development of the accident.

[0004] Therefore, a dynamic digital emergency plan simulation and analysis system for multiple scenarios is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a dynamic digital emergency plan simulation and analysis system for multiple scenarios.

[0006] According to one aspect of the present application, a dynamic digital emergency plan simulation and analysis system for multiple scenarios is provided, which includes: an accident information collection module, which is used to collect accident information in real time to obtain a time series distribution of the accident information; a digital emergency plan extraction module to be applied, which is used to extract the digital emergency plan to be applied from the digital emergency plan library; a digital emergency plan semantic embedding module to be applied, which is used to perform semantic embedding coding on the digital emergency plan to be applied to obtain semantic embedding coding features of the digital emergency plan to be applied; an accident information semantic embedding module, which is used to perform semantic embedding coding on each accident information in the time series distribution of the accident information to obtain the time series distribution of the semantic embedding coding features of the accident information; a simulation and analysis module, which is used to perform semantic embedding coding on the digital emergency plan to be applied A large-model-based simulation analysis is performed using the time series distribution of the semantic embedded coding features of the digital plan and the semantic embedded coding features of the accident information to obtain a simulation analysis result for representing a label representing the evolution trend of the disaster level, wherein the simulation analysis module comprises: a digital plan to be applied-accident information semantic fusion unit, which is used to perform a priori modal-aware cross-domain embedding fusion on the time series distribution of the semantic embedded coding features of the digital plan to be applied and the semantic embedded coding features of the accident information to obtain a cross-domain semantic fusion feature of the digital plan to be applied-accident information; and a simulation analysis result generation unit, which is used to obtain the simulation analysis result based on the cross-domain semantic fusion feature of the digital plan to be applied-accident information.

[0007] Compared with the prior art, the dynamic digital emergency plan simulation and analysis system for multiple scenarios provided by this application uses machine learning-based data processing technology to perform semantic embedding encoding on the digital emergency plan to be applied and each accident information, so as to automatically generate disaster level evolution trend labels based on the prior modal cross-domain fusion representation between the semantic embedding features of the digital emergency plan to be applied and the semantic embedding features of each accident information. In this way, by combining digital emergency plan information and real-time accident information, it is possible to make more accurate predictions on the development of accidents, which is conducive to flexible adjustment of response strategies to adapt to the actual development of accidents, and to achieve more effective emergency management. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0009] In the attached picture: Figure 1 It is a system block diagram of a dynamic digital emergency plan simulation and analysis system for multiple scenarios according to an embodiment of the present application.

[0010] Figure 2Schematic diagram of data flow of a dynamic digital emergency plan simulation and analysis system for multiple scenarios according to an embodiment of the present application.

[0011] Figure 3 It is a block diagram of a simulation and analysis module in a dynamic digital emergency plan simulation and analysis system for multiple scenarios according to an embodiment of the present application.

[0012] Figure 4 It is a block diagram of a digital plan-accident information semantic fusion unit to be applied in a dynamic digital plan simulation and analysis system for multiple scenarios according to an embodiment of the present application. DETAILED DESCRIPTION

[0013] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0014] In today's complex social and natural environment, various types of accidents are on the rise, involving a wide range of fields such as natural disasters, industrial accidents, and public health incidents. These accidents are generally characterized by suddenness, complexity, and dynamic evolution, posing severe challenges to social security and economic development.

[0015] The traditional emergency management system has exposed obvious limitations when analyzing the evolution of accidents. On the one hand, the system is difficult to provide sufficient real-time data and scientific basis, which in turn affects the scientificity and accuracy of decision-making. Specifically, the traditional decision-making process relies on limited historical experience and outdated information, which not only hinders the accurate prediction of the dynamic development of accidents, but also greatly reduces the timeliness and effectiveness of response measures. This information lag and lack of analytical ability greatly restrict the optimal effect of emergency response. On the other hand, traditional emergency management mainly relies on static plans based on fixed scenarios and historical experience. However, due to its inherent limitations, it is difficult to respond in a timely manner in the face of rapidly changing actual situations, and it is impossible to flexibly adjust the response strategy to match the actual development process of the accident.

[0016] In response to the above technical problems, the technical concept of the present application is to collect accident information in real time to obtain the time series distribution of accident information, and extract the digital plan to be applied from the digital plan library, and use machine learning parameter analysis and semantic embedding technology to semantically embed the digital plan to be applied and each accident information. Encode, so as to automatically generate a disaster level evolution trend label based on the prior modal cross-domain fusion representation between the semantic embedding features of the digital plan to be applied and the semantic embedding features of each accident information. The present application can collect accident information in real time, and automatically analyze the development process of the accident according to the time series distribution of this information. At the same time, the most suitable plan is automatically selected from the digital plan library based on real-time data, and flexibly adjusted according to the evolution of the accident to ensure that the response strategy always remains in the best state.

[0017] Figure 1 It is a system block diagram of a dynamic digital emergency plan simulation and analysis system for multiple scenarios according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of a dynamic digital emergency plan simulation and analysis system for multiple scenarios according to an embodiment of the present application. Figure 1 and Figure 2 As shown, in the dynamic digital plan simulation and analysis system 100 for multiple scenarios, it includes: an accident information collection module 110, which is used to collect accident information in real time to obtain the time series distribution of the accident information; a digital plan extraction module 120, which is used to extract the digital plan to be applied from the digital plan library; a digital plan semantic embedding module 130, which is used to perform semantic embedding coding on the digital plan to be applied to obtain the semantic embedding coding features of the digital plan to be applied; an accident information semantic embedding module 140, which is used to perform semantic embedding coding on each accident information in the time series distribution of the accident information to obtain the time series distribution of the accident information semantic embedding coding features; a simulation analysis module 150, which is used to perform simulation analysis on the time series distribution of the semantic embedding coding features of the digital plan to be applied and the semantic embedding coding features of the accident information based on a large model to obtain a simulation analysis result for representing a label for the evolution trend of the disaster level.

[0018] In the embodiment of the present application, the accident information collection module 110 is used to collect accident information in real time to obtain the time series distribution of accident information. Specifically, in the embodiment of the present application, the accident information is a simulation parameter input by the user, and the simulation parameters include geographical location, disaster level and current environmental conditions. It should be understood that the geographical location refers to the specific location where the accident occurred, which can be an accurate coordinate (longitude, latitude) or a broader description of the geographical area. The disaster level is classified according to the severity of the accident, such as small, medium or large accidents, or different levels defined according to specific standards. This information helps to assess the potential hazards and response requirements of the accident. The current environmental conditions specifically include but are not limited to weather conditions (temperature, humidity, wind speed, etc.), topographic features, surrounding building density and type, traffic flow, population distribution, etc., which will affect the development trend of the accident. The time series distribution of accident information reflects a series of state changes from the occurrence to the development of the accident. By collecting and analyzing these time series information, it can help the model identify the pattern and trend of the development of the accident. For example, if a number of small-scale fires are reported in succession in a certain area, and there is a trend of expansion over time, the system may predict that the risk of fire in the area will increase in the future, and then give a corresponding disaster level evolution trend label, such as "from small fires to medium-sized fires." It is worth noting that the accident information (i.e., the simulation parameters input by the user) can be data from any scenario. In other words, no matter what type of accident is faced, the system can be used for analysis and prediction, which makes the system very flexible and versatile, and thus can improve the utilization efficiency of the system.

[0019] In the embodiment of the present application, the digital plan extraction module 120 to be applied is used to extract the digital plan to be applied from the digital plan library. It should be understood that the digital plan to be applied is formulated based on previous accident experience and response strategies, which contains many typical situations and response measures of similar accidents. When the current accident information is input and combined with this plan, the analyzer can refer to the information in the plan to find out the pattern and development path that the current accident may follow. For example, if an urban waterlogging disaster is being handled, the extracted digital plan may contain the development and response experience of waterlogging under similar scale and terrain conditions in the past, and the analyzer can use this information to predict the development trend of the current waterlogging disaster. It is worth noting that with the continuous updating of accident information, the digital plan to be applied is not static. The system will continue to monitor the development of the situation and adjust the plan according to the new accident information. For example, if a new source of danger is found or the environmental conditions change, the system may select another plan that is more suitable for the current situation, or partially modify the existing plan. This dynamic adjustment ensures that the plan always matches the latest on-site conditions, so that the analysis results of the subsequent disaster level evolution trend are more reliable.

[0020] In an embodiment of the present application, the semantic embedding module 130 of the digital plan to be applied is used to perform semantic embedding coding on the digital plan to be applied to obtain semantic embedding coding features of the digital plan to be applied. Specifically, in an embodiment of the present application, the semantic embedding module of the digital plan to be applied is used to: use a semantic embedding encoder based on a bidirectional recurrent neural network to perform semantic embedding coding on the digital plan to be applied to obtain a semantic embedding coding vector of the digital plan to be applied as the semantic embedding coding feature of the digital plan to be applied. It should be understood that the original digital plan to be applied may contain a large amount of text content described in natural language, and these descriptions contain specific semantic meanings, such as the description statements about the rescue process in the plan. There are semantic associations between the contexts between each text fragment. Based on this, the present application performs semantic embedding coding on the digital plan to be applied to map the text information in the digital plan to a low-dimensional vector space, and mines the semantic associations between different parts of the plan to obtain the semantic embedding coding vector of the digital plan to be applied. In particular, in a specific embodiment of the present application, a semantic embedding encoder based on a bidirectional recurrent neural network is used to perform semantic embedding encoding on the digital plan to be applied to obtain a semantic embedding encoding vector of the digital plan to be applied. That is, the text information in the digital plan to be applied often has a strong contextual relevance. For example, when describing an emergency rescue process, the meaning of a certain step may depend on the description of the steps before and after it. The bidirectional recurrent neural network can simultaneously process the information from front to back (forward) and the information from back to front (reverse) of the sequence, thereby fully capturing the contextual dependencies in the text. For example, the potential contextual connection between rescue measures and disaster scene descriptions, and the bidirectional recurrent neural network can better understand the inherent logic of the plan.

[0021] In an embodiment of the present application, the accident information semantic embedding module 140 is used to perform semantic embedding encoding on each accident information in the time series distribution of the accident information to obtain the time series distribution of the accident information semantic embedding encoding features. Specifically, in an embodiment of the present application, the accident information semantic embedding module is used to: use the semantic embedding encoder based on the bidirectional recurrent neural network to perform semantic embedding encoding on each accident information in the time series distribution of the accident information to obtain the time series distribution of the accident information semantic embedding encoding vector as the time series distribution of the accident information semantic embedding encoding features. It should be understood that the original accident information contains multiple types, such as geographical location, disaster situation, etc., and there are also mutual semantic dependencies between these different types of descriptions. Therefore, in order to deeply mine the semantic features and associations of accident information, in the technical solution of the present application, semantic embedding encoding is performed on each accident information in the time series distribution of the accident information to capture and mine the semantic correlation between different accident information, capture the semantic features of the accident evolving over time, and obtain the time series distribution of the accident information semantic embedding encoding vector. In particular, in a specific embodiment of the present application, the semantic embedding encoder based on the bidirectional recurrent neural network is used to perform semantic embedding encoding on each accident information in the time series distribution of the accident information so as to utilize the bidirectional recurrent neural network to simultaneously consider the forward and reverse information of the time series, thereby fully considering the contextual semantic association and obtaining the time series distribution of the semantic embedding coding vector of the accident information.

[0022] The following is a detailed description of a specific implementation process of "using the semantic embedding encoder based on the bidirectional recurrent neural network to semantically embed the various accident information in the time series distribution of the accident information to obtain the time series distribution of the semantic embedding encoding vector of the accident information": First, in the data preprocessing stage, since the accident information (i.e., the simulation parameters input by the user) is consistent with the actual accident data, and the actual accident data comes from different sensors, monitoring equipment, etc., the data quality is uneven. Therefore, data cleaning is crucial. These data may contain a lot of noise. For example, the sensor is disturbed in a harsh environment, which will cause errors or anomalies in the transmitted data. For numerical data, it is necessary to carefully check whether its range is reasonable. For data that obviously exceeds the normal range, such as abnormally high or low temperature data monitored in a fire accident, if it exceeds the possible physical range, it needs to be identified as erroneous data and excluded. For missing data, it needs to be processed according to the characteristics of the data. For continuous data, such as ambient temperature data at different time points when a disaster occurs, if the data at a certain time point is missing, it can be filled by calculating the mean or median of the time series or using linear interpolation and other methods to ensure the integrity and continuity of the data. At the same time, data standardization or normalization is also an indispensable link, because different accident information features may have different magnitudes. Taking earthquake disasters as an example, the magnitude of an earthquake may be a larger numerical range, while the relative humidity in the weather conditions when the earthquake occurs may be a smaller numerical range. If it is not standardized, the magnitude data may have an excessive impact on the results in subsequent network processing. Here, the minimum-maximum normalization method can be used to map the data to the range of [0,1], or the Z-score normalization method can be used to convert the data into a distribution with a mean of 0 and a standard deviation of 1, so that each feature participates in subsequent calculations on the basis of equal importance.

[0023] Next, we enter the training process of the bidirectional recurrent neural network (Bi-RNN). For the accident information at each time point, it needs to be reasonably represented as a vector as the input of the network. Assuming that the accident information considered includes the geographical location (which can be accurately represented by longitude and latitude), the disaster level (which may be a discrete level classification, such as low, medium, and high, which can be mapped to numbers, such as [0,1,2]) and environmental conditions (including temperature, humidity, wind speed, air pressure, etc.), these data of different dimensions will be integrated into a vector. This vector sequence will be used as the input of the Bi-RNN, and the length of the input sequence is determined according to the actual time window. For example, if the accident information within the past 24 hours is to be analyzed, the data per hour constitutes a sequence. In the forward propagation process of the network, the input sequence starts from the first element of the sequence and passes through the forward RNN unit element by element. Each unit updates its own hidden state according to the current input and the hidden state of the previous unit, and passes the updated hidden state to the next unit. This information transmission is gradually accumulated, and it will take into account the order of the time series to capture the temporal dependency. At the same time, the back propagation process is also in progress, starting from the last element of the sequence, updating the hidden state through the reverse RNN unit and passing it forward. In the semantic embedding coding stage, the bidirectional recurrent neural network is constantly updating the hidden state, and the final hidden state contains the information of the entire sequence. For each time point, its corresponding forward and reverse hidden states are spliced, or other suitable combinations are used to form the semantic embedding coding vector of that time point. In this way, by operating on each time point in the entire time series distribution, a new vector sequence can be obtained, that is, the time series distribution of the semantic embedding coding vector of the accident information. The length of this new time series distribution is the same as the input accident information time series distribution, but each vector has a higher dimension and can more deeply reflect the semantic information of each time point, providing a richer information basis for subsequent disaster evolution analysis and decision-making.

[0024] Then comes the training and optimization part. During the training process, it is necessary to define a suitable loss function according to the specific task objectives. If the purpose is to predict the future development trend of disasters, such as predicting the change of disaster intensity over time, the mean square error (MSE) can be used as the loss function. The performance of the model is measured by calculating the difference between the predicted value and the true value to guide the model to adjust the parameters in the direction of more accurate prediction. The preprocessed data is divided into a training set, a validation set, and a test set. The training set is used to train the Bi-RNN model. The network parameters are updated according to the loss function through the back propagation algorithm and optimizers (such as stochastic gradient descent, Adagrad, Adam, etc.). After each training cycle, the validation set is used to evaluate the performance of the model to prevent overfitting. Overfitting is a common problem. It means that the model has over-learned the noise and details in the training data and lost the ability to generalize to new data, resulting in poor performance in practical applications. By observing the loss function value on the validation set and other performance indicators, such as accuracy and recall, hyperparameters can be adjusted or the structure of the model can be fine-tuned to ensure that the model can learn the inherent laws of the data during the learning process without falling into overfitting of the training data.

[0025] Finally, it is the application and evaluation stage of the model. When the trained model is applied to new accident information time series data, the new data is input into the bidirectional recurrent neural network, and the time series distribution of the semantic embedding coding vector of the new data is output. These coding vectors can be used as input for subsequent more advanced analysis to help the model predict the evolution trend of disasters. For real-time monitoring systems, its value is even more significant. Data from new time points can be continuously input into the model, and the time series distribution of the semantic embedding coding vector can be continuously updated, allowing the system to grasp the latest situation of the accident in real time and realize dynamic monitoring of the accident status.

[0026] In the embodiment of the present application, the simulation analysis module 150 is used to perform a simulation analysis based on a large model on the time series distribution of the semantic embedded coding features of the digital emergency plan to be applied and the semantic embedded coding features of the accident information to obtain a simulation analysis result for representing the disaster level evolution trend label. Specifically, Figure 3 FIG. 1 is a block diagram of a simulation analysis module in a dynamic digital emergency plan simulation and analysis system for multiple scenarios according to an embodiment of the present application. Figure 3As shown, the simulation analysis module 150 includes: a digital plan to be applied-accident information semantic fusion unit 151, which is used to perform a priori modal-aware cross-domain embedding fusion on the time series distribution of the semantic embedded coding features of the digital plan to be applied and the semantic embedded coding features of the accident information to obtain the cross-domain semantic fusion features of the digital plan to be applied-accident information; a simulation analysis result generation unit 152, which is used to obtain the simulation analysis result based on the cross-domain semantic fusion features of the digital plan to be applied-accident information.

[0027] In the embodiment of the present application, the digital emergency plan to be applied-accident information semantic fusion unit 151 is used to perform cross-domain embedding fusion of the time series distribution of the semantic embedding coding features of the digital emergency plan to be applied and the semantic embedding coding features of the accident information to obtain the cross-domain semantic fusion features of the digital emergency plan to be applied-accident information. Specifically, Figure 4 FIG. 1 is a block diagram of a digital emergency plan-accident information semantic fusion unit to be applied in a dynamic digital emergency plan simulation and analysis system for multiple scenarios according to an embodiment of the present application. Figure 4 As shown, the digital plan to be applied-accident information semantic fusion unit 151 includes: an accident information kernel feature extraction subunit 1511, which is used to perform modal kernel feature extraction on the time series distribution of the accident information semantic embedding coding vector to obtain the accident information semantic kernel feature coding vector; a priori clustering center calculation subunit 1512, which is used to determine the digital plan to be applied-accident information semantic priori clustering center coding vector based on the digital plan semantic embedding coding vector to be applied and the accident information semantic kernel feature coding vector; a cross-domain fine-grained aggregation subunit 1513, which is used to perform cross-domain fine-grained clustering aggregation on the time series distribution of the digital plan to be applied-accident information semantic priori clustering center coding vector and the accident information semantic embedding coding vector to obtain the digital plan to be applied-accident information cross-domain semantic fusion feature vector as the cross-domain semantic fusion feature of the digital plan to be applied-accident information.

[0028] It should be understood that the semantic embedding coding features of the digital plan to be applied represent the semantic features of the preset response plan, while the temporal distribution of the semantic embedding coding features of the accident information reflects the dynamic information of the actual accident. These two types of information come from different sources and each carries a unique semantic modality. Therefore, in order to integrate the prior knowledge in the plan with the real-time information of the accident, avoid making decisions based on only a single piece of information, and thus fully and comprehensively understand the accident situation, this application integrates this prior knowledge into the fusion with the real-time accident information by performing a prior modality-aware cross-domain embedding fusion of the time series distribution of the semantic embedding coding features of the digital plan to be applied and the semantic embedding coding features of the accident information, and captures and mines the core key features of the accident information to obtain a comprehensive and comprehensive cross-domain semantic fusion feature of the digital plan to be applied-accident information. For example, in a fire accident, the general process and strategy for fire fighting in the plan is a prior knowledge, while the real-time information such as the speed of fire spread and wind direction at the accident scene is reflected through the temporal distribution of the accident information. The fusion of the two allows rescue personnel to understand the general practices and make flexible adjustments according to actual conditions.

[0029] Specifically, it is first necessary to perform modal kernel feature extraction on the temporal distribution of the accident information semantic embedding coding vector to obtain the accident information semantic kernel feature coding vector. The above process can be expressed as: ;in, is the time series distribution of the semantic embedding coding vector of the accident information, and are the first, second, and third in the time series distribution of the accident information semantic embedding coding vector. , and The semantic embedding encoding vector of accident information, For Perform kernel feature extraction, The one-norm of the eigenvector represents the number of vectors in the time series distribution of the accident information semantic embedding encoding vector minus one, yes The corresponding semantic difference value of accident information, Indicated by natural constant The exponential function with base , is the number of vectors in the time series distribution of the accident information semantic embedding encoding vector, It is the semantic kernel feature encoding vector of accident information.

[0030] It should be understood that the time series distribution of the semantic embedding coding vector of the accident information contains a large amount of complex information. For example, in an earthquake accident, the information may cover the real-time changes in the magnitude of the earthquake, the real-time dynamics of the surrounding environment (such as the damage to buildings, the degree of road congestion, etc.) and the comprehensive evolution of this information over time. In order to accurately understand the development trend of the accident, it is necessary to extract key and core features from these complex information. Based on this, the present application performs modal kernel feature extraction on the time series distribution of the semantic embedding coding vector of the accident information to obtain the semantic kernel feature coding vector of the accident information. In this way, irrelevant interference information in the original feature data can be removed, making the extracted semantic kernel feature coding vector of the accident information purer. Moreover, the core features captured by kernel feature extraction can provide a clear and effective feature basis for subsequent cross-modal interaction with the digital plan data to be applied, so that data of different modalities can interact and merge at a meaningful semantic level, thereby providing strong support for generating reliable disaster level evolution trend labels.

[0031] Specifically, in the embodiment of the present application, the a priori cluster center calculation subunit is used to: cascade the semantic embedding coding vector of the digital emergency plan to be applied and the semantic kernel feature coding vector of the accident information to obtain the cascade feature vector of the digital emergency plan-accident information semantics to be applied; perform a linear transformation on the cascade feature vector of the digital emergency plan-accident information semantics to obtain the a priori cluster center coding vector of the digital emergency plan-accident information semantics to be applied. The above process can be expressed as: ;in, is the semantic kernel feature encoding vector of accident information, is the semantic embedding coding vector of the digital plan to be applied, For cascade operation, is the weight transformation matrix, is the bias vector, It is the semantic prior clustering center encoding vector of the digital emergency plan-accident information to be applied.

[0032] It should be understood that the semantic embedding coding vector of the digital plan to be applied represents the semantic features of the preset response plan, contains the experience and knowledge of dealing with similar accidents in the past, and is a kind of prior information. The semantic kernel feature coding vector of the accident information is the core feature extracted from the dynamic information of the actual accident, reflecting the real-time status of the current accident. The two modalities of information each have their own advantages. In order to be able to combine the information of these two modalities to more comprehensively understand the accident situation and provide richer information for subsequent analysis, it is necessary to determine the digital plan to be applied-accident information semantic prior clustering center coding vector based on the semantic embedding coding vector of the digital plan to be applied and the semantic kernel feature coding vector of the accident information in this application. That is, the semantic prior clustering center coding vector of the digital plan to be applied-accident information mines the core association of the two modal information, removes interference information, and can make the disaster level evolution trend label generated based on this more reliable.

[0033] Specifically, in an embodiment of the present application, the cross-domain fine-grained aggregation subunit is used to: use a semantic distance measurement module based on the acrcosh function to perform distance measurement on each accident information semantic embedding coding vector in the time series distribution of the digital emergency plan-accident information semantic prior clustering center coding vector to be applied and the accident information semantic embedding coding vector to obtain the time series distribution of the digital emergency plan-accident information semantic measurement value to be applied. The process can be expressed as: ;in, is the time series distribution of the accident information semantic embedding encoding vector The semantic embedding encoding vector of accident information, is the semantic prior clustering center encoding vector of the digital emergency plan-accident information to be applied, is the square of the Euclidean norm of the vector, is the inverse hyperbolic cosine function, for and The semantic metric value of the digital emergency plan to be applied-accident information between them.

[0034] A binary function is used to perform clustering judgment on the time series distribution of the semantic metric value of the digital emergency plan-accident information to be applied to obtain the time series distribution of the clustering coefficient of the digital emergency plan-accident information to be applied. The process can be expressed as: ;in, for and The semantic metric value of the digital emergency plan to be applied between the accident information, yes The corresponding clustering coefficient of the digital emergency plan to be applied-accident information, is the preset threshold.

[0035] Based on the time series distribution of the clustering coefficient of the digital emergency plan to be applied-accident information, the semantic prior clustering center encoding vector of the digital emergency plan to be applied-accident information and each accident information semantic embedding encoding vector in the time series distribution of the accident information semantic embedding encoding vector are weighted and fused to obtain the cross-domain semantic fusion feature vector of the digital emergency plan to be applied-accident information. This process can be expressed as: ;in, is the time series distribution of the accident information semantic embedding encoding vector The semantic embedding encoding vector of accident information, is the semantic prior clustering center encoding vector of the digital emergency plan-accident information to be applied, yes The corresponding clustering coefficient of the digital emergency plan to be applied-accident information, is the number of vectors in the time series distribution of the accident information semantic embedding encoding vector, It is the cross-domain semantic fusion feature vector of the digital emergency plan to be applied-accident information.

[0036] It should be understood that although the semantic prior clustering center encoding vector of the digital plan-accident information to be applied has combined the key features of the plan and accident information, the actual accident situation is complex and changeable, and this encoding vector alone is not enough to fully reflect all information. The time series distribution of the semantic embedding encoding vector of the accident information records the detailed dynamic information of the accident over time. By performing cross-domain fine-grained clustering aggregation on the two, the deep fusion of multimodal information can be achieved to more comprehensively characterize the accident scene. It is also worth mentioning that the accident data (i.e., the simulation parameters input by the user) comes from the real data in the real accident, but in the actual scene, the current environmental conditions in the accident data are often interfered by various factors, and there are noise or outliers. By performing cross-domain fine-grained clustering aggregation on the data, unnecessary interference factors can be filtered out, focusing on the part that best reflects the essential characteristics of the accident, thereby providing richer and more accurate data support for the prediction of the evolution trend of the disaster level.

[0037] In an embodiment of the present application, the simulation analysis result generating unit 152 is used to obtain the simulation analysis result based on the cross-domain semantic fusion feature of the digital plan to be applied-accident information. Specifically, in an embodiment of the present application, the simulation analysis result generating unit is used to: input the cross-domain semantic fusion feature vector of the digital plan to be applied-accident information into the simulation analyzer to obtain the simulation analysis result, and the simulation analysis result is a disaster level evolution trend label. More specifically, the simulation analyzer is a classifier-based simulation analyzer. It should be understood that the cross-domain semantic fusion feature vector of the digital plan to be applied-accident information is classified and processed by the simulation analyzer based on the classifier to use the classifier to analyze and judge these features according to its internal algorithm and model structure, so as to map them to different categories, and obtain the simulation analysis result, and the simulation analysis result is a disaster level evolution trend label. In this way, it can help decision makers understand the possible development direction of the disaster in advance, so as to formulate and adjust emergency strategies in a targeted manner. For example, if the label shows that the disaster level may rise, decision makers can deploy more rescue resources in advance and strengthen monitoring and early warning work; if the label shows that the disaster level tends to stabilize, decision makers can appropriately adjust the allocation of rescue forces and focus on maintaining the current control situation.

[0038] The following is a detailed description of a specific implementation process of "inputting the cross-domain semantic fusion feature vector of the digital plan to be applied-accident information into the simulation analyzer to obtain the simulation analysis result, and the simulation analysis result is the disaster level evolution trend label": input the cross-domain semantic fusion feature vector of the digital plan to be applied-accident information into the simulation analyzer based on the classifier, and its internal calculation and decision-making process is immediately started. If the classifier is built based on a traditional machine learning algorithm, such as a decision tree, the input feature vector will start from the root node and judge and branch according to the pre-set splitting rules based on different features. For example, in the analysis of fire accidents, the decision tree may first branch for the first time based on whether the flame temperature exceeds a certain threshold, and then continue to subdivide according to other features such as the burning area and smoke concentration, until the input vector is accurately divided into the corresponding category, that is, the disaster level evolution trend category to which it belongs is determined.

[0039] The simulation analyzer will eventually output the simulation analysis results, which is the disaster level evolution trend label. This label reflects the direction and trend of the future development of the disaster, and it has different specific manifestations for different types of disasters. In natural disasters, such as earthquake disasters, if the output is a "weakening trend" label, it means that according to the input feature information, including magnitude, aftershock frequency, resource allocation and rescue conditions in the emergency digital plan, the energy release of the earthquake is gradually weakening, and the possibility and scope of subsequent secondary disasters will be reduced. The emergency management team can adjust the deployment of rescue and reconstruction resources accordingly, reduce attention to certain areas, and invest more resources in the repair and reconstruction of areas that have been affected. On the contrary, if it is an "enhancing trend", it means that the earthquake situation may be more serious, which may cause more building collapses and secondary disasters. At this time, more rescue personnel, materials and equipment need to be urgently deployed, and the scope of personnel evacuation should be expanded to cope with more serious situations that may occur.

[0040] For flood disasters, when the label "water retreat trend" is obtained, it means that after combining the current water level, flow, rainfall, and drainage and flood control measures in the emergency plan, the flood water level is decreasing, the disaster situation has improved, and emergency personnel can start to organize cleanup work and reduce the emergency rescue efforts for the affected people. If it is a "water level rising trend", it means that according to the analysis, the flood may spread further, which may be due to factors such as increased rainfall or poor drainage. It is necessary to strengthen embankment construction, improve drainage capacity, and evacuate more residents under the threat of floods.

[0041] In terms of industrial accidents, taking chemical leak accidents as an example, if the label "leakage mitigation trend" appears, it means that according to the information such as the leaked material flow, diffusion range, wind direction and speed in the feature vector and the leak control and cleanup measures in the emergency plan, the leak has been controlled, and the emergency team can adjust resource allocation and focus on the follow-up treatment of the leak. However, if it is a "leakage expansion trend", it may be due to factors such as changes in wind direction and failure to effectively block the leak source. At this time, it is necessary to expand the protection range, evacuate surrounding personnel, and take more stringent protection measures to prevent further spread of the leak.

[0042] For fire accidents, the "fire weakening trend" indicates that the fire has been controlled based on information such as flame temperature, burning area, and fire-fighting resource input. The fire-fighting force can be appropriately adjusted to carry out subsequent investigation and reconstruction work; the "fire spread trend" means that there is a risk of the fire getting out of control, and more fire-fighting resources need to be urgently called upon to expand the evacuation range to prevent the fire from causing damage to more areas.

[0043] In public health events, for infectious disease epidemics, the "epidemic easing trend" means that based on information such as the number of confirmed cases, infection rate, transmission routes and prevention and control measures, the epidemic has been effectively controlled, and society can begin to consider gradually resuming normal activities; while the "epidemic outbreak trend" indicates that the epidemic is out of control, possibly due to the emergence of new mutant strains, tight prevention and control resources, etc., and prevention and control measures need to be strengthened, such as increasing isolation efforts, accelerating vaccine development and vaccination, to control the spread of the epidemic.

[0044] In particular, when the time series distribution of the semantic embedded coding vector of the digital plan to be applied and the semantic embedded coding feature vector of the accident information respectively represent the embedded coding semantic features of the digital plan to be applied and the embedded time series features of the embedded coding semantics of the accident information, when performing cross-domain fine-grained aggregation analysis based on modal prior, the insufficient cross-domain correlation correspondence of the modal prior caused by the distribution difference between the embedded coding semantics of the digital plan to be applied in the semantic space and the embedded semantics time series space of the accident information will cause sparse fine-grained correspondence, resulting in the lack of classification reasoning degree of aggregation analysis, thereby reducing the accuracy of the obtained simulation analysis results.

[0045] Therefore, the present application optimizes the digital emergency plan-accident information cross-domain semantic fusion feature vector to be applied obtained by the simulation analyzer based on the large model for the time series distribution input of the semantic embedding coding vector of the digital emergency plan to be applied and the semantic embedding coding feature vector of the accident information. The specific optimization includes the steps of: based on the time series distribution of the digital emergency plan-accident information cross-domain semantic fusion feature vector to be applied, the feature value of the digital emergency plan-accident information cross-domain semantic fusion feature vector to be applied is obtained. Distance and The distance matrix is ​​obtained by respectively obtaining the digital emergency plan to be applied-accident information embedded semantics-time series aggregation-distance matrix and the distance matrix of the digital emergency plan to be applied, the accident information embedded semantics, and the temporal aggregation .

[0046] The weighted sum of the distance matrix of the digital emergency plan to be applied-accident information embedded semantics-time series aggregation and the distance matrix of the digital emergency plan to be applied-accident information embedded semantics-time series aggregation is calculated to obtain the joint distance matrix of the digital emergency plan to be applied-accident information embedded semantics-time series aggregation. The process can be expressed as: ;in, represents the digital emergency plan to be applied-accident information embedded semantics-temporal aggregation-distance matrix, represents the distance matrix of the digital emergency plan to be applied, the accident information embedded semantics, and the time series aggregation, It means point multiplication by position. Indicates adding by position point. and represents the weighted hyperparameter, Represents the digital emergency plan to be applied-accident information embedded semantics-temporal aggregation joint distance matrix.

[0047] Determine the eigenvalues ​​of the digital emergency plan to be applied-accident information embedded semantics-time series aggregation joint distance matrix , and form the digital emergency plan to be applied-accident information embedded semantics-time series aggregation joint distance eigenvector .

[0048] The cross-domain semantic fusion feature vector of the digital emergency plan to be applied-accident information as a row vector is matrix-multiplied with the digital emergency plan to be applied-accident information embedded semantics-temporal aggregation-distance matrix to obtain the digital emergency plan to be applied-accident information embedded semantics-temporal aggregation-distance query vector. The process can be expressed as: ;in, represents the digital emergency plan to be applied-accident information embedded semantics-temporal aggregation-distance matrix, represents matrix multiplication, represents the cross-domain semantic fusion feature vector of the digital emergency plan to be applied-accident information, and is a row vector, Represents the digital emergency plan to be applied-accident information embedded semantics-time series aggregation-distance query vector.

[0049] The digital emergency plan to be applied - accident information embedded semantics - time series aggregation two distance matrices are matrix multiplied with the self-association matrix of the cross-domain semantic fusion feature vector of the digital emergency plan to be applied - accident information to be applied to obtain the digital emergency plan to be applied - accident information embedded semantics - time series aggregation two distance association matrix, and the process can be expressed as: ;in, represents the distance matrix of the digital emergency plan to be applied, the accident information embedded semantics, and the time series aggregation, represents matrix multiplication, represents the cross-domain semantic fusion feature vector of the digital emergency plan to be applied-accident information, and is a row vector, express The transposed vector of It represents the distance association matrix of the digital emergency plan to be applied, the accident information embedded semantics, and the temporal aggregation.

[0050] After matrix multiplication of the distance query vector of the digital emergency plan-accident information embedded semantics-time series aggregation to be applied and the distance association matrix of the digital emergency plan-accident information embedded semantics-time series aggregation to be applied, it is further point multiplied with the digital emergency plan-accident information embedded semantics-time series aggregation joint distance eigenvector to be applied to obtain the optimized cross-domain semantic fusion feature vector of the digital emergency plan-accident information to be applied. This process can be expressed as: ;in, represents the digital emergency plan to be applied-accident information embedded semantics-time series aggregation-distance query vector, represents the distance association matrix of the digital emergency plan to be applied, the accident information embedded semantics, and the time series aggregation, It means point multiplication by position. represents matrix multiplication, represents the eigenvector of the joint distance between the digital emergency plan to be applied, the accident information embedded semantics and the time series aggregation, Represents the optimized cross-domain semantic fusion feature vector of the digital emergency plan to be applied-accident information.

[0051] Then, based on the optimized cross-domain semantic fusion feature vector of the digital emergency plan to be applied and the accident information Get the simulation analysis results.

[0052] That is, the first distance matrix and the second distance matrix of the cross-domain semantic fusion feature vector of the digital emergency plan-accident information to be applied are used as the fine-grained metric association cluster representation of the cross-domain semantic fusion feature vector of the digital emergency plan-accident information to be applied, and the self-association representation of the cross-domain semantic fusion feature vector of the digital emergency plan-accident information to be applied and the cross-domain semantic fusion feature vector of the digital emergency plan-accident information to be applied are respectively dynamically planned for the relationship between association clusters of different association clusters to simulate the sparse activation based on neuron clusters of the association system, and the fine-grained predictable sparsity of the cross-domain semantic fusion feature vector of the digital emergency plan-accident information to be applied is coordinated with the intrinsic representation of the metric association cluster of the first distance matrix and the second distance matrix, so as to avoid the lack of classification reasoning degree affected by insufficient association caused by sparsity, and improve the accuracy of the simulation analysis results obtained by the cross-domain semantic fusion feature vector of the digital emergency plan-accident information to be applied.

[0053] In summary, a dynamic digital emergency plan simulation and analysis system 100 for multiple scenarios based on an embodiment of the present application is explained, which uses machine learning-based data processing technology to perform semantic embedding encoding on the digital emergency plan to be applied and each accident information, thereby automatically generating a disaster level evolution trend label based on the prior modal cross-domain fusion representation between the semantic embedding features of the digital emergency plan to be applied and the semantic embedding features of each accident information. In this way, by combining digital emergency plan information and real-time accident information, it is possible to more accurately predict the development dynamics of the accident, which is conducive to flexibly adjusting the response strategy to adapt to the actual development of the accident, and is conducive to achieving more effective emergency management.

[0054] As described above, the dynamic digital emergency plan simulation and analysis system 100 for multiple scenarios according to the embodiment of the present application can be implemented in various terminal devices, such as a server for dynamic digital emergency plan simulation and analysis for multiple scenarios. In one example, the dynamic digital emergency plan simulation and analysis system 100 for multiple scenarios according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the dynamic digital emergency plan simulation and analysis system 100 for multiple scenarios can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the dynamic digital emergency plan simulation and analysis system 100 for multiple scenarios can also be one of the many hardware modules of the terminal device.

[0055] Alternatively, in another example, the dynamic digital emergency plan simulation and analysis system 100 for multiple scenarios and the terminal device may also be separate devices, and the dynamic digital emergency plan simulation and analysis system 100 for multiple scenarios may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

Claims

1. A dynamic digital emergency plan simulation and analysis system for multiple scenarios, characterized in that: include: An accident information collection module is used to collect accident information in real time to obtain the time series distribution of the accident information; A digital plan extraction module to be applied is used to extract the digital plan to be applied from the digital plan library; a digital plan semantic embedding module to be applied is used to perform semantic embedding coding on the digital plan to be applied to obtain semantic embedding coding features of the digital plan to be applied; An accident information semantic embedding module is used to perform semantic embedding coding on each accident information in the time series distribution of the accident information to obtain the time series distribution of the accident information semantic embedding coding features; a simulation analysis module is used to perform simulation analysis on the time series distribution of the semantic embedding coding features of the digital plan to be applied and the semantic embedding coding features of the accident information based on a large model to obtain a simulation analysis result for representing the disaster level evolution trend label, wherein the simulation analysis module includes: a digital plan to be applied-accident information semantic fusion unit, which is used to perform a priori modal-aware cross-domain embedding fusion on the time series distribution of the semantic embedding coding features of the digital plan to be applied and the semantic embedding coding features of the accident information to obtain the cross-domain semantic fusion features of the digital plan to be applied-accident information; a simulation analysis result generation unit is used to obtain the simulation analysis result based on the cross-domain semantic fusion features of the digital plan to be applied-accident information.

2. The dynamic digital emergency plan simulation and analysis system for multiple scenarios according to claim 1 is characterized in that: The accident information is simulation parameters input by the user, and the simulation parameters include geographical location, disaster level and current environmental conditions.

3. The dynamic digital emergency plan simulation and analysis system for multiple scenarios according to claim 2 is characterized in that: The semantic embedding module of the digital plan to be applied is used to: use a semantic embedding encoder based on a bidirectional recurrent neural network to perform semantic embedding encoding on the digital plan to be applied to obtain a semantic embedding encoding vector of the digital plan to be applied as the semantic embedding encoding feature of the digital plan to be applied.

4. The dynamic digital emergency plan simulation and analysis system for multiple scenarios according to claim 3 is characterized in that: The accident information semantic embedding module is used to: use the semantic embedding encoder based on the bidirectional recurrent neural network to perform semantic embedding encoding on each accident information in the time series distribution of the accident information to obtain the time series distribution of the accident information semantic embedding encoding vector as the time series distribution of the accident information semantic embedding encoding feature.

5. The dynamic digital emergency plan simulation and analysis system for multiple scenarios according to claim 4 is characterized in that: The digital emergency plan to be applied - accident information semantic fusion unit includes: an accident information kernel feature extraction subunit, which is used to perform modal kernel feature extraction on the time series distribution of the accident information semantic embedding coding vector to obtain the accident information semantic kernel feature coding vector; a priori clustering center calculation subunit, which is used to determine the digital emergency plan to be applied - accident information semantic priori clustering center coding vector based on the digital emergency plan semantic embedding coding vector to be applied and the accident information semantic kernel feature coding vector; a cross-domain fine-grained aggregation subunit, which is used to perform cross-domain fine-grained clustering aggregation on the time series distribution of the digital emergency plan to be applied - accident information semantic priori clustering center coding vector and the accident information semantic embedding coding vector to obtain the digital emergency plan to be applied - accident information cross-domain semantic fusion feature vector as the cross-domain semantic fusion feature of the digital emergency plan to be applied - accident information.

6. The dynamic digital emergency plan simulation and analysis system for multiple scenarios according to claim 5 is characterized in that: The prior clustering center calculation subunit is used to: cascade the semantic embedding coding vector of the digital plan to be applied and the semantic kernel feature coding vector of the accident information to obtain the semantic cascade feature vector of the digital plan to be applied-accident information; and perform linear transformation on the semantic cascade feature vector of the digital plan to be applied-accident information to obtain the semantic prior clustering center coding vector of the digital plan to be applied-accident information.

7. The dynamic digital emergency plan simulation and analysis system for multiple scenarios according to claim 6 is characterized in that: The cross-domain fine-grained aggregation subunit is used to: use a semantic distance measurement module based on the acrcosh function to perform distance measurement on the digital plan-accident information semantic prior clustering center coding vector to be applied and each accident information semantic embedding coding vector in the time series distribution of the accident information semantic embedding coding vector to obtain the time series distribution of the digital plan-accident information semantic measurement value to be applied; use a binary function to perform clustering judgment on the time series distribution of the digital plan-accident information semantic measurement value to be applied to obtain the time series distribution of the digital plan-accident information clustering coefficient to be applied; based on the time series distribution of the digital plan-accident information clustering coefficient to be applied, weight and fuse each accident information semantic embedding coding vector in the time series distribution of the digital plan-accident information semantic prior clustering center coding vector to be applied and the accident information semantic embedding coding vector to obtain the cross-domain semantic fusion feature vector of the digital plan-accident information to be applied.

8. The dynamic digital emergency plan simulation and analysis system for multiple scenarios according to claim 7 is characterized in that: The simulation analysis result generating unit is used to: input the cross-domain semantic fusion feature vector of the digital emergency plan-accident information to be applied into the simulation analyzer to obtain the simulation analysis result, and the simulation analysis result is a disaster level evolution trend label.

9. The dynamic digital emergency plan simulation and analysis system for multiple scenarios according to claim 8, characterized in that: The simulation analyzer is a classifier-based simulation analyzer.

Images