Emergency event identification method based on neural network model and computer equipment

Abstract

The invention provides an emergency event identification method based on a neural network model and computer equipment, and the method comprises the steps: obtaining a multi-source perception data set covering a target region, carrying out the feature coding processing of the multi-source perception data set, obtaining a multi-dimensional feature representation set of perception data units, and carrying out the recognition of an emergency event. Calling a pre-trained emergency event recognition neural network model to perform event recognition processing on the multi-dimensional feature representation set, generating an event recognition result of the sensing data unit, and determining the type of an emergency event existing in the target area and spatial-temporal feature information of the emergency event in the sensing data unit according to the event recognition result; and based on the emergency event type and the time-space feature information, generating an emergency early warning instruction containing an event positioning identifier, and sending the emergency early warning instruction to the target emergency response terminal to start a response operation. According to the invention, the comprehensiveness and accuracy of emergency event identification are improved, and the conversion efficiency of the identification result to the actual response is enhanced, so that the timely early warning of the emergency event is effectively supported.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to an emergency event recognition method and computer equipment based on a neural network model. Background Art

[0002] With the development of information technology and sensing devices, emergency event recognition technology has been widely used in fields such as public safety and disaster warning. This technology analyzes relevant data to identify potential or ongoing emergency events, supporting timely warning and response. Currently, this approach typically acquires a single type of monitoring data, extracts a single-dimensional feature from the data, and then uses traditional classification models to analyze this feature to generate event recognition results. However, this recognition approach, which relies solely on single-source data and single-dimensional features, fails to simultaneously capture both the physical state changes of the event and the human feedback intentions, resulting in an incomplete portrayal of the event's essence. Furthermore, the generated recognition results often only contain probabilistic judgments about the event type and lack specific spatiotemporal location information, making them difficult to directly translate into actionable warning instructions, impacting the timeliness and accuracy of emergency response. Summary of the Invention

[0003] The present invention provides an emergency event recognition method and computer equipment based on a neural network model.

[0004] In the first aspect, an embodiment of the present invention provides an emergency event identification method based on a neural network model, the method comprising: obtaining a multi-source perception data set covering a target area, the multi-source perception data set comprising multiple perception data units with timestamp marks, each perception data unit consisting of environmental sensing data and interactive feedback data; performing feature encoding processing on the multi-source perception data set to obtain a multi-dimensional feature representation set of the perception data unit, the multi-dimensional feature representation set comprising temporal correlation features and semantic correlation features; calling a pre-trained emergency event identification neural network model to perform event identification processing on the multi-dimensional feature representation set to generate an event identification result of the perception data unit, the event identification result comprising a predicted probability distribution of different emergency event categories; determining the type of emergency event existing in the target area and the spatiotemporal feature information of the emergency event in the perception data unit based on the event identification result; generating an emergency warning instruction comprising an event location identifier based on the emergency event type and the spatiotemporal feature information, and sending the emergency warning instruction to the target emergency response terminal to initiate a response operation.

[0005] In a second aspect, an embodiment of the present invention provides a computer device, comprising: a memory storing a computer program; and a processor for loading the computer program to implement the emergency event identification method based on the neural network model as described above.

[0006] The emergency event recognition method based on the neural network model provided by the present invention can simultaneously cover the physical state change information and human feedback intention information of the emergency event by acquiring a multi-source perception data set including environmental sensing data and interactive feedback data, thereby avoiding the problem of incomplete coverage of event information by single-source data; by performing feature coding processing on the multi-source perception data, extracting the temporal correlation features reflecting the evolution of the continuous time state and the semantic correlation features reflecting the contextual intention of the interactive content, forming a multi-perspective characterization of the physical development process of the event and the human intervention intention, so that the subsequent model reasoning can be based on two types of key attribute information at the same time; calling the pre-trained emergency event recognition neural network model to jointly process the two-dimensional features, The model's feature fusion capability can be used to achieve complementary reasoning of the two types of features, thereby enhancing the depth of understanding of the nature of the event; based on the recognition results, the emergency event type and spatiotemporal feature information can be determined, and the probability distribution of the model output can be converted into event type identification and spatiotemporal positioning description in the specific scenario to ensure the practicality of the recognition results; emergency warning instructions containing event positioning identification are generated and sent, and the recognition results are converted into actionable response information, so that the emergency response terminal can quickly initiate corresponding operations based on the event type, spatiotemporal characteristics and positioning identification, which not only improves the comprehensiveness and accuracy of emergency event identification, but also enhances the efficiency of converting recognition results into actual responses, thereby effectively supporting timely warning and accurate response to emergency events. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 This is a flowchart of an emergency event identification method based on a neural network model provided by an embodiment of the present invention.

[0008] Figure 2 The figure is a schematic diagram of the composition of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0009] See also Figure 1 , Figure 1 A flowchart of a method for identifying emergency events based on a neural network model is provided in an embodiment of the present invention. The method can be executed by a computer device and includes the following steps:

[0010] Step S100: Acquire a multi-source perception data set covering a target area. The multi-source perception data set includes a plurality of perception data units with timestamps. Each perception data unit consists of environmental sensing data and interactive feedback data.

[0011] The target coverage area refers to the geographic area where emergency event monitoring and identification are required, such as a city block, an industrial park, or a nature reserve. A multi-source perception data set is a collection of data collected from a variety of sources, including but not limited to sensor networks, surveillance cameras, social media, and emergency alarm systems. A timestamp is added to each perception data unit to record the specific moment of data collection for subsequent time series analysis. A perception data unit is the basic data unit in a multi-source perception data set. Each unit contains two components: environmental sensing data and interactive feedback data. Environmental sensing data is collected by various environmental sensors and describes the environmental conditions of the target area, such as temperature, humidity, air pressure, air quality, and noise level. Interactive feedback data is data obtained through user interaction, communication devices, or other feedback channels, such as user comments on social media, alarm messages from emergency alarm systems, and human behavior information from surveillance cameras.

[0012] In practical applications, various types of sensor networks can be deployed to collect environmental sensing data. For example, within a city block, temperature sensors can be installed to monitor ambient temperature in real time, and air quality sensors can be installed to detect pollutant concentrations in the air. These sensors collect data at regular intervals and timestamp each data point. Interactive feedback data can be obtained from data sources such as social media platforms and emergency alarm systems through data interfaces. For example, by collaborating with social media platforms, user posts and comments related to emergency events can be retrieved through APIs, while alarm information can be obtained from emergency alarm systems, also timestamped. These sensing data units from different data sources are integrated to form a multi-source sensing data set covering the target area.

[0013] Step S200: performing feature coding processing on the multi-source perception data set to obtain a multi-dimensional feature representation set of the perception data unit, wherein the multi-dimensional feature representation set includes temporal correlation features and semantic correlation features.

[0014] Feature encoding is the process of converting raw multi-source perceptual data sets into more representative and analyzable feature representations. A multidimensional feature representation is a set of features encompassing multiple dimensions that more comprehensively describe the characteristics of perceptual data units. Temporal association features reflect the associations and changes in perceptual data units over time, such as trends in data changes between adjacent timestamps or cyclical patterns. Semantic association features reflect the semantic information and contextual associations of the interactive content within a perceptual data unit, such as the topics, intent, and sentiment in user reviews.

[0015] As an implementation manner, step S200 may specifically include the following steps S210 to S250:

[0016] Step S210: performing data cleaning processing on the multi-source perception data set to eliminate noise interference information in the perception data units to obtain a denoised perception data set.

[0017] Data cleaning is a crucial step in preprocessing raw data. It aims to remove noise, errors, missing values, and other interfering information to improve data quality and usability. This interference can arise from sensor measurement errors, interference during data transmission, and human input errors. Denoised perception data sets are clean, reliable datasets obtained through data cleaning.

[0018] In practice, missing values ​​can be filled using interpolation. For example, if the temperature data for a certain timestamp in environmental sensing data is missing, linear interpolation can be used to estimate the missing value based on the temperature values ​​of the preceding and following timestamps. Noise can be removed using filtering algorithms. For example, salt and pepper noise in image data collected by surveillance cameras can be processed using a median filter. The specific implementation of the median filter algorithm involves taking the median of the pixel values ​​in its neighborhood for each pixel in the image as the new value for that pixel, effectively removing noise points. This data cleaning method can eliminate noise interference in sensory data units, resulting in a denoised sensory data set.

[0019] Step S220: performing time series correlation extraction processing on the environmental sensing data in the denoised perception data set to generate state evolution features of the perception data units at continuous timestamps as time series correlation features, where the state evolution features include a description of the change trend between adjacent timestamps.

[0020] Temporal correlation extraction is the process of extracting temporal associations and changes from environmental sensing data within a denoised sensory data set. State evolution features describe the state changes of sensory data units at consecutive timestamps, reflecting the dynamic evolution of environmental conditions. The description of change trends between adjacent timestamps specifically describes the direction and extent of change in environmental sensing data between two adjacent timestamps, such as whether the temperature increased or decreased, and by how much.

[0021] As an implementation manner, step S220 may specifically include the following steps S221 to S225:

[0022] Step S221: extracting the environmental sensing data change area between adjacent perception data units, and analyzing the attribute parameter difference characteristics of the change area.

[0023] Adjacent sensory data units refer to two adjacent sensory data units in a temporal sequence. An environmental sensing data change region refers to the region between adjacent sensory data units where the environmental sensing data changes. Attribute parameter difference features refer to the differences between different attribute parameters within a change region, such as the temperature difference within a temperature change region or the humidity difference within a humidity change region.

[0024] In practice, extracting areas of environmental sensing data change between adjacent sensory data units can be achieved by comparing the environmental sensing data at adjacent timestamps. For example, for temperature data, the temperature difference between adjacent timestamps can be calculated, and areas with non-zero differences can be marked as temperature change areas. For humidity data, the humidity difference between adjacent timestamps can be similarly calculated to mark humidity change areas. The attribute parameter differences in these change areas are then analyzed. For example, for a temperature change area, the average temperature difference, maximum temperature difference, and minimum temperature difference within the area can be calculated to more comprehensively describe the extent and characteristics of the temperature change.

[0025] Step S222: Determine the state change direction and degree description of the perception data unit between adjacent timestamps based on the attribute parameter difference characteristics.

[0026] The direction of state change refers to the direction in which the state of the sensor data unit develops between adjacent timestamps, such as whether the temperature increases or decreases, or whether the air pressure increases or decreases. The degree of change describes the magnitude of the state change, such as how many degrees the temperature increases or how many Pascals the air pressure increases.

[0027] Determining the direction and extent of state changes based on the differences in attribute parameters can be achieved through simple comparisons and calculations. For example, for temperature data, if the temperature difference between adjacent timestamps is positive, it indicates a temperature increase, and the absolute value of the difference is the magnitude of the temperature increase; if the difference is negative, it indicates a temperature decrease, and the absolute value of the difference is the magnitude of the temperature decrease. For air pressure data, the direction of pressure change is similarly determined by the sign of the pressure difference, and the degree of pressure change is determined by the absolute value of the difference. In this way, the state changes of sensory data units between adjacent timestamps can be accurately described.

[0028] Step S223: constructing a state evolution analysis model of the perception data unit based on the description of the state change direction and change degree.

[0029] The state evolution analysis model is a mathematical model used to describe the state evolution of sensory data units at consecutive timestamps. Based on the description of the direction and degree of state change, it can predict the state changes of sensory data units at future timestamps.

[0030] When constructing a state evolution analysis model, various methods can be used. For example, the autoregressive integrated moving average (ARIMA) model from time series analysis can be used. For example, data describing the direction and degree of state change can be tested for stationarity. If the data is nonstationary, differencing is performed to transform it into a stationary series. Then, based on the graphical characteristics of the autocorrelation function (ACF) and partial autocorrelation function (PACF), the order (p, d, q) of the ARIMA model is determined. Here, p represents the order of the autoregressive term, d represents the order of differencing, and q represents the order of the moving average term. Finally, the model parameters are estimated using methods such as least squares to obtain the final state evolution analysis model.

[0031] Step S224: Calculate the state change trend consistency description of the perception data unit at continuous timestamps through the state evolution analysis model.

[0032] The state change trend consistency description refers to whether the state change trend of the perception data unit remains consistent under consecutive timestamps. For example, if the temperature shows a continuous upward trend within multiple adjacent timestamps, the state change trend of the temperature is consistent; if the temperature sometimes increases and sometimes decreases, the state change trend is inconsistent.

[0033] Specifically, a state evolution analysis model is used to predict state changes for consecutive timestamps, obtaining the predicted state change direction and degree for each timestamp. The predicted state change directions for adjacent timestamps are then compared. If the predicted state change directions for adjacent timestamps are the same, the state change trends between the two timestamps are considered consistent; if they are different, the state change trends are considered inconsistent. Finally, the number of time intervals with consistent state change trends in consecutive timestamps is counted, and their proportion to the total number of time intervals is calculated to describe the consistency of the state change trends.

[0034] Step S225: performing periodic pattern recognition processing on the state evolution process of the perception data unit, and extracting the state repetition feature of the perception data unit within a preset time window as the state evolution feature.

[0035] Periodic pattern recognition is the process of identifying periodic patterns from the state evolution of sensory data units. A preset time window refers to a pre-set time range, such as a day, a week, or a month. State repetition features refer to the repetition of states within a sensory data unit within the preset time window. For example, temperature values ​​at the same time each day are roughly the same, or air pressure change patterns are similar during the same time period each week. Specifically, the Fourier transform method can be used. The basic principle of the Fourier transform is to convert time-domain signals into frequency-domain signals and identify periodic patterns by analyzing the signal characteristics in the frequency domain. Specifically, the state evolution data of the sensory data unit is first divided into preset time windows. Then, a Fourier transform is performed on the data within each time window to obtain a spectrogram. Peaks in the spectrogram represent the primary frequency components of the signal. By analyzing the location and amplitude of these peaks, periodic patterns in the data can be identified. For example, if the spectrogram contains a distinct peak that occurs once a day, it indicates that the data has a daily periodic pattern. Finally, the state repetition features corresponding to these periodic patterns are extracted as state evolution features.

[0036] Step S230: performing semantic association extraction processing on the interaction feedback data in the denoised perception data set to generate semantic association features of the interaction content in the perception data unit, where the semantic association features include a contextual association description of the interaction intention.

[0037] Semantic association extraction is the process of extracting semantic information and contextual associations from interaction feedback data within a denoised perception data set. Interaction content refers to the specific information contained in interaction feedback data, such as user comments and alarm information. Semantic association features are features that reflect the semantic information and contextual associations of interaction content, such as the subject, intent, and sentiment of the interaction content. The contextual association description of interaction intent further illustrates the association and meaning of interaction intent within the context, such as the association between a topic in a user comment and other topics, or the association between an event in an alarm message and related background information.

[0038] As an implementation manner, step S230 may specifically include the following steps S231 to S235:

[0039] Step S231: performing word segmentation processing on the text content in the interactive feedback data to obtain a plurality of interactive word units.

[0040] Word segmentation is the process of segmenting the text content in interactive feedback data into independent word units according to specific rules. An interactive word unit is a single word or phrase obtained after word segmentation. A variety of word segmentation algorithms can be used to segment text content in interactive feedback data. For example, the Jieba word segmentation algorithm can be used for Chinese text. The Jieba word segmentation algorithm is a word segmentation algorithm based on statistical models and rules. It combines dictionary matching and machine learning methods, making it highly effective for word segmentation in Chinese text. The specific implementation process is as follows: First, the Jieba word segmentation algorithm loads a pre-built dictionary containing a large number of Chinese words. Then, for the input text content, the algorithm matches the longest word in the dictionary. If a match is successful, the word is output as an interactive word unit. If a match fails, the text is segmented and further matching attempts are made. In this way, the text content is segmented into multiple interactive word units. For English text, the word segmentation tool in the NLTK library can be used, which provides simple and effective English word segmentation.

[0041] Step S232: Calling a pre-trained semantic encoding model to perform context encoding processing on multiple interaction word units to generate a semantic vector representation of the interaction feedback data.

[0042] Pretrained semantic encoding models are pre-trained on a large-scale corpus and can learn the semantic information and contextual associations of words and sentences. Contextual encoding is the process of encoding multiple interactive word units with their contextual information to generate vector representations that better reflect their semantic meaning. Semantic vector representation converts interactive feedback data into a representation in vector space. Vector operations and comparisons facilitate semantic analysis and processing.

[0043] The BERT model can be used to call a pre-trained semantic encoding model to perform context encoding processing on multiple interaction word units. The BERT model is a pre-trained language model based on the Transformer architecture. It can learn the context information of words and sentences through the method of bidirectional encoder representations from transformers. Specifically, multiple interaction word units are input into the BERT model. The BERT model encodes the input words and generates a context embedding vector for each word. These context embedding vectors are then aggregated, for example, through average pooling or maximum pooling methods, to obtain a semantic vector representation of the entire interaction feedback data. In this way, the interaction feedback data is converted into a vector representation with more semantic information.

[0044] Step S233: Analyze the semantic association strength between different interactive word units in the semantic vector representation, and generate an association relationship graph including semantic association weights.

[0045] Semantic association strength refers to the closeness of the semantic connection between different interaction word units. For example, the semantic association strength between "fire" and "smoke" is strong, while the semantic association strength between "fire" and "apple" is weak. An association graph is a graphical structure used to represent the semantic association relationships between different interaction word units. Nodes represent interaction word units, edges represent semantic association relationships, and edge weights represent the semantic association strength.

[0046] Cosine similarity can be used to analyze the strength of semantic associations between different interaction word units in the semantic vector representation. Cosine similarity measures similarity by calculating the cosine of the angle between two vectors. A cosine value closer to 1 indicates greater similarity and stronger semantic association between the two vectors; a cosine value closer to 0 indicates less similarity and weaker semantic association between the two vectors. Specifically, the vector of each interaction word unit is first extracted from the semantic vector representation. Then, the cosine similarity between any two interaction word unit vectors is calculated. The cosine similarity value is used as a measure of semantic association strength. Finally, an association graph is generated based on these semantic association strength values. For example, a graph database (such as Neo4j) can be used to store and manage the association graph, with interaction word units as nodes and semantic association strengths as edge weights. This graph, which includes semantic association weights, is constructed.

[0047] Step S234: extracting semantic features of the core interaction intention in the interaction feedback data as semantic association features according to the association relationship graph.

[0048] Core interaction intent refers to the most important intent and information in interaction feedback data, such as the main topic in user comments and the core events in alarm information. Semantic features are features that can reflect the semantic information of core interaction intent.

[0049] Extracting the semantic features of the core interaction intent based on the association relationship graph can be implemented, for example, as follows: analyzing the association relationship graph to find the nodes with higher connectivity, which usually represent the related words of the core interaction intent. For example, in an association relationship graph of interactive feedback data about fire alarms, the connectivity of the "fire" node may be higher because it has a strong semantic association with nodes such as "smoke", "flame", and "alarm". Then, the semantic vector representation of these core nodes is extracted and used as the semantic features of the core interaction intent. Finally, these semantic features are used as semantic association features. In this way, the semantic information of the core interaction intent in the interactive feedback data can be extracted more accurately.

[0050] Step S235: Perform multiple rounds of context enhancement processing on the semantic association features to strengthen the contextual coherence description of the interaction intention.

[0051] Multi-round context enhancement involves multiple processing of semantically relevant features to further enhance their contextual coherence and semantic information. A contextually coherent description of interaction intent refers to a coherent and consistent description of the interaction intent within the context. For example, this includes whether topic transitions in user comments are natural or whether event descriptions in alarm messages are complete.

[0052] Multiple rounds of contextual enhancement of semantic association features can be performed using methods such as recurrent neural networks (RNNs) or long short-term memory networks (LSTMs). Taking LSTM as an example, semantic association features are first input into the LSTM network. The LSTM network updates its current hidden state based on the input semantic association features and the previous hidden state. The updated hidden state is then used as the new semantic association feature for the next round of processing. After multiple rounds of processing, the contextual coherence of the semantic association feature is enhanced. Finally, the semantic association feature, which has undergone multiple rounds of contextual enhancement, is output, strengthening the contextual coherence of the interaction intent.

[0053] Step S240: input the temporal correlation features and the semantic correlation features into the feature integration module for correlation analysis and processing, to obtain an integrated feature set with a unified representation form.

[0054] The feature integration module integrates and analyzes temporal and semantic correlation features. Correlation analysis is the process of analyzing the relationships between temporal and semantic correlation features. Unified representation converts different types of features into a unified representation that facilitates subsequent processing and analysis. The integrated feature set is the set of features containing the unified representation obtained after processing by the feature integration module.

[0055] As an implementation manner, step S240 may specifically include the following steps S241 to S245:

[0056] Step S241: performing feature mutual information calculation on the temporal correlation features and the semantic correlation features to generate a mutual information matrix reflecting the degree of information sharing between the two types of features.

[0057] The feature mutual information calculation process is the process of calculating the mutual information between temporal correlation features and semantic correlation features. Mutual information is a metric that measures the degree of information sharing between two random variables. It indicates the extent to which information about one variable can be predicted by the other. The mutual information matrix is ​​a matrix whose elements represent the mutual information between temporal correlation features and semantic correlation features.

[0058] The following general formula can be used to calculate the mutual information of temporal correlation features and semantic correlation features: ; Where X represents the temporal correlation feature, x is its specific value, Y represents the semantic correlation feature, y is its specific value, p(x,y) is the joint probability distribution of X and Y, and p(x) and p(y) are the marginal probability distributions of X and Y, respectively. Specifically, the joint probability distribution and marginal probability distribution of the temporal correlation feature and the semantic correlation feature are calculated. These probability distributions are estimated by the frequency of occurrence of different feature values ​​in the statistical data. Then, the mutual information value between any two features is calculated according to the above formula. Finally, these mutual information values ​​are combined into a mutual information matrix. For example, assuming there are n temporal correlation features and m semantic correlation features, the mutual information matrix is ​​an n×m matrix, and each element in the matrix represents the mutual information value between a temporal correlation feature and a semantic correlation feature.

[0059] Step S242: According to the mutual information values ​​of each feature pair in the mutual information matrix, key feature pairs whose mutual information values ​​exceed a preset threshold are screened.

[0060] Key feature pairs refer to feature pairs whose mutual information values ​​between temporal correlation features and semantic correlation features exceed a preset threshold. These feature pairs have a strong degree of information sharing. The preset threshold is a pre-set mutual information value standard used to screen key feature pairs.

[0061] Screening key feature pairs based on the mutual information values ​​of each feature pair in the mutual information matrix can be implemented, for example, by determining a preset threshold. The selection of the preset threshold can be adjusted according to the specific application scenario and data characteristics. For example, in some cases, the preset threshold can be set to the average value of all elements in the mutual information matrix, or a suitable value can be determined based on experience and experimental results. Then, each element in the mutual information matrix is ​​traversed, and the feature pairs whose mutual information values ​​exceed the preset threshold are marked as key feature pairs. Finally, these key feature pairs are extracted for subsequent processing. By screening the key feature pairs, the number of features can be reduced and the efficiency and accuracy of subsequent processing can be improved.

[0062] Step S243: performing feature space alignment processing on the key feature pairs, and mapping the temporal correlation features and the semantic correlation features to the same feature space dimension through linear transformation.

[0063] Feature space alignment is the process of mapping features from different feature spaces to the same feature space dimension. Linear transformation converts vectors from one vector space to another through linear combination. Identifying feature space dimensions means converting temporal and semantic features to the same dimensional space for subsequent fusion and analysis.

[0064] Principal component analysis (PCA) can be used to align key feature pairs in the feature space. The basic principle of PCA is to find the principal components of the data and project it into a low-dimensional feature space while preserving the data's key information. Specifically, the temporal and semantic correlation features of the key feature pairs are first combined into a matrix. This matrix is ​​then normalized to have a mean of 0 and a variance of 1. Next, the covariance matrix of the matrix is ​​calculated. The principal components of the data are obtained by calculating the eigenvalues ​​and eigenvectors of the covariance matrix. The first k principal components with the largest eigenvalues ​​are selected as the primary features (e.g., the first k principal components whose cumulative variance contribution exceeds a preset threshold (e.g., 95%)). The original data is projected onto these principal components to obtain the reduced-dimensional feature vectors. Finally, both the temporal and semantic correlation features are projected into this low-dimensional feature space to achieve feature space alignment. It is understood that the temporal and semantic features are each normalized, such as using the Z-score, before PCA.

[0065] Step S244: Based on the spatial alignment results of the key feature pairs, a feature splicing operation is used to perform dimension fusion processing on the temporal correlation features and the semantic correlation features to generate an initial integrated feature set with a unified representation form.

[0066] The feature concatenation operation concatenates temporal and semantic features, which have undergone spatial feature alignment. Dimensional fusion combines features of different dimensions into a higher-dimensional feature representation. The initial integrated feature set is the set of features with a unified representation obtained after the feature concatenation operation. Based on the spatial alignment results of key feature pairs, the feature concatenation operation can be used to perform dimensional fusion of temporal and semantic features using simple vector concatenation.

[0067] Step S245: performing feature redundancy detection on the initial integrated feature set, removing redundant feature dimensions that repeatedly express the same information, and generating an integrated feature set.

[0068] Feature redundancy detection is the process of examining the initial integrated feature set for feature dimensions that repeatedly express the same information. Redundant feature dimensions are those that do not contribute significantly to subsequent analysis and processing. These dimensions may be caused by duplication during data collection or redundancy during feature extraction. The integrated feature set is the one that has been removed from the redundant feature dimensions after feature redundancy detection.

[0069] Correlation analysis can be used to detect feature redundancy in the initial integrated feature set. Correlation analysis measures the degree of similarity between features by calculating the correlation coefficient between them. Specifically, the correlation coefficient between any two feature dimensions in the initial integrated feature set is first calculated. The Pearson correlation coefficient can be used to measure linear correlation. Then, a correlation threshold is set, and feature dimensions with correlation coefficients exceeding the threshold are marked as redundant feature dimensions. Finally, these redundant feature dimensions are removed to generate an integrated feature set. By removing redundant feature dimensions, the number of features can be reduced, computational complexity can be lowered, and the generalization ability of the model can be improved.

[0070] Step S250: Based on the association strength evaluation results of the features of different dimensions in the integrated feature set, the fusion emphasis of the temporal association features and the semantic association features in the feature encoding process is dynamically adjusted to generate a multi-dimensional feature representation set.

[0071] The association strength assessment results evaluate and analyze the strength of associations between features of different dimensions within the integrated feature set, such as the degree of association between features obtained through methods such as mutual information calculation and correlation analysis. Dynamic adjustment refers to the real-time alteration of the ratio and method of fusion of temporal and semantic features in the feature encoding process based on the association strength assessment results. Fusion emphasis refers to the preference for fusing temporal or semantic features, or a balanced combination of both, in the feature encoding process. A multidimensional feature representation set is a set of features containing multiple dimensions, resulting from dynamic adjustment of the fusion emphasis.

[0072] Based on the association strength evaluation results of the different dimensional features in the integrated feature set, the dynamic adjustment of the fusion emphasis of the temporal association features and the semantic association features in the feature encoding process can be implemented, for example, as follows: based on the association strength evaluation results, the association relationship between the different dimensional features is determined. For example, if the association strength of a temporal association feature is high with multiple semantic association features, it means that the temporal association feature plays an important role in the fusion. Then, based on these association relationships, the weights of the temporal association features and the semantic association features in the feature encoding process are adjusted. For example, if it is found that the semantic association feature has a greater impact on the final result in some cases, the weight of the semantic association feature is appropriately increased and the weight of the temporal association feature is reduced. Finally, based on the adjusted weights, the temporal association features and the semantic association features are fused to generate a multidimensional feature representation set. By dynamically adjusting the fusion emphasis, the multidimensional feature representation set can more accurately reflect the characteristics of the perceptual data unit.

[0073] Step S300: Calling a pre-trained emergency event recognition neural network model to perform event recognition processing on a multi-dimensional feature representation set, generating an event recognition result of a perception data unit, wherein the event recognition result includes a predicted probability distribution of different emergency event categories.

[0074] A pretrained emergency event recognition neural network model is a neural network model pre-trained on large-scale emergency event data. It can learn the characteristics and patterns of emergency events and is used to identify events in new sensory data. Event recognition processing is the process of inputting a set of multidimensional feature representations into the emergency event recognition neural network model and, through the model's calculations and reasoning, determining the type of emergency event that may exist in the sensory data unit. The event recognition result is the model's output of the identification of the emergency event type in the sensory data unit, including the predicted probability distribution of different emergency event categories. The predicted probability distribution refers to the probability that each emergency event category is predicted to occur by the model.

[0075] As an implementation manner, step S300 may specifically include the following steps S310 to S350:

[0076] Step S310: Input the multi-dimensional feature representation set into the feature processing layer of the emergency event recognition neural network model, perform hierarchical feature fusion processing, and generate a fused feature vector with multi-granularity representation.

[0077] The feature processing layer processes and fuses the input multidimensional feature representation set. Hierarchical feature fusion involves fusing features from different dimensions within the multidimensional feature representation set at multiple levels to extract richer feature information. Multi-granularity representation means that the fused feature vector can represent feature information at different granularity levels, such as from local details to global features. The fused feature vector is a comprehensive feature vector obtained through hierarchical feature fusion, containing the key information from the multidimensional feature representation set.

[0078] A convolutional neural network (CNN) architecture can be used to input a multidimensional feature representation set into the feature processing layer of an emergency event recognition neural network model for hierarchical feature fusion. The basic principle of a CNN is to extract and fuse features from input data through a combination of convolutional, pooling, and fully connected layers. Specifically, the multidimensional feature representation set is first input into the convolutional layer of the CNN. The convolutional layer slides the convolution kernel over the input data to extract local features. Different kernel sizes can extract feature information of different granularities; for example, small kernels can extract local details, while large kernels can extract global features. The output of the convolutional layer is then downsampled by the pooling layer to reduce the feature dimensionality while retaining the key feature information. Methods such as max pooling and average pooling can be used. Finally, the output of the pooling layer is combined and fused through a fully connected layer to generate a fused feature vector with a multi-granular representation. For example, for a multidimensional feature representation set containing multiple temporal and semantic correlation features, the hierarchical feature fusion of a CNN can produce a fused feature vector that comprehensively reflects these features.

[0079] Step S320: Perform context dependency modeling on the fused feature vector through the context modeling layer of the emergency event recognition neural network model to generate context enhanced features in the time dimension and semantic dimension.

[0080] The context modeling layer is used to model and analyze the contextual dependencies of the fused feature vector. Contextual dependencies refer to the temporal and semantic associations and dependencies between different elements in the fused feature vector. For example, they include the associations between the states of an emergency event at different points in time, as well as the associations between different semantic information. Context-enhanced features in the temporal and semantic dimensions are derived through contextual dependency modeling. They more comprehensively reflect the temporal and semantic context of the emergency event.

[0081] As an implementation manner, step S320 may specifically include the following steps S321 to S325:

[0082] Step S321: input the fused feature vector into the time dependency sublayer of the context modeling layer, perform front-to-back dependency analysis on the time series, and generate context dependency features in the time dimension.

[0083] The temporal dependency sublayer, a sublayer within the context modeling layer, is specifically designed to process the contextual dependencies of time series data. This process analyzes the time-related information in the fused feature vectors to identify associations and dependencies between different time points. The temporal contextual dependency features, derived from this process, can reflect the temporal development and changes of emergency events.

[0084] The fused feature vector is input into the time dependency sublayer of the context modeling layer to analyze the forward and backward dependencies of the time series. Long-short-term memory (LSTM) networks can be used. LSTMs effectively handle long-term dependencies in time series data, controlling the flow and memory of information through a gating mechanism. Specifically, the fused feature vector is first input into the LSTM network in chronological order. The LSTM network updates the current hidden state based on the current input and the previous hidden state. The hidden state contains both historical and contextual information about the time series. The hidden state output by the LSTM network is then used as the contextual dependency feature in the time dimension. For example, for emergency event data containing multiple timestamps, the LSTM network can process the contextual dependency features corresponding to each timestamp in the time dimension. These features can reflect the temporal trends and patterns of change of the emergency event.

[0085] Step S322: Input the fused feature vector into the semantic dependency sublayer of the context modeling layer, perform context-dependent relationship analysis on the semantic content, and generate context-dependent features of the semantic dimension.

[0086] The semantic dependency sublayer, another sublayer within the context modeling layer, processes the contextual relationships of the semantic content within the fused feature vector. This contextual relationship analysis analyzes the semantically relevant information within the fused feature vector to identify the relationships and dependencies between different semantic elements. The contextual dependency feature of the semantic dimension, derived from this contextual relationship analysis, reflects the semantic coherence and relevance of emergency events.

[0087] The fused feature vector is input into the semantic dependency sublayer of the context modeling layer for contextual analysis of semantic content. The multi-head attention mechanism in the Transformer architecture can be used to analyze the contextual relationships of semantic content. The basic principle of the multi-head attention mechanism is to use multiple attention heads to simultaneously focus on different parts of the input data and calculate the importance weight of each part, thereby modeling the contextual relationships of semantic content. Specifically, the fused feature vector is first input into the multi-head attention mechanism. The multi-head attention mechanism calculates the query vector, key vector, and value vector for each position. It then calculates the attention score by taking the dot product of the query vector and key vector, and performs a weighted sum on the value vector to produce the output for each position. The outputs of multiple attention heads are concatenated and linearly transformed to obtain the final contextual dependency features of the semantic dimension. For example, for a fused feature vector containing multiple semantic dependency features, the multi-head attention mechanism can produce contextual dependency features of the semantic dimension that reflect the contextual relationships of semantic content.

[0088] Step S323: performing cross-dimensional correlation analysis on the context-dependent features of the time dimension and the context-dependent features of the semantic dimension to generate cross-dimensional context-dependent features.

[0089] Cross-dimensional correlation analysis analyzes context-dependent features in the temporal dimension and the semantic dimension, identifying the associations and interactions between them. Cross-dimensional context-dependent features are derived from this process and comprehensively reflect the contextual associations between the temporal and semantic dimensions.

[0090] A combination of a fully connected layer and an attention mechanism can be used to perform cross-dimensional correlation analysis on the context-dependent features of the time dimension and the context-dependent features of the semantic dimension. Specifically, the context-dependent features of the time dimension and the context-dependent features of the semantic dimension are first concatenated together. Then, the concatenated vector is input into the fully connected layer and linearly transformed to obtain a new feature vector. Next, the attention mechanism is used to process this new feature vector and calculate the importance weight of each dimension. Through the weighted summation of the attention mechanism, cross-dimensional context-dependent features are obtained. For example, for a vector containing context-dependent features of the time dimension and the semantic dimension, through the processing of the fully connected layer and the attention mechanism, a cross-dimensional context-dependent feature that can reflect the correlation between the two dimensions can be obtained.

[0091] Step S324: performing weighted fusion processing on the context-dependent features of the time dimension and the context-dependent features of the semantic dimension based on the cross-dimensional context association features to generate fused context-dependent features.

[0092] Weighted fusion is the process of weightedly combining temporal and semantic context-dependent features based on cross-dimensional contextual association features. The fused context-dependent feature is a comprehensive feature obtained through weighted fusion, encompassing contextual information from both temporal and semantic dimensions, as well as the association between them.

[0093] The weighted fusion processing of the context-dependent features of the time dimension and the context-dependent features of the semantic dimension based on the cross-dimensional context-related features can be implemented as follows: the weights of the context-dependent features of the time dimension and the context-dependent features of the semantic dimension are calculated according to the cross-dimensional context-related features. The cross-dimensional context-related features can be converted into weight values ​​using a softmax function to ensure that the sum of the weight values ​​is 1. Then, the context-dependent features of the time dimension and the context-dependent features of the semantic dimension are multiplied by the corresponding weights respectively. Finally, the weighted features are added together to generate a fused context-dependent feature. For example, if the cross-dimensional context-related features indicate that the context-dependent features of the time dimension are more important in the current situation, the context-dependent features of the time dimension are given a higher weight, thereby highlighting the information of the time dimension in the fusion.

[0094] Step S325: performing dimension alignment processing on the fused context-dependent features to generate context-enhanced features in the time dimension and the semantic dimension.

[0095] Dimension alignment adjusts the dimensions of the fused context-dependent features to ensure consistent representation and dimensionality across both the temporal and semantic dimensions. Context-enhanced features in the temporal and semantic dimensions are derived from this dimension alignment. They more accurately reflect the temporal and semantic context of emergency events and maintain dimensional consistency, facilitating subsequent processing and analysis.

[0096] A linear transformation method can be used to align the dimensions of the fused context-dependent features. Specifically, the target dimension sizes of the temporal dimension and the semantic dimension are first determined. The target dimension can be set according to the design of the model and the needs of subsequent processing. Then, a linear transformation matrix is ​​used to project the fused context-dependent features onto the target dimension. The linear transformation matrix can be obtained through training so that the projected features can retain the information of the original features as much as possible. Finally, context-enhanced features with consistent dimensions in the temporal dimension and the semantic dimension are obtained.

[0097] Step S330: performing event probability prediction processing on the context enhancement features based on the probability prediction layer of the emergency event recognition neural network model to generate a probability distribution set containing the predicted probability values ​​corresponding to each emergency event category.

[0098] The probability prediction layer is a layer within the emergency event recognition neural network model that predicts event probabilities based on context-enhanced features. This process calculates the probability of each emergency event category based on the context-enhanced features. The probability distribution set is the model's output, containing the predicted probability values ​​for each emergency event category.

[0099] The probability prediction layer of the emergency event recognition neural network model uses the softmax function to predict event probabilities based on context-enhanced features. The basic principle of the softmax function is to convert the input vector into a probability distribution so that the sum of the probabilities of all elements is 1. Specifically, the context-enhanced features are first input into the fully connected layer of the probability prediction layer, where a linear transformation is performed to produce an output vector. This output vector is then input into the softmax function to calculate the probability value corresponding to each emergency event category.

[0100] Step S340: performing category confidence calibration processing on the probability distribution set, adjusting the confidence weights of the predicted probability values ​​of different emergency event categories, and generating a calibrated probability distribution set.

[0101] Category confidence calibration is the process of adjusting and calibrating the confidence levels of the predicted probabilities for different emergency event categories within a probability distribution set. Confidence weights represent the degree of confidence in the predicted probability values ​​for each emergency event category. By adjusting these confidence weights, the predicted probabilities can more accurately reflect the actual situation. The calibrated probability distribution set, obtained after category confidence calibration, results in more reliable and trustworthy predicted probabilities.

[0102] The Platt scaling method can be used to calibrate the probability distribution set through a logistic regression model to calibrate the probability distribution set. Specifically, a calibration data set is first prepared, which contains samples of known emergency event categories and their corresponding predicted probability values. Then, the calibration data set is trained using a logistic regression model to learn the relationship between the predicted probability value and the true category. The output of the logistic regression model is a calibrated probability value. Finally, each predicted probability value in the probability distribution set is input into the trained logistic regression model to obtain the calibrated probability value and generate a calibrated probability distribution set.

[0103] Step S350: Generate an event recognition result of the perception data unit based on the calibrated probability distribution set, where the event recognition result includes a predicted probability distribution of each emergency event category.

[0104] Generating event recognition results for a perception data unit based on a calibrated probability distribution set allows the calibrated probability distribution set to be directly used as the event recognition result. This is because the calibrated probability distribution set has been confidence-calibrated and more accurately reflects the probability of each emergency event category. For example, if the predicted probability of a fire event in the calibrated probability distribution set is 0.7, the predicted probability of an earthquake event is 0.2, and the predicted probability of other events is 0.1, then the event recognition result is the predicted probability distribution containing these probability values, indicating that in the current perception data unit, the fire event is most likely to occur, followed by the earthquake event, and the other events are less likely to occur.

[0105] Step S400: Determine the type of emergency event existing in the target area and the spatiotemporal feature information of the emergency event in the perception data unit according to the event recognition result.

[0106] Event recognition results are generated using the emergency event recognition neural network model, including the predicted probability distribution of each emergency event category. Emergency event types refer to the different types of emergencies that may occur in the target area, such as fire, earthquake, and flood. Spatiotemporal feature information refers to the temporal and spatial characteristics of emergency events within the perception data unit, such as the time of occurrence, duration, location, and direction of spread of the emergency event.

[0107] As an implementation manner, step S400 may specifically include the following steps S410 to S450:

[0108] Step S410: Analyze the predicted probability distribution in the event recognition result, detect the probability peak point exceeding the preset confidence threshold and its corresponding perception data unit timestamp information.

[0109] Parsing the predicted probability distribution involves analyzing and interpreting the probability distribution set within the event recognition results to identify key information. The preset confidence threshold is a pre-set probability value standard used to select emergency event prediction results with high credibility. The probability peak point refers to the point where the probability value in the predicted probability distribution reaches its peak; these points generally indicate a high probability of an emergency event. The sensory data unit timestamp information refers to the time stamp of the sensory data unit corresponding to the probability peak point, which is used to determine the time when the emergency event occurred.

[0110] Parsing the predicted probability distribution in the event recognition results and detecting probability peak points exceeding a preset confidence threshold and their corresponding sensory data unit timestamp information can be implemented, for example, by setting a preset confidence threshold, then traversing the predicted probability distribution in the event recognition results to find points whose probability values ​​exceed the preset confidence threshold. For each point exceeding the threshold, check whether it is a probability peak point, that is, the probability value of the point is greater than the probability values ​​of its adjacent points. If it is a probability peak point, record the sensory data unit timestamp information corresponding to the point.

[0111] Step S420: Determine the start and end time points of the emergency event based on the timestamp information corresponding to the probability peak point and the continuous change trend of the predicted probabilities of the previous and next perception data units, and generate time boundary description information.

[0112] The timestamp information corresponding to the probability peak point refers to the time stamp of the perception data unit corresponding to the probability peak point detected in step S410. The continuous change trend of the predicted probability of the previous and next perception data units refers to the change of the predicted probability value in the perception data units before and after the probability peak point, for example, whether it gradually increases or decreases. The starting time point and the ending time point of the emergency event refer to the time points when the emergency event begins and ends. The time boundary description information is a detailed description of the starting time point and the ending time point of the emergency event.

[0113] According to the timestamp information corresponding to the probability peak point, combined with the continuous change trend of the predicted probability of the previous and next perception data units, the starting time point and the ending time point of the emergency event are determined, and the time boundary description information is generated. For example, it can be implemented as follows: starting from the probability peak point, trace back the predicted probability value of the perception data unit, find the time point when the predicted probability value begins to rise significantly, and use it as the starting time point of the emergency event. The obvious rise can be judged by setting a rising threshold, for example, the predicted probability value rises by more than 0.2 in several consecutive timestamps. Then, starting from the probability peak point, search the predicted probability value of the perception data unit backward, find the time point when the predicted probability value begins to drop significantly, and use it as the end time point of the emergency event. The obvious drop can also be judged by setting a dropping threshold, for example, the predicted probability value drops by more than 0.2 in several consecutive timestamps. Finally, the starting time point and the ending time point are recorded to generate the time boundary description information.

[0114] Step S430: extracting environmental sensing data of the perception data unit corresponding to the probability peak point, and determining the initial occurrence area identifier and diffusion direction description of the emergency event in the perception data unit.

[0115] The environmental sensing data of the perception data unit corresponding to the probability peak point refers to the environmental sensing data contained in the perception data unit corresponding to the probability peak point detected in step S410, such as temperature, humidity, and air pressure. The initial occurrence area identifier refers to the identifier of the area where the emergency event first occurred, such as the number of a block or a building. The diffusion direction description refers to the direction in which the emergency event spreads from the initial occurrence area to other areas, such as east, west, southeast, etc.

[0116] Extracting the environmental sensing data of the perception data unit corresponding to the probability peak point, determining the initial occurrence area identification and diffusion direction description of the emergency event in the perception data unit can be implemented as follows: extracting the environmental sensing data of the perception data unit corresponding to the probability peak point. The corresponding perception data unit can be found from the multi-source perception data set according to the timestamp information, and then the environmental sensing data therein is extracted. Next, the abnormal change area in the environmental sensing data is analyzed to determine the initial occurrence area of ​​the emergency event. For example, if the temperature data in a certain area has an abnormal increase, and other environmental sensing data in the area have corresponding changes, then the area is marked as the initial occurrence area, and the identification information of the area is obtained. Finally, by comparing the environmental sensing data of adjacent timestamps, the expansion of the abnormal change area is analyzed to determine the diffusion direction of the emergency event.

[0117] Step S440: constructing a development trajectory model of the emergency event in the continuous perception data unit based on the initial occurrence area identification, diffusion direction description and time boundary description information.

[0118] The development trajectory model describes the development and evolution of an emergency event within continuous sensory data units. It can predict the future trend and location of an emergency event. The fundamental data for constructing the development trajectory model is the identification of the initial occurrence area, the description of the diffusion direction, and the temporal boundary. These data provide the starting location, development direction, and time range of the emergency event, respectively.

[0119] As an implementation manner, step S440 may specifically include the following steps S441 to S445:

[0120] Step S441: extracting a historical development state sequence of an emergency event from a continuous sensing data unit, where the historical development state sequence includes a historical timestamp, a historical region identifier, and a historical diffusion direction description.

[0121] A continuous sensory data unit refers to multiple sensory data units that are continuous in time and record the status information of an emergency event at different points in time. A historical development status sequence is a sequence of the development status information of an emergency event in the past, extracted from the continuous sensory data unit. It contains historical timestamps, historical region identifiers, and historical diffusion direction descriptions. Historical timestamps are time stamps of emergency events at different points in the past. Historical region identifiers are identifiers of the areas affected by the emergency event at different points in the past. Historical diffusion direction descriptions are the diffusion directions of the emergency event at different points in the past.

[0122] Extracting the historical development state sequence of emergency events from continuous perception data units can be implemented, for example, as follows: determining the time range of the continuous perception data units based on the time boundary description information. Then, within this time range, extracting relevant information for each perception data unit from the multi-source perception data set. For each perception data unit, extract its timestamp as the historical timestamp, determine the area identifier affected by the emergency event as the historical area identifier based on the environmental sensing data and event recognition results, and determine the historical diffusion direction description by comparing the regional changes of adjacent timestamps. Finally, this information is organized into a historical development state sequence in chronological order.

[0123] Step S442: Count the state transition probabilities of adjacent time stamps in the historical development state sequence, and construct a Markov state transition matrix that reflects the evolution law of the emergency event state.

[0124] The state transition probability is the probability that an emergency event will transition from one state to another at adjacent timestamps. The Markov state transition matrix is ​​a matrix whose elements represent the probability of an emergency event transitioning between different states. It reflects the evolution of the emergency event state.

[0125] Counting the state transition probabilities between adjacent timestamps in the historical development state sequence to construct a Markov state transition matrix that reflects the evolution of the emergency event state can be implemented as follows: For example, define the state space of the emergency event, combining different regional identifiers and diffusion directions as different states. Then, traverse the historical development state sequence and count the number of each state transition between adjacent timestamps. For example, the number of transitions from state A (block A, spreading eastward) to state B (block B, spreading eastward). Next, calculate the probability of each state transition by dividing the number of transitions by the total number of occurrences of that state. Finally, these state transition probabilities are combined to form a Markov state transition matrix. For example, if state A appears 10 times in the historical development state sequence, and 8 of those transitions lead to state B, the probability of transitioning from state A to state B is 0.8. Fill the matrix with the transition probabilities between all states to obtain the Markov state transition matrix.

[0126] If possible, a verification process can also be included, that is, calculating the autocorrelation coefficient of the state sequence. If the correlation coefficient of the lag order 2 or more is significantly non-zero (such as |ρ|>0.3), a high-order Markov model or RNN is used instead.

[0127] Step S443: The initial occurrence area identifier, diffusion direction description and time boundary description information are used as initial state parameters and input into the Markov state transfer matrix.

[0128] The initial occurrence area identifier, diffusion direction description, and temporal boundary description information represent the initial information of the emergency event determined in steps S430 and S420. The initial state parameters are a set of parameters that combine this initial information and serve as input to the Markov state transition matrix to begin predicting the development of the emergency event.

[0129] Using the initial occurrence area identifier, diffusion direction description, and time boundary description information as initial state parameters, the input of the Markov state transition matrix can be implemented as follows: the initial occurrence area identifier and diffusion direction description are combined into an initial state. For example, if the initial occurrence area identifier is Block A and the diffusion direction is eastward, the initial state is (Block A, diffusion eastward). Then, based on the state space of the Markov state transition matrix, the matrix row corresponding to the initial state is found. Finally, the initial state is used as the matrix input to begin the state transition calculation. For example, in the Markov state transition matrix, the row corresponding to (Block A, diffusion eastward) is found, and the state of the next timestamp is predicted based on the transition probability of this row.

[0130] Step S444: Calculate the possible development states of the emergency event in subsequent perception data units through the state transition probability, and generate a state prediction set containing multi-path development possibilities.

[0131] The state transition probability is the probability of an emergency event transitioning between different states, as recorded in the Markov state transition matrix. Subsequent sensory data units refer to sensory data units after the information describing the time boundary ends. Possible development states refer to the possible states of the emergency event in subsequent sensory data units, such as different combinations of regional identifiers and diffusion directions. The state prediction set is a set of possible development paths of the emergency event in subsequent sensory data units, which displays the possible development paths of the emergency event and the corresponding probabilities.

[0132] Using state transition probabilities to calculate the possible development states of an emergency event in subsequent sensory data units, generating a state prediction set encompassing multiple development paths can be implemented, for example, as follows: Starting from the initial state, the possible states and their probabilities for the next timestamp are calculated based on the state transition probabilities in the Markov state transition matrix. For example, if the initial state is (Block A, spreading eastward), and the probability of transitioning from this state to (Block B, spreading eastward) in the Markov state transition matrix is ​​0.8, and the probability of transitioning to (Block C, spreading southward) is 0.2, then the possible states for the next timestamp are (Block B, spreading eastward) and (Block C, spreading southward), with corresponding probabilities of 0.8 and 0.2, respectively. Then, for each possible state, the possible state and its probability for the next timestamp are calculated based on the state transition probabilities, and so on, until a predetermined number of time steps are reached. Finally, all possible development states and their corresponding probabilities are combined into a state prediction set.

[0133] Step S445: Confidence screening is performed on the state prediction set, and development state paths with prediction probabilities exceeding a preset threshold are retained to generate a development trajectory model.

[0134] Confidence screening processes the development state paths within the state prediction set, retaining only those with high predicted probabilities to improve the reliability and accuracy of the development trajectory model. The preset threshold is a pre-set probability value used to filter development state paths. The development trajectory model, derived after confidence screening, captures the primary development trajectory of an emergency event within continuous sensory data units.

[0135] The state prediction set is subjected to confidence screening, and the development state paths with predicted probabilities exceeding a preset threshold are retained. The generation of a development trajectory model can be implemented, for example, as follows: a preset threshold is set, such as 0.5. Then, each development state path in the state prediction set is traversed to check whether its predicted probability exceeds the preset threshold. The development state paths with predicted probabilities exceeding the threshold are retained, and the paths with predicted probabilities below the threshold are eliminated. Finally, the retained development state paths are sorted according to chronological order and logical relationships to generate a development trajectory model. For example, in a state prediction set, there are 3 development state paths, of which 2 paths have predicted probabilities of 0.6 and 0.8 respectively, exceeding the preset threshold of 0.5, and the predicted probability of the other path is 0.3, which is below the preset threshold. The first two paths are retained and connected in chronological order to generate a development trajectory model of the emergency event. This model can more reliably demonstrate the development trend of the emergency event in the continuous perception data unit.

[0136] Step S450: Generate spatiotemporal feature information including regional coverage and development trend prediction through the development trajectory model.

[0137] Regional coverage refers to the size of the geographic area affected by an emergency at different points in time. Development trend forecasts predict the future direction and speed of an emergency. Spatiotemporal feature information integrates the temporal and spatial characteristics of an emergency, providing a more comprehensive description of its development.

[0138] As an implementation manner, step S450 may specifically include the following steps S451 to S456:

[0139] Step S451: Perform density cluster analysis on the development trajectory identifier sequence output by the development trajectory model to identify the boundaries of dense areas where trajectory points are distributed, calculate coverage parameters based on the center identifiers of the dense area boundaries, and generate a regional coverage description with the center identifiers and coverage parameters as elements.

[0140] The trajectory marker sequence is a series of markers output by the trajectory model, representing the location and status of emergency events at different points in time. Density cluster analysis is a data clustering method that divides data points into different clusters based on the density of data points. Dense regions of trajectory point distribution refer to areas within the trajectory marker sequence where trajectory points are relatively concentrated. Dense region boundaries refer to the edges of these dense regions. Center markers are markers representing the center of dense regions, and coverage parameters, such as radius and area, describe the size of dense regions. The DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm can be used to perform density cluster analysis on the trajectory marker sequence output by the trajectory model. The basic idea of ​​the DBSCAN algorithm is to divide data points into core points, boundary points, and noise points by defining a neighborhood radius and a minimum number of points. A core point is a data point whose neighborhood contains at least the minimum number of points. A boundary point is a data point whose neighborhood contains fewer than the minimum number of points but intersects the neighborhood of a core point. A noise point is a data point that is neither a core point nor a boundary point. Specifically, first set the neighborhood radius and the minimum number of points. Then, traverse each data point in the development trajectory identification sequence and calculate the number of data points in its neighborhood. If the number of data points in the neighborhood of a data point is greater than or equal to the minimum number of points, it is marked as a core point; if the number of data points in the neighborhood of a data point is less than the minimum number of points, but it intersects with the neighborhood of a core point, it is marked as a boundary point; otherwise, it is marked as a noise point. Then, all core points and the boundary points in their neighborhoods are merged into a cluster to obtain different dense areas. For each dense area, calculate its boundary and center identifier. The center identifier can be obtained by calculating the average coordinate of all data points in the dense area. Finally, the coverage parameters are calculated based on the boundary of the dense area. For example, the radius of the circumscribed circle of the dense area can be calculated as the coverage parameter. The center identifier and coverage parameters are combined to generate a regional coverage description. For example, for a fire event development trajectory identification sequence, several dense areas are identified through the DBSCAN algorithm, and the center coordinates and circumscribed circle radius of each dense area are calculated to generate an area coverage description such as "center coordinates: (x1, y1), coverage radius: r1; center coordinates: (x2, y2), coverage radius: r2".

[0141] Step S452: Extract the direction identification set of the development trajectory identification sequence under continuous time stamps, calculate the angle change rate and identification movement gradient of the direction identification of adjacent time stamps based on the time interval of each unit, and construct a development trend direction stability evaluation function based on the time unit.

[0142] The direction identifier set refers to the identifier of the emergency event's development direction corresponding to each timestamp in the development trajectory identifier sequence. The angular change rate is the ratio of the angular change between the directional identifiers of adjacent timestamps to the time interval, reflecting the speed of change in the emergency event's development direction. The identifier movement gradient is the ratio of the displacement between the development trajectory identifiers of adjacent timestamps to the time interval, reflecting the speed of the emergency event's development. The development trend directional stability assessment function is used to assess the stability of the emergency event's development direction. It uses time units as a benchmark and comprehensively considers the angular change rate and identifier movement gradient.

[0143] The direction identification set of the development trajectory identification sequence under continuous time stamps can be determined by analyzing the position changes of adjacent time stamps in the development trajectory identification sequence. For example, for a development trajectory identification sequence of an emergency event, each identification contains the position coordinates of the emergency event at a certain time stamp. By calculating the vectors between the position coordinates of adjacent time stamps, the moving direction of the emergency event in each time interval is obtained, and these moving directions are used as direction identifications. The angle change rate of the direction identifications of adjacent time stamps based on the time interval of each unit can be calculated, for example, as follows: calculating the angle between the direction identifications of adjacent time stamps. The dot product formula of the vector can be used to calculate the cosine value of the angle between the two vectors, and then the angle is obtained by the inverse cosine function. Then, the angle is divided by the time interval to obtain the angle change rate. For the identification movement gradient, the displacement distance between the development trajectory identifications of adjacent time stamps is calculated, and the displacement distance is divided by the time interval to obtain the identification movement gradient.

[0144] Constructing a development trend direction stability evaluation function based on time units can comprehensively consider the angle change rate and the marker movement gradient, such as normalizing them separately to eliminate the dimension and then performing weighted summation.

[0145] Step S453: performing sliding window weighted averaging processing on the direction identification set based on the direction stability evaluation function to generate a trend prediction direction sequence after direction smoothing, and calculating the cumulative amount of slope change of the trend prediction direction sequence within the preset time window.

[0146] Sliding window weighted averaging reduces data fluctuations and smooths out directional indicators by sliding a fixed-size window across the set of directional indicators and performing a weighted average of the data within the window (understandably, normalization is performed before weighting to eliminate dimensional differences). The trend forecast direction sequence, derived from the sliding window weighted averaging process, more accurately reflects the development trend of the emergency. The cumulative slope change refers to the sum of the slope changes of the trend forecast direction sequence within a preset time window, reflecting the degree of change in the development trend of the emergency.

[0147] The sliding window weighted averaging of the direction identification set based on the direction stability evaluation function can be carried out in the following steps. First, determine the size w of the sliding window. Then, starting from the first element of the direction identification set, select windows of length w in sequence. For the direction identification in each window, calculate the weight of each identification according to the direction stability evaluation function. The weight can be adjusted according to the value of the direction stability evaluation function. For example, the smaller the value of the direction stability evaluation function, the more stable the direction, and the larger the corresponding weight can be set. Then, perform weighted averaging on the direction identification in the window to obtain the smoothed direction identification at the center position of the window. Combine the smoothed direction identifications at the center positions of all windows to generate a trend prediction direction sequence after direction smoothing.

[0148] Calculating the cumulative amount of slope changes in a trend prediction direction sequence within a preset time window can be implemented, for example, as follows: determine the size T of the preset time window. Then, select a subsequence of length T in the trend prediction direction sequence. For each adjacent data point in the subsequence, calculate its slope change. The slope change can be obtained by calculating the ratio of the directional difference between adjacent data points to the time interval. Finally, add up all the slope changes in the subsequence to obtain the cumulative amount of slope changes. For example, in a trend prediction direction sequence of a flood event, the preset time window is 10 time units, select a subsequence of length 10, calculate the directional slope changes of adjacent data points in the subsequence, and add up these slope changes to obtain the cumulative amount of slope changes. This cumulative amount can reflect the degree of change in the direction of the development trend of the flood within these 10 time units.

[0149] Step S454: constructing a dynamic area expansion model based on the coverage parameters and the cumulative amount of slope changes in the trend prediction direction sequence, and predicting the expansion direction and expansion rate description of the area coverage in subsequent perception data units.

[0150] The dynamic area expansion model describes the dynamic changes in emergency area coverage over time. It comprehensively considers coverage parameters and the cumulative slope changes of the trend prediction direction sequence. The expansion direction refers to the direction in which the emergency area coverage expands in subsequent perception data units, and the expansion rate describes the speed at which the emergency area coverage expands, such as the increase in coverage area per unit time.

[0151] The following method can be used to construct a dynamic regional expansion model based on the coverage parameter and the cumulative slope change of the trend forecast direction series. First, analyze the relationship between the coverage parameter and the cumulative slope change of the trend forecast direction series. Generally speaking, a larger cumulative slope change indicates a more dramatic change in the emergency's development trend and potentially less stable regional coverage expansion. A larger coverage parameter indicates a larger current impact area of ​​the emergency, potentially affecting the expansion rate. This relationship can be described by building a regression model. For example, assume a linear relationship exists between the expansion rate v and the coverage parameter r and the cumulative slope change s: v = a × r + b × s + c, where a, b, and c are coefficients to be determined (it is understood that r and s should be normalized, for example, to [0, 1], before addition to eliminate dimensionality issues). This regression model can be fitted using historical data to determine the coefficients. The expansion direction can be determined based on the trend forecast direction series. The average direction of the trend forecast direction series can be used as the basis for the expansion direction, while also considering the influence of the cumulative slope change on the expansion direction. If the cumulative slope change is large, a random offset can be made based on the average direction to reflect the uncertainty of the expansion direction. Alternatively, an XGBoost regression model can be trained using historical data, with input r and s used to predict v.

[0152] The predicted expansion direction and rate of the regional coverage in subsequent sensing data units can be calculated by inputting the current coverage parameters and the cumulative slope change of the trend prediction direction sequence into the dynamic regional expansion model. For example, for a fire incident, the current coverage radius r and the cumulative slope change s are known. These are substituted into the regression model v = a × r + b × s + c to calculate the expansion rate v. Simultaneously, the expansion direction is determined based on the trend prediction direction sequence, generating a description of the expansion direction and rate, such as "Expansion direction: 30° east by south, expansion rate: 2 square kilometers per hour."

[0153] Step S455: Input the expansion direction and expansion rate description into the trajectory prediction function, generate a probability distribution identification map containing the trajectory path in the subsequent time window, and extract the maximum connected area identification in the probability distribution identification map with a confidence level exceeding a preset threshold as the predicted development path boundary.

[0154] The trajectory prediction function is a function that predicts the future trajectory of an emergency event based on its expansion direction and expansion rate. The probability distribution identification map is the output of the trajectory prediction function, which represents the possible trajectory paths of the emergency event in the subsequent time window and its corresponding probability distribution. The confidence level refers to the probability value corresponding to each area in the probability distribution identification map, indicating the likelihood that the area is the actual trajectory path of the emergency event. The preset threshold is a pre-set probability value standard used to filter out areas with higher confidence levels. The maximum connected area identification refers to the identification of the largest interconnected area in the probability distribution identification map with a confidence level exceeding the preset threshold. The predicted development path boundary refers to the boundary of this maximum connected area, which represents the most likely range of development paths for the emergency event in the subsequent time window.

[0155] The Monte Carlo simulation method can be used to input the expansion direction and expansion rate description into the trajectory prediction function. The basic idea of ​​Monte Carlo simulation is to simulate the development process of an emergency event through random sampling. Specifically, based on the expansion direction and expansion rate description, the possible movement range of the emergency event at each time step is determined. For example, if the expansion direction is 30° east-southeast and the expansion rate is 2 square kilometers per hour, the emergency event may move a certain distance in this direction within each hour. Then, random sampling is performed within the possible movement range to generate multiple possible trajectory paths. For each trajectory path, its probability of occurrence is calculated. The probability value can be adjusted based on the uncertainty of the expansion rate and direction. All possible trajectory paths and their probabilities are combined to generate a probability distribution map.

[0156] Extracting the identifier of the largest connected region in the probability distribution identification graph with a confidence level exceeding a preset threshold can be implemented, for example, by setting a preset threshold. Then, each region in the probability distribution identification graph is traversed, and regions with a confidence level exceeding the preset threshold are marked as valid regions. Next, a connected region analysis algorithm, such as a depth-first search (DFS) or breadth-first search (BFS) algorithm, is used to identify all interconnected valid regions. Finally, the region with the largest area is selected from these connected regions, and its identifier is extracted as the boundary of the predicted development path. For example, for a probability distribution identification graph of an earthquake event, the preset threshold is 0.6. Through traversal and connected region analysis, the largest connected region with a confidence level exceeding 0.6 is found, and its boundary identifier is "boundary coordinates: [(x1, y1), (x2, y2), …, (xn, yn)]." This boundary identifier represents the most likely range of the earthquake's development path within the subsequent time window.

[0157] Step S456: Based on the regional coverage description and the predicted development path boundary, spatiotemporal feature information including a spatial coverage identification sequence and a trend direction change vector is generated. The spatial coverage identification sequence is used to mark the real-time coverage area of ​​the emergency event in the perception data unit, and the trend direction change vector is used to describe the predicted offset angle and speed change description of the development direction of the emergency event.

[0158] The regional coverage description is generated in step S451, which includes the center identifier and coverage parameters of the emergency event at different time points. The predicted development path boundary is extracted in step S455, which represents the most likely development path range of the emergency event in the subsequent time window. The spatial coverage identification sequence is a series of identifiers used to mark the real-time coverage area of ​​the emergency event in the perception data unit, and can be generated based on the regional coverage description and the predicted development path boundary. The trend direction change vector is a vector whose direction represents the predicted offset angle of the development direction of the emergency event, and its magnitude represents a description of the change in the development speed of the emergency event.

[0159] Generating a spatial coverage identifier sequence based on a regional coverage description and predicted development path boundaries can, for example, be implemented as follows: Based on the regional coverage description, determine the coverage area of ​​the emergency event at the current time point. Then, combined with the predicted development path boundaries, predict the coverage area of ​​the emergency event at subsequent time points. The location and size of the coverage area can be gradually updated based on the expansion direction and expansion rate descriptions. The coverage area identifiers at each time point are combined to generate a spatial coverage identifier sequence. For example, for a typhoon event, the current typhoon's center position and coverage radius are determined based on the regional coverage description. Combined with the predicted development path boundaries, the typhoon's movement path and coverage changes over the next few hours are predicted, generating a spatial coverage identifier sequence such as "time: t1, center coordinates: (x1, y1), coverage radius: r1; time: t2, center coordinates: (x2, y2), coverage radius: r2;..." Generating a trend direction change vector can be achieved by analyzing the trend prediction direction sequence and the expansion rate description. First, calculate the direction difference between the trend prediction direction sequence at the current time point and the subsequent time point to obtain the predicted offset angle. Then, calculate the change in the emergency event's development speed based on the expansion rate description. The predicted deviation angle and speed change are combined into a single vector, which serves as the trend direction change vector. For example, for a forest fire incident, the trend direction sequence indicates that the fire's direction will shift from due east to 20° east-southeast over the next period of time, and the expansion rate will increase from 1 square kilometer per hour to 1.5 square kilometers per hour. The trend direction change vector can be expressed as "direction deviation: 20°, speed change: 0.5 square kilometers per hour." Combining the spatial coverage identifier sequence and the trend direction change vector generates spatiotemporal feature information containing both the spatial coverage identifier sequence and the trend direction change vector. This information can comprehensively describe the temporal and spatial development of the emergency incident.

[0160] Step S500: Generate an emergency warning instruction including an event location identifier based on the emergency event type and temporal and spatial feature information, and send the emergency warning instruction to the target emergency response terminal to initiate a response operation.

[0161] The emergency event type is the type of emergency event existing in the target area determined in step S400, such as fire, earthquake, flood, etc. The spatiotemporal feature information is the feature information generated in step S450 and contains information such as regional coverage and development trend forecast. The event location identifier is an identifier used to accurately identify the location and scope of the emergency event. The emergency warning instruction is an instruction generated based on the emergency event type and spatiotemporal feature information, which is used to convey relevant information of the emergency event to the target emergency response terminal and initiate response operations. The target emergency response terminal refers to the terminal equipment responsible for responding to and handling emergency events, such as the command center of the fire department, the emergency system of the earthquake monitoring department, etc.

[0162] As an implementation manner, step S500 may specifically include the following steps S510 to S560:

[0163] Step S510: Parse the preset warning rule set corresponding to the emergency event type, and extract the warning level identifier and response strategy code associated with the emergency event type.

[0164] The preset warning rule set is a collection of pre-defined warning rules for different emergency event types. Each rule includes information such as the warning level and response strategy for that emergency event type. Warning level identifiers, such as Level 1 and Level 2, are used to distinguish the severity of emergency events. Response strategy codes encode the emergency response strategy, with different codes representing different response measures, such as evacuating residents or deploying relief supplies.

[0165] Parsing the preset warning rule set corresponding to the emergency event type can be implemented, for example, as follows: establishing a mapping relationship between the emergency event type and the preset warning rule set. For example, mapping a fire event to a fire warning rule set, mapping an earthquake event to an earthquake warning rule set, and so on. Then, based on the determined emergency event type, find the corresponding preset warning rule set. In the preset warning rule set, the warning level identification and response strategy code are determined according to the specific circumstances of the emergency event, such as the regional coverage and development trend in the spatiotemporal feature information. For example, for a fire event, if its regional coverage is large and its development trend is rapid, the corresponding warning level identification may be a level one warning, and the response strategy code may be "evacuating residents + deploying a large number of firefighting resources."

[0166] Step S520: Calculate the coverage area center identifier and boundary extension parameters of the emergency event within the current perception data unit according to the area coverage range in the spatiotemporal feature information.

[0167] Regional coverage refers to the spatial and temporal characteristics of the geographic area affected by an emergency at different points in time. The coverage area center identifier is the identifier, such as coordinates, of the center of the emergency coverage area within the current sensing data unit. Boundary expansion parameters describe the expansion of the emergency coverage area boundary, such as expansion speed and direction.

[0168] Calculating the center identifier of the coverage area based on the area coverage in the spatiotemporal feature information can be achieved by calculating the average coordinates of all location points within the area coverage. For example, if the area coverage is a polygonal area represented by a series of coordinate points, the x-coordinates and y-coordinates of all coordinate points in the polygon are added together, and then divided by the number of coordinate points to obtain the center coordinates of the coverage area. The boundary expansion parameters can be determined based on the description of the expansion direction and expansion rate in the spatiotemporal feature information. For example, if the expansion direction given in the spatiotemporal feature information is 30° east by south and the expansion rate is 2 square kilometers per hour, this information can be used as a boundary expansion parameter to describe that the boundary of the emergency coverage area expands at a rate of 2 square kilometers per hour in this direction.

[0169] Step S530: extracting the development trend forecast from the spatiotemporal feature information, combining it with the center mark of the coverage area, and constructing a dynamic development direction prediction function based on the linear fitting model of the historical development trajectory.

[0170] Development trend forecasting is the prediction of the future direction and speed of emergency events based on spatiotemporal characteristic information. Historical development trajectories refer to the path of emergency events over time. A linear fitting model uses a linear equation to fit historical development trajectory data, identifying linear trends within the data. The dynamic development direction prediction function is constructed based on development trend forecasts, coverage area center markers, and a linear fitting model of historical development trajectories. It is used to predict the direction of emergency events at different future time points.

[0171] Extracting development trend forecasts from spatiotemporal feature information allows direct access to relevant information, such as the predicted trend direction sequence, expansion direction, and expansion rate. Combining the coverage area center identifier with a linear fitting model of historical development trajectories to construct a dynamic development direction prediction function can be implemented, for example, by fitting the historical development trajectory data using linear regression to obtain a linear fitting equation. The basic idea of ​​linear regression is to find the optimal fitting line by minimizing the sum of the squared distances from the data points to the fitted line. For example, for the location coordinates of emergency events in the past, linear regression can be used to find a linear equation y=ax+b, where x represents time and y represents the location coordinates. The parameters of the linear fitting equation are then adjusted based on the development trend forecast and the coverage area center identifier. For example, if the development trend forecast indicates a change in the direction of the emergency event, the slope and intercept of the linear fitting equation can be adjusted based on the degree and magnitude of the change. Finally, the adjusted linear fitting equation is used as the dynamic development direction prediction function to predict the direction of the emergency event at different future time points.

[0172] Step S540: Input the boundary extension parameters into the predefined coordinate mapping rules to generate a standardized positioning identifier set consistent with the target area geographic identifier system.

[0173] Predefined coordinate mapping rules are pre-set rules for converting boundary extension parameters into coordinate representations consistent with the target area's geographic identification system (GIS). A target area GIS is a system used to identify geographic locations within the target area, such as a latitude and longitude coordinate system or a geographic information system (GIS). A standardized location identifier set is a set of location identifiers converted using these coordinate mapping rules. These identifiers have a unified representation within the target area's GIS, facilitating subsequent processing and use.

[0174] The following steps can be used to input boundary expansion parameters into predefined coordinate mapping rules. First, a predefined coordinate mapping rule is determined. The coordinate mapping rule can be designed based on the characteristics of the target area's geographic identification system. For example, the boundary expansion distance in kilometers is converted into changes in longitude and latitude coordinates. Then, information such as the expansion speed and direction in the boundary expansion parameter is converted into coordinate changes. For example, if the boundary expansion parameter indicates an expansion speed of 2 kilometers per hour in a due east direction, the coordinate conversion rule of the target area's geographic identification system is used to convert the 2 kilometers per hour distance into changes in longitude and latitude. Finally, these coordinate changes are applied to the center identifier of the coverage area to generate a series of standardized location identifiers consistent with the target area's geographic identification system. These location identifiers are combined to form a standardized location identifier set. For example, for a city flood event, the boundary expansion parameters are converted into changes in longitude and latitude coordinates using the coordinate mapping rule. These changes are then applied to the longitude and latitude coordinates of the center of the flood coverage area, generating a standardized location identifier set such as "(longitude 116.40, latitude 39.90), (longitude 116.41, latitude 39.90),..."

[0175] Step S550: performing time dimension interpolation processing on the standardized positioning identifier set according to the dynamic development direction prediction function to generate a prediction identifier sequence of the emergency event development path within a subsequent preset time period.

[0176] Temporal interpolation, based on a standardized set of location identifiers, inserts additional location identifiers within a pre-set time period based on a dynamic development direction prediction function to provide a more detailed description of the emergency's development path. A pre-set time period is a predetermined time range, such as the next 24 hours or the next week. The predicted identifier sequence is a series of location identifiers generated through temporal interpolation, representing the predicted development path of the emergency within the pre-set time period.

[0177] As an implementation manner, step S550 may specifically include the following steps S551 to S555:

[0178] Step S551: extracting the positioning identification points of adjacent timestamps from the standardized positioning identification set, and constructing a two-dimensional coordinate point set including timestamps and positioning identifications.

[0179] The location identifiers in the standardized location identifier set are location identifiers generated in step S540 and consistent with the target area's geographic identification system. Each location identifier corresponds to a timestamp. The two-dimensional coordinate point set is a set of two-dimensional coordinate points consisting of a timestamp and a location identifier, where the horizontal axis represents the timestamp and the vertical axis represents the location identifier.

[0180] To extract location markers with adjacent timestamps from a standardized location marker set, traverse the set and select adjacent location markers in the order of their timestamps. For example, if the standardized location marker set contains location markers p1, p2, and p3 corresponding to timestamps t1, t2, and t3, extract the location marker pairs (p1, p2) and (p2, p3) with adjacent timestamps. Combine each location marker with its corresponding timestamp to form a two-dimensional coordinate point, such as (t1, p1), (t2, p2), and (t3, p3). Combine all these two-dimensional coordinate points to construct a two-dimensional coordinate point set containing both timestamps and location markers.

[0181] Step S552: Using a cubic spline interpolation algorithm to perform curve fitting processing on the two-dimensional coordinate point set, generating an interpolation curve reflecting the continuity of the emergency event development path.

[0182] The cubic spline interpolation algorithm is used for curve fitting. By constructing a cubic polynomial function between each adjacent data point, the entire curve has continuous first- and second-order derivatives at each data point, thus ensuring smoothness and continuity. The interpolation curve is obtained by fitting a set of two-dimensional coordinate points using the cubic spline interpolation algorithm. It can reflect the continuity of the development path of an emergency event.

[0183] Using the cubic spline interpolation algorithm to perform curve fitting on a set of two-dimensional coordinate points can be implemented, for example, by sorting the two-dimensional coordinate points in the order of their abscissas (timestamps). Then, based on the principles of the cubic spline interpolation algorithm, a cubic polynomial function is constructed between each adjacent data point. The general form of a cubic polynomial function is S(x)=a+bx+cx 2 +dx 3, where x is the timestamp and a, b, c, and d are the coefficients of the polynomial. By satisfying the continuity conditions (equal function values, equal first-order derivatives, and equal second-order derivatives) of adjacent polynomials at data points and boundary conditions (such as the first-order or second-order derivatives at the two endpoints being known), the coefficients of the cubic polynomial for each interval can be solved. Finally, the cubic polynomials for all intervals are combined to form an interpolation curve. For example, for a set of two-dimensional coordinate points for a hurricane event, an interpolation curve can be constructed using the cubic spline interpolation algorithm. This curve smoothly connects the location markers at each time point, reflecting the continuity of the hurricane's development path.

[0184] Step S553: ​​determining the curvature adjustment parameter of the interpolation curve according to the dynamic development direction prediction function, and adaptively correcting the local curvature of the interpolation curve.

[0185] The curvature adjustment parameter is used to adjust the curvature of the interpolation curve. It can be determined based on the dynamic development direction prediction function. The local curvature refers to the degree of curvature of the interpolation curve in a certain local area. By adaptively correcting the local curvature, the interpolation curve can be made more consistent with the actual development path of the emergency event.

[0186] Determining the curvature adjustment parameter of the interpolation curve based on the dynamic development direction prediction function can be implemented, for example, by analyzing the changes in the dynamic development direction prediction function at different time points. The dynamic development direction prediction function can provide information on the development direction of the emergency event at different future time points, such as the angle and speed of the direction change. If the dynamic development direction prediction function shows that the development direction of the emergency event will undergo a significant change at a certain time point, it means that the local curvature of the interpolation curve near that time point should be larger; conversely, if the development direction changes slightly, the local curvature can be smaller. Then, based on the changes in the dynamic development direction prediction function, the curvature adjustment parameter is determined. By setting a mapping relationship, the change value of the dynamic development direction prediction function can be mapped to the curvature adjustment parameter. Finally, the curvature adjustment parameter is used to correct the local curvature of the interpolation curve. The adjustment of the local curvature can be achieved by adjusting the coefficients of the cubic polynomial of the interpolation curve in the corresponding area.

[0187] Step S554: uniformly sampling at preset time intervals on the interpolation curve to generate predicted positioning identifiers corresponding to each time point in a subsequent preset time period.

[0188] The preset time interval is a pre-set time interval, such as every hour, every half day, etc. The predicted location identifier is a location identifier obtained by sampling at the preset time interval on the interpolation curve, representing the predicted location of the emergency event at each time point in the subsequent preset time period.

[0189] For example, uniform sampling at preset time intervals on the interpolation curve can be implemented by determining the start and end times of a subsequent preset time period. Then, based on the preset time intervals, a series of time points are uniformly selected between the start and end times. For each selected time point, a corresponding location marker is found on the interpolation curve. The coordinates of the corresponding location marker can be calculated by substituting the time point into the cubic polynomial function of the interpolation curve. All of these predicted location markers are combined to generate a sequence of predicted location markers corresponding to each time point in the subsequent preset time period.

[0190] Step S555: performing smoothing filtering on the predicted positioning identifier to eliminate abnormal fluctuation identifier points generated during the interpolation process and generate a predicted identifier sequence.

[0191] Smoothing filtering is a method used to remove abnormal fluctuations in data, making it smoother and more stable. Abnormal fluctuation markers are those points that deviate from the normal development path due to data noise, fitting errors, and other factors during the interpolation process. The predicted marker sequence is a series of predicted location markers obtained through smoothing filtering, which can more accurately reflect the development path of the emergency event within a predetermined time period.

[0192] The moving average filtering method can be used to smooth the predicted location markers. The basic idea of ​​moving average filtering is to smooth the data by calculating the average value of the data within a certain window. Specifically, the size of the moving average window is first determined. The window size can be adjusted according to the specific situation, for example, it can be set to 3, 5, etc. Then, the predicted location marker sequence is traversed. For each marker point, a certain number of marker points before and after it (the window size) are selected, and the average value of these marker points is calculated. The average value is used as the smoothed value of the marker point. The smoothed values ​​of all marker points are combined to generate a smoothed predicted location marker sequence. For example, for a predicted location marker sequence for a forest fire, the moving average window size is set to 3. For each marker point in the sequence, the average value of the two marker points before and after it and the average value of the marker point itself is calculated to obtain a smoothed predicted location marker sequence, eliminating the abnormal fluctuating marker points generated during the interpolation process.

[0193] Step S560: The warning level identifier, response strategy code and prediction identifier sequence are integrated to generate an emergency warning instruction containing an event location identifier chain with timestamp alignment. The representation of each identifier node in the event location identifier chain is consistent with the representation of the standardized location identifier set.

[0194] The warning level identifier is the warning level identifier associated with the emergency event type extracted in step S510, the response strategy code is the code of the corresponding response strategy, and the prediction identifier sequence is the prediction location identifier sequence of the emergency event within the subsequent preset time period generated in step S555. Timestamp alignment refers to associating each identification point in the prediction identifier sequence with the corresponding timestamp to ensure that each identification point has clear time information. The event location identifier chain is a chain composed of a series of event location identifiers arranged in chronological order, and each identification node represents the location of the emergency event at a certain point in time. The emergency warning instruction is an instruction that integrates the warning level identifier, the response strategy code and the event location identifier chain, and is used to convey relevant information of the emergency event to the target emergency response terminal and initiate response operations.

[0195] As an implementation manner, step S560 may specifically include the following steps S561 to S566:

[0196] Step S561: input the warning level identifier, response strategy code and prediction identifier sequence into the instruction generation module, perform information association matching processing, and generate the initial warning instruction content.

[0197] The instruction generation module is responsible for integrating and processing the warning level identifier, response strategy code, and prediction identifier sequence. Information association and matching involves associating and matching the warning level identifier, response strategy code, and prediction identifier sequence, ensuring a clear correspondence between them within the initial warning instruction content. The initial warning instruction content, generated after information association and matching, contains relevant information about the warning level identifier, response strategy code, and prediction identifier sequence.

[0198] Inputting the warning level identifier, response strategy code, and prediction identifier sequence into the instruction generation module for information association and matching processing can be implemented, for example, by establishing an association table of the warning level identifier, response strategy code, and prediction identifier sequence in the instruction generation module. The association table can record the correspondence between the warning level identifier, response strategy code, and each identifier point in the prediction identifier sequence. For example, for a level one warning level identifier, the corresponding response strategy code is "full rescue + evacuation of residents." This correspondence can be associated with each identifier point in the prediction identifier sequence. Then, based on the association table, the warning level identifier, response strategy code, and prediction identifier sequence are combined to form the initial warning instruction content. The initial warning instruction content can adopt a structured format, such as "Warning Level: Level One; Response Strategy: Full Rescue + Evacuation of Residents; Prediction Identifier Sequence: [(t1, p1), (t2, p2), ...]".

[0199] Step S562: Perform timestamp alignment processing on the prediction identification sequence in the initial warning instruction content to ensure that each prediction identification node corresponds to the corresponding timestamp information one by one.

[0200] Timestamp alignment is the process of accurately associating each prediction identifier node in a prediction identifier sequence with its corresponding timestamp information. A prediction identifier node is each location identifier in the prediction identifier sequence, and the timestamp information is the time information corresponding to each prediction identifier node. Timestamp alignment ensures that the initial warning instructions are more accurate and clear, making them easier for target emergency response terminals to understand and process.

[0201] The timestamp alignment processing of the prediction identification sequence in the initial warning instruction content can be implemented as follows: check whether each prediction identification node in the prediction identification sequence already contains the corresponding timestamp information. If the prediction identification node only contains the positioning identification but no timestamp information, it is necessary to obtain the corresponding timestamp from the time point information used when sampling in step S554. Then, the timestamp information is associated with the prediction identification node to ensure that each prediction identification node has a unique corresponding timestamp. For example, if the prediction identification sequence is [p1, p2, p3] and the corresponding timestamp information is [t1, t2, t3], they are combined into the form of [(t1, p1), (t2, p2), (t3, p3)] to achieve timestamp alignment.

[0202] Step S563: extract key information elements from the initial warning instruction content, including the warning level identifier, the response strategy code, and the prediction identifier sequence after timestamp alignment.

[0203] Key information elements are the most important information in the initial warning instruction content, including the warning level identifier, response strategy code, and the predicted identification sequence after timestamp alignment. This information plays a key role in the emergency response operation of the target emergency response terminal.

[0204] Extracting key information elements from the initial warning instructions can be achieved by parsing their structured format. For example, if the initial warning instructions use the format "Warning Level: Level 1; Response Strategy: Comprehensive Rescue + Evacuation; Prediction Identifier Sequence: [(t1, p1), (t2, p2), ...]," string matching and segmentation can be used to extract "Level 1" as the warning level identifier, "Comprehensive Rescue + Evacuation" as the response strategy code, and [(t1, p1), (t2, p2), ...] as the timestamp-aligned prediction identifier sequence.

[0205] Step S564: Standardize the format of key information elements, unify the representation form and data structure of information elements, and generate standardized early warning instruction content.

[0206] Format standardization is the process of unifying and standardizing the representation and data structure of key information elements. This unified representation and data structure makes warning instructions easier to process and parse by target emergency response terminals. Standardized warning instructions are generated after format standardization and have a unified format and data structure.

[0207] Standardizing the format of key information elements can be implemented, for example, by establishing a unified representation format and data structure. For example, warning level identifiers can be uniformly represented by numbers, such as 1 for a Level 1 warning; response strategy codes can be encoded using unified coding rules; and prediction identifier sequences can be represented using a unified coordinate system and data type. Then, the key information elements are converted according to the unified representation format and data structure. For example, the warning level identifier "Level 1" is converted to the number 1, and the response strategy code "Full rescue + evacuation of residents" is converted to the corresponding coded value. Finally, the converted key information elements are combined into standardized warning instruction content. For example, the standardized warning instruction content can be represented in JSON format: {"Warning Level": 1, "Response Strategy Code": "001", "Prediction Identifier Sequence": [(t1, p1), (t2, p2), ...]}.

[0208] Step S565: Verify the content of the standardized warning instruction, verify the integrity and logical consistency of the information elements, and generate an emergency warning instruction that passes the verification.

[0209] Verification is the process of checking and verifying the content of standardized warning instructions, primarily verifying the integrity and logical consistency of information elements. Information element integrity means that the standardized warning instructions contain all necessary information elements, such as warning level identifiers, response strategy codes, and forecast identifier sequences. Logical consistency refers to the logical relationship between information elements, such as the correct correspondence between warning level identifiers and response strategy codes. Emergency warning instructions that pass verification are those that, after verification, demonstrate complete and logically consistent information elements.

[0210] Verification of the content of the standardized warning instruction can be implemented, for example, as follows: checking whether the content of the standardized warning instruction contains all the necessary information elements. This can be achieved by checking the key-value pairs of the standardized warning instruction content in JSON format. For example, check whether it contains keys such as "warning level", "response strategy code" and "forecast identification sequence". Then, verify the logical consistency between the information elements. The correspondence between the warning level identifier and the response strategy code can be verified by querying the predefined association table. For example, if the response strategy code corresponding to the first-level warning level identifier should be "001", then check whether the content of the standardized warning instruction complies with this correspondence. If the information elements in the content of the standardized warning instruction are complete and logically consistent, an emergency warning instruction that passes the verification is generated; otherwise, the content of the standardized warning instruction needs to be revised.

[0211] Step S566: Encrypt the verified emergency warning instruction to generate a final emergency warning instruction containing encrypted information.

[0212] Encryption is the process of encrypting verified emergency warning instructions to ensure the security and confidentiality of the instruction content. Encrypted information is encrypted information, which prevents the instruction content from being eavesdropped or tampered with during transmission. The final emergency warning instruction, which contains encrypted information, can be securely transmitted to the target emergency response terminal.

[0213] Symmetric encryption algorithms, such as the Advanced Encryption Standard (AES) algorithm, can be used to encrypt verified emergency warning instructions. The AES algorithm is a symmetric key encryption algorithm that uses the same key for encryption and decryption. Specifically, an encryption key is first generated. The encryption key can be a randomly generated binary sequence. Then, the verified emergency warning instructions are encrypted using the AES algorithm and the encryption key. The encryption process converts the verified emergency warning instructions into a string of ciphertext. Finally, the encrypted ciphertext is used as the encrypted information and combined with necessary encryption metadata (such as the encryption algorithm identifier and key length) to form the final emergency warning instruction containing the encrypted information. For example, the final emergency warning instruction can use the following format: {"Encryption Algorithm": "AES", "Key Length": 128, "Encryption Information": "xxxxxx"}, where "xxxxxx" is the encrypted ciphertext.

[0214] Transmitting the final emergency warning instruction to the target emergency response terminal can be accomplished through network communication technologies. For example, wireless communication networks (such as 4G and 5G) can be used to transmit the final emergency warning instruction to the communication module of the target emergency response terminal. After receiving the final emergency warning instruction, the target emergency response terminal uses the corresponding decryption algorithm and key to decrypt the encrypted information, obtaining the original emergency warning instruction content, and then initiating response operations, such as deploying rescue resources and organizing personnel evacuation, to address the emergency event in the target area.

[0215] It is understandable that the various algorithms involved in the above-mentioned introductions of the embodiments of the present invention, such as word segmentation algorithms, filtering algorithms, cubic spline interpolation algorithms, symmetric encryption algorithms, etc., can all be learned from the relevant content in the prior art. In order to save space, they will not be expanded too much in the embodiments of the present invention. In addition, when implementing the scheme of the present invention, those skilled in the art can supplement the details according to the common knowledge in this field. For example, according to the common knowledge in this field, normalization can be used to eliminate dimensional conflicts before feature fusion, interpolation can be used to eliminate dimensional differences, and thresholds can be reasonably set based on historical data, experience or business scenario requirements. The model can be trained based on a general model training method, and the number of layers in the model structure can be set based on actual needs, the activation function can be selected, etc. The present invention will no longer provide a complicated introduction to the overly detailed implementation process.

[0216] See also Figure 2 , Figure 2 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. The computer device includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 may be connected via a bus or other means. The processor 101 (also known as the Central Processing Unit (CPU)) is the computing and control core of the computer device, capable of parsing various instructions within the computer device and processing various data within the computer device. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, a mobile communication interface, etc.), which can be used to send and receive data under the control of the processor 101. The communication interface 102 may also be used for data transmission and interaction within the computer device. The memory 103 is a storage device within the computer device, used to store programs and data. It is understood that the memory 103 herein may include both the built-in memory of the computer device and, of course, the extended memory supported by the computer device. The memory 103 provides storage space, which stores the operating system of the computer device, but this is not limited by the present invention.

[0217] In one embodiment, the processor 101 executes the emergency event identification method based on the neural network model provided in the above embodiment of the present invention by running the computer program in the memory 103.

Claims

1. A method for identifying emergency events based on a neural network model, characterized in that: The method comprises: Acquire a multi-source perception data set covering a target area, wherein the multi-source perception data set includes a plurality of perception data units with timestamps, each perception data unit consisting of environmental sensing data and interaction feedback data; Performing feature encoding processing on the multi-source perception data set to obtain a multi-dimensional feature representation set of the perception data unit, wherein the multi-dimensional feature representation set includes a temporal correlation feature and a semantic correlation feature; Calling a pre-trained emergency event recognition neural network model to perform event recognition processing on the multidimensional feature representation set to generate an event recognition result for the perception data unit, wherein the event recognition result includes a predicted probability distribution of different emergency event categories; Determine the type of emergency event present in the target area and the spatiotemporal feature information of the emergency event in the perception data unit according to the event recognition result; An emergency warning instruction including an event location identifier is generated based on the emergency event type and the spatiotemporal feature information, and the emergency warning instruction is sent to a target emergency response terminal to initiate a response operation.

2. The method for identifying emergency events based on a neural network model according to claim 1, characterized in that: The performing feature encoding processing on the multi-source perception data set to obtain a multi-dimensional feature representation set of the perception data unit includes: Performing data cleaning on the multi-source perception data set to eliminate noise interference information in the perception data units to obtain a denoised perception data set; Performing time series correlation extraction processing on the environmental sensing data in the denoised sensing data set to generate state evolution features of the sensing data units at consecutive timestamps as time series correlation features, wherein the state evolution features include a description of a change trend between adjacent timestamps; Performing semantic association extraction processing on the interaction feedback data in the denoised perception data set to generate semantic association features of the interaction content in the perception data unit, wherein the semantic association features include a contextual association description of the interaction intention; Inputting the temporal correlation features and the semantic correlation features into a feature integration module for correlation analysis and processing to obtain an integrated feature set with a unified representation form; Based on the association strength evaluation results of the features of different dimensions in the integrated feature set, the fusion emphasis direction of the temporal association features and the semantic association features in the feature encoding process is dynamically adjusted to generate the multidimensional feature representation set.

3. The method for identifying emergency events based on a neural network model according to claim 1, characterized in that: The calling of the pre-trained emergency event recognition neural network model to perform event recognition processing on the multi-dimensional feature representation set to generate an event recognition result of the perception data unit includes: Inputting the multidimensional feature representation set into the feature processing layer of the emergency event recognition neural network model to perform hierarchical feature fusion processing to generate a fused feature vector with multi-granularity representation; Performing context dependency modeling processing on the fused feature vector through the context modeling layer of the emergency event recognition neural network model to generate context enhancement features in the time dimension and the semantic dimension; Performing event probability prediction processing on the context enhancement features based on the probability prediction layer of the emergency event recognition neural network model to generate a probability distribution set containing predicted probability values ​​corresponding to each emergency event category; Performing category confidence calibration processing on the probability distribution set, adjusting the confidence weights of the predicted probability values ​​of different emergency event categories, and generating a calibrated probability distribution set; An event recognition result of the perception data unit is generated based on the calibrated probability distribution set, where the event recognition result includes a predicted probability distribution of each emergency event category.

4. The method for identifying emergency events based on a neural network model according to claim 1, wherein: The determining, based on the event recognition result, the type of emergency event existing in the target area and the spatiotemporal feature information of the emergency event in the perception data unit includes: Analyze the predicted probability distribution in the event recognition result, detect the probability peak point exceeding the preset confidence threshold and the corresponding perception data unit timestamp information; According to the timestamp information corresponding to the probability peak point, combined with the continuous change trend of the predicted probability of the previous and next perception data units, the start and end time points of the emergency event are determined, and time boundary description information is generated; Extracting environmental sensing data of the perception data unit corresponding to the probability peak point, and determining the initial occurrence area identifier and diffusion direction description of the emergency event in the perception data unit; Constructing a development trajectory model of the emergency event in continuous perception data units based on the initial occurrence area identifier, the diffusion direction description, and the time boundary description information; The development trajectory model is used to generate spatiotemporal feature information including regional coverage and development trend prediction.

5. The method for identifying emergency events based on a neural network model according to claim 1, wherein: The generating of the emergency warning instruction including the event location identifier based on the emergency event type and the spatiotemporal feature information includes: Parsing the preset warning rule set corresponding to the emergency event type, and extracting the warning level identifier and response strategy code associated with the emergency event type; Calculate the coverage area center identifier and boundary extension parameters of the emergency event within the current perception data unit according to the area coverage in the spatiotemporal feature information; Extracting the development trend forecast from the spatiotemporal feature information, combining it with the coverage area center identifier, and constructing a dynamic development direction prediction function based on a linear fitting model of historical development trajectory; Inputting the boundary extension parameters into a predefined coordinate mapping rule to generate a standardized location identifier set consistent with the target area's geographic identifier system; Performing time dimension interpolation processing on the standardized positioning identifier set according to the dynamic development direction prediction function to generate a prediction identifier sequence of the emergency event development path within a subsequent preset time period; The warning level identifier, response strategy code and prediction identifier sequence are integrated to generate an emergency warning instruction containing an event location identifier chain with timestamp alignment, and the representation of each identifier node in the event location identifier chain is consistent with the representation of the standardized location identifier set.

6. The method for identifying emergency events based on a neural network model according to claim 2, characterized in that: The performing time series correlation extraction processing on the environmental sensing data in the denoised sensing data set to generate state evolution features of the sensing data units at continuous timestamps as time series correlation features includes: Extracting the environmental sensing data change area between adjacent perception data units and analyzing the attribute parameter difference characteristics of the change area; Determining a description of a state change direction and a degree of change of the perception data unit between adjacent timestamps according to the attribute parameter difference characteristics; Constructing a state evolution analysis model of the perception data unit based on the description of the state change direction and change degree; Calculating a consistent description of the state change trend of the perception data unit at continuous timestamps using the state evolution analysis model; A periodic pattern recognition process is performed on the state evolution process of the perception data unit, and a state repetition feature of the perception data unit within a preset time window is extracted as a state evolution feature.

7. The method for identifying emergency events based on a neural network model according to claim 2, characterized in that: The performing semantic association extraction processing on the interactive feedback data in the denoised perception data set to generate semantic association features of the interactive content in the perception data unit includes: Performing word segmentation processing on the text content in the interactive feedback data to obtain a plurality of interactive word units; Calling a pre-trained semantic encoding model to perform context encoding processing on the multiple interaction word units to generate a semantic vector representation of the interaction feedback data; Analyzing the semantic association strengths between different interactive word units in the semantic vector representation to generate an association relationship graph including semantic association weights; The semantic features of the core interaction intention in the interaction feedback data are extracted according to the association relationship graph as semantic association features.

8. The method for identifying emergency events based on a neural network model according to claim 3, characterized in that: The process of performing context dependency modeling on the fused feature vector by the context modeling layer of the emergency event recognition neural network model to generate context enhanced features in the time dimension and the semantic dimension includes: Inputting the fused feature vector into the time dependency sublayer of the context modeling layer to perform time series dependency analysis and processing to generate context dependency features of the time dimension; Inputting the fused feature vector into the semantic dependency sublayer of the context modeling layer, performing context-dependent relationship analysis on the semantic content, and generating context-dependent features of the semantic dimension; Performing cross-dimensional correlation analysis on the context-dependent features of the time dimension and the context-dependent features of the semantic dimension to generate cross-dimensional context-dependent features; performing weighted fusion processing on the context-dependent features of the time dimension and the context-dependent features of the semantic dimension based on the cross-dimensional context association features to generate a fused context-dependent feature; Dimension alignment processing is performed on the fused context-dependent features to generate context-enhanced features in the time dimension and the semantic dimension.

9. The method for identifying emergency events based on a neural network model according to claim 4, characterized in that: The generation of spatiotemporal feature information including regional coverage and development trend predictions through the development trajectory model includes: Performing density cluster analysis on the development trajectory identifier sequence output by the development trajectory model to identify the boundaries of dense regions where trajectory points are distributed, and calculating coverage parameters based on the center identifiers of the dense region boundaries to generate a regional coverage description with the center identifiers and coverage parameters as elements; Extracting the direction identification set of the development trajectory identification sequence at continuous time stamps, calculating the angle change rate and identification movement gradient of the direction identification of adjacent time stamps based on the time interval of each unit, and constructing a development trend direction stability evaluation function based on the time unit; Performing sliding window weighted averaging processing on the direction identification set based on the direction stability evaluation function to generate a trend prediction direction sequence after direction smoothing, and calculating the cumulative amount of slope change of the trend prediction direction sequence within a preset time window; Constructing a dynamic area expansion model based on the coverage parameters and the cumulative amount of slope changes of the trend prediction direction sequence, and predicting the expansion direction and expansion rate description of the area coverage in subsequent perception data units; Inputting the expansion direction and expansion rate description into the trajectory prediction function, generating a probability distribution identification map containing the trajectory path in the subsequent time window, and extracting the maximum connected area identification with a confidence level exceeding a preset threshold in the probability distribution identification map as the predicted development path boundary; Based on the regional coverage description and the predicted development path boundary, spatiotemporal feature information including a spatial coverage identification sequence and a trend direction change vector is generated. The spatial coverage identification sequence is used to mark the real-time coverage area of ​​the emergency event in the perception data unit, and the trend direction change vector is used to describe the predicted offset angle and speed change description of the development direction of the emergency event.

10. A computer device, characterized in that: include: a memory storing a computer program; A processor is used to load the computer program to implement the emergency event identification method based on the neural network model as described in any one of claims 1 to 9.

Images