Emergency event visualization cooperation method and terminal device

Abstract

The invention provides an emergency event visualization cooperation method and terminal equipment, and the method comprises the steps: obtaining a historical monitoring data set of an emergency event in a target region, carrying out the preprocessing, and generating a standardized event feature set; training a time sequence prediction model according to the dynamic environment parameters to obtain an event evolution prediction model; and constructing a geographic region matching model based on the static geographic parameters, and generating a visual layer template associated with the target region. And collecting a current environment data flow of the target area in real time, performing classification through the event evolution prediction model to obtain an event classification result, and calling a corresponding visual layer template from the geographic area matching model according to the event classification result. And fusing the diffusion path, the intensity level and the called visual layer template in the event classification result to generate a dynamic visual instruction set, and rendering and outputting an emergency event evolution situation map through an interactive interface. According to the invention, the timeliness of emergency response and decision reliability can be improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and particularly to a method and a terminal device for visual collaborative emergency events. Background Art

[0002] In the field of emergency event processing, bottleneck problems such as data islands, disconnection between model prediction and geographical constraints, and lagged visual response generally exist in the existing technologies. Traditional methods usually process dynamic environment monitoring data and static geographical information separately, making it difficult for the prediction model to capture the physical constraints of geographical space on event diffusion. In addition, mainstream time series prediction models (such as single LSTM) cannot synchronously analyze spatio-temporal coupling features, and the prediction results lack the dynamic correlation of event intensity spreading with space, resulting in insufficient basis for emergency resource scheduling.

[0003] At the visual level, existing GIS systems mostly use static layer rendering and cannot dynamically adjust display elements according to the evolution of real-time events. At the same time, traditional rendering engines have problems such as high latency and layer overlay conflicts when processing multi-source heterogeneous data, and it is difficult to support real-time interactive analysis of the evolution process of emergency events. A more prominent problem is the disconnection between the prediction and disposal links: the output results of the model need to be manually converted into geographically interpretable information, resulting in an overly long response delay. These problems seriously restrict the timeliness of emergency response and the reliability of decision-making. Summary of the Invention

[0004] The present invention provides a method and a terminal device for visual collaborative emergency events.

[0005] According to one aspect of the present invention, there is provided a method for visual collaborative emergency events, the method comprising: Obtaining a historical monitoring data set of emergency events in a target area, and preprocessing the historical monitoring data set to generate a standardized event feature set; the standardized event feature set includes dynamic environment parameters and static geographical parameters associated with various types of emergency events; Training a time series prediction model according to the dynamic environment parameters in the standardized event feature set to obtain an event evolution prediction model; the output of the event evolution prediction model includes the diffusion path and intensity level of the emergency event within a preset time window; Based on the static geographical parameters in the standardized event feature set, constructing a geographical area matching model to generate a visual layer template associated with the target area; the visual layer template includes terrain distribution, infrastructure coordinates, and emergency resource deployment information; Collect the current environmental data stream of the target area in real time, classify the current environmental data stream through the event evolution prediction model to obtain an event classification result, and call a corresponding visual layer template from the geographical area matching model according to the event classification result; Fuse the diffusion path and intensity level in the event classification result with the called visual layer template to generate a dynamic visualization instruction set, and render and output an emergency event evolution situation map through an interaction interface.

[0006] According to another aspect of the present invention, there is provided a terminal device, including: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method described above.

[0007] The present invention has at least the following beneficial effects: The emergency event visualization method provided by the present invention constructs an intelligent decision-making closed loop of prediction-matching-rendering through the collaborative processing of dynamic environmental parameters and static geographical data. In this way, dynamic environmental parameters can capture real-time evolution laws such as meteorological changes and surface displacements, and static geographical parameters can accurately reflect spatial constraint conditions such as terrain structure and resource layout. Based on the fusion analysis of multi-dimensional parameters, the limitations of the traditional visualization system's spatio-temporal dimension fragmentation can be broken through. Moreover, through the probability distribution of the diffusion path output by the time series prediction model and the temporal prediction of the intensity level, combined with the customized layer template generated by the geographical matching model, the accurate superposition of the disaster evolution trend and the geographical real scene can be realized in the visualization interface. By dynamically fusing the spatio-temporal evolution characteristics of the prediction model and the static constraint characteristics of the geographical model, the generated thermal diffusion trajectory can adapt to the terrain undulation changes, and the intensity level rendering can reflect the resource deployment density in real time, thus forming a complete decision-making chain of "prediction and early warning - path simulation - resource allocation" in emergency command. In addition, based on the multi-source data fusion mechanism of the standardized feature set, the visualization system can still maintain a stable situation deduction ability when the data collection is incomplete or there is noise interference, significantly enhancing the robustness of emergency response decision-making in complex environments.

[0008] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. Description of the Drawings

[0009] The accompanying drawings exemplarily illustrate embodiments and form a part of the specification, and are used together with the textual description of the specification to explain the exemplary implementation manners of the embodiments. The illustrated embodiments are for illustrative purposes only and do not limit the scope of the claims. In all the drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0010] Figure 1 The schematic diagram of the application scenario of the emergency event visualization collaboration method according to an embodiment of the present invention is shown.

[0011] Figure 2 The flowchart of an emergency event visualization collaboration method according to an embodiment of the present invention is shown.

[0012] Figure 3 The schematic diagram of the composition of a terminal device according to an embodiment of the present invention is shown. Detailed implementation manners

[0013] The following makes an explanation of the exemplary embodiments of the present invention with reference to the accompanying drawings. Various details of the embodiments of the present invention are included to facilitate understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the present invention. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted below.

[0014] Figure 1 The schematic diagram of the application scenario provided according to an embodiment of the present invention is shown. The application scenario includes one or more emergency monitoring devices 101, a terminal device 120, and one or more communication networks 110 that couple the one or more emergency monitoring devices 101 to the terminal device 120. The emergency monitoring device 101 is, for example, a device for monitoring data related to emergency events, such as relevant sensor devices in a meteorological observation station, sensing devices in an air-space-ground integrated sensing network, etc., and is not specifically limited.

[0015] In Figure 1 the configuration shown, the terminal device 120 can be a server or a computer, and it includes one or more components that implement the functions executed by the terminal device 120. These components can include software components, hardware components, or a combination thereof that can be executed by one or more processors. The user operating the emergency monitoring device 101 can sequentially use one or more application programs to interact with the terminal device 120 to utilize the services provided by these components. It should be understood that various different system configurations are possible, which can be different from the application scenario. Therefore, Figure 1 is an example of a system for implementing the various methods described herein and is not intended to be limiting.

[0016] Network 110 can be any type of network well-known to those skilled in the art, which can support data communication using any one of a variety of available protocols (including but not limited to TCP / IP, SNA, IPX, etc.). By way of example only, one or more networks 110 can be a local area network (LAN), an Ethernet-based network, Token Ring, a wide area network (WAN), the Internet, a virtual network, a virtual private network (VPN), an intranet, an extranet, a blockchain network, a public switched telephone network (PSTN), an infrared network, a wireless network (such as Bluetooth, WIFI), and / or any combination of these and / or other networks.

[0017] Terminal device 120 can include one or more general-purpose computers, dedicated server computers (such as PC (personal computer) servers, UNIX servers, midrange servers), blade servers, mainframes, server clusters, or any other suitable arrangement and / or combination. The computing unit in terminal device 120 can run one or more operating systems including any of the above operating systems as well as any commercially available server operating system.

[0018] The above application scenario provided by the embodiments of the present invention may further include one or more databases 130. In some embodiments, these databases can be used to store data and other information. For example, one or more of databases 130 can be used to store historical data. Databases 130 can reside in various locations.

[0019] Please refer to Figure 2 , the emergency event visualization collaboration method provided by the embodiments of the present invention may include the following steps: Step S100: Obtain a historical monitoring data set of emergency events in a target area, and preprocess the historical monitoring data set to generate a standardized event feature set; the standardized event feature set includes dynamic environmental parameters and static geographical parameters associated with multiple types of emergency events.

[0020] The historical monitoring dataset refers to the set of original data related to emergency events recorded in the target area during a past time period. Its data sources include, for example, environmental sensor networks, geographic information systems, historical event report databases, and manual inspection logs. Dynamic environmental parameters refer to environmental factors that change over time and have a direct impact on the evolution process of emergency events, such as real-time wind speed, rainfall intensity, temperature gradient changes, fluctuations in air pollutant concentrations, and the thermal distribution of crowd movement; static geographic parameters refer to the geographic spatial attributes that remain relatively stable on a longer time scale, such as terrain elevation data, geological structure characteristics, the distribution density of permanent buildings, the topological structure of the transportation road network, and the coordinates of fixed emergency shelters. The preprocessing process includes cleaning, alignment, normalization, and feature extraction operations on the historical monitoring dataset. For example: filling in missing sensor values through an interpolation algorithm, eliminating abnormal noise using a sliding window filter, aligning the timestamps of multi-source heterogeneous data based on event type tags, mapping parameters with different dimensions to a unified numerical interval using the maximum-minimum normalization method, and finally screening out dynamic environmental parameters and static geographic parameters that are strongly correlated with the evolution of emergency events through feature engineering to form a standardized event feature set. The generation of this set makes the data collected at different times and by different monitoring devices consistent and comparable, laying a data foundation for subsequent model training and visualization analysis. For example, for emergency events such as forest fires, the dynamic environmental parameters can extract the wind speed sequence in the direction of fire spread and the curve of combustible humidity change, and the static geographic parameters include the mountain slope distribution map, the location of firebreaks, and the coordinates of fire fighting water storage ponds.

[0021] Step S200: Train a time series prediction model based on the dynamic environmental parameters in the standardized event feature set to obtain an event evolution prediction model; the output of the event evolution prediction model includes the diffusion path and intensity level of the emergency event within a preset time window.

[0022] Exemplarily, the time series prediction model adopts a long short-term memory neural network architecture. Its input layer receives a multi-dimensional vector of dynamic environmental parameters in the standardized event feature set arranged by time steps. The hidden layer captures the temporal dependence relationship and non-linear interaction between parameters through a gating mechanism. The output layer maps to the prediction results of event evolution within a preset time window. The preset time window refers to the future prediction duration set according to the emergency response level, such as 6 hours, 24 hours, or 72 hours, and its length is associated with the event type and regional risk level. The diffusion path is, for example, the spreading trajectory of an emergency event in the spatial dimension, specifically manifested as the direction angle, diffusion speed, and blocked boundary conditions extending outward from the event center point; the intensity level can quantify the harm degree of the event through a discretized grading index. For example, the fire intensity is divided into levels 1-5, corresponding to different combustion rates and heat radiation ranges. During the training process, the model optimizes the loss function through the backpropagation algorithm, and the loss function is composed of the Euclidean distance error of the diffusion path and the cross-entropy of the intensity level weighted. After multiple rounds of iteration, the model can accurately capture the mapping law between dynamic environmental parameters and event evolution trends. For example, in the flood prediction scenario, the model can predict the area where the flood peak arrives and the inundation depth level within the next 12 hours based on real-time rainfall, river water level, and soil permeability data.

[0023] Step S300: Based on the static geographical parameters in the standardized event feature set, construct a geographical area matching model to generate a visualization layer template associated with the target area; the visualization layer template includes terrain distribution, infrastructure coordinates, and emergency resource deployment information.

[0024] Exemplarily, the geographical area matching model can adopt a graph convolutional neural network structure. Its nodes represent geographical units with spatial continuity within the target area, and the edge weights are calculated from the adjacency relationship and attribute similarity between the units. The terrain distribution data is obtained through rasterization processing of a digital elevation model (DEM). Each raster unit contains elevation values, slope, and aspect information; the infrastructure coordinates cover the longitude and latitude positioning data of key buildings such as hospitals, schools, and transportation hubs, and additional attributes such as building structure type, occupancy, and seismic resistance level are attached; the emergency resource deployment information includes the locations of fire stations, the capacities of material storage warehouses, the distribution of UAV takeoff and landing points, and the coverage range of communication base stations. The model aggregates the static geographical parameters into geographical feature vectors with spatial relevance through a hierarchical clustering algorithm, and then generates a multi-scale visualization layer template through a feature fusion layer. For example, for the earthquake disaster scenario, the basic layer can display the distribution of fault zones and the seismic resistance levels of buildings, and the overlay layer marks the locations of rescue teams and the medical supply distribution routes. The visualization layer template can adopt a hierarchical rendering technique, allowing dynamic adjustment of the display level details to ensure that key geographical elements can be clearly presented at different zoom levels.

[0025] Step S400: Continuously collect the current environmental data stream of the target area in real time, classify the current environmental data stream through the event evolution prediction model to obtain an event classification result, and call the corresponding visualization layer template from the geographical area matching model according to the event classification result.

[0026] Exemplarily, the current environmental data stream can be continuously collected by Internet of Things devices deployed in the target area, including real-time temperature, humidity, wind speed and wind direction data transmitted by weather stations, river water level changes reported by hydrological monitoring stations, and crustal deformation monitoring data released by seismic networks. The event evolution prediction model performs sliding window sampling on the data stream, inputs the multi-dimensional time series data within the window into the trained neural network, and the output layer calculates the probability distribution belonging to each emergency event type through the Softmax function, and selects the event type with a probability exceeding the threshold as the classification result. For example, when the model detects continuous heavy rainfall, the river water level exceeding the warning line and the soil water saturation being too high, the classification result is marked as "Level II flood disaster". The geographical area matching model retrieves the visualization template with the highest spatial matching degree from the pre-generated layer template library according to the event type code in the classification result. Specifically, the template corresponding to the flood disaster needs to highlight low-lying areas, the distribution of drainage pipe networks and the locations of sandbag stacks; for chemical leakage events, a special template marking the locations of dangerous goods warehouses, wind rose diagrams and the coordinates of decontamination stations needs to be called. This process realizes fast retrieval by establishing a mapping relationship table between event types and layer elements, ensuring response timeliness.

[0027] Step S500: Integrate the diffusion path, intensity level in the event classification result and the called visualization layer template to generate a dynamic visualization instruction set, and render and output an emergency event evolution situation map through the interaction interface.

[0028] Exemplarily, the dynamic visualization instruction set can be composed of spatial overlay instructions, symbol rendering instructions, animation generation instructions, and so on. The spatial overlay instructions perform coordinate system unification alignment on the geometric polygon data of the diffusion path and the geographical features in the visualization layer template. For example, the predicted area of fire spread is overlaid on the forest vegetation distribution map; the symbol rendering instructions assign color gradient schemes and icon styles to different regions according to the intensity level. For example, dark red is used to represent the influence range of a Category 5 hurricane, and a rotating wind field symbol is added at the map marker; the animation generation instructions render the spatial morphological changes in the event evolution process frame by frame according to the time window division. The interactive interface uses WebGL technology to implement three-dimensional scene rendering. Users can rotate the viewing angle, switch the layer transparency, or click on the hotspot area to view detailed attribute data through gesture operations. The finally output emergency event evolution situation map can dynamically display the changes in the disaster influence range at multiple future time points, the optimized resource scheduling route plan, and the personnel evacuation path plan. For example, it shows the epidemic spread risk areas within 24 hours in the form of a heat map, and at the same time marks the predicted values of the bed occupancy rate of the mobile cabin hospitals and the queuing duration at the nucleic acid testing points. This visualization result provides intuitive spatial decision support for command and decision-making, and significantly improves the cross-departmental collaborative disposal efficiency.

[0029] As an implementation manner, step S100 of preprocessing the historical monitoring data set to generate a standardized event feature set may specifically include: Step S110: Interpolate and fill the missing data in the historical monitoring data set, and filter out the noise data according to a preset anomaly threshold to generate a cleaned monitoring data subset.

[0030] Exemplarily, missing data interpolation and filling refers to the operation of numerically estimating and complementing the missing parameter values in the historical monitoring dataset due to sensor failures or transmission interruptions. The specific implementation methods include linear interpolation based on time proximity, Kriging interpolation based on spatial similarity, and multiple regression filling based on feature correlation. For example, in the flood event monitoring scenario, if a hydrological station lacks rainfall data for 3 consecutive hours due to equipment failure, the missing data can be interpolated and complemented based on the linear trend of the data from adjacent stations during the same period and the data of the station before and after that period. The preset anomaly threshold is the boundary of the reasonable range of parameters set according to historical statistical distributions or domain expert experience, and is used to identify and filter out noise data that exceeds the normal fluctuation range; the noise data filtering uses the moving window standard deviation method, that is, calculating the mean and standard deviation of the parameters within a fixed time window, and determining the data points that deviate from the mean by more than 3 times the standard deviation as outliers and removing them. For example, an abnormal record of the surface displacement suddenly increasing to 500 mm / s in earthquake monitoring data, if it exceeds the historical maximum displacement threshold of 400 mm / s in this area, is marked as noise data and removed. The cleaned monitoring data subset eliminates data inconsistencies and interference factors through the above operations, forming a standardized data set that can be used for feature extraction. For example, in the case of forest fires, the cleaned data subset contains continuous and complete sequences of temperature, humidity, wind speed, and fire point coordinates, and there is no abnormal mutation noise.

[0031] Step S120: Extract target feature parameters strongly related to the type of emergency event from the cleaned monitoring data subset; the target feature parameters include the meteorological change rate and surface displacement in dynamic environmental parameters, as well as the altitude and geological structure stability index in static geographical parameters.

[0032] Exemplarily, the target characteristic parameters strongly related to the emergency event type refer to physical quantities or derived indicators that can significantly distinguish different event categories and play a decisive role in the event evolution process. The meteorological change rate is defined as the ratio of the change amplitude of meteorological parameters (such as wind speed, rainfall intensity, and pressure gradient) within a specific time period to the time interval. For example, in a typhoon event, the central wind speed increases from 20 m / s to 45 m / s within 24 hours, and its change rate is (45 - 20) / 24 ≈ 1.04 m / s²; the surface displacement amount is obtained through satellite synthetic aperture radar interferometry and represents the absolute value of the surface displacement in the horizontal or vertical direction. For example, in an earthquake event, the vertical surface displacement amount in the epicenter area reaches 2.3 m. The altitude in the static geographical parameters is extracted through a digital elevation model, reflecting the blocking effect of terrain undulation on event diffusion. For example, the area above 800 m in mountain elevation forms a natural barrier to flood spread; the geological structure stability index is calculated by weighting the shear strength of rock layers, the fault activity frequency, and the soil permeability coefficient, and is used to evaluate the risk level of landslide disasters. For example, a red warning is triggered when the stability index of a certain slope area is lower than 0.5. The extraction of target characteristic parameters needs to combine the physical mechanism of the event and the statistical characteristics of the data. For example, in debris flow warning, the change rate of hourly rainfall, slope inclination angle, and thickness of loose deposits are selected as the core characteristic parameters.

[0033] Step S130: According to the physical constraint conditions corresponding to each emergency event type, normalize the target characteristic parameters to generate the standardized event feature set; wherein, the normalization process includes mapping parameters with different dimensions to a unified numerical interval and matching the corresponding non-linear scaling function according to the event type.

[0034] Normalization aims to eliminate the scale differences of feature parameters with different dimensions, making them comparable and meeting the model input requirements. For unified numerical interval mapping, methods such as min-max normalization can be used to linearly transform the original parameters into the range of [0,1]. For example, the altitude ranging from 0 - 5000 meters is mapped to 0 - 1, and the ground displacement ranging from 0 - 10 meters is mapped to 0 - 1. For the physical constraint conditions of different emergency event types, non-linear scaling functions are used to enhance the resolution ability of key features. For example, for earthquake events, the logarithmic function is used to process the ground displacement to compress the influence of extreme values; for flood disasters, the Sigmoid function is used to perform non-linear transformation on the rainfall intensity change rate to highlight the threshold effect; for fire events, the exponential function is used to scale the wind speed parameter to amplify its sensitivity to the spread of the fire. For example, when the original value of the geological structure stability index of an earthquake event is 0.3, it is mapped to the standardized interval 0.75 after logarithmic transformation, and the change rate of hourly rainfall of 50 mm / h in a flood event is mapped to 0.88 after Sigmoid transformation. The standardized event feature set forms a multi-dimensional feature matrix in a unified format through the above processing. Each row corresponds to an event sample, and each column represents a standardized feature parameter. For example, a row vector in the matrix can be expressed as [0.72, 0.65, 0.91, 0.43], corresponding to the normalized meteorological change rate, ground displacement, altitude, and geological structure stability index respectively.

[0035] As an implementation manner, in step S120, extracting target feature parameters strongly related to the emergency event type includes the following steps: Step S121: According to the event type labels marked in the historical emergency event case library, calculate the information gain of each feature parameter in the subset of the cleaned monitoring data.

[0036] The information gain is used to quantify the contribution degree of a single feature parameter to the classification of the emergency event type. Its calculation can be based on the information entropy theory by comparing the reduction in classification uncertainty with the presence or absence of the feature. The historical emergency event case library contains complete monitoring data records with marked event types, such as 1000 earthquake event data, 800 flood event data, and 500 fire event data. The information entropy calculation method is as follows: Let the entropy of the event type set C be H(C) = -Σp(c)log 2 p(c), where p(c) is the occurrence probability of event type c in the case library; for the feature parameter F, its conditional entropy H(C|F) = Σp(f)Σp(c|f)log 2p(c|f), the information gain IG(F) = H(C) - H(C|F). For example, when calculating the information gain of the meteorological change rate, the case base is first discretized into several intervals according to this parameter, and the distribution frequencies of different event types in each interval are counted, and then the conditional entropy and information gain are calculated. If the proportion of earthquake events in the high meteorological change rate interval is significantly higher than other types, the information gain of this feature is higher, indicating that it plays an important role in distinguishing earthquakes from other events.

[0037] Step S122: Screen out the meteorological change rate and surface displacement amount with information gain greater than the first threshold from the dynamic environment parameters as the first candidate feature set.

[0038] Exemplarily, the first threshold is a screening criterion set according to the model performance requirements and feature dimension limitations. For example, the optimal value can be determined through cross-validation. For example, set the first threshold to 0.15. When the information gain of the meteorological change rate is 0.22 and the surface displacement amount is 0.18, both exceed the threshold and are thus selected into the first candidate feature set; while the information gain of the temperature parameter is only 0.08 and is excluded. The first candidate feature set focuses on the time-series change indicators strongly related to the event type in the dynamic environment parameters. For example, in the landslide warning scenario, the hourly rainfall change rate (IG = 0.21) and the slope inclination angle change amount (IG = 0.17) are selected as the core dynamic features.

[0039] Step S123: Screen out the geological structure stability index and altitude with information gain greater than the second threshold from the static geographical parameters as the second candidate feature set.

[0040] Exemplarily, the second threshold is set lower than the first threshold to retain more static features, for example, set to 0.10. The calculation of the information gain of the geological structure stability index needs to consider its spatial correlation with different event types. For example, in earthquake cases, this index shows a strong negative correlation with the magnitude (IG = 0.25); the information gain of altitude in flood events is 0.12, reflecting that low-altitude areas are more vulnerable to disasters. The second candidate feature set screens out the key spatial attributes in the static parameters. For example, the geological structure stability index of a certain area is 0.35 (IG = 0.18) and the altitude is 200 meters (IG = 0.13), both of which exceed the second threshold and are thus included in the candidate set.

[0041] Step S124: Perform a multicollinearity test on the first candidate feature set and the second candidate feature set, and delete the redundant features linearly correlated with other features to obtain an optimized target feature parameter set.

[0042] Exemplarily, the multicollinearity detection can evaluate the linear correlation between features by calculating the variance inflation factor (VIF). A VIF value greater than 10 indicates severe collinearity and redundant features need to be removed. For example, the VIF of the ground displacement amount and the seismic wave propagation speed in the first candidate feature set is 12.5, indicating a high linear correlation between the two. Therefore, the seismic wave propagation speed is removed; the VIF of the altitude and the slope angle in the second candidate feature set is 8.7, which is below the threshold, so it is retained. The optimized target feature parameter set reduces the risk of model overfitting by removing redundant features. For example, the meteorological change rate, the geological structure stability index, and the altitude are retained, and the features collinear with the ground displacement amount are removed.

[0043] Step S125: Match the optimized target feature parameter set with a preset emergency event type - feature mapping table, and dynamically adjust the feature weights according to the matching results to generate the final target feature parameters.

[0044] Exemplarily, the emergency event type - feature mapping table can be predefined by domain experts. For example, the key features and their importance weights for different event types are specified. For example, in the mapping table of earthquake events, the weight of the geological structure stability index is 0.6, and the weight of the ground displacement amount is 0.4; in flood events, the weight of the meteorological change rate is 0.7, and the weight of the altitude is 0.3. During the matching process, the weight coefficients are dynamically adjusted according to the consistency degree between the parameters in the optimized feature set and the mapping table entries. For example, if the optimized features in a certain earthquake case include the geological structure stability index (mapping table weight 0.6) and the meteorological change rate (not in the mapping table), the weight of the latter is automatically reduced to 0.1, and the former remains 0.6. Finally, a weighted target feature parameter vector is generated for model training and prediction.

[0045] As an implementation manner, step 200, training a time - series prediction model according to the dynamic environment parameters in the standardized event feature set to obtain an event evolution prediction model, includes: Step 210: Divide the dynamic environment parameters into a training sequence and a validation sequence in chronological order, where the training sequence is used for updating model parameters, and the validation sequence is used for evaluating the model prediction accuracy.

[0046] Exemplarily, the chronological division of dynamic environmental parameters can follow the principle of forward and backward dependence of time series data to ensure that the training sequence and the validation sequence do not overlap on the time axis and maintain the continuity of event evolution. The training sequence usually occupies 80% of the total time span and is used for the model to learn the evolution law through historical data; the validation sequence occupies the remaining 20% and is used to verify the generalization ability of the model on unknown data. For example, in the typhoon event prediction scenario, wind speed, air pressure, and sea surface temperature data for 30 consecutive days are selected. The data for the first 24 days are used as the training sequence, and the data for the last 6 days are used as the validation sequence. Random sampling should be avoided during the division to prevent future information leakage. The specific implementation method is to strictly segment according to the time stamp. For example, the dynamic environmental parameters from August 1st to August 24th, 2023 are used as the training sequence, and the data from August 25th to August 30th are used as the validation sequence. Each data point in the training sequence contains multi-dimensional dynamic parameters. For example, the instantaneous wind speed at a certain moment is 28 m / s, the pressure gradient is 5 hPa / km, and the sea surface temperature is 29 °C; the validation sequence contains the true observed values of the same parameters at subsequent time points, which are used to calculate the error index by comparing with the model prediction results.

[0047] Step 220: Construct a hybrid model architecture that includes a long short-term memory network and a convolutional neural network, and perform segmented sampling on the training sequence through a sliding window mechanism to generate multiple time segment samples.

[0048] The long short-term memory network (LSTM) is used to capture the long-term temporal dependence relationship of dynamic environmental parameters. Its hidden layer units regulate the information flow through an input gate, a forget gate, and an output gate. For example, when processing a 72-hour wind speed sequence, the LSTM can identify the correlation pattern between the periodic change of the wind speed and the typhoon movement path. The convolutional neural network (CNN) uses a one-dimensional convolutional kernel to extract local fluctuation features. For example, the sudden event of a sharp drop in air pressure is captured through the convolutional operation of a 3-hour time window. The sliding window mechanism intercepts the training sequence at a fixed step size. For example, the window length is set to 24 hours and the sliding step size is set to 1 hour, and the training sequence is divided into multiple consecutive or partially overlapping time segment samples. Each time segment sample contains the dynamic parameter matrix of all time points within the window. For example, a 24-hour window sample contains a wind speed sequence [20, 22, 25, …, 35] m / s, an air pressure sequence [1005, 1002, 998, …, 985] hPa, and a sea surface temperature sequence [28, 28.5, 29, …, 30] °C, forming an input matrix with a dimension of 24×3 for the hybrid model architecture to perform feature learning.

[0049] Step 230: Add event type labels and intensity level labels to each time segment sample, and input it into the hybrid model architecture for multi-task joint training to obtain an initial prediction model.

[0050] Exemplarily, the event type labels are matched according to the classification annotations in the historical emergency event case library. For example, the time segment sample is labeled as "typhoon", "flood" or "earthquake" type; the intensity level labels adopt a discretized grading standard. For example, the typhoon intensity is divided into levels 1-5, corresponding to the central wind speed ranges of 17.2-24.4 m / s, 24.5-32.6 m / s, 32.7-41.4 m / s, 41.5-50.9 m / s, and greater than 51.0 m / s respectively. Multi-task joint training means that the model synchronously learns the event type classification and intensity level regression tasks. The implementation method is to parallelly set a fully connected classification layer and a regression layer at the end of the hybrid model architecture. For example, input a 24-hour time segment sample labeled as "typhoon type - intensity level 3". After the LSTM-CNN hybrid model extracts spatio-temporal features, the classification layer outputs the probability distribution of typhoon, flood, and earthquake [0.92, 0.05, 0.03], and the regression layer outputs the predicted intensity level value of 3.2. Calculate the cross-entropy loss and mean squared error loss by comparing with the true label to drive the update of model parameters.

[0051] Step 240: Calculate the mean squared error and classification accuracy of the initial prediction model on each time segment through the validation sequence, and use the adaptive learning rate algorithm to optimize the model parameters until convergence.

[0052] The mean squared error (MSE) is used to evaluate the accuracy of the intensity level regression task. The calculation method is the mean of the squared differences between the predicted intensity level and the true level. For example, if the true intensity level of a certain time segment is 4 and the model prediction value is 3.8, then the error of a single sample is (4 - 3.8)² = 0.04; the classification accuracy is obtained by counting the proportion of samples with correct event type predictions. For example, if 92 out of 100 validation samples are correctly classified, the accuracy is 92%. The adaptive learning rate algorithm such as the Adam optimizer dynamically adjusts the learning rate according to the first and second moment estimates of the gradient. For example, the initial learning rate is set to 0.001. When the validation loss does not decrease for 5 consecutive epochs, the learning rate decays to 0.0001. The convergence condition is set that the fluctuation range of the validation loss is less than 1e-5 and lasts for 3 epochs. For example, the MSE of the typhoon intensity regression task gradually decreases from 0.25 to 0.18 and stabilizes near this value. At the same time, the classification accuracy increases from 85% to 92% and no longer changes significantly, determining that the model training is completed.

[0053] Step 250: Conduct a matching test between the optimized model and the latency tolerance of the real-time data stream. Adjust the number of model layers and neurons according to the test results to generate an event evolution prediction model that meets the real-time requirements.

[0054] The latency tolerance of the real-time data stream refers to the maximum data processing and prediction time allowed by the system. For example, it is determined by the timeliness requirements of emergency response. For example, flood prediction needs to complete the entire process from data input to result output within 10 seconds. The matching test measures the model inference time and resource occupancy rate by simulating the injection of real-time data streams. For example, the test shows that the optimized model takes 12 seconds to process a 24-hour time segment, exceeding the tolerance threshold. Adjustment measures include reducing the number of LSTM layers from 3 to 2 and reducing the number of CNN convolutional kernels from 64 to 32, which shortens the inference time to 8 seconds. At the same time, verify the change in model accuracy. If the mean squared error (MSE) of intensity regression slightly increases from 0.18 to 0.20 but is still lower than the acceptable threshold of 0.25, the adjustment is considered effective. The finally generated event evolution prediction model achieves a balance between computing resources and prediction accuracy. For example, when deployed on an edge server, it can process 5 time segment samples per second, meeting the real-time warning requirements 30 minutes before the flood peak arrives.

[0055] As an implementation, the training process of the hybrid model architecture further includes the following steps: Step S2301: Input the global temporal features output by the long short-term memory network into the attention mechanism layer to generate the event evolution contribution weights corresponding to each time step, and perform dynamic weighted summation on the global temporal features according to the contribution weights to obtain a time attention feature vector.

[0056] The attention mechanism layer calculates the importance scores of the features at each time step for the current prediction target through learnable parameters. For example, when processing a 72-hour typhoon wind speed sequence, the model may assign higher weights to the wind speed mutations in the 24 hours before landing. In a specific implementation, the output dimension of the LSTM is, for example, a global temporal feature matrix of 24×128. The attention layer generates a 24-dimensional weight vector [0.02, 0.03, …, 0.15], indicating the contribution degree of each hour's feature to the prediction of the typhoon path. After dynamic weighted summation, a 128-dimensional time attention feature vector is obtained, and its values centrally reflect the influence of key periods. For example, the weight of 0.15 corresponds to the features in the typhoon eyewall replacement stage. This vector can highlight the key turning points in the evolution process.

[0057] Step S2302: Align and splice the local fluctuation patterns extracted by the convolutional neural network with the time attention feature vector to form a spliced feature matrix containing temporal global dependencies and spatial local details.

[0058] Exemplarily, the dimension of the local fluctuation pattern extracted by the convolutional neural network is set to 24×64, for example, which represents the local change characteristics of hourly data. For example, a sudden drop in air pressure event recognized by a 3-hour convolutional window; the time attention feature vector is replicated and extended into a 24×128 matrix to achieve dimension alignment. The concatenation operation is performed along the feature channel dimension, generating a concatenated feature matrix of 24×(128 + 64) = 24×192. This matrix simultaneously retains the long-term trends captured by the LSTM (such as the stability of the typhoon movement direction) and the short-term anomalies detected by the CNN (such as sudden storm surges). For example, the feature at the 12th hour in the matrix contains the joint representation of the wind speed trend in the previous 12 hours and the current 3-hour air pressure fluctuation, providing comprehensive spatio-temporal information input for the subsequent gated recurrent unit.

[0059] Step S2303: Input the concatenated feature matrix into the gated recurrent unit, control the feature transfer path through the reset gate and the update gate, filter out the noise data irrelevant to the diffusion path of the emergency event, and output the filtered high-dimensional spatio-temporal feature sequence.

[0060] The reset gate of the gated recurrent unit (GRU) determines the degree of forgetting of historical features, and the update gate controls the proportion of new features incorporated. For example, when it is detected that the rainfall data at a certain time step has an outlier due to sensor failure, the reset gate closes the transmission of historical information at this time step to prevent noise from contaminating subsequent predictions. After inputting the 24×192 concatenated feature matrix, the GRU processes it step by step in time and outputs a high-dimensional spatio-temporal feature sequence of 24×256, where the numerical value of the feature vector corresponding to the time step of the abnormal data is significantly reduced. For example, in flood prediction, the false water level rise signal caused by instrument error is suppressed, while the feature of increased runoff caused by real continuous rainfall is retained, ensuring the reliability of the output feature sequence.

[0061] Step S2304: Input the high-dimensional spatio-temporal feature sequence into the diffusion path prediction branch and the intensity level regression branch respectively, where the diffusion path prediction branch outputs the probability distribution map of each grid cell being covered by diffusion through the softmax function, and the intensity level regression branch outputs the intensity level value at each moment within a preset time window through the linear activation function.

[0062] Exemplarily, the diffusion path prediction branch can map the 256-dimensional feature to the geographical grid space. For example, the target area is divided into grid cells of 1km×1km, and each cell corresponds to an output node. The softmax function calculates the probability that each cell will be covered by the event in the next time step. For example, the probability that the typhoon center will be located at grid (15, 32) in the next 6 hours is 0.93. The intensity level regression branch converts the 256-dimensional feature into a scalar value through a fully connected layer. For example, it outputs that the typhoon center pressure will be 935 hPa in the next 6 hours, corresponding to intensity level 4. The parallel processing of the two branches enables the model to simultaneously predict the spatial diffusion range and the temporal intensity change of the event. For example, it outputs the probability map of the wildfire spreading direction and the predicted value of the fire line intensity at the same time.

[0063] Step S2305: Calculate the cross entropy between the probability distribution map and the grid cell coordinates of the true diffusion path in the validation sequence. At the same time, calculate the mean squared error between the intensity level value and the labeled level label in the validation sequence, and fuse the two loss values to generate a joint optimization objective.

[0064] The cross entropy loss quantifies the difference between the predicted probability distribution and the true diffusion path. For example, if the true path covers grid (15, 32) and the predicted probability of this grid by the model is 0.93, then the loss of a single sample is -ln(0.93)=0.072; the mean squared error loss calculates the squared difference between the intensity prediction value and the true level. For example, if the true level is 4 and the predicted value is 3.8, then the loss is (4 - 3.8)² = 0.04. The joint optimization objective integrates the two losses through weighted summation. For example, if the cross entropy weight is set to 0.6 and the mean squared error weight is set to 0.4, then the total loss is 0.6×0.072 + 0.4×0.04 = 0.059. This design forces the model to balance the spatial diffusion accuracy and the intensity prediction accuracy during the optimization process, and avoids overfitting of a single task.

[0065] Step S2306: Backpropagate the error signal according to the joint optimization objective, and synchronously adjust the weight parameters of the long short-term memory network, convolutional neural network, and gated recurrent unit until the classification accuracy of the diffusion path prediction branch and the error rate of the intensity level regression branch both reach the convergence threshold.

[0066] During the backpropagation process, the error signal is reversely transmitted along the diffusion path prediction branch and the intensity level regression branch to the GRU, CNN, and LSTM networks, and the gradients of the parameters of each layer are calculated through the chain rule. For example, the gradient of the cross-entropy loss with respect to the GRU parameters indicates how to adjust the gating mechanism to better capture the key features of the diffusion path; the gradient of the mean squared error loss with respect to the CNN convolution kernel guides how to optimize the local fluctuation detection ability. The parameter update adopts an adaptive learning rate algorithm. When the diffusion path prediction accuracy remains above 95% for 5 consecutive epochs and the intensity regression MSE is lower than 0.05, the model is determined to converge. Finally, the optimized model achieves a diffusion path prediction F1 score of 0.92 and an average absolute error of 0.15 for intensity level prediction on the test set, meeting the accuracy requirements for emergency response decision-making.

[0067] As an implementation manner, in step S300, based on the static geographic parameters in the standardized event feature set, a geographic area matching model is constructed, including: Step S310: Convert the terrain distribution data in the static geographic parameters into a rasterized elevation map, and mark the raster cells corresponding to the infrastructure coordinates in the elevation map.

[0068] Exemplarily, the terrain distribution data can be rasterized through a Digital Elevation Model (DEM), discretizing the continuous geographical space into a two-dimensional raster matrix with a fixed resolution. Each raster cell stores the elevation value corresponding to the geographical location. The raster resolution is set according to the emergency response accuracy requirements. For example, a raster size of 1 km × 1 km is adopted to generate a 300 × 200 raster matrix in the target area, where the elevation value of each cell is accurate to the order of 0.1 m. The infrastructure coordinates are obtained based on a Geographic Information System (GIS) database, including the longitude and latitude information of key facilities such as hospitals, fire stations, and bridges. Through a coordinate conversion algorithm, they are mapped to the raster cell indices. For example, a certain hospital is located at 120.35 degrees east longitude and 30.28 degrees north latitude. After coordinate conversion, it corresponds to the cell in the 152nd row and 78th column of the raster matrix. This cell is marked as "Medical Facility - Grade III Class A Hospital" and additional attributes such as the number of beds and ambulance capacity are attached. In the scenario of a forest fire, after the terrain distribution data is converted into a rasterized elevation map, the raster cells corresponding to the coordinates of fire lookout towers are marked red, forming a spatial association base of the terrain and infrastructure.

[0069] Step S320: According to the historical emergency resource dispatch records, associate the emergency resource deployment information in the raster cells to generate an initial geographic layer.

[0070] Exemplarily, historical emergency resource dispatch records are stored in a structured database, and the record content includes event type, resource type, dispatch time, quantity, and target grid unit code. Emergency resource deployment information is bound to the grid unit through spatial association operations, such as associating the inventory of flood control sandbags to the grid unit along the river bank, and associating the take-off and landing points of fire helicopters to the grid unit in the mountainous area. The initial geographic layer adopts a multi-layer superposition structure, with the bottom layer being a rasterized elevation map and the upper layer being a resource density heat map. For example, in the case of a typhoon disaster, the number of assault boats associated with the historical dispatch record of a grid unit in a coastal area is 50, and the inventory of life jackets is 2,000. When generating a resource distribution heat map, the color depth of the unit is proportional to the resource sufficiency rate. The resource sufficiency rate calculation formula is the ratio of the current inventory to the historical maximum demand. When the inventory of assault boats in a grid unit is 30 and the historical maximum demand is 50, the sufficiency rate is 60%, and the corresponding color code is orange.

[0071] Step S330: Perform a topological structure analysis on the initial geographic layer, identify key path nodes and potential risk areas, and add corresponding topological relationship labels.

[0072] For example, topological structure analysis can be implemented based on graph theory algorithms, where grid cells are abstracted as nodes, and the adjacency relationship between cells is abstracted as edges, and indicators such as node betweenness centrality and edge connectivity strength are calculated. Critical path nodes refer to grid cells that must be passed through for emergency evacuation or resource transportation, such as cells where cross-river bridges are located and tunnel entrance cells, whose betweenness centrality exceeds the threshold of 0.7; potential risk areas refer to clusters of grid cells with unstable geological structures or frequent historical disasters, such as high-risk areas for landslides with slopes greater than 25 degrees and areas with liquefied soil layers. Topological relationship labels include categories such as "core hub node", "secondary transit node", and "high-risk isolation zone". For example, the grid cell where a cross-sea bridge is located is marked as "core hub node-transportation network level 1", and the surrounding buffer area cells are marked as "secondary transit node-transportation network level 2". In earthquake emergency scenarios, potential landslide areas on both sides of the fault zone are identified through topological analysis, and the label "high-risk isolation zone-geological stability level D" is added to restrict rescue path planning from passing through this area.

[0073] Step S340: performing region segmentation on the initial geographic layer based on the topological relationship labels to obtain a plurality of sub-region templates, and configuring an independent data rendering channel for each sub-region template.

[0074] Exemplarily, region segmentation can be implemented using the Voronoi Diagram algorithm. Taking critical path nodes as seed points, Thiessen polygons are generated, and the grid cells covered by each polygon form a sub-region template. For example, a sub-region template generated with 5 core hospitals as seed points, each template contains 15 - 20 grid cells, covering a radius of 10 kilometers around the hospital. The data rendering channel assigns an independent color mapping table, transparency parameter, and layer overlay order to each sub-region. For example, the "high-risk isolation area" sub-region is filled with red semi-transparency, and the "core hub node" sub-region is highlighted with a golden border. In the flood disaster scenario, the sub-region template segmented along the river channel is configured with a blue gradient rendering channel to show the rising trend of the water level; the mountain area template is configured with a gray contour rendering channel to highlight the influence of the terrain on the flood flow direction.

[0075] Step S350: Dynamically bind each sub-region template to the intensity level in the event classification result to generate a visualization layer template that can automatically adjust the display ratio according to the intensity level.

[0076] Exemplarily, dynamic binding can be achieved by establishing an intensity level - display ratio mapping rule. For example, typhoon intensity levels 1 - 5 respectively correspond to sub-region display ratios of 30%, 50%, 70%, 90%, and 100%. When the event classification result is "typhoon - intensity level 4", only 90% of the core hub node sub-region template is displayed, and the edge low-risk area is hidden; if the intensity level rises to 5, all sub-region templates are fully displayed. The automatic adjustment of the visualization layer template is achieved through a spatial index acceleration algorithm, such as using an R-tree index to quickly retrieve the set of grid cells to be rendered. In the chemical leakage event, intensity level 3 corresponds to highlighting the sub-region within a 3-kilometer radius around the leakage source, and the emergency resource deployment information only renders the positions of the chemical defense forces within this range; when the predicted intensity rises to 4, the display radius expands to 5 kilometers, and a wind rose sub-region template is superimposed to indicate the direction of toxic gas diffusion.

[0077] As an implementation, in step S500, fuse the diffusion path, intensity level in the event classification result, and the called visualization layer template to generate a dynamic visualization instruction set, including: Step S510: According to the diffusion path in the event classification result, draw a thermal diffusion trajectory in the called visualization layer template, and represent the change in intensity level through a color gradient.

[0078] Exemplarily, the thermal diffusion trajectory can be generated based on the grid cell probability distribution output by the diffusion path prediction model. The inverse distance weighted interpolation algorithm is used to convert discrete probability values into continuous color patches. The color gradient scheme follows industry standards. For example, red represents intensity level 5 (highest risk), orange is level 4, yellow is level 3, green is level 2, and blue is level 1. In the forest fire scenario, the grid cells covered by the diffusion path have probability values decreasing from 0.9 to 0.1 outward from the ignition point, corresponding to a color change from dark red to light blue, forming a heat diffusion map of the fire spread. The trajectory drawing uses a Bezier curve to smooth the transition between cells. For example, the center points of adjacent cells are connected as a curved path, and the line width is proportional to the probability value. The line width of the highest probability path is set to 5 pixels, and that of the low probability path is 1 pixel.

[0079] Step S520: Identify the grid cells covered by the diffusion path, and calculate the resource gap ratio based on the emergency resource deployment information, and generate a resource warning mark.

[0080] Exemplarily, the calculation method of the resource gap ratio is, for example: (current resource inventory - predicted demand) / predicted demand × 100%, where the predicted demand is estimated based on the population density covered by the diffusion path and the historical per capita resource consumption. For example, a flood diffusion path covers 10 grid cells with a total population of 50,000 people. According to historical data, each person needs 0.5 bottles of drinking water, and the predicted demand is 25,000 bottles. If the current inventory is only 18,000 bottles, the gap ratio is (1.8 - 2.5) / 2.5 × 100% = -28%, and a warning mark of "28% shortage of drinking water" is generated. The resource warning mark is displayed in an icon overlay manner. For example, a flashing exclamation mark icon is added above the grid cells with a gap exceeding 20%. The icon colors are divided into three levels: red, orange, and yellow according to the severity of the gap. In the earthquake rescue scenario, a certain landslide area grid cell is predicted to require 50 hydraulic jacking devices, and 30 are actually deployed, with a gap of 40%. The warning mark is a red equipment icon and is marked with "40% gap".

[0081] Step S530: Encode the thermal diffusion trajectory, intensity level color gradient, and resource warning mark into vector graphic instructions.

[0082] Exemplarily, the vector graphic instructions are defined in the Scalable Vector Graphics (SVG) format. The thermal diffusion trajectory is encoded as <path>Elements, with attributes including d (path coordinates), stroke (color), and stroke-width (line width); the intensity level color gradient is encoded as <lineargradient>Element, defining start and end colors and gradient direction; Resource warning mark coded as <g>Combined elements, including icon reference and <text>Annotated text. For example, the typhoon diffusion path instruction is <path d="M152,78 L155,80..." stroke="#FF4500" stroke-width="3" / > , and the resource gap marker instruction is <g transform="translate(152,78)"> <img xlink:href="warning.png"> <text x="10" y="5">Notch 28%< / text> < / g> . During the encoding process, an instruction tree is dynamically generated through a DOM operation library to ensure rendering efficiency and editability.

[0083] As an implementation manner, in step S530, encoding the thermal diffusion trajectory, intensity level color gradient, and resource warning marker into vector graphic instructions includes: Step S531: Parse the continuous coordinate point sequence of the thermal diffusion trajectory, generate a set of control points for a piecewise Bézier curve with an alpha attenuation attribute according to the timestamp information of the diffusion path, and map the intensity level color gradient to a curve width gradient parameter to form a trace vector primitive that can be traced in the time dimension.

[0084] Exemplarily, the continuous coordinate point sequence of the thermal diffusion trajectory is composed of the longitude and latitude coordinates of the center points of raster cells output by an event evolution prediction model. For example, typhoon path prediction data contains a set of center point coordinates updated hourly [(120.35, 30.28), (120.41, 30.31), …, (120.78, 30.65)]. The timestamp information is bound to each coordinate point to record the prediction moment corresponding to this position. For example, there is one point per hour from 09:00 to 12:00 on August 25, 2023. The alpha attenuation attribute can be dynamically calculated based on the interval between the timestamp and the current display moment. For example, it is set that the transparency of the trajectory within 24 hours linearly decays from 1.0 to 0.2, and data beyond 24 hours is completely transparent. The set of control points for the piecewise Bézier curve can be generated, for example, through a cubic Bézier interpolation algorithm, converting adjacent coordinate points into curve segments containing two control points. For example, the control points for the curve between coordinate points P0(120.35, 30.28) and P1(120.41, 30.31) are C0(120.37, 30.29) and C1(120.39, 30.30), forming a smooth path. When the intensity level color gradient is mapped to a curve width gradient parameter, it is set that intensity level 1 corresponds to a line width of 2 pixels, level 5 corresponds to a line width of 10 pixels, and intermediate levels are interpolated linearly. The finally generated trace vector primitive contains a coordinate sequence, transparency parameters, and line width data, such as an SVG path instruction <path d="M120.3530.28 C120.37 30.29 120.39 30.30 120.41 30.31…" stroke-opacity="0.8" stroke-width="5" / > , realizing a visually expressive form that can be traced in the time dimension.

[0085] Step S532: Extract the gamut distribution range of the intensity level color gradient, discretize the continuous color scale into polygon filling area definition instructions based on the raster cell boundary coordinates, and add a gradient color band attribute label containing intensity numerical metadata to each polygon.

[0086] Exemplarily, the gamut distribution range of the intensity level color gradient can be from red (RGB 255, 0, 0) to blue (RGB 0, 0, 255), divided into 5 equally spaced color levels. The grid cell boundary coordinates are obtained through geographic coordinate transformation. For example, the four vertex coordinates of a 1km×1km grid are (120.30, 30.25), (120.30, 30.26), (120.31, 30.26), (120.31, 30.25). The discretization process maps the continuous color levels to the grid cells. For example, it is predicted that intensity level 3 corresponds to RGB(255, 165, 0), and this color is filled into all grid cells marked as level 3. The filling area definition instruction uses SVG's <polygon>Elements, such as <polygon points="120.30,30.25 120.30,30.26 120.31,30.26120.31,30.25" fill="#FFA500" / > The gradient color band attribute tags are embedded as custom data attributes. For example, adding data-intensity-level="3" data-max-wind="41.4m / s" allows the interactive interface to query the intensity details. In the flood prediction scenario, the raster cells with intensity level 2 are filled with light blue (RGB 173, 216, 230) and labeled with data-water-depth="1.2m", forming a complete instruction for visualizing and associating data.

[0087] Step S533: Decompose the icon elements of the resource warning mark into a set of basic geometric primitives, convert the warning level text into vector font outline data, and encode the material gap ratio value as the primitive scaling factor and rotation angle parameters.

[0088] Exemplarily, the resource warning mark icon can adopt a standardized design. For example, the "medical supplies shortage" icon consists of a cross (two rectangles orthogonally superimposed) and an exclamation triangle. The decomposition process converts the cross into four line primitives: <line x1="10" y1="5" x2="10" y2="15" / > and <line x1="5" y1="10" x2="15" y2="10" / > The exclamation triangle is converted to <path d="M10,0 L15,5 L5,5 Z" / > . The warning level text "Level 3" is converted into vector outline data through the FreeType library, generating <path d="M20,30 L25,30…" / > A series of line and curve instructions. The material gap ratio is encoded as the scaling factor. For example, a 30% gap corresponds to the icon being scaled to 130% of the original size. The rotation angle parameter is used to dynamically adjust the icon direction. For example, the wind direction indicator icon rotates 45 degrees with the predicted wind direction. In the earthquake rescue scenario, the exclamation icon of the hydraulic equipment gap mark is enlarged to 1.5 times and displayed at a 10-degree tilt angle to enhance the visual warning effect.

[0089] Step S534: Align the trajectory vector primitives, gradient color band attribute tags, and the set of geometric primitives in the spatial coordinate system, establish the mapping relationship between the time axis control and the animation attributes of the vector elements, and generate an uncompressed set of original vector instructions.

[0090] Exemplarily, the spatial coordinate system alignment can be achieved by uniformly adopting the WGS84 geographic coordinate system, and the coordinate parameters of all vector elements are converted into screen pixel coordinates. For example, the geographic coordinates (120.35, 30.28) are converted into canvas coordinates (532, 784) through the Mercator projection. The mapping relationship between the time axis control and the vector elements is implemented through a JavaScript object. For example, it is defined that the transparency of the trajectory segment corresponding to the timestamp 2023-08-25T09:00:00 is 0.8, and the transparency of the segment corresponding to 10:00:00 is 0.7. The set of original vector instructions contains unoptimized complete drawing commands. For example, it contains 5000 <polygon>Element, 200 pieces <path>Instructions and 150 <g>Combined elements, each carrying independent spatio-temporal attribute parameters. In chemical leakage visualization, each trajectory segment of the gas diffusion path is bound with a starting time and a concentration value, and when the time axis slides, it triggers the transparency and color changes of the corresponding segment.

[0091] Step S535: Optimize the topological structure of the original vector instruction set, merge polygon filling instructions with the same attributes in adjacent raster cells, compress the coordinate sequences of repeatedly occurring geometric primitives, and generate a refined composite vector instruction set.

[0092] Exemplarily, the topological structure optimization can use the Douglas-Peucker algorithm to simplify the polygon boundary coordinate points. For example, a complex polygon with 100 vertices can be simplified to 20 key points, keeping the shape error less than 1 pixel. Polygons with the same attributes are merged into a single drawing instruction. For example, 10 adjacent raster cells with intensity level 3 are merged into <path d="M120.30 30.25 H120.40 V30.30 H120.30 Z" / > , reducing the number of drawing calls. The coordinate sequence compression of geometric primitives uses differential encoding. For example, the continuous coordinate points (120.30, 30.25), (120.30, 30.26), (120.31, 30.26) are encoded in the "+0.01 latitude +0.0 longitude" increment format. After optimization, the capacity of the composite vector instruction set is reduced by 60%. For example, it is reduced from the original 50MB to 20MB, improving the transmission and rendering efficiency. In the forest fire scenario, the jagged polygons at the fire edge are converted into smooth curves after optimization, reducing the computational load of the GPU vertex shader.

[0093] Step S536: Embed interaction response tags associated with emergency resource deployment information in the composite vector instruction set, and attach a material type code and an inventory query interface to each resource warning mark.

[0094] Exemplarily, the interaction response tags can be embedded in the form of HTML5 custom data attributes. For example, add data-resource-type="medical" data-inventory="1200" to the medical supply icon element. The inventory query interface is implemented through an AJAX request. When the user clicks on the icon, it triggers fetch("api / inventory?type=medical&grid=152-78") to obtain real-time data. The material type code uses the ISO standard. For example, "MED1" represents a first aid kit, and "WAT2" represents bottled drinking water. In the typhoon emergency scenario, the inflatable boat icon is attached with data-resource-type="boat" data-capacity="20 people", and when clicked, it displays the positions and capacity lists of all available boats within 5 kilometers around.

[0095] Step S537: Sort according to the rendering priority of the visualization layer template, hierarchically index and mark the composite vector instruction set, and generate a final vector graphic instruction package containing the layer stacking order and display / hide control parameters.

[0096] Exemplarily, the rendering priority sorting can be based on the visual importance rule. Set the heat diffusion trajectory layer as the highest level (z-index = 1000), the resource warning mark next (z-index = 900), and the intensity level color band as the bottom layer (z-index = 800). The display / hide control parameters are implemented through CSS class names. For example, add class="geology-layer visibility-toggle" to the earthquake fault zone layer, and the user can control the display status through the interface switch. The hierarchical index marking adopts a quadtree spatial index structure. For example, divide the screen into four quadrants, and each quadrant stores the vector instruction pointers of the corresponding area to accelerate data retrieval during local rendering. The final vector graphic instruction package is encapsulated in JSON format, including layer definitions, spatial indexes, and instruction data. For example, {"layers": [{"name": "heat-path", "zIndex": 1000, "commands": [...]}]}, for the rendering engine to parse and execute.

[0097] Step S540: Dynamically adjust the rendering precision of the vector graphic instructions according to the resolution of the interaction interface, and add a timeline control to achieve historical playback and prediction deduction of the evolution situation map.

[0098] Exemplarily, the rendering precision adjustment can be achieved through the Level of Detail (LOD) algorithm. When the interaction interface is zoomed to the full-area view, simplified vector instructions are used (such as reducing the number of path nodes and merging adjacent color areas); when zoomed to the grid cell level, high-precision instructions are loaded (such as displaying complete path nodes and resource annotation details). The timeline control integrates HTML5 <input type="range"> elements, bind timestamp metadata, and the slider step size is aligned with the prediction time window. For example, in the 72-hour typhoon prediction, when the timeline moves 1 hour, it triggers the vector instruction update function to redraw the heat diffusion trajectory and resource status at the corresponding moment. The historical playback function is implemented by caching the instruction sets at past time points. When the user drags the timeline to a historical position, the corresponding instructions are extracted from the cache queue for re-rendering.

[0099] Step S550: Convert the vector graphic instructions into pixel data streams through a parallel rendering engine and output them to the interaction interface for real-time refresh.

[0100] Exemplarily, the parallel rendering engine can be implemented based on WebGL technology, parsing SVG instructions into WebGL Shading Language (GLSL), and performing parallel calculations on vertex coordinates and fragment colors through a Graphics Processing Unit (GPU). For example, the vector path of the heat diffusion trajectory is converted into a triangular mesh vertex buffer, and the color gradient is generated by fragment shader interpolation based on uv coordinates. The pixel data stream is output through a double-buffering mechanism, and the front and back buffers are immediately swapped after the current frame rendering is completed to ensure a refresh rate of no less than 60fps. In the mobile interaction interface, the engine automatically switches to the Canvas 2D rendering mode, pre-generating a static frame sequence through an off-screen canvas, and loading frame data as needed when sliding the timeline to balance performance and visual effects. The real-time refresh function receives the latest sensor data every 5 seconds and updates the vector instructions during the chemical leakage deduction, allowing users to observe the dynamic change process of the poison gas diffusion boundary.

[0101] As an implementation manner, step S550 of converting the vector graphic instruction into a pixel data stream through the parallel rendering engine and outputting it to the interaction interface for real-time refresh may include: Step S551: Splitting the vector graphic instruction into a heat diffusion trajectory vector sub-instruction set, an intensity level color gradient vector sub-instruction set, and a resource warning mark vector sub-instruction set according to the layer type.

[0102] Exemplarily, the layer splitting is based on the hierarchical definition in the vector instruction package. For example, all <path>The element is classified into the subset of the thermal diffusion trajectory with z-index = 800 <polygon>The elements are classified into the color gradient subset. The resource warning marker subset filters those containing the data-resource-type attribute <g>Elements. In GPU video memory allocation, the trajectory subset is stored in Buffer Object 0, the color gradient subset is stored in Buffer Object 1, and the marker subset is stored in Buffer Object 2 to achieve data isolation and parallel access. For example, in a typhoon visualization scenario, the trajectory subset contains 1,200 Bezier curve instructions, the color subset contains 5,000 polygons, and the marker subset contains 200 icon combinations.

[0103] Step S552: According to the spatial resolution of the current window of the interaction interface, allocate an independent graphics rendering pipeline for each vector sub-instruction set, and map the rendering tasks of each pipeline to the parallel computing unit of the GPU.

[0104] Exemplarily, the graphics rendering pipeline is divided into a geometry processing pipeline and a fragment processing pipeline. The thermal diffusion trajectory subset is allocated to pipeline 0, and a tessellation shader is used to process the Bezier curve; the color gradient subset is allocated to pipeline 1, and a compute shader is used to batch process polygon filling; the resource marker subset is allocated to pipeline 2, and a geometry shader is called to generate icon instances. The GPU computing unit is divided according to the NVIDIA CUDA core architecture, and the 5,000 polygon filling tasks are decomposed into parallel computing tasks of 512 threads / block. At 4K resolution, pipeline 0 allocates 8 SM units to process trajectory anti-aliasing, and pipeline 1 occupies 12 SM units for color gamut conversion.

[0105] Step S553: Perform Bezier curve rasterization processing on the thermal diffusion trajectory vector sub-instruction set to generate a trajectory pixel lattice with transparency gradient attributes, and dynamically clip the trajectory segments that exceed the time window according to the current progress parameter of the time axis control.

[0106] Exemplarily, the Loop-Blinn algorithm can be used for Bezier curve rasterization to convert a cubic Bezier curve into a triangle strip. For example, the instruction <path d="M0,0 C10,20 30,40 50,50" / > is decomposed into 32 triangular patches. The transparency gradient is calculated by the fragment shader according to the curve parameter t value. For example, when t = 0, alpha = 1.0, and when t = 1, alpha = 0.2. The time axis clipping function is implemented by setting the stencil buffer. When the time window is set to 24 hours, only the trajectory segments with parameter t ∈ [0, 0.8] (corresponding to the first 24 hours of data) are retained, and the subsequent parts are excluded by the stencil test. In the flood evolution visualization, the 72-hour prediction trajectory is clipped to the data within the current display of 48 hours, and the excess part does not participate in rasterization.

[0107] Step S554: Perform color gamut space conversion on the intensity level color gradient vector sub-instruction set, map the HSL color model to the RGB pixel matrix for screen display, and add an anti-aliasing color transition band in combination with the grid cell boundary data.

[0108] Exemplarily, the HSL to RGB conversion can be implemented in the fragment shader. For example, the conversion of HSL(30°, 100%, 50%) to RGB(255, 128, 0). Anti-aliasing can be performed using multi-sample anti-aliasing (MSAA 4x), calculating the color coverage at 4 sub-sample points per pixel. The color transition band is generated by linear interpolation. For example, at the grid boundary between intensity level 3 (HSL(60°, 100%, 50%)) and level 4 (HSL(30°, 100%, 50%)), a gradient transition zone with a width of 2 pixels is generated. In seismic intensity visualization, a smooth transition is added at the boundary between degree IX area (HSL(0°, 100%, 50%)) and degree X area (HSL(30°, 100%, 50%)) to avoid jagged edges of color blocks.

[0109] Step S555: Decompose the resource warning marker vector instruction set into icon outline geometry data and text annotation data, and use a geometry shader to generate multi-resolution icon bitmaps respectively, and generate an adaptive-sized warning text texture through a font rendering engine.

[0110] Exemplarily, after the icon outline geometry data is input into the geometry shader, an LOD model can be dynamically generated according to the viewing distance. Specifically, for example, when the icon screen size is less than 16×16 pixels, a simplified octagon is generated; when it is greater than 64×64 pixels, a complete 128-sided polygon is generated. The font rendering engine uses the Signed Distance Field (SDF) technology to generate a 512×512 pixel distance field texture for the text "Gap 28%", and smooths the edges through the fragment shader during scaling. When displayed on mobile devices, the icon bitmaps are pre-generated at three resolutions of 256×256, 128×128, and 64×64, and are automatically switched according to the zoom level.

[0111] Step S556: Detect the output delay status of each graphics rendering pipeline. When the difference in the generation progress between the trajectory pixel dot matrix and the color transition band exceeds the threshold, start the pixel cache synchronization mechanism to align the spatial coordinate reference points of each layer.

[0112] Exemplarily, the delay status can be monitored through an OpenGL fence object (GL_FENCE). Assume that the set threshold ΔT = 16ms (corresponding to 60fps). When the trajectory rendering of pipeline 0 takes 18ms and the color band rendering of pipeline 1 takes 12ms, the synchronization mechanism is triggered to pause the output of pipeline 1 and wait for pipeline 0 to complete the trajectory rasterization of the current frame. The alignment of the spatial coordinate reference points is implemented using an atomic counter to ensure that the pixel data of all layers is synthesized based on the same world coordinate system origin (such as the screen center point (0, 0)). In the typhoon eyewall replacement animation, when the offset error between the trajectory layer and the color band layer exceeds 2 pixels, the coordinate system is automatically resynchronized.

[0113] Step S557: Perform layer fusion on the synchronized track pixel dot matrix, RGB pixel matrix, and icon bitmap, resolve the display priority conflicts in the overlapping areas of graphic elements according to the depth buffer data, and generate complete frame buffer data.

[0114] Exemplarily, alpha blending formula is adopted for layer fusion: C out =C src *α src +C dst *(1 - α src ). GL_DEPTH_TEST is enabled for depth buffer test, the depth value of the track layer is set to 0.5, the icon layer to 0.7, and the color band layer to 0.3, ensuring that the color band is displayed at the bottom and the icon covers the track. In the overlapping area of the medical resource icon and the flood color band, the depth test makes the icon always displayed in the front end. The frame buffer object (FBO) stores the fused RGBA pixel data, with a resolution of 3840×2160, 32-bit color depth, and occupying 31.6MB of video memory.

[0115] Step S558: Cut the frame buffer data into pixel data streams matching the vertical synchronization signal according to the refresh rate of the interaction interface, and send them to the video memory mapping area of the display device through multi-threaded transmission channels respectively.

[0116] Exemplarily, the interval of the vertical synchronization signal is 16.67ms (60Hz refresh rate), the frame buffer data is divided into 4 horizontal strips (540 rows per strip), and transmitted in parallel through the PCIe 4.0×16 channel. Video memory mapping adopts NVIDIA GPUDirect technology and directly writes to the Frame Buffer Object of the graphics card. In the 8K resolution scenario, 4-way SLI parallel transmission is enabled, each path processes 2160×3840 pixel data, and the total bandwidth requirement is 12.8GB / s.

[0117] Step S559: Real-time monitor the view transformation parameters triggered by user interaction events. When detecting zooming or panning operations, interrupt the current pixel data stream transmission and preferentially render a low-precision preview frame of the center area of the viewport.

[0118] Exemplarily, the view transformation parameters can be captured through the Render Transform matrix of WPF. For example, when the zoom factor is set to be greater than 1.5, the low-precision mode is triggered. It is assumed that the preview frame adopts the dichotomy downsampling method to compress the 4K frame buffer to a resolution of 1024×576, and the texture filtering mode is set to NEAREST. The center area is defined as the central 50% range of the view window, and computing resources are preferentially allocated to render the high-precision content of this area. When the user quickly zooms the map, the system preferentially displays the downsampled full-map overview and progressively loads the high-definition details after the operation stops.

[0119] Step S5510: When data transmission is delayed due to network bandwidth fluctuations, generate a motion blur compensation image based on historical frame buffer data to fill the display abnormal area until the new pixel data stream completely covers the current window.

[0120] Exemplarily, motion blur compensation can calculate the pixel displacement vector through Optical Flow of the previous two frames to generate a dynamic blur effect. For example, in real-time typhoon path tracking, when the new data is delayed by 200 ms, based on the previous frame position (120.50, 30.40) and speed of 25 km / h, it is predicted that the current position should be (120.53, 30.42), and a directional blur filter is added to this area. Abnormal area detection triggers compensated rendering for areas where the SSIM structural similarity of the frame sequence differs by more than 0.3. The compensated image is encoded and compressed using JPEG-XS, and the bit rate is reduced to 30% of the original data to prioritize visual continuity in key areas.

[0121] As an implementation, the method provided by the embodiments of the present invention may further include the following steps: Step S600: In response to a region selection instruction triggered by the user through the interaction interface, intercept a target sub-region from the visualization layer template according to the coordinate range included in the region selection instruction, and extract the terrain distribution, infrastructure coordinates, and emergency resource deployment information corresponding to the target sub-region.

[0122] Exemplarily, the region selection instruction captures the geographical coordinate range input by the user through the rectangular selection tool or free polygon drawing tool of the interaction interface. For example, the user selects a rectangular area of 120.30° - 120.45° east longitude and 30.25° - 30.35° north latitude on the map interface. The system extracts the corresponding rasterized elevation map subset from the visualization layer template according to this coordinate range. For example, a sub-matrix with row numbers 150 - 200 and column numbers 80 - 120 in the raster matrix is intercepted as the target sub-region. Terrain distribution data extraction includes elevation values, slope, and aspect parameters within this sub-region, such as a continuous change gradient of altitude from 50 meters to 300 meters. Infrastructure coordinate extraction covers the longitude and latitude positioning information of all hospitals, fire stations, and transportation hubs within the selected range. For example, a certain tertiary hospital is located at 120.38° east longitude and 30.30° north latitude, and its raster cell index is (175, 95). Emergency resource deployment information is associated with each infrastructure coordinate. For example, the inventory of medical supplies corresponding to this hospital is 1200 first aid kits and 800 bags of plasma. In the emergency scenario of a forest fire, target sub-region extraction includes the coordinates of fire breaks and the capacity data of fire fighting reservoirs within a 10-kilometer radius around the fire site, forming a basic spatial data set for emergency decision-making.

[0123] Step S700: Superimpose the terrain distribution of the target sub-region on the real-time meteorological parameters in the current environmental data stream to generate an enhanced terrain layer with wind speed vector arrows and precipitation intensity contour lines.

[0124] Exemplarily, the real-time meteorological parameters can be updated at a second-level frequency through Internet of Things sensors, including wind speed, wind direction, precipitation amount, and temperature data. The spatial superposition of the terrain distribution data and the meteorological parameters adopts a raster cell binding method. For example, each 1km×1km raster cell is associated with meteorological observation values of a current wind speed of 12m / s, a wind direction of 270°, and an hourly precipitation amount of 15mm. The wind speed vector arrows are generated through SVG's <marker>Element implementation: The length of the arrow is proportional to the wind speed, and the direction is aligned with the wind direction angle. For example, a wind speed of 12 m / s corresponds to an arrow length of 40 pixels, and a wind direction of 270° points due west. The precipitation intensity contour lines are generated based on the inverse distance weighted interpolation algorithm, which converts discrete precipitation observations into continuous contour lines. For example, three contour lines of 5 mm, 10 mm, and 15 mm are drawn in the target sub-region, and the line width increases with the intensity. The enhanced terrain layer integrates elevation color levels, vector arrows, and contour lines. For example, in a 3D rendering view, the area with an altitude of 300 meters is shown in dark brown, with northwest wind speed arrows and a thick red outline of the 15 mm contour line superimposed, forming a visual base for the collaborative expression of multi-dimensional data.

[0125] Step S800: Locate the positions of the buildings affected by the diffusion path in the enhanced terrain layer according to the infrastructure coordinates, and retrieve the material inventory in the associated emergency resource deployment information to generate a resource distribution coverage map representing the inventory adequacy with icon sizes.

[0126] Exemplarily, the buildings affected by the diffusion path are identified through a spatial intersection algorithm. For example, the grid cells covered by the thermal diffusion trajectory are judged for point-in-polygon inclusion with the hospital coordinates. If the diffusion probability of the grid cell where the hospital is located exceeds 50%, it is marked as an affected target. The material inventory adequacy is calculated as the ratio of the current inventory to the historical peak demand. For example, a fire station has 200 fire extinguishers in stock, and the historical peak demand is 300, so the adequacy is 66.7%. The icon size mapping rule sets that when the adequacy ≥ 80%, it is the original size, and for every 10% reduction, the size is reduced by 15%. For example, an adequacy of 66.7% corresponds to the icon being scaled to 77.5% of its original size. The resource distribution coverage map uses a hierarchical symbol system. For example, the hospital icon is represented by a cross, with a size gradient from 30×30 pixels to 15×15 pixels; the fire station is represented by a flame icon, and the size is dynamically adjusted according to the fire extinguisher inventory. In a chemical leakage incident, the icon size of the gas mask inventory in the affected chemical plant is reduced to 60%, and a red exclamation mark warning label is superimposed below the icon.

[0127] Step S900: Based on the spatial intersection area of the resource distribution coverage map and the thermal diffusion trajectory, calculate the redundant coordinate points of the emergency resources not covered by the diffusion path, and convert them into resource supply path planning points that can be dynamically scheduled.

[0128] Exemplarily, the spatial intersection region can be determined through the logical AND operation of grid cells. For example, when the grid cells where the icons are located in the resource distribution coverage map are not covered by any cells with a probability exceeding 30% in the thermal diffusion trajectory, they are determined as redundant resource points. After extracting the redundant coordinate points, they are converted into a set of longitude and latitude coordinates of path planning points. For example, a material warehouse is located at 120.40° east longitude and 30.32° north latitude, and the diffusion probability within 1 kilometer around it is lower than 10%, and it is marked as a schedulable starting point. The attributes of path planning points include resource type, inventory quantity, and maximum transportation capacity. For example, the warehouse point is appended with {"type":"medical","capacity":"3 trucks with a capacity of 5 tons each"}. In the flood scenario, the coordinates of the emergency material reserve warehouses that are not flooded are converted into the starting points of ship supply paths, and scheduling task nodes for transporting sandbags to the disaster-stricken areas are generated.

[0129] Step S1000: Topologically match the resource supply path planning points with the traffic network data in the terrain distribution of the target sub-region, eliminate the path nodes marked as impassable by the surface displacement data, and generate an optimized distribution route set with detour suggestion marks.

[0130] Exemplarily, the traffic network data includes, for example, road grade, width, and real-time traffic status. For example, highways, national roads, and county roads are respectively coded as L1, L2, and L3 levels. When performing topological matching, the Dijkstra algorithm can be used to calculate the shortest path from the planning point to the disaster-stricken area. For example, the optimal route from the warehouse to the hospital is to drive 8 kilometers along the L1 road and then turn into the L2 road for 3 kilometers. The surface displacement data is monitored by InSAR technology, and the road segments with a displacement exceeding 50 cm are marked as impassable nodes. The detour suggestions are generated based on the A* algorithm. For example, when the displacement of a certain bridge in the original path reaches 80 cm, the system recommends detouring to an alternative road 5 kilometers away, and adds a yellow dashed arrow and the text annotation "Suggested Detour" to the route map. The optimized distribution route set is output as a GPX format file, which includes a sequence of path points, detour tips, and estimated travel time for the navigation device to import and execute.

[0131] Step S1100: According to the length and surface displacement of each route in the optimized distribution route set, overlay and display a route safety level color bar on the enhanced terrain layer, and compare the travel speed of the path with the highest safety level in the distribution route set with the thermal diffusion trajectory to generate a resource scheduling timeliness warning signal.

[0132] Exemplarily, the route safety level can be calculated by a linear weighting formula. For example, safety score = 0.6×(1 - displacement / 100 cm) + 0.4×(1 - route length / 50 km), and the result is mapped to a three-color bar of red (0 - 0.3), yellow (0.3 - 0.6), and green (0.6 - 1). For example, for a route with a length of 20 km and a maximum displacement of 30 cm, the safety score is 0.71, which is displayed as a green bar. The travel speed comparison is based on the predicted speed of the diffusion model and the average vehicle speed. For example, the thermal diffusion trajectory advances 2 km per hour, while the optimal delivery route speed is 60 km / h, generating a time - efficiency warning signal "remaining safety time 3 hours". The warning signal is encoded as a JSON message body, containing the path ID, the remaining time threshold, and the recommended action. For example, {"route_id":"R002", "time_left":"180min", "action":"prioritized scheduling"}, and is distributed to the command terminal through a message queue.

[0133] Step S1200: Real - time associate the resource scheduling time - efficiency warning signal with the coordinates of new monitoring points in the real - time environmental data stream. When a new monitoring point falls into the warning signal coverage area, automatically trigger the recalculation of the optimized delivery route set, and lock the view zoom ratio of the current enhanced terrain layer to prevent path display overlap.

[0134] Exemplarily, the coordinates of new monitoring points can be uploaded in real - time through drone inspections or ground sensors. For example, a new smoke concentration monitoring point at 120.42° east longitude and 30.31° north latitude is added. The spatial association judgment uses a point - in - polygon algorithm. If the point is located within the buffer zone covered by any warning signal (such as a radius of 2 km), the path recalculation is triggered. The view zoom ratio is locked by saving the center coordinates and zoom level of the current viewport. For example, maintaining a scale of 1:50000 ensures that the path lines and icons do not shift or misalign during redrawing. In the scenario of typhoon path update, when a new monitoring point causes the original optimal path to enter the 10 - level wind circle, the system immediately replans a detour route and keeps the map view stable to avoid operation interference.

[0135] Step S1300: In response to the user's click operation on the resource distribution coverage map, dynamically filter all redundant coordinate points of emergency resources within the preset range of the click coordinates, and drive the interactive interface to highlight the icons of materials that can be prioritized for scheduling in a pulsed flashing mode.

[0136] Exemplarily, the predefined click coordinate range is defined as a circular area with a radius of 5 km centered on the click point. For example, when the user clicks at 120.38° east longitude and 30.29° north latitude, the coordinates of all warehouses with a sufficient inventory level ≥ 70% within this range are filtered. The pulse flashing mode is implemented through CSS animation, where the icon transparency periodically changes between 0.3 and 1.0 at a frequency of 2 Hz, and the border color is synchronously switched to gold. The highlighting strategy follows the principles of spatial proximity and inventory priority. For example, among 10 candidate warehouses, the icons of the 3 warehouses closest to the disaster area continuously flash, and the rest are adjusted to a semi-transparent state. In the dispatching of epidemic prevention materials, after clicking on the icon of the mobile cabin hospital, the icons of suppliers with sufficient protective clothing inventory within 5 km around enter the flashing state, prompting the commander to contact them first.

[0137] Step S1400: When it is detected that the intensity level color gradient of the heat diffusion trajectory undergoes a cross-level change, refresh the safety level color bars of all distribution routes intersecting with the changed area in the enhanced terrain layer, and adjust the priority sorting of the resource dispatching plan list according to the refreshed color bars.

[0138] Exemplarily, the cross-level change determination can be based on the intensity level difference between adjacent grid cells ≥ 2 levels. For example, a certain area suddenly changes from level 2 (yellow) to level 4 (orange). The intersecting distribution routes are identified through spatial overlay analysis. For example, a 3-km section of the original safe green route enters the intensity level 4 area, triggering the update of the color bar of this section to orange. The priority of the resource dispatching plan is dynamically sorted according to the path safety score. For example, the route with a score of 0.7 in the original plan list is downgraded to the third place, and the newly calculated route with a score of 0.85 rises to the first place. The update of the plan list is pushed to the mobile terminal in real time through WebSocket to ensure that on-site personnel obtain the latest operation guidelines.

[0139] Step S1500: When synchronously playing the resource dispatching time limit warning signal, the optimized distribution route set, and the evolution process of the heat diffusion trajectory in the interaction interface, dynamically insert tweening animations between frames according to the real-time movement speed of the graphic elements to maintain the visual continuity of the multi-layer collaborative evolution.

[0140] Exemplarily, inter-frame interpolation animation can use a linear interpolation algorithm, for example, the heat diffusion trajectory moves 5 pixels per frame, and 3 intermediate positions are inserted between two frames to make the trajectory advancement process smooth and without jumps. The location update frequency of the resource scheduling route icon is synchronized with the vehicle GPS signal. When the data transmission interval exceeds 500ms, the intermediate trajectory point is calculated based on the velocity vector. Visual continuity is guaranteed through a unified time base, and the animation progress of all layers is bound to the same timer, for example, a global redraw is triggered every 100ms. In the visualization of the flood peak evolution, the resource transportation vehicle icon moves dynamically along the path, and the proportional relationship between its displacement speed and the flood diffusion speed is adjusted through the animation frame rate to ensure that the commander can intuitively perceive the urgency of the time window.

[0141] As an implementation mode, the method further includes a model online updating process, which may specifically include the following steps: Step S1600: monitor the prediction error of the event evolution prediction model on the latest environmental data stream in real time, and trigger the model retraining process when the prediction error exceeds a preset tolerance.

[0142] Exemplarily, the prediction error of the event evolution prediction model is quantified by calculating the root mean square error (RMSE) between the model output value and the actual observation value, and the preset tolerance sets a dynamic threshold based on different emergency event types. For example, in the typhoon path prediction scenario, when the prediction deviation of the center position of three consecutive time segments exceeds 15 kilometers (RMSE ≥ 15km), the prediction error is determined to exceed the tolerance threshold, triggering the model retraining process. The real-time monitoring module is embedded in the data pipeline processing node. Every time a batch of the latest environmental data stream is received (such as wind speed and air pressure data updated every minute), the error calculation is immediately performed and compared with the threshold. In the prediction of forest fire spread, if the relative error between the actual advancement speed of the fire line and the model prediction value exceeds 20% for five consecutive times, the system automatically generates a model failure alarm log and starts the parameter update task queue.

[0143] Step S1700: extract continuous time segments from the current environment data stream as new training data, and perform distribution consistency detection with the historical monitoring data set.

[0144] Exemplarily, continuous time segment truncation can adopt a sliding window mechanism, and the window length is aligned with the model input dimension. For example, a typhoon prediction model requires 72 hours of continuous data as input. The newly added training data truncates the sea surface temperature and wind speed matrices updated every second within the most recent 72 hours and stores them as a tensor structure with a shape of (72×3600, 12). The distribution consistency detection is achieved by comparing the statistical characteristics of the newly added data and the historical data in key parameters. For example, the mean value of the air pressure in the historical dataset is calculated to be 1013 hPa and the standard deviation is 5 hPa. If the newly added data shows a distribution shift with a mean value of 1018 hPa and a standard deviation of 12 hPa, a data anomaly flag is triggered. In the flood prediction scenario, the proportion of samples in the newly added data with hourly rainfall exceeding 200% of the historical maximum record reaches 15%, indicating a significant change in the rainfall intensity distribution and subsequent processing is required.

[0145] As an implementation manner, in step S1700, the process of distribution consistency detection includes: Step S1710: Calculate the KL divergence between the newly added training data and the historical monitoring data in dynamic environmental parameters, and determine whether the data distribution has a significant shift.

[0146] The KL divergence (Kullback-Leibler Divergence) is used to measure the asymmetric difference between two probability distributions, and the dynamic environmental parameters select the feature items that have the greatest impact on model prediction. For example, in typhoon prediction, three parameters, namely the central air pressure, the maximum wind speed, and the moving direction angle, are selected, and the KL divergence values of their probability density functions are calculated respectively. The significance threshold is set as KL≥0.1. When the KL value of the air pressure distribution of the newly added data is 0.15, it is determined that a significant shift has occurred. Taking flood prediction as an example, the KL value of the hourly rainfall distribution of the newly added dataset is 0.25, which is significantly higher than the historical data distribution, triggering the data augmentation process.

[0147] Step S1720: If a significant shift occurs, start the data augmentation process and synthesize simulated data consistent with the historical distribution through a generative adversarial network.

[0148] Exemplarily, the generator of the generative adversarial network (Generative Adversarial Network, GAN) learns the distribution characteristics of historical data. For example, in the typhoon scenario, it generates simulated data that conforms to the historical air pressure mean of 1013 hPa and the Weibull distribution shape parameter k = 2 of the wind speed. The discriminator network distinguishes real and synthetic samples with a confidence level of 0.95, and stops training when the misjudgment rate of the generated data passing through the discriminator exceeds 30%. In the heavy rain prediction case, the GAN generates 1000 groups of hourly rainfall data, and the difference between its distribution mean and variance and the historical data is less than 5%, effectively compensating for the distribution deviation of the newly added data.

[0149] Step S1730: Perform physical constraint verification on the simulated data and eliminate abnormal samples that do not conform to the laws of the actual environment.

[0150] Exemplarily, the physical constraint verification can define a reasonable range of parameters based on domain knowledge. For example, the central pressure of a typhoon should not be lower than 870 hPa or higher than 1100 hPa, and the wind speed and pressure satisfy the gradient wind equation relationship. Use the symbolic regression algorithm to detect contradictory terms in the generated data. For example, if the wind speed of a certain simulated sample is 50 m / s corresponding to a pressure of 980 hPa, which violates the wind speed-pressure empirical formula Vmax = 14.5√(1010 - Pmin), it is marked as an abnormal sample. In the verification of flood data, a sample of hourly rainfall of 250 mm generated is detected that the terrain slope in the area is > 30 degrees, while actually the soil saturation runoff generation time under this slope is less than 1 hour and no effective runoff can be formed, so it is eliminated.

[0151] Step S1740: Mix the verified simulated data with the newly added training data to generate a balanced training dataset for model retraining.

[0152] Exemplarily, the mixing ratio can be dynamically adjusted according to the KL divergence value. When the KL of the newly added data is 0.15, a balanced dataset is constructed according to the ratio of historical data: newly added data: simulated data = 6:2:2. The dataset balancing process uses the SMOTE (Synthetic Minority Oversampling Technique) algorithm to oversample the minority class samples (such as super typhoon data). When training an earthquake prediction model, mix the simulated earthquake cases of magnitude 7 with the newly added actual earthquake cases of magnitude 5, so that the magnitude distribution changes from the original skewed distribution to a uniform distribution, improving the model's prediction ability for rare events.

[0153] Step S1750: After the model update is completed, evaluate the improvement in the performance of the updated model by comparing the differences in the prediction results of the new and old models on the validation set.

[0154] Exemplarily, the validation set is constructed using the time series cross-validation method and contains data that has not participated in training in the past 30 days. The performance evaluation metrics include the prediction accuracy improvement rate (such as the typhoon path error drops from 15 km to 12 km, an increase of 20%) and the classification F1 score (such as the F1 for heavy rain level prediction increases from 0.85 to 0.89). In the case of forest fires, the prediction accuracy of the updated model for the direction of fire spread increases from 78% to 85%, and the false alarm rate decreases by 3 percentage points. The system generates a performance report and stores it in the model version library.

[0155] Step S1800: Assign dynamic weights to the newly added training data that passes the detection, and use the incremental learning algorithm to update the parameters of the event evolution prediction model.

[0156] Exemplarily, the dynamic weight allocation is adjusted inversely based on data timeliness and prediction error. The weight of the data in the most recent 72 hours is set to 1.2, and the weight of historical data is reduced to 0.8. The incremental learning adopts the Online Sequential Extreme Learning Machine (OS-ELM) algorithm, which only updates the weight matrix of the network output layer and retains the feature extraction ability of the hidden layer. For example, the number of hidden layer nodes of the typhoon prediction model is fixed at 500. The output layer weights are adjusted with a learning rate of 0.01 for new data, and the parameter update on the GPU computing unit takes no more than 200 ms, meeting the real-time requirement.

[0157] Step S1900: Perform a compatibility test on the updated model with the geographical area matching model to ensure that the parameter mapping relationship of the visualization layer template is adjusted synchronously.

[0158] Exemplarily, the compatibility test focuses on verifying the spatial alignment accuracy between the model output and the layer features. For example, the coordinate system of the diffusion path raster cells output by the updated model needs to be consistent with the tile index of the geographical area matching model. When the resolution of the path probability matrix output by the model is increased from 1 km to 500 m, the geographical area matching model automatically adjusts the raster division rule and subdivides the original raster cells into 4 sub-units for parameter mapping. In the test phase, 100 groups of test cases are injected into the system, and it is required that the deviation between the path coverage raster and the layer display coordinates does not exceed 2 pixels, otherwise the coordinate conversion parameter calibration process is triggered.

[0159] Step S2000: Gradually deploy the updated model to the production environment through the gray release mechanism, and prompt the model version change information in the interactive interface.

[0160] Exemplarily, the gray release adopts the traffic splitting technology. For example, on the first day, 5% of the prediction requests are routed to the new model, and its prediction error and system load are continuously monitored. When the new model performs stably within 24 hours (such as the error rate is lower than 90% of the old model), the traffic switching ratio is gradually increased to 100%. The interactive interface adds a version identifier (such as "Typhoon Prediction Model V2.1.5") in the upper right status bar, and displays the update log through the Tooltip component, including the performance improvement data and the main parameter change items. Users can click on the version number to view the comparison chart of the prediction results of the new and old models. For example, the difference band diagram of the 72-hour path predictions of the V2.1.4 and V2.1.5 models for the same typhoon event is displayed side by side.

[0161] Please refer to Figure 3 , is a structural block diagram of the terminal device 120 of the present invention. The terminal device 120 includes a computing unit 1001, which can execute various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 1002 or the computer program loaded from the storage unit 1008 into the random access memory (RAM) 1003. In the RAM 1003, various programs and data required for the operation of the terminal device 120 can also be stored. The computing unit 1001, the ROM 1002, and the RAM 1003 are connected to each other through a bus 1004. The input / output (I / O) interface 1005 is also connected to the bus 1004.

[0162] Multiple components in the terminal device 120 are connected to the I / O interface 1005, including: an input unit 1006, an output unit 1007, a storage unit 1008, and a communication unit 1009. The input unit 1006 can be any type of device that can input information into the terminal device 120. The input unit 1006 can receive input digital or character information, and generate key signal inputs related to user settings and / or function controls of the server, and can include but are not limited to a mouse, a keyboard, a touch screen, a trackpad, a trackball, a joystick, a microphone, and / or a remote control. The output unit 1007 can be any type of device that can present information, and can include but are not limited to a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 1008 can include but are not limited to a magnetic disk, an optical disk. The communication unit 1009 allows the terminal device 120 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks, and can include but are not limited to a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset, such as a Bluetooth™ device, an 802.11 device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0163] The computing unit 1001 can be various general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 executes the various methods and processes described above, such as the emergency event visualization collaboration method. For example, in some embodiments, the emergency event visualization collaboration method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 1008. In some embodiments, part or all of the computer program can be loaded and / or installed onto the terminal device 120 via the ROM 1002 and / or the communication unit 1009. When the computer program is loaded into the RAM 1003 and executed by the computing unit 1001, one or more steps of the emergency event visualization collaboration method described above can be executed. Alternatively, in other embodiments, the computing unit 1001 can be configured to execute the emergency event visualization collaboration method by any other suitable means (e.g., by means of firmware).

[0164] Various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field-programmable gate arrays (FPGA), application-specific integrated circuits (ASIC), application-specific standard products (ASSP), system-on-a-chip systems (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special or general programmable processor that receives data and instructions from a storage system, at least one input device, and at least one output device, and transmits the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0165] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), and the Internet.

[0166] It should be understood that the various forms of the processes shown above can be used, steps can be reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present invention can be achieved, and no limitations are imposed herein.

[0167] Although embodiments or examples of the present invention have been described with reference to the accompanying drawings, it should be understood that the above methods, systems, and devices are merely exemplary embodiments or examples, and the scope of the present invention is not limited by these embodiments or examples, but is only defined by the authorized claims and their equivalent scope. Various elements in the embodiments or examples can be omitted or replaced by their equivalent elements. In addition, the steps can be executed in an order different from that described in the present invention. Further, the various elements in the embodiments or examples can be combined in various ways. Importantly, with the evolution of technology, many of the elements described herein can be replaced by equivalent elements that emerge after the present invention.< / marker> < / g> < / polygon> < / path> < / g> < / path> < / polygon> < / polygon> < / text> < / g> < / lineargradient> < / path>

Claims

1. A visual collaborative method for emergency events, characterized in that: The method comprises: Acquire a historical monitoring data set of emergency events in a target area, and preprocess the historical monitoring data set to generate a standardized event feature set; the standardized event feature set includes dynamic environmental parameters and static geographic parameters associated with multiple emergency event types; According to the dynamic environmental parameters in the standardized event feature set, a time series prediction model is trained to obtain an event evolution prediction model; the output of the event evolution prediction model includes a diffusion path and intensity level of the emergency event within a preset time window; Based on the static geographic parameters in the standardized event feature set, a geographic area matching model is constructed to generate a visualization layer template associated with the target area; the visualization layer template includes terrain distribution, infrastructure coordinates and emergency resource deployment information; The current environment data stream of the target area is collected in real time, the current environment data stream is classified by the event evolution prediction model to obtain an event classification result, and the corresponding visualization layer template is called from the geographic area matching model according to the event classification result; The diffusion path, intensity level and the called visualization layer template in the event classification result are integrated to generate a dynamic visualization instruction set, and the emergency event evolution situation diagram is rendered and output through an interactive interface.

2. The method according to claim 1, characterized in that The preprocessing of the historical monitoring data set to generate a standardized event feature set includes: Interpolate and fill missing data in the historical monitoring data set, and filter noise data according to a preset abnormal threshold to generate a cleaned monitoring data subset; Extracting target characteristic parameters strongly related to the type of emergency event from the cleaned monitoring data subset; the target characteristic parameters include the meteorological change rate and surface displacement in the dynamic environmental parameters, and the altitude and geological structure stability index in the static geographical parameters; According to the physical constraints corresponding to each type of emergency event, the target feature parameters are normalized to generate the standardized event feature set; wherein the normalization includes mapping parameters of different dimensions to a uniform numerical interval and matching the corresponding nonlinear scaling function according to the event type.

3. The method according to claim 2, characterized in that The extraction of target feature parameters strongly related to the emergency event type includes: Calculate the information gain of each characteristic parameter in the cleaned monitoring data subset according to the event type label marked in the historical emergency event case library; Selecting the meteorological change rate and the surface displacement whose information gain is greater than the first threshold from the dynamic environmental parameters as the first candidate feature set; Selecting geological structure stability indicators and altitudes with information gain greater than a second threshold from the static geographic parameters as a second candidate feature set; Performing multicollinearity detection on the first candidate feature set and the second candidate feature set, deleting redundant features that are linearly correlated with other features, and obtaining an optimized target feature parameter set; The optimized target feature parameter set is matched with a preset emergency event type-feature mapping table, and the feature weights are dynamically adjusted according to the matching results to generate the final target feature parameters.

4. The method according to claim 1, characterized in that: The step of training a time series prediction model according to the dynamic environment parameters in the standardized event feature set to obtain an event evolution prediction model includes: Dividing the dynamic environment parameters into a training sequence and a verification sequence in chronological order, wherein the training sequence is used for updating model parameters, and the verification sequence is used for evaluating model prediction accuracy; Constructing a hybrid model architecture including a long short-term memory network and a convolutional neural network, and performing segmented sampling on the training sequence through a sliding window mechanism to generate multiple time segment samples; Adding event type labels and intensity level labels to each time segment sample, inputting the hybrid model architecture to perform multi-task joint training, and obtaining an initial prediction model; The mean square error and classification accuracy of the initial prediction model in each time segment are calculated through the verification sequence, and the model parameters are optimized by an adaptive learning rate algorithm until convergence; The optimized model is tested to match the delay tolerance of the real-time data stream, and the number of model layers and neurons is adjusted according to the test results to generate an event evolution prediction model that meets real-time requirements.

5. The method according to claim 4, characterized in that The training process of the hybrid model architecture also includes: Input the global temporal features output by the long short-term memory network into the attention mechanism layer, generate the event evolution contribution weight corresponding to each time step, and dynamically weight the global temporal features according to the contribution weight to obtain the temporal attention feature vector; Aligning the local fluctuation pattern extracted by the convolutional neural network with the temporal attention feature vector to form a concatenated feature matrix containing temporal global dependency and spatial local details; Input the concatenated feature matrix into a gated recurrent unit, control the feature transfer path by resetting the gate and updating the gate, filter out noise data irrelevant to the diffusion path of the emergency event, and output a filtered high-dimensional spatiotemporal feature sequence; The high-dimensional spatiotemporal feature sequence is input into the diffusion path prediction branch and the intensity level regression branch respectively, wherein the diffusion path prediction branch outputs a probability distribution map of each grid unit being diffusely covered by a softmax function, and the intensity level regression branch outputs an intensity level value at each moment in a preset time window by a linear activation function; The probability distribution graph and the grid unit coordinates of the real diffusion path in the verification sequence are cross-entropy calculated, and the intensity level value and the level label marked in the verification sequence are mean square error calculated, and the two loss values ​​are combined to generate a joint optimization target; According to the joint optimization objective, the back-propagation error signal is used to synchronously adjust the weight parameters of the long short-term memory network, the convolutional neural network and the gated recurrent unit until the classification accuracy of the diffusion path prediction branch and the error rate of the intensity level regression branch reach the convergence threshold at the same time.

6. The method according to claim 1, characterized in that The step of constructing a geographic area matching model based on the static geographic parameters in the standardized event feature set includes: Converting the terrain distribution data in the static geographic parameters into a rasterized elevation map, and marking the grid cells corresponding to the infrastructure coordinates in the elevation map; According to the historical emergency resource dispatch records, the emergency resource deployment information is associated within the grid unit to generate an initial geographic layer; Performing a topological structure analysis on the initial geographic layer, identifying key path nodes and potential risk areas, and adding corresponding topological relationship labels; Performing regional segmentation on the initial geographic layer based on the topological relationship labels to obtain multiple sub-region templates, and configuring an independent data rendering channel for each sub-region template; Each sub-area template is dynamically bound to the intensity level in the event classification result to generate a visualization layer template that can automatically adjust the display ratio according to the intensity level.

7. The method according to claim 6, characterized in that The step of fusing the diffusion path, intensity level and the called visualization layer template in the event classification result to generate a dynamic visualization instruction set includes: According to the diffusion path in the event classification result, a heat diffusion trajectory is drawn in the called visualization layer template, and the intensity level change is represented by a color gradient; Identify the grid cells covered by the diffusion path, calculate the resource gap ratio based on the emergency resource deployment information, and generate a resource warning mark; Encoding the heat diffusion trajectory, intensity level color gradient and resource warning mark into vector graphics instructions; Dynamically adjust the rendering accuracy of vector graphics instructions according to the resolution of the interactive interface, and add timeline controls to achieve historical playback and prediction of evolutionary situation diagrams; The vector graphics instructions are converted into pixel data streams through a parallel rendering engine and output to the interactive interface for real-time refresh.

8. The method according to claim 7, characterized in that The method further comprises: In response to a region selection instruction triggered by a user through an interactive interface, a target sub-region is intercepted from the visualization layer template according to a coordinate range contained in the region selection instruction, and terrain distribution, infrastructure coordinates and emergency resource deployment information corresponding to the target sub-region are extracted; Superimposing the terrain distribution of the target sub-area with the real-time meteorological parameters in the current environmental data stream to generate an enhanced terrain layer with wind speed vector arrows and precipitation intensity contour lines; Locating the locations of buildings affected by the diffusion path in the enhanced terrain layer according to the infrastructure coordinates, and retrieving the material inventory in the emergency resource deployment information associated therewith, and generating a resource distribution overlay map with icon size indicating the inventory adequacy; Based on the spatial intersection area of ​​the resource distribution coverage map and the thermal diffusion trajectory, redundant coordinate points of emergency resources not covered by the diffusion path are calculated, and converted into resource replenishment path planning points that can be dynamically scheduled; Topologically match the resource replenishment path planning points with the traffic network data in the terrain distribution of the target sub-area, remove the path nodes marked as inaccessible by the surface displacement data, and generate an optimized distribution route set with detour suggestion marks; According to the length and surface displacement of each route in the optimized delivery route set, a route safety level color bar is superimposed on the enhanced terrain layer, and the path with the highest safety level in the delivery route set is compared with the thermal diffusion trajectory in terms of travel speed to generate a resource scheduling timeliness warning signal; The resource scheduling timeliness warning signal is associated with the coordinates of the new monitoring point in the real-time environmental data stream in real time. When the new monitoring point falls into the coverage area of ​​the warning signal, the recalculation of the optimized distribution route set is automatically triggered, and the view zoom ratio of the current enhanced terrain layer is locked to prevent the path display from overlapping; In response to the user's click operation on the resource distribution overlay map, all redundant coordinate points of emergency resources within the preset coordinate range are dynamically screened and clicked, and the interactive interface is driven to highlight the icons of materials that can be dispatched with priority in a pulse flashing mode; When it is detected that the intensity level color gradient of the heat diffusion trajectory changes across levels, the safety level color bars of all delivery routes intersecting the changed area in the enhanced terrain layer are refreshed, and the priority sorting of the resource scheduling plan list is adjusted according to the refreshed color bars; When synchronously playing the evolution process of resource scheduling time warning signals, optimized distribution route sets and heat diffusion trajectories in the interactive interface, interframe interpolation animations are dynamically inserted according to the real-time movement speed of graphic elements to maintain the visual continuity of the collaborative evolution of multiple layers.

9. The method according to claim 1, characterized in that: The method also includes a model online updating process, including: Monitor the prediction error of the event evolution prediction model on the latest environmental data stream in real time, and trigger the model retraining process when the prediction error exceeds a preset tolerance; Extract continuous time segments from the current environmental data stream as new training data, and perform distribution consistency check with historical monitoring data sets; Assigning dynamic weights to newly added training data that have passed the test, and using an incremental learning algorithm to update the parameters of the event evolution prediction model; Perform compatibility tests on the updated model and the geographic region matching model to ensure that the parameter mapping relationship of the visualization layer template is adjusted synchronously; The updated model is gradually deployed to the production environment through the grayscale release mechanism, and the model version change information is prompted in the interactive interface.

10. A terminal device, characterized in that: include: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method described in any one of claims 1 to 9.

Images