Regional risk density analysis method and system

Abstract

The invention discloses a regional risk density analysis method and system, and solves the problem of inaccurate risk based on static data in the prior art. The method comprises the following steps: acquiring multi-source dynamic region data; carrying out preprocessing and fusion on the multi-source dynamic regional data by adopting an unstructured text risk quantification and dynamic weight distribution algorithm to obtain a fused feature matrix; establishing a risk density analysis engine by adopting a spatio-temporal joint modeling and dynamic clustering parameter adjustment algorithm, and performing regional risk density analysis on the fused feature matrix; and carrying out interactive display and dynamic early warning by adopting thermodynamic diagram dynamic layering. Through the scheme, the method has the advantages of simple logic, accuracy, reliability and the like, and has very high practical value and popularization value in the technical field of risk analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of risk analysis, and in particular to a method and system for regional risk density analysis. Background Art

[0002] Regional risk analysis refers to a series of processes that use a certain evaluation method to conduct hazard assessments on environmental pollution, transportation, power transmission safety, medical care, tourism and other aspects in a certain area, or to conduct automated quantitative assessments of these aspects from the perspective of losses.

[0003] Currently, existing regional risk analysis systems often perform risk assessments based on static data (such as historical disaster records and population density statistics) or a single data source (such as meteorological data). These systems typically employ fixed thresholds or linear weighting to generate risk heat maps and rely on manual parameter adjustments to update risk levels, resulting in numerous drawbacks. First, relying on static data results in a long update cycle and is unable to dynamically respond to changes in risk density caused by emergencies (such as earthquakes and epidemic outbreaks). Second, existing systems struggle to integrate structured data (such as sensor data) with unstructured data (such as social media sentiment), resulting in one-sided risk predictions. Furthermore, risk presentation is limited to two-dimensional heat maps, lacking dynamic layered rendering (such as timeline changes and risk type overlays) and interactive analysis capabilities.

[0004] For example, the public technology "Publication No. CN115018318A, named "A Social Regional Risk Prediction and Analysis Method and System" adopts the method of "performing kernel density analysis on the current system risk source or event based on the kernel density analysis weight of the risk source or event", and does not implement the dynamic weight allocation algorithm described in this patent; and this technology uses a trained BP neural network for predictive analysis. The P neural network is sensitive to the initial weight. If the initial weight is too small, it may cause the gradient to disappear, making the network training speed very slow or even unable to learn effectively.

[0005] Therefore, there is an urgent need to propose a logically simple, accurate and reliable regional risk density analysis method and system. Summary of the Invention

[0006] In view of the above problems, the present invention aims to provide a method and system for regional risk density analysis. The technical solutions adopted by the present invention are as follows: A regional risk density analysis method comprises the following steps: Obtain multi-source dynamic regional data of the area to be analyzed; Unstructured text risk quantification and dynamic weight allocation algorithms are used to preprocess and fuse multi-source dynamic regional data to obtain the fused feature matrix. A risk density analysis engine is built using spatiotemporal joint modeling and dynamic clustering parameter adjustment algorithms to perform regional risk density analysis on the fused feature matrix. Use dynamic layering of heat maps for interactive display and dynamic warning.

[0007] Furthermore, the unstructured text risk quantification and dynamic weight allocation algorithm is used to preprocess and fuse the multi-source dynamic regional data to obtain the fused feature matrix, which includes the following steps: Build domain knowledge graphs based on natural language processing algorithms; Use domain knowledge graphs to perform named entity recognition, risk entity library matching selection, and unstructured text data judgment on multi-source dynamic regional data; The risk entity database is matched to selected multi-source dynamic regional data, which is then injected with geographic context to generate a risk relevance score. The risk relevance score is a quantitative indicator that measures the degree of match between unstructured text data (such as social media posts) and geographic risk events (such as floods and epidemics) in the analyzed area. Its value range is [0, 1], with higher values ​​indicating stronger correlation. Risk entities (such as "ponding water" and "epidemic") are extracted from the text and matched with the pre-built risk entity database. Geographic context (such as topography, population density, and historical events) is then injected to calculate the risk relevance score. The score is based on a statistical analysis of the proportion of different risk types in a historical case database (such as a historical case database of emergencies).

[0008] Based on the entropy weight method and TOPSIS model, an improved entropy weight method with data credibility attenuation factor is introduced to perform weighted normalization on the multi-source dynamic regional data after the geographic context operation is injected to obtain the fused feature matrix.

[0009] Furthermore, the data credibility attenuation factor The expression is: in, Indicates the information entropy value of the data source; represents the adjustment coefficient; Indicates the accuracy of historical calculations based on data sources; n indicates the number of historical data sources.

[0010] Furthermore, a risk density analysis engine is built using spatiotemporal joint modeling and dynamic clustering parameter adjustment algorithms to perform regional risk density analysis on the fused feature matrix, including the following steps: Obtain a historical case library of emergencies and construct a dynamic risk model of an LSTM neural network; train the dynamic risk model using the historical case library of emergencies; the historical case library of emergencies has a built-in emergency trigger model; Inputting the fused feature matrix into the trained dynamic risk model; the fused feature matrix includes the historical data of the time sliding window and the spatial neighborhood features of the data mean; A spatial clustering algorithm is used to identify high-risk density areas, output the results, update the risk level in real time, and adaptively adjust the parameters of the emergency trigger model. The high-risk density areas are geographical areas with abnormally concentrated risk values ​​identified by the spatial clustering algorithm. The risk level is a dynamic classification of risk values ​​based on the output of the LSTM neural network. The results of the spatial clustering algorithm include the geometric center coordinates of the high-risk cluster, the risk level, and the persistence index. The geometric center coordinates of the high-risk cluster are used to locate the warning focus area. The persistence index (Pt) is a quantitative value that predicts the probability of future risk persistence, with a value range of [0,1].

[0011] Furthermore, the emergency trigger model performs parameter adaptive adjustment, and its expression is: ;

[0012] in, Represents the dynamic search radius of the spatial clustering algorithm; Indicates the base radius preset according to geographical units; Indicates the maximum risk value of the current area; represents the global average risk value; Indicates the radius after recalculation based on the emergency; Indicates the correction factor.

[0013] Furthermore, the dynamic early warning includes the following steps: A dynamic threshold adjustment algorithm is used to adaptively update the warning threshold based on the results of spatiotemporal clustering.

[0014] A system using a regional risk density analysis method, comprising: A data acquisition module is set up in the area to be monitored and collects dynamic regional data from multiple sources; The fusion processing module is connected to the data acquisition module to obtain multi-source dynamic regional data. It uses unstructured text risk quantification and dynamic weight allocation algorithms to preprocess and fuse the multi-source dynamic regional data to obtain a fused feature matrix. The risk density analysis module is connected to the fusion processing module. It uses spatiotemporal joint modeling and dynamic clustering parameter adjustment algorithms to build a risk density analysis engine and conduct regional risk density analysis on the fused feature matrix. The interactive and dynamic early warning module is connected to the risk density analysis module and uses dynamic layering of heat maps for interactive display and dynamic early warning.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention obtains multi-source dynamic regional data and adopts unstructured text risk quantification and dynamic weight allocation algorithm to preprocess and fuse the multi-source dynamic regional data to obtain a fused feature matrix. The benefit of this method is that it enhances the comprehensiveness of risk features and effectively improves the accuracy of risk prediction.

[0016] (2) The present invention uses spatiotemporal joint modeling and dynamic clustering parameter adjustment algorithm to build a risk density analysis engine, and performs regional risk density analysis on the fused feature matrix. The advantage is that high-risk geographical clusters can be identified in real time, and sudden events can automatically trigger the adjustment of clustering parameters, thus reducing manual intervention.

[0017] (3) The present invention adopts a dynamic threshold adjustment algorithm to adaptively update the warning threshold based on the spatiotemporal clustering results. The advantage is that the warning threshold is automatically updated according to the spatiotemporal clustering results (such as the area of ​​high-risk clusters and the persistence index).

[0018] In summary, the present invention has the advantages of simple logic, accuracy and reliability, and has high practical value and promotion value in the field of risk analysis technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope of protection. For those skilled in the art, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 It is a logic flow chart of the present invention.

[0021] Figure 2 A heat map is shown for the case of the present invention.

[0022] Figure 3 The risk density distribution is shown for the case of the present invention. DETAILED DESCRIPTION

[0023] To make the purpose, technical solutions, and advantages of this application more clear, the present invention is further described below with reference to the accompanying drawings and examples. Implementation methods of the present invention include, but are not limited to, the following examples. All other embodiments obtained by persons of ordinary skill in the art based on the examples in this application without creative effort are within the scope of protection of this application.

[0024] like Figures 1 to 3 As shown, this embodiment provides a regional risk density analysis method, which includes the following steps: The first step is to dynamically collect multi-source data through heterogeneous data interface standardization and streaming data priority scheduling technology, thereby acquiring dynamic regional data from multiple sources. This dynamic regional data includes both structured and unstructured (text) data. Structured data includes sensor data (such as water level sensors, heavy rain warnings from the Meteorological Bureau), and historical disaster records. Unstructured (text) data includes social media content and public opinion reports.

[0025] Here, structured and unstructured data are acquired in real time through APIs. Secondly, a distributed message queue (Kafka) is used to buffer highly concurrent data streams, prioritizing data in a dynamic queue based on the data source's SLA (Service Level Agreement). A token bucket algorithm is then used to control the throughput of different data sources. Furthermore, a unified data encapsulation protocol is designed for protocol conversion, outputting standardized data streams that are written to the top partition of the distributed message queue.

[0026] In the second step, the unstructured text risk quantification and dynamic weight allocation algorithm are used to preprocess and fuse the multi-source dynamic regional data to obtain the fused feature matrix.

[0027] In step S21, a domain knowledge graph is constructed based on a natural language processing algorithm to achieve semantic enhancement.

[0028] Step S22: Use the domain knowledge graph to perform named entity recognition, risk entity library matching selection, and unstructured text data judgment on multi-source dynamic regional data.

[0029] In step S23, the multi-source dynamic regional data selected by the risk entity database is injected into the geographic context operation and a risk relevance score is generated. Otherwise, the data is discarded. Geographic context refers to various information related to the geographic location, including its physical characteristics (such as topography and climate), cultural characteristics (such as population density and economic development level), and spatial relationships (such as distance and proximity). This information can provide richer context for the data, enabling more accurate analysis and interpretation.

[0030] In step S24, based on the entropy weight method and the TOPSIS model, an improved entropy weight method that introduces a data credibility attenuation factor is used to perform weighted normalization on the multi-source dynamic regional data after the geographic context operation is injected to obtain a fused feature matrix. Here, the expression of the data credibility attenuation factor is: in, Indicates the information entropy value of the data source (the higher the entropy value, the lower the data credibility); It represents the adjustment coefficient, and its value is 0.3; Indicates the accuracy of historical calculations based on data sources; n indicates the number of historical data sources.

[0031] The third step is to build a risk density analysis engine using spatiotemporal joint modeling and dynamic clustering parameter adjustment algorithms to perform regional risk density analysis on the fused feature matrix. The specific steps are as follows: Obtain a historical case library of emergencies and construct a dynamic risk model of an LSTM neural network; train the dynamic risk model using the historical case library of emergencies; the historical case library of emergencies has a built-in emergency trigger model; The fused feature matrix is ​​input into the trained dynamic risk model. Here, the dynamic risk model inputs a temporal sliding window consisting of 24 hours of historical data and spatial neighborhood features consisting of the mean of data within a 5km radius. Furthermore, an attention mechanism is introduced into the hidden layer of the LSTM network to automatically focus on key time periods.

[0032] A spatial clustering algorithm is used to identify high-risk density areas, update risk levels in real time, and adaptively adjust parameters of the emergency trigger model to reduce the need for manual intervention. Here, emergency characteristics (time and space dimensions) are extracted from a historical emergency case library (such as earthquakes, floods, and epidemics). When an emergency is triggered, the radius is restarted and recalculated. The expression is:

[0033] in, Represents the dynamic search radius of the spatial clustering algorithm; Indicates the base radius preset according to the geographical unit, with a city-level value of 0.5 and a block-level value of 0.1; Indicates the maximum risk value of the current area; represents the global average risk value; Indicates the radius after recalculation based on the emergency; Indicates the correction coefficient, which is generally taken as 0.2 in cities.

[0034] The fourth step is to use dynamic stratification of heat maps for interactive display and dynamic warning.

[0035] In color channel calculations, RGB values ​​are determined by the risk type (e.g., fire, flood, or epidemic). Transparency is positively correlated with risk, with the alpha channel value equal to the square of the risk value. A WebGL-based 3D geographic information rendering engine employs GPU acceleration algorithms, encoding geographic grid data into textures. Shaders enable parallel computing and dynamically layered overlays of risk heat maps.

[0036] In this embodiment, a dynamic threshold adjustment algorithm is used to adaptively update the warning threshold based on the results of spatiotemporal clustering. Here, the result of spatiotemporal clustering is a structured data set, which mainly includes three types of key parameters: the warning focus area is located by the geometric center of the high-risk cluster ( Figure 2 Heatmap rendering is used as a basis), risk levels trigger different levels of warning (red / orange / yellow / blue), and the persistence index dynamically predicts the probability of future risk persistence. The input is the risk value sequence within the current time window, and the initial threshold is determined by historical data statistics. After the threshold update rule is used, the updated threshold is output and written into the risk analysis engine's rule base.

[0037] Here, take a certain park as an example: Step 1: Basic data preparation: The park experienced a sudden and continuous heavy rainfall (rainfall > 100 mm in 3 hours). The system collected multi-source data in real time: (1) Structured data: Park water level sensors (risk value 0.85), Meteorological Bureau's red alert for heavy rain; (2) Unstructured data: Internet text "The road in Area A of the park is flooded 50 cm deep" (determined as a flood risk by natural language processing, with a correlation score of 0.92). The results of spatiotemporal clustering: Output high-risk cluster center coordinates, risk level R = 0.88 (major risk / high risk), persistence index = 0.95 (LSTM predicts the probability of risk persistence in the next three hours). Here, the risk level is defined as four levels (red / orange / yellow / blue). For example, the classification logic is: a risk value > 0.75 is red (significant risk), a risk value between [0.6, 0.75] is orange (high risk), a risk value between [0.45, 0.6] is orange (medium risk), and a risk value between [0.3, 0.45] is orange (low risk).

[0038] Step 2: Dynamic layered rendering of heat map: Risk type mapping RGB color calculation: flood risk → blue channel dominant → RGB = (0, 0, 255); superimpose rainstorm warning → add red warning → final RGB = (70, 0, 220) (harmonious color). Actual effect: Area A of the park is displayed as a semi-transparent dark blue area (darkest in the center), and the surrounding roads gradually turn light blue, such as Figure 2 .

[0039] Step 3: Dynamic threshold adjustment warning: Historical water level (initial threshold ): Based on the statistics of the past 100 floods, the safe water level line is determined (the park =0.75). Heavy rain forecast (persistence index) ): Weather radar predicts the probability of rainfall in the next 3 hours ( =0.95). State flood discharge (threshold lowered): To avoid dam breach, the water level warning line is lowered to =0.61.

[0040] The above calculation method is as follows:

[0041] Threshold update: This is equivalent to increasing the warning sensitivity by 24% (originally, an alarm was triggered when the risk value was >0.75, but now it is triggered when it is >0.61).

[0042] The on-site effect of the park: When the park risk value R=0.88> =0.61: The center coordinates of the heat map display a red pulse aperture (warning radius of about 1.2km); the transparency simultaneously deepens or becomes lighter, forming a strong visual reminder (such as Figure 2 、 Figure 3 ), the park management system automatically receives evacuation instructions (SMS + large screen pop-up window). It should be noted that, Figure 3 The numerical marks ②, ③, ④, ⑱, ⑳, etc. in the figure represent the number of accumulated risk points in the area.

[0043] The above embodiments are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any changes that adopt the design principles of the present invention and any changes made through non-creative work on this basis should fall within the scope of protection of the present invention.

Claims

1. A regional risk density analysis method, characterized in that: The following steps are involved: Obtain multi-source dynamic regional data of the area to be analyzed; Unstructured text risk quantification and dynamic weight allocation algorithms are used to preprocess and fuse multi-source dynamic regional data to obtain the fused feature matrix. A risk density analysis engine is built using spatiotemporal joint modeling and dynamic clustering parameter adjustment algorithms to perform regional risk density analysis on the fused feature matrix. Use dynamic layering of heat maps for interactive display and dynamic warning.

2. A regional risk density analysis method according to claim 1, characterized in that: The unstructured text risk quantification and dynamic weight allocation algorithm is used to preprocess and fuse the multi-source dynamic regional data to obtain the fused feature matrix, which includes the following steps: Build domain knowledge graphs based on natural language processing algorithms; Use domain knowledge graphs to perform named entity recognition, risk entity library matching selection, and unstructured text data judgment on multi-source dynamic regional data; Injecting the multi-source dynamic regional data selected by the risk entity library into the geographic context operation and generating a risk relevance score; the risk relevance score is the degree of match between the unstructured text data and the geographic risk events of the area to be analyzed; Based on the entropy weight method and TOPSIS model, and introducing the improved entropy weight method with data credibility attenuation factor, the multi-source dynamic regional data after the injection of geographic context operation are weighted and normalized to obtain the fused feature matrix.

3. A regional risk density analysis method according to claim 2, characterized in that: The data credibility attenuation factor The expression is: ,in, Indicates the information entropy value of the data source; represents the adjustment coefficient; Indicates the accuracy of historical calculations based on data sources; n indicates the number of historical data sources.

4. A regional risk density analysis method according to claim 1, 2 or 3, characterized in that: A risk density analysis engine is built using spatiotemporal joint modeling and dynamic clustering parameter adjustment algorithms. Regional risk density analysis is performed on the fused feature matrix, which includes the following steps: Obtain a historical case library of emergencies and construct a dynamic risk model of an LSTM neural network; train the dynamic risk model using the historical case library of emergencies; the historical case library of emergencies has a built-in emergency trigger model; Inputting the fused feature matrix into the trained dynamic risk model; the fused feature matrix includes the historical data of the time sliding window and the spatial neighborhood features of the data mean; A spatial clustering algorithm is used to identify high-risk density areas, the results of the spatial clustering algorithm are output, the risk level is updated in real time, and the parameters of the emergency trigger model are adaptively adjusted; the high-risk density area is a geographical area where risk values ​​are abnormally concentrated as identified by the spatial clustering algorithm; the risk level is a dynamic division of risk values ​​based on the output of the LSTM neural network; the results of the spatial clustering algorithm include the geometric center coordinates of the high-risk cluster, the risk level and the persistence index.

5. A regional risk density analysis method according to claim 4, characterized in that: The emergency trigger model performs parameter adaptive adjustment, and its expression is: ; ,in, Represents the dynamic search radius of the spatial clustering algorithm; Indicates the base radius preset according to geographical units; Indicates the maximum risk value of the current area; represents the global average risk value; Indicates the radius after recalculation based on the emergency; Indicates the correction factor.

6. A regional risk density analysis method according to claim 5, characterized in that: The dynamic early warning comprises the following steps: A dynamic threshold adjustment algorithm is used to adaptively update the warning threshold based on the results of spatiotemporal clustering.

7. A system using the regional risk density analysis method according to any one of claims 1 to 6, characterized in that: include: A data acquisition module is set up in the area to be monitored and collects dynamic regional data from multiple sources; The fusion processing module is connected to the data acquisition module to obtain multi-source dynamic regional data. It uses unstructured text risk quantification and dynamic weight allocation algorithms to preprocess and fuse the multi-source dynamic regional data to obtain a fused feature matrix. The risk density analysis module is connected to the fusion processing module. It uses spatiotemporal joint modeling and dynamic clustering parameter adjustment algorithms to build a risk density analysis engine and conduct regional risk density analysis on the fused feature matrix. The interactive and dynamic early warning module is connected to the risk density analysis module and uses dynamic layering of heat maps for interactive display and dynamic early warning.

Images