Flood prevention consultation intelligent assistance method based on large language model and hydrological knowledge platform

Abstract

This invention relates to the field of flood control consultation technology, specifically to an intelligent auxiliary method for flood control consultation based on a large language model and a hydrological knowledge platform. The method includes: extracting all structured data of the monitored water area based on the structured description output by the large language model, and clustering them according to the location of the monitoring stations to output location clusters; clustering the structured data in the location clusters according to data categories to output category clusters; determining the water situation of the monitored water area through comparative analysis of the structured data in the category clusters, thereby classifying and integrating all plans in the emergency plan library of the hydrological knowledge platform into a group of plans with relatively mild water conditions and a group of plans with relatively severe water conditions; and determining the flood control dispatch plan based on the group of plans with relatively mild water conditions and the group of plans with relatively severe water conditions, combined with the unstructured description output by the large language model. This invention quantifies the systematic nature of water situation anomalies from two dimensions: spatial region and parameter type, reflecting the water situation of the monitored water area.

Description

Technical Field

[0001] This invention relates to the field of flood control consultation technology, specifically to an intelligent auxiliary method for flood control consultation based on a large language model and hydrological knowledge platform. Background Technology

[0002] Flood control work is related to the safety of people's lives and property and social stability. Flood control consultation is an important part of flood control decision-making, aiming to gather information from all parties, scientifically analyze the water situation, and formulate reasonable flood control scheduling plans.

[0003] However, current flood control consultation work faces many challenges. On the one hand, flood control data comes from a wide range of sources, covering multiple monitoring stations, and the data formats are complex, making efficient integration and analysis of this data difficult. On the other hand, existing flood control decision support systems cannot fully utilize the structured and unstructured data from large-scale prediction models, and cannot combine them with the scheduling schemes of hydrological knowledge platforms, resulting in insufficient in-depth analysis of the water situation and a lack of targetedness and scientific rigor in the generated scheduling schemes.

[0004] To this end, we propose an intelligent auxiliary method for flood control consultation based on a large language model and a hydrological knowledge platform. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent auxiliary method for flood control consultation based on a large language model and a hydrological knowledge platform, so as to solve at least one of the above-mentioned problems in the prior art.

[0006] This invention provides an intelligent auxiliary method for flood control consultation based on a large language model and a hydrological knowledge platform, comprising: Based on the structured description output by the large language model, all structured data of the monitored water area are extracted and clustered according to the location of the monitoring station, and the location clusters are output. Cluster the structured data in the location clusters according to the data categories, and output the category clusters; By comparing and analyzing the structured data in the category clusters, the water conditions of the monitored water area are determined, thereby classifying and integrating all the plans in the emergency plan library of the hydrological knowledge platform into a plan group with mild water conditions and a plan group with severe water conditions. Based on the mild and severe flood scenarios, and combined with the unstructured descriptions output by the large language model, the flood control scheduling plan is determined.

[0007] As a further technical solution of the present invention: the structured data includes monitoring longitude, monitoring latitude, water level, and flow rate; the process of obtaining location clusters is as follows: The monitoring longitude and latitude are extracted from the structured data. Using the DBSCAN clustering algorithm, all data monitored in the monitored water area are clustered according to the location of the monitoring station. All points that are density-reachable are grouped together to form a cluster based on the location of the monitoring station, which is called the location cluster.

[0008] As a further technical solution of the present invention: the process of obtaining the category clusters is as follows: Based on any location cluster, water level and flow rate are extracted from all structured data. The DBSCAN clustering algorithm is used to cluster all structured data according to data category, and all mutually density-reachable points are grouped together to form a cluster based on data category, which is called the category cluster.

[0009] As a further technical solution of the present invention: the specific process of determining the water conditions of the monitored water area is as follows: Analyze the data values ​​of all structured data in any category cluster and output the outlier ratio; the category clusters with an outlier ratio greater than or equal to the threshold are recorded as outlier category clusters; Calculate the proportion of outlier clusters in the location clusters to obtain structured outliers; If the structured outlier is less than the structured outlier threshold, it indicates that the water condition in the monitored area is relatively mild; otherwise, it indicates that the water condition in the monitored area is relatively severe.

[0010] As a further technical solution of the present invention: the process of obtaining the abnormal data value ratio is as follows: Extract the data values ​​of all structured data in the category clusters and integrate them into single-point data groups; If the data value of the structured data in a single data point group is greater than or equal to the data standard value, then the corresponding data value is recorded as an outlier. Calculate the percentage of outlier data values ​​in a single data set to obtain the outlier ratio.

[0011] As a further technical solution of the present invention: the process of obtaining the less severe water situation scheme group is as follows: Extract all plans applicable to the relatively mild water conditions in the monitored water area from the emergency response plan database of the hydrological knowledge platform and integrate them into a plan group for the relatively mild water conditions.

[0012] As a further technical solution of the present invention: the process of obtaining the more severe water situation scheme group is as follows: Extract all plans applicable to the water conditions in the monitored water area from the emergency response plan database of the hydrological knowledge platform and integrate them into a plan group for severe water conditions.

[0013] As a further technical solution of the present invention: the method for obtaining the unstructured matching value is as follows: Extract keywords from the unstructured descriptions output by the large language model and denote them as text keywords; Extract keywords describing the applicable scenario of one solution in any solution group and record them as preset keywords. Integrate all preset keywords into a preset keyword group. Based on any text keyword; combine the text keyword with preset keywords in a preset keyword group to obtain a recombined keyword group; Analyze the recombined keyword phrases and output matching keywords; Calculate the percentage of matching keywords among the preset keywords to obtain the unstructured matching value.

[0014] As a further technical solution of the present invention: the process of obtaining the matching keywords is as follows: Calculate the cosine similarity value between the text keywords in the recombined keywords and the preset keywords; If the cosine similarity value is greater than or equal to the cosine similarity threshold, then the preset keywords in the recombined keyword group will be recorded as the matching keywords.

[0015] As a further technical solution of the present invention: the process of obtaining the flood control dispatching scheme is as follows: Schemes that are greater than or equal to the unstructured matching threshold are denoted as matching schemes; Extract all matching schemes in the current scheme group, sort the unstructured matching values ​​corresponding to the matching schemes from largest to smallest, and generate a priority list; Extract the matching scheme with the largest unstructured matching value from the priority list and use it as the flood control dispatch scheme.

[0016] The beneficial effects of this invention are: This invention extracts all structured data of the monitored water area based on the structured description output by the large language model, and clusters them according to the location of the monitoring station, outputting location clusters. Then, it clusters the structured data in the location clusters according to the data category, outputting category clusters. Through two-layer DBSCAN clustering, the monitoring data in the flood control scenario can be transformed into structured clusters with clear spatial boundaries and parameter characteristics, laying a solid data foundation for subsequent anomaly detection and contingency plan matching, and helping to improve the scientificity and efficiency of flood control and disaster reduction.

[0017] This invention determines the hydrological conditions of monitored water areas by comparing and analyzing structured data within category clusters. This allows for the categorization and integration of all emergency response plans in the hydrological knowledge platform's emergency plan database into two groups: those with milder water conditions and those with more severe water conditions. Based on these two groups, and combined with the unstructured descriptions output by a large language model, a flood control dispatch plan is determined. Furthermore, the invention quantifies the systematic nature of hydrological anomalies from two dimensions: spatial region (location clusters) and parameter type (category clusters). This provides a more comprehensive and accurate reflection of the hydrological conditions in the monitored water areas, offering a more reliable risk assessment basis for flood control decisions. Depending on the situation, applicable plans are extracted and integrated from the emergency plan database of the hydrological knowledge platform to form plans for milder and more severe flooding, providing a targeted plan basis for the determination of subsequent dispatch plans and improving the efficiency and effectiveness of responding to different flood conditions. When determining the flood control dispatch plan, structured and unstructured information are comprehensively considered, making the determination of the dispatch plan more comprehensive and reasonable. Matching plans with high similarity are extracted and a priority list is generated. Finally, the matching plan with the largest unstructured matching value is selected as the dispatch plan, which improves the matching degree between the dispatch plan and the actual flood situation and enhances the scientificity and effectiveness of flood control decision-making. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the intelligent auxiliary method for flood control consultation based on a large language model and hydrological knowledge platform according to an embodiment of the present invention. Figure 2 This is a system block diagram of the intelligent auxiliary system for flood control consultation based on a large language model and hydrological knowledge platform according to an embodiment of the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0021] Example 1 like Figure 1As shown in the figure, the intelligent auxiliary method for flood control consultation based on a large language model and hydrological knowledge platform provided in this embodiment of the invention specifically includes the following steps: Step 1: Based on the structured description output by the large language model, extract all structured data of the monitored water area, and cluster them according to the location of the monitoring station to output the location clusters; Large language models are used to identify flood control consultation information, which includes structured and unstructured descriptions. For the structured description output by the large language model, extract all structured data monitored in the monitored water area from the structured description. The structured data includes, but is not limited to, monitoring longitude, monitoring latitude, water level, and flow rate. Understandably, in flood control scenarios, structured data is quantitative information with clearly defined fields and standardized formats extracted from consultation information by large language models, such as monitoring longitude, monitoring latitude, water level, flow rate, etc. Understandably, the unstructured description output by the large language model refers to text information generated by the large language model that does not have a fixed format or organizational rules. Unlike structured data, unstructured descriptions usually exist in the form of natural language text. It can be a narrative about flood control, containing various detailed descriptions, subjective judgments, causal relationships, etc., without strict predefined format requirements. For example, the large language model may output a text such as "Recently, continuous rainfall in the upper reaches of the basin has led to a significant increase in river flow. The water level in some sections of the river has approached the warning level, and there is a certain risk of flooding in the surrounding low-lying areas." This text is an unstructured description. The DBSCAN clustering algorithm is used to cluster all structured data from the monitored water area according to the location of the monitoring station, and outputs location clusters based on the location of the monitoring station. The specific clustering process is as follows: Extract the monitoring longitude and latitude from the structured data. Based on any given set of structured data, use the corresponding monitoring longitude and latitude as a sample point. Where i = 1, 2, ..., n, and n represents the total number of structured data. This represents the monitoring dimension of the i-th structured data. This represents the monitoring longitude of the i-th structured data point; Calculate sample points to sample point European distance The specific calculation formula is as follows: mark The point has been visited; set the minimum number of samples for the core point. Use the K-distance graph to select an appropriate neighborhood radius. ,calculate of Neighborhood The specific calculation formula is as follows: It should be noted that the minimum number of samples for the core points It is set based on data density experience. The purpose is to determine the local density of prominent points, providing a basis for judging whether a prominent point is a core point; statistics The number of sample points in the sample is denoted as ; Will Minimum number of samples for core points The comparison process is as follows: like ≥ This indicates that the sample points It is the core point; like This indicates that the sample points Not the core point; It should be noted that the core element is the "seed" of clustering, which is used to expand and form high-density regions (clusters). Based on any core point, if we find If all points within the neighborhood are also core points, then... All points within the neighborhood are directly density-reachable from the initial core point; if All points within the neighborhood are not core points, but are core points. Within the neighborhood, the density is also achievable; Repeat this process, and for each newly found core point, continue to find more. The set of points within a neighborhood whose density continuously expands to reach the desired density. All points that are mutually density-reachable are grouped together to form a cluster based on the location of the monitoring station, which is called the location cluster. It should be noted that if some points do not belong to any cluster, they are considered noise points. The clustering process ends when all core points have been visited and the density of the core points has been assigned to the corresponding clusters. Step 2: Cluster the structured data in the location clusters according to data categories, and output the category clusters; Based on any location cluster, extract all structured data from the location cluster, and then use the DBSCAN clustering algorithm to cluster all structured data according to data categories, outputting the category clusters divided according to data categories. The specific clustering process is as follows: Use water level or flow rate in structured data as a sample point; Calculate the Euclidean distance between two sample points, mark the sample points as visited, set the minimum number of samples for the core point, and select an appropriate neighborhood radius using a K-distance graph. Calculate sample points Neighborhood; Statistical sample points The number of sample points in the neighborhood, if the sample points If the number of sample points in the neighborhood is greater than or equal to the minimum number of samples for the core point, then the sample point is a core point; otherwise, the sample point is not a core point. Based on any core point, if we find If all points within the neighborhood are also core points, then... All points within the neighborhood are directly density-reachable from the initial core point; if All points within the neighborhood are not core points, but are core points. Within the neighborhood, the density is also achievable; Repeat this process, and for each newly found core point, continue to find more. The set of points within a neighborhood whose density continuously expands to reach the desired density. All points that are density-reachable from each other are grouped together to form a cluster based on data categories, denoted as a category cluster. It should be noted that if some points do not belong to any cluster, they are considered noise points. The clustering process ends when all core points have been visited and the density of the core points has been assigned to the corresponding clusters. It should be noted that the purpose of using the DBSCAN clustering algorithm to cluster all structured data according to data categories is to organize historical flood control consultation information for the large language model, so as to provide reliable support for subsequent judgment of water conditions in the monitored water area; The technical solution of this embodiment is as follows: Based on the structured description output by the large language model, all structured data of the monitored water area are extracted, and clustered according to the location of the monitoring station to output location clusters. Then, the structured data in the location clusters are clustered according to the data category to output category clusters. This invention uses two-layer DBSCAN clustering to transform the monitoring data in the flood control scenario into structured clusters with clear spatial boundaries and parameter characteristics, laying a solid data foundation for subsequent anomaly detection and contingency plan matching, which is conducive to improving the scientificity and efficiency of flood control and disaster reduction. Example 2 like Figure 1 As shown in the figure, the intelligent auxiliary method for flood control consultation based on a large language model and hydrological knowledge platform provided in this embodiment of the invention specifically includes the following steps: Step 3: By comparing and analyzing the structured data in the category clusters, the water conditions of the monitored water area are determined, thereby classifying and integrating all plans in the emergency plan library of the hydrological knowledge platform into a plan group with mild water conditions and a plan group with severe water conditions. Based on any category cluster, extract the data values ​​of all structured data in the category cluster and integrate them into a single point data group; The data values ​​of structured data in a single data set are compared with the data standard values. If the data value is greater than or equal to the data standard value, the corresponding data value is recorded as an abnormal data value. If the data value is less than the data standard value, the corresponding data value is recorded as a normal data value. It is understandable that the data standard values ​​are set by those skilled in the art based on historical experience; Calculate the percentage of outlier data values ​​in a single data set to obtain the outlier ratio; If the ratio of outlier data values ​​is greater than or equal to the outlier data value ratio threshold, the corresponding category cluster is recorded as an outlier category cluster; if the ratio of outlier data values ​​is less than the outlier data value ratio threshold, the corresponding category cluster is recorded as a normal category cluster. Calculate the proportion of outlier clusters in the location clusters to obtain structured outliers; If the structured outlier is less than the structured outlier threshold, it indicates that the water situation in the monitored water area is relatively mild in the structured description of the historical flood control consultation information identified by the big language model. If the structured outlier is greater than or equal to the structured outlier threshold, it indicates that the water situation in the monitored water area is relatively severe in the structured description of the historical flood control consultation information identified by the big language model. It is understandable that structured outliers reflect the proportion of anomalies in hydrological parameter categories (category clusters) within a specific geographical area (location cluster). Their core function is to quantify the systematic nature of hydrological anomalies from two dimensions: spatial region and parameter type, providing regional and multi-parameter collaborative risk assessment basis for flood control decisions. Based on the relatively mild water conditions in the monitored water area, all plans applicable to the relatively mild water conditions in the emergency plan library of the hydrological knowledge platform are extracted and integrated into a plan group for relatively mild water conditions. Based on the severe water conditions in the monitored water area, all plans applicable to the severe water conditions in the monitored water area were extracted from the emergency plan database of the hydrological knowledge platform and integrated into a severe water condition plan group. Step 4: Based on the mild flood control scheme group and the severe flood control scheme group, and combined with the unstructured description output by the large language model, determine the flood control dispatch scheme; For the unstructured descriptions output by the large language model, we use text analysis tools in natural language processing to extract keywords from the unstructured descriptions and record them as text keywords. Extract the keywords from the "applicable scenario description" of one solution in any solution group, and record them as preset keywords. Integrate all preset keywords into a preset keyword group. Understandably, the plan group includes a plan group for milder flooding and a plan group for more severe flooding. Based on any text keyword; Combine text keywords with preset keywords in preset keyword groups to obtain recombined keyword groups; Calculate the cosine similarity between the text keywords in the recombined keywords and the preset keywords. The specific calculation formula is as follows: Where A is the text keyword vector in the recombined keyword group, B is the preset keyword vector of the recombined keyword group, A・B is the dot product of the two vectors, and ||A|| and ||B|| are the moduli of vector A and vector B, respectively; Compare the cosine similarity value with the cosine similarity threshold; If the cosine similarity value is less than the cosine similarity threshold, the recombined keyword group will be canceled. If the cosine similarity value is greater than or equal to the cosine similarity threshold, it means that the text keywords in the recombined keyword group are similar to the preset keywords, and the preset keywords in the recombined keyword group are recorded as the matching keywords. It should be noted that if no preset keywords or no text keywords remain after the reorganization is completed, the reorganization operation will stop; if the number of remaining text keywords is not 1 after the reorganization is completed, the reorganization operation will continue. Calculate the percentage of matching keywords among the preset keywords to obtain the unstructured matching value; Compare the unstructured match value with the unstructured match threshold: If the unstructured matching value is greater than or equal to the unstructured matching threshold, it means that the "applicable scenario description" of the current solution is highly similar to the unstructured description output by the large language model, and the corresponding solution is recorded as the matching solution. If the unstructured matching value is less than the unstructured matching threshold, it means that the "applicable scenario description" of the current solution is not very similar to the unstructured description output by the large language model, and the corresponding solution is recorded as a non-matching solution. Extract all matching schemes in the current scheme group, sort the unstructured matching values ​​corresponding to the matching schemes from largest to smallest, and generate a priority list; Extract the matching scheme with the largest unstructured matching value from the priority list, and use the matching scheme with the largest unstructured matching value as the flood control dispatch scheme; It is understandable that, given the existence of multiple matching schemes with the largest unstructured matching values, a flood control dispatch scheme can be selected based on flood control costs. The technical solution of this embodiment is as follows: By comparing and analyzing the structured data in the category clusters, the hydrological conditions of the monitored water area are determined. This allows for the classification and integration of all plans in the emergency response plan library of the hydrological knowledge platform into two groups: plans with milder water conditions and plans with more severe water conditions. Based on these two groups, and combined with the unstructured descriptions output by the large language model, a flood control dispatch plan is determined. This invention, through the analysis of data in the category clusters, quantifies the systematic nature of hydrological anomalies from two dimensions: spatial region (location clusters) and parameter type (category clusters). This provides a more comprehensive and accurate reflection of the hydrological conditions of the monitored water area, offering a more reliable risk assessment for flood control decision-making. Based on the different situations of light and heavy flooding, applicable plans were extracted from the emergency plan database of the hydrological knowledge platform and integrated to form plans for light and heavy flooding. This provided a targeted plan basis for the subsequent determination of dispatching plans, improving the efficiency and effectiveness of responding to different flood conditions. When determining the flood control dispatching plan, structured and unstructured information were comprehensively considered, making the determination of the dispatching plan more comprehensive and reasonable. Matching plans with high similarity were extracted and a priority list was generated. Finally, the matching plan with the largest unstructured matching value was selected as the dispatching plan, improving the matching degree between the dispatching plan and the actual flood situation, and enhancing the scientificity and effectiveness of flood control decision-making.

[0022] Example 3 like Figure 2 As shown, the intelligent auxiliary system for flood control consultation based on a large language model and hydrological knowledge platform provided in this embodiment of the invention specifically includes: Location clustering module: Based on the structured description output by the large language model, extract all structured data of the monitored water area, and cluster them according to the location of the monitoring station, outputting location clusters; Category clustering module: Clusters structured data in location clusters according to data categories and outputs category clusters; Hydrological Analysis Module: By comparing and analyzing structured data in category clusters, the hydrological conditions of the monitored water area are determined, thereby classifying and integrating all plans in the emergency plan library of the hydrological knowledge platform into groups with milder hydrological conditions and groups with more severe hydrological conditions. Scheme Determination Module: Based on the scheme groups with lighter flood conditions and those with heavier flood conditions, and combined with the unstructured descriptions output by the large language model, the flood control dispatch scheme is determined.

[0023] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0024] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for intelligent assistance in flood control consultation based on a large language model and hydrological knowledge platform, characterized in that: include: Based on the structured description output by the large language model, all structured data of the monitored water area are extracted and clustered according to the location of the monitoring station, and the location clusters are output. The structured data includes monitoring longitude, monitoring latitude, water level, and flow rate; The monitoring longitude and latitude are extracted from the structured data. Using the DBSCAN clustering algorithm, all data from the monitored water area are clustered according to the monitoring station location. All mutually reachable points are grouped together to form a cluster based on the monitoring station location, denoted as the location cluster. Based on any location cluster, water level and flow rate are extracted from all structured data. The DBSCAN clustering algorithm is used to cluster all structured data according to data category. All points that are density-reachable are grouped together to form a cluster based on data category, which is called the category cluster. Analyze the data values ​​of all structured data in any category cluster, extract the data values ​​of all structured data in the category cluster, and integrate them into a single data group; if the data value of the structured data in the single data group is greater than or equal to the data standard value, the corresponding data value is recorded as an outlier. Calculate the percentage of outlier data values ​​in a single data set to obtain the outlier ratio; Clusters of categories whose outlier values ​​are greater than or equal to a threshold are categorized as outlier clusters; the proportion of outlier clusters in the location clusters is calculated to obtain structured outliers. If the structured outlier is less than the structured outlier threshold, it indicates that the water condition in the monitored area is relatively mild; otherwise, it indicates that the water condition in the monitored area is relatively severe. Extract all plans applicable to the monitored water area with mild water conditions from the emergency response plan database of the hydrological knowledge platform and integrate them into a mild water condition plan group; extract all plans applicable to the monitored water area with severe water conditions from the emergency response plan database of the hydrological knowledge platform and integrate them into a severe water condition plan group. Based on the mild and severe flood control scenarios, and combined with the unstructured descriptions output by the large language model, flood control dispatching plans are determined, including: Extract keywords from the unstructured description output by the large language model and denote them as text keywords; extract keywords from the applicable scenario description of one solution in the current solution group and denote them as preset keywords; integrate all preset keywords into a preset keyword group; based on any text keyword, combine the text keyword with the preset keywords in the preset keyword group to obtain a recombined keyword group; analyze the recombined keyword group and output matching keywords; calculate the proportion of matching keywords in the preset keywords to obtain the unstructured matching value; Schemes in the current scheme group that are greater than or equal to the unstructured matching threshold are recorded as matching schemes. All matching schemes in the current scheme group are extracted, and the unstructured matching values ​​corresponding to the matching schemes are sorted from largest to smallest to generate a priority list. The matching scheme with the largest unstructured matching value in the priority list is extracted as the flood control dispatch scheme.

2. The flood prevention consultation intelligent assistance method based on a large language model and a hydrological knowledge platform according to claim 1, characterized in that, The process of obtaining the matching keywords is as follows: Calculate the cosine similarity value between the text keywords in the recombined keywords and the preset keywords; If the cosine similarity value is greater than or equal to the cosine similarity threshold, then the preset keywords in the recombined keyword group will be recorded as the matching keywords.

Images