Early warning method and system for urban inland inundation points with extreme rainstorm disasters based on hybrid intelligence

Abstract

The invention discloses an extreme rainstorm disaster urban waterlogging point early warning method and system based on hybrid intelligence. The method comprises the following steps: collecting waterlogging points of which the waterlogging risk needs to be assessed in a current city; performing risk prediction on each waterlogging point to obtain a machine score by adopting a machine learning model and based on the waterlogging-related multi-source heterogeneous data; according to the performance measurement of the machine learning model, generating the confidence coefficient of the machine score of each waterlogging point; performing risk assessment on the waterlogging point set based on expert experience to obtain an expert score of each waterlogging point; based on the evaluation consensus degree of the expert individuals and the expert groups, calculating the confidence coefficient of expert scoring of each waterlogging point; corresponding weights are dynamically optimized and generated according to confidence coefficients of scoring of machines and experts of the waterlogging points; and finally, according to the man-machine score and the respective weight, performing fusion calculation to obtain a waterlogging risk value of each waterlogging point, and further providing risk early warning. According to the method, the comprehensiveness and robustness of waterlogging point early warning under the extreme rainstorm are improved through dynamic weight optimization and man-machine cooperation.

Description

Technical Field

[0001] The present invention belongs to the field of human-machine collaborative decision-making, and specifically relates to a hybrid intelligence-based method and system for early warning of urban waterlogging points caused by extreme rainstorm disasters. Background Art

[0002] Traditional waterlogging early warning methods primarily rely on historical data or empirical models. However, due to the dynamic, complex, and hidden nature of urban waterlogging sites, single data-driven or empirical approaches are unable to comprehensively and accurately predict waterlogging risks. Especially in extreme rainstorm scenarios, waterlogging sites are influenced by multiple factors, such as topography, drainage systems, and rainfall intensity. The complex interactions between these factors result in insufficient early warning accuracy and timeliness for traditional methods. Furthermore, data on urban waterlogging sites often comes from multiple sources and are heterogeneous, including meteorological data, geographic information, and real-time monitoring data. Effectively integrating this data and identifying key waterlogging sites presents a technical challenge. In this context, a hybrid early warning method that combines human expert experience with machine intelligence analysis is urgently needed to improve the accuracy and reliability of urban waterlogging early warnings during extreme rainstorms. Through human-machine collaboration, this approach can fully leverage the intuitive judgment of human experts based on implicit knowledge and complex scenarios, as well as the rapid processing capabilities of machine intelligence for massive amounts of data, enabling in-depth identification of waterlogging sites and dynamic early warning.

[0003] Existing urban flooding early warning technologies fall into two main categories: qualitative assessments based on human expert experience, and data-driven algorithmic analysis. The former relies on experts' historical knowledge and subjective judgment of flooding sites. While this can capture implicit knowledge, it is limited by the number of experts and cognitive biases, making it difficult to cover large, dynamically changing flooding sites. The latter, which uses machine learning models to analyze historical data, can process massive amounts of information but lacks the ability to integrate heterogeneous data from multiple sources and adapts to extreme scenarios. Both approaches suffer from significant drawbacks: human expert assessments lack scalability and cannot rapidly respond to sudden rainstorms; data-driven approaches, however, are limited in their credibility due to issues with data quality and model generalization. Furthermore, existing technologies lack a mechanism for dynamically integrating human and machine assessment results, making it difficult to balance the differences between human subjective judgment and objective machine analysis. For example, human experts may overestimate the risk in familiar areas, while machine models may overlook rare but high-impact flooding sites. These issues stem from the limitations of a single agent (human or machine) and the interplay of multiple data sources and complex scenarios. Summary of the Invention

[0004] The present invention provides a hybrid intelligence-based method and system for warning urban waterlogging points in extreme rainstorm disasters. Through dynamic weight optimization and human-machine collaboration, the advantages of both parties are integrated to improve the comprehensiveness and robustness of urban waterlogging point warnings under extreme rainstorms.

[0005] In order to achieve the above technical objectives, the present invention adopts the following technical solutions: A hybrid intelligence-based early warning method for urban waterlogging during extreme rainstorm disasters, including: Collect the waterlogging points in the city that need to be assessed for waterlogging risk, forming a set of waterlogging points to be assessed; Using a machine learning model and multi-source heterogeneous data related to waterlogging, we predict the risk of each waterlogging point and generate a machine score. Based on the performance measurement of the machine learning model, we generate the confidence level of each waterlogging point's machine score. Based on expert experience, a risk assessment of the waterlogging point set was conducted to obtain expert scores for each waterlogging point. Furthermore, based on the consensus between individual experts and the expert group, the confidence level of the expert scores for each waterlogging point was calculated. According to the confidence of the machine score and expert score of each waterlogging point, the corresponding weight is dynamically optimized and generated; The waterlogging risk value of each waterlogging point is obtained by fusion calculation based on the machine score, expert score and corresponding weight of each waterlogging point; Risk warnings are issued for each waterlogging point based on the waterlogging risk value.

[0006] Furthermore, the machine learning model adopts LSTM.

[0007] Furthermore, the multi-source heterogeneous data includes meteorological data, geographic information data, pipeline network data and real-time monitoring data. The meteorological data includes rainfall amount, rainfall intensity and duration. The geographic information data includes elevation, slope, surface permeability and land use type. The pipeline network data includes pipe diameter, pipe age, drainage capacity and pump station status. The real-time monitoring data includes manhole water level, road surface water depth and flow rate.

[0008] Furthermore, for each waterlogged point, the confidence assessment area is calculated based on the machine learning model. The performance of the machine learning model is measured by using the AUC and F1 indicators to predict other waterlogged points within the area. The confidence level of the machine score of the waterlogged points is calculated using the AUC and F1 indicators. Specifically, First, determine the Waterlogging point Confidence assessment region : ; in, is the spatial Haversine distance function, is the preset space radius, is the time-distance Euclidean function, is the preset time window; Indicates the target waterlogging point The associated neighboring sample points, is the true label of the sample, To cause waterlogging, Represents sample points The observation timestamp, Indicates the target waterlogging point Current assessment time; Then, the confidence evaluation region , calculate the waterlogging point The local AUC indicator and F1 indicator : ; ; in, Confidence assessment region The set of positive samples in the set, the true label of the positive sample ; Confidence assessment region The negative sample set in the , the true label of the negative sample ; Aligning positive samples for machine learning models The risk prediction probability, Negative samples for machine learning models The risk prediction probability, is the indicator function; is the precision, indicating the confidence assessment area The proportion of actual waterlogging in waterlogged areas; is the recall rate, indicating the confidence assessment area The proportion of correctly predicted real flooding samples; Finally, the local AUC index and F1 index of all waterlogging points are normalized to obtain the local AUC index and F1 index of each waterlogging point. Normalized index of and The confidence level of the machine score of the waterlogging point is obtained by fusing the two normalized indicators through the indicator weight coefficient: ; in, For waterlogging points The confidence level of the machine rating, is the indicator weight coefficient.

[0009] Furthermore, the risk assessment of the waterlogging point set based on expert experience is performed to obtain expert scores for each waterlogging point, including: First, collect the risk assessment comparison matrix of each expert individual for all P waterlogging points , recorded as: ; in, Indicates the Expert individuals Consider waterlogging point Relative to waterlogging points the importance of Then, the risk assessment comparison matrix of all individual experts is aggregated to obtain the risk assessment matrix of the expert group : ; in, It means that the expert group is targeting waterlogging points Relative to waterlogging points The importance fusion value of is the number of individual experts in the expert group; Finally, a risk assessment comparison matrix based on expert groups Calculate the risk assessment vector for all waterlogging points: ; in, Indicates that the expert group has a good understanding of the waterlogging points The waterlogging risk assessment value is the waterlogging point Expert ratings, and .

[0010] Furthermore, the confidence level of the expert ratings for each waterlogging point is calculated based on the consensus between the individual experts and the expert group, including: First, calculate the consensus between each expert's individual risk assessment of each waterlogging point and the expert group's assessment: ; in, For the Expert individuals For the first Waterlogging point The risk assessment vector is Targeting waterlogging areas The row vector of ; Experts on waterlogging points The risk assessment vector is Targeting waterlogging areas The row vector of ; For individual experts For waterlogging points The consensus between the risk assessment and the expert group's assessment; Then, the consensus of all individual experts’ evaluations is combined to obtain the consensus of the expert group’s evaluation of each waterlogging point, which is recorded as the confidence of the expert score of each waterlogging point: ; in, Indicates that the expert group has a good understanding of the waterlogging points The consensus of the assessment, i.e. the waterlogging point The confidence level of the expert rating.

[0011] Furthermore, the corresponding weights are dynamically optimized and generated based on the confidence of the machine score and expert score of each waterlogging point, including: First, the human-machine distribution weight is calculated based on the confidence of the machine score and expert score at each flooding point: ; in, and Representing experts and machine models respectively, Represents one of the evaluation subjects, i.e., an expert or a machine model, Indicates the evaluation subject For waterlogging points The distribution weight of and Waterlogging points Based on the maximum and minimum normalized values ​​of the confidence of expert scoring and machine scoring; Secondly, the human-machine ranking weight is calculated based on the confidence level of the machine score and expert score for each flooding point: ; in, Indicates the evaluation subject For waterlogging points The ranking weight of and Waterlogging points The normalized value of the ranking position based on the confidence of expert ratings and machine ratings, i.e. , , is the indicator function; Then, an optimization model is established to minimize the combined deviation of the two weights: ; ; Where, Indicates the evaluation subject For waterlogging points The weight of Finally, the above optimization model is solved to obtain the weights of each flood point assigned by the experts and the machine. and .

[0012] Furthermore, the waterlogging risk value of each waterlogging point is obtained by integrating the machine score and the expert score of each waterlogging point and the corresponding weight, including: First, based on the machine scores and expert scores of all waterlogging points and their corresponding weights, an evaluation system for the waterlogging point set is established as follows: ; Where, is the expert scoring vector composed of all waterlogging points, is the machine scoring vector composed of all waterlogging points, is the expert weight vector composed of all waterlogging points, The machine weight vector composed of all waterlogging points; Then, quantile normalization is performed on the expert rating vector and the machine rating vector respectively: ; Where, represents quantile normalization, are the expert rating vector and machine rating vector after quantile normalization respectively; Finally, use and respectively Perform weighted processing to obtain the final waterlogging risk value of each waterlogging point : ; Where, and Waterlogging points Expert and machine ratings after quantile normalization, and Waterlogging points The weights of machine and expert ratings.

[0013] A hybrid intelligence-based urban waterlogging warning system for extreme rainstorm disasters includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor implements the above-mentioned urban waterlogging warning method for extreme rainstorm disasters.

[0014] Aiming at the problem of early warning of urban waterlogging points under extreme rainstorm scenarios, the present invention proposes a method and system for early warning of urban waterlogging points under extreme rainstorm disasters based on hybrid intelligence. The method can integrate human subjective cognition with the objective analysis of machine algorithms, retaining the intuitive judgment advantage of human experts on complex scenarios while giving full play to the efficiency of machine learning in processing massive data, thus providing support and guarantee for early warning of urban waterlogging points under extreme rainstorm disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is the overall framework of the hybrid intelligence-based urban waterlogging warning method for extreme rainstorm disasters described in the embodiment of the present application. DETAILED DESCRIPTION

[0016] The following is a detailed description of an embodiment of the present invention. This embodiment is based on the technical solution of the present invention, provides a detailed implementation method and a specific operation process, and further explains the technical solution of the present invention.

[0017] The present invention provides a hybrid intelligent-based urban waterlogging early warning method and system for extreme rainstorm disasters. The overall framework is shown in Figure 1. First, the early warning system receives the waterlogging points in the current city that need to be assessed for waterlogging risk, forming a set of waterlogging points to be assessed. Secondly, we constructed a machine-intelligence urban waterlogging warning module based on the LSTM algorithm. Furthermore, we established a waterlogging risk warning module based on human experience. Finally, we measured the confidence level of human-machine warning results, established a dynamic optimization model for human-machine weighting, and integrated the quantitative assessment results of machine and human assessments of urban waterlogging warnings under extreme rainstorms, ultimately achieving hybrid intelligent warnings for waterlogging risk points.

[0018] This embodiment provides a hybrid intelligent early warning method for urban waterlogging caused by extreme rainstorms. Figure 1 As shown, the following steps are included: Step 1: Collect the waterlogging points in the city that need to be assessed for waterlogging risk and form a set of waterlogging points to be assessed. .

[0019] Step 2: Use a machine learning model and multi-source heterogeneous data related to waterlogging to predict the risk of each waterlogging point and obtain a machine score; and generate the confidence level of the machine score of each waterlogging point based on the performance measurement of the machine learning model.

[0020] In step 2.1, a machine learning model is used to predict the risk of each waterlogging point based on multi-source heterogeneous data related to waterlogging to obtain a machine score.

[0021] The machine learning model needs to input key features related to urban waterlogging. This embodiment collects multi-source heterogeneous data related to urban waterlogging in real time, including meteorological data, geographic information, pipe network data, and real-time monitoring data. Through data cleaning, normalization, and spatiotemporal alignment, a standardized time series dataset suitable for LSTM model input is constructed. The input tensor with unified timestamps is recorded as ,in is the time step, is the feature dimension.

[0022] ; in, is the eigenvector merging function; Meteorological data, including rainfall amount, rainfall intensity, duration, etc., are obtained from meteorological radar and ground observation stations; Geographic information data, including elevation, slope, surface permeability, land use type, etc., are derived from satellite remote sensing and GIS databases; Pipeline network data, including pipe diameter, age, drainage capacity, pumping station status, etc., comes from the municipal drainage system database; Real-time monitoring data, including manhole water level, road water depth, flow rate, etc., comes from IoT sensors and cameras. Finally, a standardized time series dataset of multi-source heterogeneous data Store to memory.

[0023] The machine learning model described in this embodiment uses the long short-term memory network (LSTM) as its core algorithm, and uses its temporal modeling advantages to capture the dynamic evolution of urban flooding risks. The machine learning model learns the nonlinear relationship between the input multi-source heterogeneous data features by stacking multiple layers of LSTM units, and completes supervised training based on historical urban flooding event labeled data, optimizing the loss function to improve prediction accuracy. The multi-layer LSTM structure is used to capture temporal dependencies, and the hidden state The update formula is: ; ; ; ; ; ; in 、 、 Represent the forget gate, input gate and output gate respectively, is the candidate cell state, is the cell state, is the predicted output of waterlogging risk. The model optimizes the cross entropy loss function through the back propagation algorithm, and the loss function is set as .

[0024] The trained LSTM model uses real-time collected and inputted rainstorm monitoring data (multi-source heterogeneous data) to infer and output the probability score of waterlogging risk at each flooding point, which is recorded as the machine score. The model uses a sliding window mechanism to achieve dynamic early warning and generate a machine scoring vector for waterlogging points. ,Will Store to memory.

[0025] In step 2.2, based on the performance measurement of the machine learning model, the confidence level of the machine score of each flooding point is generated.

[0026] In this embodiment, for each waterlogging point, the confidence assessment area is calculated based on the machine learning model. The AUC and F1 indicators of other waterlogged points in the prediction are used to measure the performance of the machine learning model and generate the confidence level of the machine score of the waterlogged point. Target waterlogging points Confidence assessment region : ; in, is the spatial Haversine distance function, is the preset space radius, is the time-distance Euclidean function, A preset time window ensures that only recent relevant data is used in the evaluation; Indicates the target waterlogging point The associated neighboring sample points, is the true label of the sample, To cause waterlogging, Represents sample points The observation timestamp, Indicates the target waterlogging point The current evaluation time.

[0027] Then, the confidence evaluation region , calculate the waterlogging point The local AUC indicator and F1 indicator , the measurement model is The ability of the surrounding area to distinguish positive and negative samples.

[0028] ; ; in, Confidence assessment region The set of positive samples in the set, the true label of the positive sample ; Confidence assessment region The negative sample set in the , the true label of the negative sample ; Aligning positive samples for machine learning models The risk prediction probability, Negative samples for machine learning models The risk prediction probability, is the indicator function; is the precision, indicating the confidence assessment area The proportion of actual waterlogging in waterlogged areas; is the recall rate, indicating the confidence assessment area The proportion of correctly predicted real flooding samples. Confidence assessment region The number of samples where real flooding occurred and the model correctly warned; Confidence assessment region The number of samples that did not experience waterlogging but were falsely reported by the model; Confidence assessment region The number of samples where real flooding occurred but was missed by the model.

[0029] Finally, the local AUC index and F1 index of all waterlogging points are normalized to eliminate the dimension effect, and the maximum and minimum normalization of each waterlogging point is obtained. Normalized index of and The confidence level of the machine score of the waterlogging point is obtained by fusing the two normalized indicators through the indicator weight coefficient: ; in, Reflecting the machine intelligence model on waterlogging points The confidence level of the prediction is the waterlogging point The confidence level of the machine rating, is an adjustable indicator weight coefficient.

[0030] Calculate the confidence of the machine scores of all waterlogging points to get ,Will The results are stored in memory.

[0031] Step 3: Based on expert experience, a risk assessment is conducted on the set of waterlogging points to obtain an expert score for each waterlogging point; and based on the consensus of the evaluations of individual experts and the expert group, the confidence of the expert score for each waterlogging point is calculated.

[0032] Urban flood risk assessment Expert groups , interact with the system through input and output devices.

[0033] Step 3.1: Based on expert experience, risk assessment is performed on the set of waterlogging points to obtain expert scores for each waterlogging point.

[0034] First, each expert individual , the human evaluation module of the login system, by using the language terminology , for all flood risk points in Pairwise comparisons were conducted for each waterlogged area. After the first comparison, the urban waterlogging risk assessment comparison matrix of the human assessor individual is obtained. Expert individuals The risk assessment comparison matrix for all P waterlogging points is recorded as: ; in, Indicates the Expert individuals Consider waterlogging point Relative to waterlogging points the importance of . Will collect all The risk assessment results of each expert are stored in the memory.

[0035] Then, read all The individual assessment results are stored in the running memory, and the risk assessment matrix of the expert individuals is aggregated and the score of the waterlogging point risk is calculated through the geometric average operator. The risk assessment comparison matrix of all expert individuals is aggregated to obtain the risk assessment matrix of the expert group. : ; in, It means that the expert group is targeting waterlogging points Relative to waterlogging points The importance fusion value of is the number of individual experts in the expert group.

[0036] Finally, a risk assessment comparison matrix based on expert groups Calculate the risk assessment vector for all waterlogging points: ; in, Indicates that the expert group has a good understanding of the waterlogging points The waterlogging risk assessment value is the waterlogging point Expert ratings, and . The calculated and The results are stored in memory.

[0037] In step 3.2, based on the consensus between the evaluation of individual experts and the expert group, the confidence of the expert score of each waterlogging point is calculated.

[0038] The confidence level of human evaluation results is measured by group consensus of waterlogging risk assessment. Read from memory , and all The evaluation results of individual experts. , the degree of convergence between individual evaluation vectors and group evaluation vectors is used to measure human consensus. The specific process is: First, calculate the consensus between each expert's individual risk assessment of each waterlogging point and the expert group's assessment: ; in, For the Expert individuals For the first Waterlogging point The risk assessment vector is Experts on waterlogging points The risk assessment vector is For individual experts For waterlogging points The risk assessment and the consensus of the expert group.

[0039] Then, the consensus of all individual experts’ evaluations is combined to obtain the consensus of the expert group’s evaluation of each waterlogging point, which is recorded as the confidence of the expert score of each waterlogging point: ; in, Indicates that the expert group has a good understanding of the waterlogging points The consensus of the assessment, i.e. the waterlogging point The confidence level of the expert rating.

[0040] The confidence vector of expert ratings for all waterlogged points is recorded as . The calculated The results are stored in memory.

[0041] Step 4: Dynamically optimize and generate corresponding weights based on the confidence of the machine score and expert score of each flooding point.

[0042] First, the human-machine distribution weight is calculated based on the confidence of the machine score and expert score at each flooding point: ; in, and Representing experts and machine models respectively, Represents one of the evaluation subjects, i.e., an expert or a machine model, Indicates the evaluation subject For waterlogging points The distribution weight of and Waterlogging points Based on the maximum and minimum normalized values ​​of the confidence of expert scoring and machine scoring, that is, , ; Secondly, the human-machine ranking weight is calculated based on the confidence level of the machine score and expert score for each flooding point: ; in, Indicates the evaluation subject For waterlogging points The ranking weight of and Waterlogging points The normalized value of the ranking position based on the confidence of expert ratings and machine ratings, i.e. , , is the indicator function.

[0043] Distribution weights can reduce the impact of outliers, and ranking weights can preserve the linear relationship between indicators. Therefore, the following optimization model is established to minimize the combined deviation of the two weights: ; ; Where, Indicates the evaluation subject For waterlogging points The weight of Finally, the above optimization model is solved to obtain the weights of each flood point assigned by the experts and the machine. and .

[0044] The expert's weight vector for all waterlogging points , and the machine's weight vector for all flood points , stored in memory.

[0045] Step 5: Based on the machine score and expert score of each waterlogging point and the corresponding weight, the waterlogging risk value of each waterlogging point is obtained through fusion calculation.

[0046] First, read from memory 、 、 、 , targeting urban waterlogging points , establish a four-element evaluation system as follows: ; Where, is the expert scoring vector composed of all waterlogging points, is the machine scoring vector composed of all waterlogging points, is the expert weight vector composed of all waterlogging points, is the machine weight vector composed of all waterlogging points.

[0047] Then, due to the heterogeneity of human and machine evaluations, quantile normalization is performed on the expert rating vector and the machine rating vector respectively: ; Where, are the expert rating vector and machine rating vector after quantile normalization, to achieve comparability between human and machine ratings. is the quantile normalization process, is the indicator function, hour ,otherwise .

[0048] Finally, use and respectively Perform weighted processing to obtain the final waterlogging risk value of each waterlogging point : ; Where, and Waterlogging points Comparable expert and machine ratings after quantile normalization. and Waterlogging points The weights of machine and expert ratings.

[0049] The waterlogging risk values ​​of all waterlogged points can be further Arrange in descending order to obtain the ranking results of hybrid intelligent assessment of waterlogging risk .Will , The data are stored in a storage device and the warning order of waterlogging risk is outputted through a display device.

[0050] The above embodiments are preferred embodiments of the present application. Ordinary technicians in this field can also make various changes or improvements on this basis. Without departing from the overall concept of the present application, these changes or improvements should fall within the scope of protection required by the present application.

Claims

1. A hybrid intelligence-based method for warning urban waterlogging points due to extreme rainstorm disasters, characterized by: include: Collect the waterlogging points in the city that need to be assessed for waterlogging risk, forming a set of waterlogging points to be assessed; Using a machine learning model and multi-source heterogeneous data related to waterlogging, we predict the risk of each waterlogging point and generate a machine score. Based on the performance measurement of the machine learning model, we generate the confidence level of each waterlogging point's machine score. Based on expert experience, a risk assessment of the waterlogging point set was conducted to obtain expert scores for each waterlogging point. Furthermore, based on the consensus between individual experts and the expert group, the confidence level of the expert scores for each waterlogging point was calculated. According to the confidence of the machine score and expert score of each waterlogging point, the corresponding weight is dynamically optimized and generated; The waterlogging risk value of each waterlogging point is obtained by fusion calculation based on the machine score, expert score and corresponding weight of each waterlogging point; Risk warnings are issued for each waterlogging point based on the waterlogging risk value.

2. The method for early warning of urban waterlogging points caused by extreme rainstorm disasters according to claim 1, characterized in that: The machine learning model adopts LSTM.

3. The method for early warning of urban waterlogging points caused by extreme rainstorm disasters according to claim 1, characterized in that: The multi-source heterogeneous data includes meteorological data, geographic information data, pipeline network data and real-time monitoring data. Meteorological data includes rainfall amount, rainfall intensity and duration. Geographic information data includes elevation, slope, surface permeability and land use type. Pipeline network data includes pipe diameter, pipe age, drainage capacity and pump station status. Real-time monitoring data includes manhole water level, road surface water depth and flow rate.

4. The method for early warning of urban waterlogging points caused by extreme rainstorm disasters according to claim 1, characterized in that: For each waterlogged point, its confidence assessment area is calculated based on the machine learning model. The performance of the machine learning model is measured by using the AUC and F1 indicators to predict other waterlogged points within the area. The confidence level of the machine score of the waterlogged points is calculated using the AUC and F1 indicators. Specifically, First, determine the Waterlogging point Confidence assessment region : ； in, is the spatial Haversine distance function, is the preset space radius, is the time-distance Euclidean function, is the preset time window; Indicates the target waterlogging point The associated neighboring sample points, is the true label of the sample, To cause waterlogging, Represents sample points The observation timestamp, Indicates the target waterlogging point Current assessment time; Then, the confidence evaluation region , calculate the waterlogging point The local AUC indicator and F1 indicator : ； ； in, Confidence assessment region The set of positive samples in the set, the true label of the positive sample ; Confidence assessment region The negative sample set in , the true label of the negative sample ; Aligning positive samples for machine learning models The risk prediction probability, Negative samples for machine learning models The risk prediction probability, is the indicator function; is the precision, indicating the confidence assessment area The proportion of actual waterlogging in waterlogged areas; is the recall rate, indicating the confidence assessment area The proportion of correctly predicted real flooding samples; Finally, the local AUC index and F1 index of all waterlogging points are normalized to obtain the local AUC index and F1 index of each waterlogging point. Normalized index of and The confidence level of the machine score of the waterlogging point is obtained by fusing the two normalized indicators through the indicator weight coefficient: ； in, For waterlogging points The confidence level of the machine rating, is the indicator weight coefficient.

5. The method for early warning of urban waterlogging points caused by extreme rainstorm disasters according to claim 1, characterized in that: The risk assessment of the waterlogging point set based on expert experience is performed to obtain expert scores for each waterlogging point, including: First, collect the risk assessment comparison matrix of each expert individual for all P waterlogging points , recorded as: ； in, Indicates the Expert individuals Consider waterlogging point Relative to waterlogging points the importance of Then, the risk assessment comparison matrix of all individual experts is aggregated to obtain the risk assessment matrix of the expert group : ； in, It means that the expert group is targeting waterlogging points Relative to waterlogging points The importance fusion value of is the number of individual experts in the expert group; Finally, a risk assessment comparison matrix based on expert groups Calculate the risk assessment vector for all waterlogging points: ； in, Indicates that the expert group has a good understanding of the waterlogging points The waterlogging risk assessment value is the waterlogging point Expert ratings, and .

6. The method for early warning of urban waterlogging points caused by extreme rainstorm disasters according to claim 5, characterized in that: The confidence level of the expert scores for each waterlogging point is calculated based on the consensus between the individual experts and the expert group, including: First, calculate the consensus between each expert's individual risk assessment of each waterlogging point and the expert group's assessment: ； in, For the Expert individuals For the first Waterlogging point The risk assessment vector is Targeting waterlogging areas The row vector of ; Experts on waterlogging points The risk assessment vector is Targeting waterlogging areas The row vector of ; For individual experts For waterlogging points The consensus between the risk assessment and the expert group's assessment; Then, the consensus of all individual experts’ evaluations is combined to obtain the consensus of the expert group’s evaluation of each waterlogging point, which is recorded as the confidence of the expert score of each waterlogging point: ； in, Indicates that the expert group has a good understanding of the waterlogging points The consensus of the assessment, i.e. the waterlogging point The confidence level of the expert rating.

7. The method for early warning of urban waterlogging points caused by extreme rainstorm disasters according to claim 1, characterized in that: The corresponding weights are dynamically optimized and generated based on the confidence of the machine score and expert score of each waterlogging point, including: First, the human-machine distribution weight is calculated based on the confidence of the machine score and expert score at each flooding point: ； in, and Representing experts and machine models respectively, Represents one of the evaluation subjects, i.e., an expert or a machine model, Indicates the evaluation subject For waterlogging points The distribution weight of and Waterlogging points Based on the maximum and minimum normalized values ​​of the confidence of expert scoring and machine scoring; Secondly, the human-machine ranking weight is calculated based on the confidence level of the machine score and expert score for each flooding point: ； in, Indicates the evaluation subject For waterlogging points The ranking weight of and Waterlogging points The normalized value of the ranking position based on the confidence of expert ratings and machine ratings, i.e. , , is the indicator function; Then, an optimization model is established to minimize the combined deviation of the two weights: ； ； Where, Indicates the evaluation subject For waterlogging points The weight of Finally, the above optimization model is solved to obtain the weights of each flood point assigned by the experts and the machine. and .

8. The method for early warning of urban waterlogging points caused by extreme rainstorm disasters according to claim 1, characterized in that: The waterlogging risk value of each waterlogging point is obtained by integrating the machine score, expert score and corresponding weight of each waterlogging point, including: First, based on the machine scores and expert scores of all waterlogging points and their corresponding weights, an evaluation system for the waterlogging point set is established as follows: ； Where, is the expert scoring vector composed of all waterlogging points, is the machine scoring vector composed of all waterlogging points, is the expert weight vector composed of all waterlogging points, is the machine weight vector composed of all waterlogging points; Then, quantile normalization is performed on the expert rating vector and the machine rating vector respectively: ； Where, represents quantile normalization, are the expert rating vector and machine rating vector after quantile normalization respectively; Finally, use and respectively Perform weighted processing to obtain the final waterlogging risk value of each waterlogging point : ； Where, and Waterlogging points Expert and machine ratings after quantile normalization, and Waterlogging points The weights of machine and expert ratings.

9. A hybrid intelligent urban waterlogging warning system for extreme rainstorm disasters, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the computer program is executed by the processor, the processor is caused to implement the method according to any one of claims 1 to 8.

Images