A scenario simulation method for urban waterlogging hazards integrating volunteer geographic information

Abstract

The present invention provides a method for simulating urban waterlogging danger scenarios by integrating volunteer geographic information, the purpose of which is to integrate multi-source volunteer geographic information and explore a method for simulating urban waterlogging danger, the method comprising the following steps: data collection; urban space unit division; volunteer geographic information data classification; residents' travel pattern mining; underlying surface "work-residence-rest" structure division; waterlogging model construction; rainstorm waterlogging scenario simulation; risk characteristic factor analysis; disaster factor evaluation index selection; disaster-pregnant environment evaluation index selection; risk assessment model construction and risk level classification. The beneficial effect of the present invention is that it integrates multi-source volunteer geographic information data, integrates multiple machine learning methods to scientifically evaluate the degree of danger caused by rainstorms and floods, and strongly supports the construction of sponge cities in flood-prone areas and flood control and drainage decisions.

Description

Technical Field

[0001] The present invention belongs to the technical field of urban waterlogging in the field of hydrology and water resources, and in particular relates to a method for simulating dangerous urban waterlogging scenarios by integrating geographic information of volunteers. Background Art

[0002] In recent years, urban flood disasters caused by sudden rainstorms have become increasingly severe, threatening people's lives and property. Waterlogging models are the cornerstone of urban flood forecasting, but most current studies do not consider the semantic knowledge of the model and its universality with urban multi-source geographic data. Volunteered Geographic Information (VGI) is created by the public in their daily production and life, whether actively or unintentionally, and is used to collect, disseminate and share map data through the Internet. It is a systematic description of data produced by the masses with geographic tags. In recent years, with the rapid development of geospatial data acquisition and processing technology, volunteer geographic information data has shown an increasing trend. However, the current urban waterlogging modeling process does not fully take into account such volunteer geographic information.

[0003] Scientifically assessing the risk of flood events has gradually become an important part of disaster prevention and mitigation in today's society. However, there is no unified standard for flood risk assessment methods and model selection, and there is a lack of quantitative research on flood risk under complex underlying surface backgrounds.

[0004] Therefore, the present invention attempts to explore a method for simulating urban waterlogging hazards by integrating multiple machine learning methods based on volunteer geographic information. Based on this purpose, the present invention proposes a method for simulating urban waterlogging hazards scenarios by integrating volunteer geographic information. Summary of the invention

[0005] In view of the problems existing in the above technologies, the present invention provides a method for simulating urban waterlogging risk scenarios by integrating volunteer geographic information, with the purpose of exploring a method for simulating urban waterlogging risk. The present invention is implemented by the following technical solutions:

[0006] The present invention discloses a method for simulating urban waterlogging risk scenarios by integrating volunteer geographic information, which mainly includes the following steps:

[0007] S1: Data collection, collecting volunteer geographic information data such as OpenStreetMap (OSM), Point of Interest (POI) and mobile phone signaling, as well as multi-source geographic data of land use, drainage network and digital elevation;

[0008] S2: Use OSM volunteer geographic information data to divide urban space units. Use OSM road network data to divide regional space units;

[0009] S3: Classify POI data and mobile phone signaling volunteer geographic information data based on the TextCNN model. Use the TextCNN model to fuse POI data and mobile phone signaling volunteer geographic information data, perform text multi-classification tasks, and obtain different land cover categories in the city;

[0010] S4: Mining residents’ travel patterns based on multi-source geographic data. Taking taxi trajectories as the basic data source, we construct a departure-arrival time series, and comprehensively utilize mobile phone signaling data and POI data. Through machine learning methods, we analyze the travel patterns of residents in different regions according to the “occupation, residence and leisure” structure;

[0011] S5: Classification of the urban underlying surface "work-residence-rest" structure by integrating multi-source geographic data. The land use data with natural attributes and the land cover data with social attributes are superimposed, and combined with the analysis results of residents' travel patterns based on taxi trajectory data, the urban underlying surface categories are divided into three categories according to the "work-residence-rest" structure;

[0012] S6: Construction of urban waterlogging model. Distribute the parameter thresholds of the urban waterlogging model to each catchment area according to the urban underlying surface category, determine the optimal parameter values ​​based on the measured values ​​of rainstorm and flood events in the experimental area, and complete the construction of the urban waterlogging model integrating the geographic information of volunteers;

[0013] S7: Simulation of rainstorm waterlogging scenarios. Using the Chicago Rain Pattern Generator, we designed rainstorms with a return period of 10, 20, 30, 40, and 50 years, respectively, and input them into the waterlogging model to simulate waterlogging under multiple rainstorm scenarios.

[0014] S8: Analysis of characteristic factors of waterlogging risk based on Multilayer Perceptron (MLP). Using the MLP model, the importance ranking of waterlogging risk results and their characteristic factors is determined;

[0015] S9: Selection of evaluation indexes for flood risk factors. Disaster factors refer to factors that lead to flood disasters. The longer the rainfall time and the more concentrated the rainfall, the greater the harm of floods. Therefore, the present invention selects precipitation and rainfall duration as disaster factors;

[0016] S10: Selection of evaluation indicators for flood-prone environment. Disaster-prone refers to an environment that breeds flood disasters, which is generally related to topography, vegetation coverage, etc. Therefore, the present invention selects elevation, flood depth, and the underlying surface type of the "work-residence-rest" structure as evaluation indicators for disaster-prone environments;

[0017] S11: Construction of waterlogging risk assessment model. The assessment model is:

[0018] ,

[0019] In the formula, It represents the waterlogging risk index. represents the weight of each indicator, Indicates the corresponding indicator, , , , and They are precipitation, rainfall duration, elevation, underlying surface type and flooding depth;

[0020] S12: Classification of risk levels. Based on the waterlogging risk assessment model, the waterlogging risk is divided into different levels: low risk area, relatively low risk area, relatively high risk area and high risk area. The risk level less than 0.5 is a low risk area. For lower risk areas, It is a relatively high risk area, and a value greater than 0.6 is a high risk area.

[0021] Further, the S2 comprises the following steps:

[0022] S21: In the OSM road network data, urban-level roads and county roads were selected as the basis for the division of research units;

[0023] S22: Merge the road network and obtain the road centerline;

[0024] S23: Use geographic information system tools to convert features into surfaces and realize the division of urban spatial units based on OSM volunteer geographic information data.

[0025] Furthermore, S3 comprises the following steps:

[0026] Based on POI data and mobile phone signaling data, two types of volunteer geographic information data, the TextCNN model is used to classify social planning land, thereby reflecting the social attributes of the urban underlying surface. POI data studies urban social planning land by reflecting the agglomeration of regional economic activities, while mobile phone signaling data recognizes the spatial differences in urban social planning land from the perspective of regional population distribution and mobility. The present invention regards the processing of this type of text data as a natural language text multi-classification task.

[0027] The label with the largest probability value is obtained from the output of the model, which corresponds to the category of social planning land, thereby completing the classification of POI and mobile phone signaling volunteer geographic information data. Based on the text classification method of the TextCNN model, the accuracy of different social planning land categories integrating social attribute data is 0.7, indicating the scientific nature of the text classification method based on the TextCNN model of the present invention.

[0028] Further, the S4 comprises the following steps:

[0029] S41: Cleaning of abnormal trajectory data. Redundant data, abnormal position data, and abnormal speed data in the taxi trajectory data are removed;

[0030] S42: Time series construction. Obtain the starting and ending points of residents' trips and pair the trajectory data by geographic location and time;

[0031] S43: Preliminary analysis of travel characteristics based on taxi trajectory data. Based on the morning peak, evening peak, weekdays and weekends, the trajectory time series data is preliminarily analyzed according to the "occupation, residence, leisure" structure to analyze the travel patterns of residents in different areas;

[0032] S44: Comprehensively utilize mobile phone signaling data and POI data to comprehensively and accurately identify the user's workplace, residence and leisure location, and ultimately generate a detailed travel pattern information database.

[0033] Further, the S44 comprises the following steps:

[0034] S441: Mapping mobile phone signaling data to S23 urban space units. Use geocoding or nearest neighbor algorithm to map signaling data to urban space units. For geocoding technology, it is achieved by converting longitude and latitude coordinates into urban space unit IDs. The nearest neighbor algorithm calculates the distance between the signaling point and the known urban space unit and selects the urban space unit with the smallest distance as the mapping location of the signaling point. Each signaling point corresponds to an urban space unit;

[0035] S442: Data cleaning and noise reduction. Clean the signaling data of a single user on a daily basis, use DBSCAN clustering analysis to remove abnormal and noisy data, and obtain the user's effective stay trajectory chain;

[0036] S443: Identify the workplace. Extract spatial unit data on working days (Monday to Friday) and during working hours (such as 9:00-18:00) from the effective stay trajectory chain. Combine POI data and stay duration to identify the attributes of spatial units and identify spatial units of office nature. Use the K-means clustering algorithm to identify the main workplaces, and refer to the resident travel pattern mining table of S43 to calculate the stay duration weight and output it to the stay occupation location and weight table;

[0037] S444: Identify the place of residence. Extract the spatial unit data of non-working time periods (such as 18:00-9:00) and the spatial unit classification attribute of residential nature from the effective stay trajectory chain. Combine POI data and stay duration to identify the attributes of spatial units and identify the spatial units of residential nature. Use the K-means clustering algorithm to identify the main residential places, and refer to the residents' travel pattern mining table of S43 to calculate the stay duration weight and output it to the stay place and weight table;

[0038] S445: Identify leisure places. Extract spatial unit data with leisure and entertainment attributes during holiday travel periods (including weekends and statutory holidays) from the effective stay trajectory chain. Combine POI data and stay duration to identify the attributes of spatial units and identify the leisure and entertainment spatial units. Use the K-means clustering algorithm to identify the main leisure and entertainment locations, and refer to the residents' travel pattern mining table in S43 to calculate the stay duration weight and output it to the stay and leisure place and weight table;

[0039] S446: Generate a database of "work-residence-rest" travel rules for residents. Based on the user's consecutive n-day stay space units and weight table, the work place, residence place and the top five leisure places with the highest weight are selected to generate the user's "work-residence-rest" travel rules information database.

[0040] Further, the S5 comprises the following steps:

[0041] S51: Reflect the natural attributes of the urban surface through land use types. ArcGIS software combined with manual identification was used to classify the natural attributes of the underlying surface, and a total of 12 land use types were divided, including cultivated land, forest land, grassland and other land;

[0042] S52: Based on the results of social planning land classification in S3 and land use type classification in S51, the natural attributes are combined with the social attributes, and the travel rules of "work-residence-rest" in S446 are integrated to classify the urban underlying surface. The present invention divides the urban underlying surface into three categories: "work-residence-rest", and the specific principles are as follows:

[0043] S521: The corresponding occupational spatial structure is basically hardened road surface, the surface is flat and dense, and the depression storage, surface Manning coefficient, infiltration rate and attenuation coefficient are all minimum;

[0044] S522: The type corresponds to the residential space structure, with hardened roads and roofs as well as a small amount of green space on the underlying surface. The surface is rougher and more undulating, with better permeability, and moderate depression storage, surface Manning coefficient, infiltration rate and attenuation coefficient;

[0045] S523: The corresponding leisure space structure is mainly gardening and greening. The land use is mainly grassland and woodland, the surface roughness is the largest, with good permeability and water storage capacity, and the depression storage capacity, surface Manning coefficient, infiltration rate and attenuation coefficient are all the largest.

[0046] Further, the S6 comprises the following steps:

[0047] S61: According to the results of parameter sensitivity analysis, the threshold of sensitive parameters is assigned to each sub-catchment area according to the distribution law of the urban underlying surface;

[0048] S62: Select multiple rainfall runoff events and use the observed urban rainstorm flood events to determine the optimal values ​​of sensitive parameters of the waterlogging model;

[0049] S63: Water depth Nash efficiency coefficient using surface water depth monitoring stations NSE , Root Mean Square Error RMSE Peak Time Difference Evaluate the model performance. NSE and RMSE The calculation method is as follows:

[0050] ,

[0051] ,

[0052] In the formula, is the total duration, for Ideal water depth at all times, for Simulate water depth at all times. NSE It is one of the important indicators for evaluating the quality of model simulation results. NSE The closer the value is to 1, the more credible the waterlogging evolution process simulated by the model is;

[0053] S64: Substitute the optimal values ​​of uncertainty sensitive parameters into the model according to the sub-catchment categories for verification.

[0054] Furthermore, the S8 specifically includes the following steps:

[0055] S81: Feature factor selection. Generate a training sample data set based on flooded points and non-flooded points. When inputting feature factors, eight indicator factors, namely, precipitation, rainfall duration, elevation, slope, building density, flooded area, flooded water depth, and waterlogging duration, are selected as the input features of the model;

[0056] S82: Adjust the neural network hyperparameters to complete the MLP model optimization. To verify the model, 80% of the data in the sample is set as the training set, and the other 20% is set as the test set. The deep learning model is trained based on the divided training and test data sets, and the hyperparameters are continuously adjusted and optimized.

[0057] Furthermore, the S11 specifically includes the following steps:

[0058] S111: There is no definite quantitative standard for the “work-residence-rest” underlying surface type. The present invention determines the impact degree classification standard for different underlying surface types based on the principle that the impact degree is in the (0, 1) interval;

[0059] S112: Normalize and calculate the disaster-pregnant factor precipitation, rainfall duration, disaster-causing factor elevation, underlying surface type and flooding depth;

[0060] S113: Use the expert experience method to determine the weights of each factor. The weight of the precipitation index is 0.2, the weight of the rainfall duration is 0.15, the weight of the elevation index is 0.15, the weight of the underlying surface type index is 0.15, and the weight of the flood depth is 0.35.

[0061] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0062] 1. Using multi-source volunteer geographic information data, combined with GIS theory and technology, and through a scientific and reasonable model structure, a more refined modeling of the urban waterlogging process is carried out;

[0063] 2. Integrate multiple machine learning methods to explore the universal laws between model parameters and urban waterlogging risk, providing new ideas for rapid simulation of urban waterlogging risk;

[0064] 3. The urban waterlogging hazard scenario simulation method proposed in the present invention integrates multi-source volunteer geographic information data, scientifically assesses the degree of danger caused by flood events, and strongly supports the construction of sponge cities in flood-prone areas and flood control and drainage decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a framework diagram of the method of the present invention;

[0066] Figure 2 It is a distribution map of regional waterlogging risk index in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0068] Embodiment: Now take a certain area as an embodiment. The rainwater drainage network in the area is relatively independent, and has various artificial ditches and lakes and other water storage units. Therefore, it can be determined that the area of ​​this embodiment belongs to an urban basin with relatively clear boundary conditions. The climate type of the embodiment area is a northern subtropical monsoon humid climate, with distinct seasons, short spring and autumn and long winter and summer, an average annual temperature of 15°C, a relative humidity of 76%, an average rainfall of 1106 mm, and an average annual rainfall of 110 days. There are many typhoons and rainstorms in the area, and the precipitation process is short and strong, which is easy to cause rainstorm waterlogging disasters.

[0069] like Figure 1 As shown, an urban waterlogging risk scenario simulation method integrating volunteer geographic information disclosed in an embodiment of the present invention mainly includes the following steps:

[0070] S1: Data collection. Collect volunteer geographic information data such as OSM, POI and mobile phone signaling, as well as multi-source geographic data such as land use, drainage network and digital elevation. The main geographic information data are shown in Table 1:

[0071] Table 1: Geographic information data:

[0072]

[0073] S2: Use OSM volunteer geographic information data to divide urban space units. Use OSM road network data to divide regional space units;

[0074] S3: Classify POI data and mobile phone signaling volunteer geographic information data based on the TextCNN model. Use the TextCNN model to fuse POI data and mobile phone signaling volunteer geographic information data, perform text multi-classification tasks, and obtain different land cover categories in the city;

[0075] S4: Mining residents’ travel patterns based on multi-source geographic data. Taking taxi trajectories as the basic data source, we construct a departure-arrival time series, and comprehensively utilize mobile phone signaling data and POI data. Through machine learning methods, we analyze the travel patterns of residents in different regions according to the “occupation, residence and leisure” structure;

[0076] S5: Urban underlying surface "work-residence-rest" structure division based on integration of multi-source geographic data. The land use data with natural attributes and the land cover data with social attributes are superimposed, and combined with the results of S4 residents' travel pattern analysis, the urban underlying surface categories are divided into three categories according to the "work-residence-rest" structure;

[0077] S6: Construction of urban waterlogging model. Distribute the parameter thresholds of the urban waterlogging model to each catchment area according to the urban underlying surface category, determine the optimal parameter values ​​based on the measured values ​​of rainstorm and flood events in the experimental area, and complete the construction of the urban waterlogging model integrating the geographic information of volunteers;

[0078] S7: Simulation of rainstorm waterlogging scenarios. Using the Chicago Rain Pattern Generator, we designed rainstorms with a return period of 10, 20, 30, 40, and 50 years, respectively, and input them into the waterlogging model to simulate waterlogging under multiple rainstorm scenarios.

[0079] S8: Analysis of characteristic factors of waterlogging risk based on multi-layer perceptron MLP. Using the MLP model, determine the importance ranking of waterlogging risk results and their characteristic factors;

[0080] S9: Selection of evaluation indexes for flood risk factors. Disaster factors refer to factors that lead to flood disasters. The longer the rainfall time and the more concentrated the rainfall, the greater the harm of floods. Therefore, the present invention selects precipitation and rainfall duration as disaster factors;

[0081] S10: Selection of evaluation indicators for flood-prone environment. Disaster-prone refers to an environment that breeds flood disasters, which is generally related to topography, vegetation coverage, etc. Therefore, the present invention selects elevation, flood depth, and the underlying surface type of the "work-residence-rest" structure as evaluation indicators for disaster-prone environments;

[0082] S11: Construction of waterlogging risk assessment model. The assessment model is:

[0083] ,

[0084] In the formula, It represents the waterlogging risk index. represents the weight of each indicator, Indicates the corresponding indicator, , , , and They are precipitation, rainfall duration, elevation, underlying surface type and flooding depth;

[0085] S12: Classification of risk levels. Based on the waterlogging risk assessment model, the waterlogging risk is divided into different levels: low risk area, relatively low risk area, relatively high risk area and high risk area. The risk level less than 0.5 is a low risk area. For lower risk areas, It is a relatively high risk area, and a value greater than 0.6 is a high risk area.

[0086] Further, the S2 comprises the following steps:

[0087] S21: In the OSM road network data, urban-level roads and county roads were selected as the basis for the division of research units;

[0088] S22: Merge the road network and obtain the road centerline;

[0089] S23: Use geographic information system tools to convert features into surfaces and realize the division of urban spatial units based on OSM volunteer geographic information data.

[0090] Furthermore, S3 comprises the following steps:

[0091] Based on POI data and mobile phone signaling data, two types of volunteer geographic information data, the TextCNN model is used to classify social planning land, thereby reflecting the social attributes of the urban underlying surface. POI data studies urban social planning land by reflecting the agglomeration of regional economic activities, while mobile phone signaling data recognizes the spatial differences in urban social planning land from the perspective of regional population distribution and mobility. The present invention regards the processing of this type of text data as a natural language text multi-classification task.

[0092] The label with the largest probability value is obtained from the output of the model, which corresponds to the category of social planning land, thereby completing the classification of POI and mobile phone signaling volunteer geographic information data. Based on the text classification method of the TextCNN model, the accuracy of different social planning land categories integrating social attribute data is 0.7, indicating the scientific nature of the text classification method based on the TextCNN model of the present invention.

[0093] Further, the S4 comprises the following steps:

[0094] S41: Cleaning of abnormal trajectory data. Redundant data, abnormal position data, and abnormal speed data in the taxi trajectory data are removed;

[0095] S42: Time series construction. Obtain the starting and ending points of residents' trips and pair the trajectory data by geographic location and time;

[0096] S43: Preliminary analysis of travel characteristics based on taxi trajectory data. Based on the morning peak, evening peak, weekdays and weekends, the trajectory time series data is preliminarily analyzed according to the "occupation, residence, leisure" structure to analyze the travel patterns of residents in different areas;

[0097] S44: Comprehensively utilize mobile phone signaling data and POI data to comprehensively and accurately identify the user's workplace, residence and leisure location, and ultimately generate detailed travel pattern information.

[0098] Further, the S44 comprises the following steps:

[0099] S441: Mapping mobile phone signaling data to S23 urban space units. Use geocoding or nearest neighbor algorithm to map signaling data to urban space units. For geocoding technology, it is achieved by converting longitude and latitude coordinates into urban space unit IDs. The nearest neighbor algorithm calculates the distance between the signaling point and the known urban space unit and selects the urban space unit with the smallest distance as the mapping location of the signaling point. Each signaling point corresponds to an urban space unit;

[0100] S442: Data cleaning and noise reduction. Clean the signaling data of a single user on a daily basis, use DBSCAN clustering analysis to remove abnormal and noisy data, and obtain the user's effective stay trajectory chain;

[0101] S443: Identify the workplace. Extract spatial unit data on working days (Monday to Friday) and during working hours (such as 9:00-18:00) from the effective stay trajectory chain. Combine POI data and stay duration to identify the attributes of spatial units and identify spatial units of office nature. Use the K-means clustering algorithm to identify the main workplaces, and refer to the resident travel pattern mining table of S43 to calculate the stay duration weight and output it to the stay occupation location and weight table;

[0102] S444: Identify the place of residence. Extract the spatial unit data of non-working time periods (such as 18:00-9:00) and the spatial unit classification attribute of residential nature from the effective stay trajectory chain. Combine POI data and stay duration to identify the attributes of spatial units and identify the spatial units of residential nature. Use the K-means clustering algorithm to identify the main residential places, and refer to the residents' travel pattern mining table of S43 to calculate the stay duration weight and output it to the stay place and weight table;

[0103] S445: Identify leisure places. Extract spatial unit data with leisure and entertainment attributes during holiday travel periods (including weekends and statutory holidays) from the effective stay trajectory chain. Combine POI data and stay duration to identify the attributes of spatial units and identify the leisure and entertainment spatial units. Use the K-means clustering algorithm to identify the main leisure and entertainment locations, and refer to the residents' travel pattern mining table in S43 to calculate the stay duration weight and output it to the stay and leisure place and weight table;

[0104] S446: Generate a database of "work-residence-rest" travel rules for residents. Based on the user's consecutive n-day stay space units and weight table, the work place, residence place and the top five leisure places with the highest weight are selected to generate the user's "work-residence-rest" travel rules information database.

[0105] Further, the S5 comprises the following steps:

[0106] S51: Reflecting the natural attributes of the urban surface through land use types. The embodiment of the present invention obtains land use type data from the city's Planning and Natural Resources Bureau, uses ArcGIS software combined with manual identification to classify the natural attributes of the underlying surface, and divides a total of 12 land use types, including cultivated land, forest land, grassland and other land;

[0107] S52: Based on the results of social planning land classification in S3 and land use type classification in S51, the natural attributes are combined with the social attributes, and the travel rules of "work-residence-rest" in S446 are integrated to classify the urban underlying surface. The present invention divides the urban underlying surface into three categories: "work-residence-rest", and the specific principles are as follows:

[0108] S521: The corresponding occupational spatial structure is basically hardened road surface, the surface is flat and dense, and the depression storage, surface Manning coefficient, infiltration rate and attenuation coefficient are all minimum;

[0109] S522: The type corresponds to the residential space structure, with hardened roads and roofs as well as a small amount of green space on the underlying surface. The surface is rougher and more undulating, with better permeability, and moderate depression storage, surface Manning coefficient, infiltration rate and attenuation coefficient;

[0110] S523: The corresponding leisure space structure is mainly gardening and greening. The land use is mainly grassland and woodland, the surface roughness is the largest, with good permeability and water storage capacity, and the depression storage capacity, surface Manning coefficient, infiltration rate and attenuation coefficient are all the largest.

[0111] Further, the S6 comprises the following steps:

[0112] S61: According to the parameter sensitivity analysis results, in the embodiment, only N-Perv, MinRate and Decay are sensitive parameters, and the thresholds of the three sensitive parameters are assigned to each sub-catchment area according to the distribution law of the urban underlying surface;

[0113] S62: Select multiple rainfall runoff events and use the observed urban rainstorm flood events to determine the optimal values ​​of sensitive parameters of the waterlogging model;

[0114] S63: Water depth Nash efficiency coefficient using surface water depth monitoring stations NSE , Root Mean Square Error RMSE Peak Time Difference Evaluate the model performance. NSE and RMSE The calculation method is as follows:

[0115] ,

[0116] ,

[0117] In the formula, is the total duration, for Ideal water depth at all times, for Simulate water depth at all times. NSE It is one of the important indicators for evaluating the quality of model simulation results. NSE The closer the value is to 1, the more credible the waterlogging evolution process simulated by the model is;

[0118] S64: Substitute the optimal value of the uncertainty sensitive parameter into the model according to the sub-catchment category for verification. The embodiment uses two observed rainfall events to simulate the constructed urban waterlogging model. In order to verify the feasibility of the method of the present invention, the model construction method that does not take into account the geographic information of the volunteers is used as a comparison, and the water depth Nash efficiency coefficient is selected NSE , Root Mean Square Error RMSE Peak Time Difference right and The simulation results of the two monitoring stations were evaluated, and the evaluation index statistics of the two methods are shown in Table 2:

[0119] Table 2 Statistical table of flood simulation results evaluation indicators

[0120]

[0121] From the evaluation index, the water depth Nash efficiency coefficient of the method proposed in this invention is NSE Basically above 0.7, RMSE The average value is about 6, and the average peak time difference is also controlled at about 45 minutes. Compared with the traditional method, the simulated water depth Nash efficiency coefficient based on the method of the present invention is NSE , Root Mean Square Error RMSE Peak Time Difference The mean values ​​increased by 0.20, 2.68 and 0.9 h respectively. It is not difficult to analyze that compared with the urban waterlogging model construction method that does not take into account the geographic information of volunteers, the simulated water depth of the method of the present invention is more consistent with the observed water depth, and can be used as the optimal model for simulating urban waterlogging danger scenarios.

[0122] Furthermore, the S8 specifically includes the following steps:

[0123] S81: Feature factor selection. Generate a training sample data set based on flooded points and non-flooded points. When inputting feature factors, eight indicator factors, namely, precipitation, rainfall duration, elevation, slope, building density, flooded area, flooded water depth, and waterlogging duration, are selected as the input features of the model;

[0124] S82: Adjust the neural network hyperparameters to complete the MLP model optimization. To verify the model, 80% of the data in the sample is set as the training set, and the other 20% is set as the test set. The deep learning model is trained based on the divided training and test data sets, and the hyperparameters are continuously adjusted and optimized.

[0125] Furthermore, the S11 specifically includes the following steps:

[0126] S111: There is no definite quantitative standard for the “work-residence-rest” underlying surface type. The present invention determines the impact classification standard for different underlying surface types based on the principle that the impact degree is in the (0, 1) interval:

[0127] Table 3: Classification criteria for the impact of underlying surface types:

[0128]

[0129] S112: Normalize and calculate the disaster-pregnant factor precipitation, rainfall duration, disaster-causing factor elevation, underlying surface type and flooding depth;

[0130] S113: Use the expert experience method to determine the weights of each factor. The weight of the precipitation index is 0.2, the weight of the rainfall duration is 0.15, the weight of the elevation index is 0.15, the weight of the underlying surface type index is 0.15, and the weight of the flood depth is 0.35.

[0131] The results of the implementation example show that there are significant differences in the degree of influence of each indicator on flood risk. Finally, five characteristic factors are selected, namely precipitation, rainfall duration, elevation, underlying surface type and flood depth.

[0132] Example The distribution of the hazard index of a certain area is shown in the attached Figure 2 As shown, four test points A, B, C, and D in the embodiment area are selected, and the risk assessment results are shown in Table 4:

[0133] Table 4: Hazard assessment results table:

[0134]

[0135] The method of the present invention can perform relatively accurate simulation calculations for waterlogging under different rainfall intensities, and thus scientifically and rationally assess the risk of waterlogging. The risk assessment results are consistent with the actual situation, which verifies the effectiveness of the scenario simulation method for urban waterlogging risk that integrates volunteer geographic information. According to the assessment results, the region in this embodiment is not capable of dealing with urban waterlogging under heavy rain, providing support for urban waterlogging prevention management and emergency rescue.

[0136] The above embodiments are only used to help understand the method and core idea of ​​the present invention; at the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for simulating urban waterlogging risk scenarios by integrating volunteer geographic information, characterized in that: The following steps are involved: S1: Data collection, collecting open street maps, points of interest and mobile phone signaling volunteer geographic information data, as well as land use, drainage network and digital elevation multi-source geographic data; S2: Use the geographic information data of Open Street Map volunteers to divide urban spatial units, and use the road network data of Open Street Map to divide regional spatial units; S3: Classify the POI data and mobile phone signaling volunteer geographic information data based on the TextCNN model, and use the TextCNN model to fuse the POI data and mobile phone signaling volunteer geographic information data to perform text multi-classification tasks and obtain different land cover categories in the city; S4: Mining residents’ travel patterns based on multi-source geographic data, using taxi trajectories as the basic data source, constructing departure-arrival time series, and comprehensively utilizing mobile phone signaling data and point of interest data, and using machine learning methods to analyze the travel patterns of residents in different regions according to the "occupation, residence and leisure" structure; S5: The urban underlying surface "work-residence-rest" structure division is integrated with multi-source geographic data. The land use data with natural attributes and the land cover data with social attributes are superimposed, and the analysis results of residents' travel patterns based on taxi trajectory data are combined to divide the urban underlying surface into three categories according to the "work-residence-rest" structure. S6: Construction of urban waterlogging model: Distribute the parameter thresholds of urban waterlogging model to each catchment area according to the urban underlying surface category, determine the optimal parameter values ​​based on the measured values ​​of rainstorm and flood events in the experimental area, and complete the construction of urban waterlogging model integrating the geographic information of volunteers; S7: Simulation of rainstorm waterlogging scenarios: using the Chicago rain pattern generation tool, design rainstorms with a return period of 10, 20, 30, 40 and 50 years, and input them into the waterlogging model to simulate waterlogging under multiple rainstorm scenarios; S8: Analysis of characteristic factors of waterlogging risk based on multi-layer perceptron. Using the multi-layer perceptron model, the importance ranking of waterlogging risk results and their characteristic factors is determined; S9: Selection of evaluation indicators for waterlogging risk factors. Disaster factors refer to factors that lead to flood disasters. Precipitation and duration of rainfall are selected as disaster factors. S10: Selection of evaluation indicators for the disaster-prone environment of waterlogging hazard. The elevation, flood depth and the underlying surface type of the "work-residence-rest" structure are selected as evaluation indicators for the disaster-prone environment; S11: Construction of waterlogging risk assessment model. The assessment model is: Where D represents the waterlogging risk index, W i Indicates the weight of each indicator, X i represents the corresponding index, X1, X2, X3, X4 and X5 are precipitation, rainfall duration, elevation, underlying surface type and flooding depth respectively; S12: Classification of risk levels. Based on the waterlogging risk assessment model, the waterlogging risk is divided into different levels: low risk area, lower risk area, higher risk area and high risk area. Among them, less than 0.5 is a low risk area, (0.5, 0.55] is a lower risk area, (0.55, 0.6] is a higher risk area, and greater than 0.6 is a high risk area. Wherein, S4 comprises the following steps: S41: cleaning of abnormal trajectory data, removing redundant data, abnormal position data, and abnormal speed data in the taxi trajectory data; S42: Time series construction, obtaining the starting and ending points of residents' travel, and matching trajectory data by geographic location and time; S43: Preliminary analysis of travel characteristics based on taxi trajectory data. Based on morning peak, evening peak, weekdays and weekends, the trajectory time series data is preliminarily analyzed according to the "occupation, residence, leisure" structure to analyze the travel patterns of residents in different areas; S44: Comprehensively utilize mobile phone signaling data and point of interest data to comprehensively and accurately identify the user's work, residence and leisure place, and ultimately generate a detailed travel pattern information database; Wherein, S44 comprises the following steps: S441: Mapping mobile phone signaling data to S23 urban space units, using geocoding or nearest neighbor algorithm to map signaling data to urban space units. For geocoding technology, this is achieved by converting longitude and latitude coordinates into urban space unit IDs; and the nearest neighbor algorithm calculates the distance between the signaling point and the known urban space unit, and selects the urban space unit with the smallest distance as the mapping position of the signaling point. Each signaling point corresponds to one urban space unit. S442: Data cleaning and noise reduction: cleaning the signaling data of a single user on a daily basis, using DBSCAN clustering analysis to remove abnormal and noisy data, and obtaining the user's effective stay trajectory chain; S443: Identify the workplace, extract the spatial unit data of working days (i.e. Monday to Friday) and working hours from the effective stay trajectory chain, combine the data of points of interest and stay duration, identify the attributes of the spatial unit, identify the spatial unit of office nature, use the K-means clustering algorithm to identify the main workplace, and refer to the resident travel pattern mining table of S43 to calculate the stay duration weight, and output it to the stay occupation location and weight table; S444: Identify the place of residence. Extract the spatial unit data of non-working time periods and the spatial unit classification attribute of residential nature from the effective stay trajectory chain. Combine the data of points of interest and the length of stay to identify the spatial units, identify the spatial units of residential nature, use the K-means clustering algorithm to identify the main residential places, and refer to the residents' travel pattern mining table of S43 to calculate the length of stay weight and output it to the stay place and weight table; S445: Identify leisure places, extract the spatial unit data of holiday travel periods and the spatial unit classification attribute of leisure and entertainment nature from the effective stay trajectory chain, combine the interest point data and stay duration, identify the spatial unit attributes, identify the spatial unit of leisure and entertainment nature, use the K-means clustering algorithm to identify the main leisure and entertainment locations, and refer to the residents' travel pattern mining table of S43 to calculate the stay duration weight and output it to the stay leisure place and weight table; S446: Generate a "work-residence-rest" resident travel regularity information database, integrate the user's stay space units and weight table for n consecutive days, take the workplace and residence with the highest weight and the top five leisure places with the highest weight, and generate the user's "work-residence-rest" travel regularity information database; Wherein, S5 comprises the following steps: S51: The natural attributes of the urban surface are reflected by land use types. ArcGIS software combined with manual identification is used to classify the natural attributes of the underlying surface. A total of 12 land use types are divided, including cultivated land, forest land, garden land, commercial and service land, industrial and mining storage land, residential land, special land, transportation land, water area and water conservancy facilities land, public management and public service land, grassland and other land. S52: Based on the results of social planning land classification and land use type classification, the natural attributes are combined with the social attributes, and the S446 "work-residence-rest" travel pattern is integrated to classify the urban underlying surface, and the urban underlying surface is divided into three categories: "work-residence-rest"; Wherein, S52 comprises the following steps: S521: Class I corresponds to occupational space structures, with the underlying surface being basically hardened pavement, the surface being flat and compact, and the depression storage, surface Manning coefficient, infiltration rate and attenuation coefficient being the smallest; S522: Class II corresponds to residential space structures, with hardened roads and roofs on the underlying surface, some green space, and depression storage, surface Manning coefficient, infiltration rate and attenuation coefficient within the middle range; S523: Class III corresponds to leisure space structure, the underlying surface is mainly gardening and greening, the land use is mainly grassland and forest land, and the depression storage capacity, surface Manning coefficient, infiltration rate and attenuation coefficient are all the largest; Wherein, S6 comprises the following steps: S61: According to the results of parameter sensitivity analysis, the threshold of sensitive parameters is assigned to each sub-catchment area according to the distribution law of the urban underlying surface; S62: Select multiple rainfall runoff events and use the observed urban rainstorm flood events to determine the optimal values ​​of sensitive parameters of the waterlogging model; S63: The model performance is evaluated using the Nash efficiency coefficient NSE, root mean square error RMSE and peak-to-peak time difference Δt of the surface water depth monitoring station. The calculation methods of NSE and RMSE are as follows: In the formula, T is the total time, D obs is the ideal water depth at time t, D sim is the simulated water depth at time t. NSE is one of the important indicators for evaluating the quality of model simulation results. The closer the NSE value is to 1, the more credible the waterlogging evolution process simulated by the model is. S64: Substitute the optimal values ​​of uncertainty sensitive parameters into the model according to the sub-catchment categories for verification.

2. The method for simulating urban waterlogging risk scenarios by integrating volunteer geographic information according to claim 1, characterized in that: S8 specifically includes the following steps: S81: Feature factor selection: Generate training sample data set based on flooded points and non-flooded points. When inputting feature factors, eight indicator factors, namely precipitation, rainfall duration, elevation, slope, building density, flooded area, flooded water depth and waterlogging duration, are selected as input features of the model. S82: Adjust the neural network hyperparameters and complete the multi-layer perceptron model optimization. To verify the model, set 80% of the data in the sample as the training set and the other 20% as the test set. Train the deep learning model based on the divided training data set and test data set, and continuously adjust and optimize the hyperparameters.

3. The method for simulating urban waterlogging risk scenarios by integrating volunteer geographic information according to claim 1, characterized in that: S2 includes the following steps: S21: In the Open Street Map road network data, urban-level roads and county roads were selected as the basis for dividing the research units; S22: Merge the road network and obtain the road centerline; S23: Use geographic information system tools to convert features into polygons and realize the division of urban spatial units based on the geographic information data of open street map volunteers.

4. The method for simulating urban waterlogging risk scenarios by integrating volunteer geographic information according to claim 1, characterized in that: S3 includes the following steps: Based on two types of volunteer geographic information data, point of interest data and mobile phone signaling data, the TextCNN model is used to classify social planning land. Point of interest data studies urban social planning land by reflecting the agglomeration of regional economic activities, while mobile phone signaling data recognizes the spatial differences in urban social planning land from the perspective of regional population distribution and mobility. The processing of text data is regarded as a natural language text multi-classification task. The label with the largest probability value is obtained from the output of the model, which corresponds to the category of social planning land, thereby completing the classification of two types of volunteer geographic information data: points of interest and mobile phone signaling. Based on the text classification method of the TextCNN model, the accuracy of different social planning land categories that integrate social attribute data is 0.

7.

5. The method for simulating urban waterlogging risk scenarios by integrating volunteer geographic information according to claim 1, characterized in that: S11 includes the following steps: S111: There is no definite quantitative standard for the "work-residence-rest" underlying surface type. Based on the principle that the impact degree is in the (0, 1) interval, the impact degree classification standard for different underlying surface types is determined; S112: Normalize and calculate the disaster-pregnant factor precipitation, rainfall duration, disaster-causing factor elevation, underlying surface type and flooding depth; S113: Determine the weight of each factor. The weight of the precipitation index is 0.2, the weight of the rainfall duration is 0.15, the weight of the elevation index is 0.15, the weight of the underlying surface type index is 0.15, and the weight of the flood depth is 0.35.

Images