Waterlogging model creation method and system for urban water management

By using deep learning and IoT technologies, the parameters of the urban flooding model are dynamically adjusted, which solves the problems of staticity and insufficient data integration of existing models, and realizes high-precision and timely flooding risk assessment and decision support, thereby improving the efficiency of water management.

CN118627407BActive Publication Date: 2024-11-15SUZHOU VORTEX INFORMATION TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411115906.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-15
Estimated Expiration
2044-08-14

AI Technical Summary

Technical Problem

Existing urban flooding models have static parameter settings, insufficient real-time data integration, and are not closely integrated with water management business processes, resulting in insufficient prediction accuracy and real-time performance, as well as limited adaptability and flexibility.

Method used

The model is trained using deep learning algorithms and combined with IoT technology to achieve real-time data acquisition and online calibration. A dynamic parameter adjustment mechanism is designed to closely integrate the model output with the urban flooding business process and support real-time decision-making.

Benefits of technology

It significantly improves the accuracy and real-time performance of urban flooding prediction models, enabling dynamic parameter adjustment to adapt to environmental changes, providing timely urban flooding risk assessment and decision support, improving water management efficiency, and reducing economic losses.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118627407B_ABST
    Figure CN118627407B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for creating a waterlogging model for use in urban water management, and relates to the technical field of urban water management. The present invention comprises the following steps: data collection and preprocessing: collecting data such as urban terrain, land use, and meteorology, and performing preprocessing to provide input for the model; collecting urban terrain data using geographic information systems and topographic surveying techniques; the present invention significantly improves the accuracy and real-time performance of the urban waterlogging prediction model by integrating deep learning algorithms and real-time data monitoring techniques. Compared with traditional mechanism-based static models, the model parameters can be dynamically adjusted to respond to environmental changes in real time, such as changes in rainfall intensity and terrain features, thereby providing a more accurate and timely waterlogging risk assessment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of urban water management, and in particular to a method and system for creating a waterlogging model applied to urban water management. Background Art

[0002] The acceleration of urbanization has led to increasingly serious urban waterlogging problems. Traditional urban drainage systems are unable to cope with extreme weather events and the increase in surface runoff caused by rapid urbanization. Urban waterlogging not only affects the efficiency of urban operations, but may also pose a serious threat to people's lives and property. Existing waterlogging models are mostly based on mechanism models, with numerous and sensitive parameters, which have a great impact on the operation results. It is difficult to obtain accurate parameter values, and the model parameters are static, which makes it difficult to reflect actual dynamic changes, resulting in insufficient prediction accuracy and real-time performance.

[0003] Among existing technologies, waterlogging models for urban water management are usually not closely integrated with actual water services. Model operation cannot be dynamically adjusted according to real-time monitoring data, resulting in deviations between model predictions and actual conditions. Model training and parameter adjustment mostly rely on historical data and offline processing. There is a lack of real-time data integration and online calibration capabilities, which limits the adaptability and flexibility of the model in practical applications. These problems limit the effectiveness and decision-making support capabilities of waterlogging models in urban water management.

[0004] In order to solve the above defects, a technical solution is now provided. Summary of the invention

[0005] The purpose of the present invention is to solve the problems of static parameter settings, insufficient real-time data integration, and loose integration with water business processes in existing urban waterlogging prediction models, and to propose a waterlogging model creation method and system for urban water management.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] The method for creating a waterlogging model for urban water management includes the following steps:

[0008] a. Data collection and preprocessing: collect urban topography, land use and meteorological data, and preprocess them to provide input for the model;

[0009] b. Deep learning model training: Use deep learning algorithms to train models and learn the patterns of waterlogging through historical data;

[0010] c. Dynamic parameter adjustment: Design a dynamic parameter adjustment mechanism to dynamically adjust model parameters based on real-time monitoring data;

[0011] d. Real-time data integration and model calibration: Integrate IoT technology to achieve real-time data collection and online calibration of the model. The specific process is as follows:

[0012] Integrate IoT devices into monitoring systems to collect data in real time;

[0013] Implement data stream processing to enable data to be transmitted and processed in real time;

[0014] Calibrate the model online based on real-time data and adjust model parameters;

[0015] Update the model regularly or as needed to adapt to new data and environmental changes;

[0016] Monitor model performance to make real-time data integration and online calibration effective;

[0017] e. Waterlogging business integration: Integrate model output with waterlogging business processes to achieve real-time decision support.

[0018] Furthermore, the specific process of step a is as follows:

[0019] Collect urban topographic data using geographic information systems and topographic survey techniques;

[0020] Collecting land use classification data, including satellite remote sensing data and urban planning data;

[0021] Obtain weather data on rainfall, wind speed, and humidity from weather stations or weather services;

[0022] Check the completeness of the data and remove or correct missing values ​​and outliers;

[0023] Standardize the data to eliminate the impact of different dimensions and magnitudes;

[0024] Extract features related to waterlogging prediction from raw data;

[0025] The data is divided into training set, validation set and test set to prepare for model training and evaluation.

[0026] Furthermore, the specific steps for checking the integrity of the data in step a and removing or correcting missing values ​​and outliers are as follows:

[0027] Verify the integrity of the data set, including number of records, field integrity, and data format consistency;

[0028] Perform logic checks to ensure that the relationships between data records meet pre-set criteria;

[0029] Apply descriptive statistical analysis and graphical visualization to assess data distribution and identify outliers;

[0030] For missing data, select deletion or filling strategies based on the proportion and pattern of missing values, where filling strategies include using mean, median, mode, time series filling, regression prediction, or interpolation methods;

[0031] For outliers, identify them through box plots, standard deviation methods, or quantile methods, and determine their actual significance based on the business context;

[0032] Correct, delete or isolate outliers, or use robust statistical methods to reduce their impact;

[0033] Before determining the processing method, evaluate the impact of missing or abnormal data on model performance, and experimentally test different processing methods to select the optimal strategy based on the contribution of the data to the model prediction results.

[0034] Furthermore, the specific operation steps of dividing the data into a training set, a validation set and a test set in step a are as follows:

[0035] Determine the ratio of training set, validation set and test set according to the data volume and model requirements, and choose a split ratio of 70% training set, 15% validation set and 15% test set or 80% training set, 10% validation set and 10% test set;

[0036] Randomly shuffle the entire data set to avoid bias caused by data sorting;

[0037] Split the training set, validation set, and test set from the shuffled dataset according to the determined ratio;

[0038] Check the similarity of the statistical characteristics of the segmented data sets to ensure there is no distribution bias;

[0039] For supervised learning tasks, separate features and labels to ensure that the model learns correctly;

[0040] Store the segmented dataset in a CSV file or database;

[0041] Verify the segmented data set to avoid data leakage;

[0042] When the data set is smaller than the preset standard, the k-fold cross-validation method is used to improve the evaluation stability;

[0043] Back up the original data set and the split data set;

[0044] During model training, update the dataset as needed, including adding new data or adjusting the scale;

[0045] Record the segmentation process and results, including proportions and distribution, for team communication and model reproduction.

[0046] Furthermore, the specific operation steps of step b are as follows:

[0047] b1: Identify the problem type and understand the data characteristics, including the data type, distribution, feature dimension and sample size;

[0048] b2: Investigate existing solutions and select appropriate deep learning model architecture based on the problem type;

[0049] b3: Design the experimental framework for model comparison, hyperparameter adjustment, and performance evaluation;

[0050] b4: Determine the optimal model architecture, configure loss function, optimizer and evaluation metrics;

[0051] b5: Use the training set data to train the model and optimize the network weights through the back propagation algorithm;

[0052] b6: Adjust hyperparameters, including learning rate and batch size, to improve model performance;

[0053] b7: Use the validation set to evaluate model performance and adjust the model architecture or parameters;

[0054] b8: Save the trained model for subsequent use and deployment.

[0055] Furthermore, the specific operation steps for determining the optimal model architecture in b4 are as follows:

[0056] According to the characteristics of the problem and the initial understanding, several model architectures are preliminarily screened and tested. The basic factors and influencing factors of the test are collected for comprehensive analysis to determine the priority of all selected model architectures. The basic factors include: accuracy, recall, mean square error and F1 score. After normalization, the basic evaluation value JPZ is obtained.

[0057] The influencing factors include: training time, inference time and model size. After normalization, the film review value YPZ is calculated by the formula;

[0058] Then normalize the obtained basic evaluation value and film review value and enter them into the formula: To obtain the comprehensive evaluation value ZPZ, and use this comprehensive evaluation value ZPZ as the evaluation standard for measuring the priority of the model architecture;

[0059] The comprehensive evaluation values ​​ZPZ obtained by all selected model architectures are sorted by size, and the model architecture with the largest comprehensive evaluation value ZPZ is selected as the optimal choice.

[0060] Furthermore, the specific operation steps of step c are as follows:

[0061] Determine which model parameters can be dynamically adjusted based on real-time data through a confirmation mechanism;

[0062] Analyze the relationship between real-time data and model parameters and establish adjustment rules;

[0063] Develop parameter adjustment strategies, including threshold triggering and adaptive adjustment mechanisms;

[0064] Utilize the parameter update mechanism to enable the model to dynamically adjust parameters using real-time data;

[0065] Design feedback loops to automatically adjust parameters based on how far the model predictions deviate from actual observations.

[0066] Furthermore, the specific process of the confirmation mechanism in step c is as follows:

[0067] Determine key model parameters, including learning rate, regularization coefficient, number of network layers and number of neurons;

[0068] Conduct parameter sensitivity analysis to identify parameters that have a significant impact on model output through experimental or theoretical analysis;

[0069] Analyze data characteristics, including range, periodicity, and seasonality, to determine parameters that require dynamic adjustment;

[0070] Assess the impact of real-time data on model predictions and determine parameters associated with data changes;

[0071] Use historical data to simulate real-time data flow and analyze performance changes under different parameter settings;

[0072] Apply model diagnostic tools to identify model performance bottlenecks;

[0073] Conduct A / B testing on models with different parameter settings to compare their performance in processing real-time data;

[0074] Design feedback mechanisms based on the deviation between prediction and observation to dynamically adjust parameters;

[0075] Test the model’s response to real-time data in a simulated environment and identify parameters that need to be adjusted;

[0076] Apply hyperparameter optimization algorithms to find the optimal adjustment strategy;

[0077] Establish a real-time monitoring system to track model performance and trigger corresponding parameter adjustments;

[0078] Evaluate the robustness of the model under different parameter settings and determine the parameter adjustment conditions;

[0079] Apply continuous learning strategies to enable the model to automatically adjust parameters over time and with data accumulation.

[0080] Furthermore, the specific process of step e is as follows:

[0081] Analyze the business process of waterlogging management and determine how model outputs can support decision making;

[0082] Integrate the model into existing water management platforms;

[0083] Design user interfaces for managers to access model outputs;

[0084] Develop early warning systems and risk assessment tools to assist decision making;

[0085] Conduct simulation exercises to evaluate the impact of model outputs on business processes;

[0086] Continuously optimize models and integrated systems based on user feedback and business needs.

[0087] The waterlogging model creation system used in urban water management includes:

[0088] Data collection module, used to collect input data of the model;

[0089] Deep learning module, used to train deep learning models;

[0090] Dynamic parameter adjustment module, used to adjust model parameters according to real-time monitoring data;

[0091] Real-time data integration module, used to integrate real-time monitoring data;

[0092] The waterlogging business integration module is used to embed the model output into the waterlogging business for output.

[0093] Compared with the prior art, the present invention has the following beneficial effects:

[0094] The present invention significantly improves the accuracy and real-time performance of the urban waterlogging prediction model by integrating deep learning algorithms and real-time data monitoring technology. Compared with the traditional mechanism-based static model, it can dynamically adjust model parameters and respond to environmental changes in real time, such as changes in rainfall intensity and terrain characteristics, thereby providing a more accurate and timely waterlogging risk assessment.

[0095] The present invention achieves real-time decision support for waterlogging by closely integrating the model output with the urban water management business process. The model output can be directly used to assist in triggering the early warning system, allocating emergency resources, and planning post-disaster recovery work, greatly improving the efficiency and effectiveness of water management personnel in responding to waterlogging events and reducing the economic losses and social impacts caused by waterlogging.

[0096] The present invention designs a flexible parameter adjustment mechanism and a real-time data integration module for the waterlogging model system, so that the model can not only adapt to the specific conditions of different cities, but also continuously optimize itself over time. The system's dynamic parameter adjustment and continuous learning strategy ensure that the model can automatically adjust to maintain the best prediction performance when facing new data and environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings;

[0098] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0099] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0100] It should be understood that the terms "include" and "comprising" used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0101] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and claims, the singular forms of "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0102] like Figure 1 As shown in the figure, the method for creating a waterlogging model for urban water management is as follows:

[0103] Data collection and preprocessing: Collect data on urban topography, land use, meteorology, etc., and preprocess them to provide input for the model;

[0104] Collect urban topographic data using geographic information systems (GIS) and topographic surveying techniques; collect land use classification data, which may include satellite remote sensing data and urban planning data; obtain meteorological data such as rainfall, wind speed, humidity, etc. from weather stations or weather services; check the integrity of the data and remove or correct missing values ​​and outliers. The specific process is as follows:

[0105] Check the size and number of records in the data set to ensure that they meet expectations, confirm that each record contains all necessary fields and that the fields are not empty, ensure that data records remain consistent in format and type, and perform logical checks to ensure that the relationship between data records meets expectations, such as the order of timestamps; use descriptive statistical analysis (such as mean, median, standard deviation) to evaluate data distribution, and use charts (such as histograms, box plots) to visually check the distribution and outliers of the data;

[0106] After determining that the data is missing, determine the proportion and pattern of missing values, understand the randomness or systematicity of data missing, and directly delete the records with missing values ​​if the proportion of missing values ​​in the data set is lower than the preset proportion; when the proportion of missing values ​​in the data set is greater than the preset proportion, use the mean, median, mode or other statistical measures to fill in the missing values. For time series data, forward filling or backward filling can be used;

[0107] Use regression, decision trees, or other machine learning models to predict missing values. For continuous data, use linear interpolation or more complex interpolation methods (such as spline interpolation). Generate multiple complete data sets through simulation, analyze each data set, and finally synthesize the results. After determining that the data is an abnormal problem, use box plots, standard deviation methods, or quantile-based methods to identify outliers, understand the business or field background of the data, and determine whether the outliers are of practical significance.

[0108] If the outliers are due to data entry errors, correct them; if the outliers are not meaningful and there are not many of them, delete them; transform the data (such as logarithmic transformation) to reduce the impact of outliers, keep the outliers in the data set, but perform special treatment during analysis to isolate them; use robust statistical methods that are not affected by outliers for analysis;

[0109] Before determining the above processing methods, evaluate the potential impact of missing or abnormal data on model performance, conduct experimental tests on different processing methods, and decide on the processing strategy based on the importance of the data in the model and its contribution to the prediction results.

[0110] Standardize the data to eliminate the impact of different dimensions and magnitudes;

[0111] Extract features useful for waterlogging prediction from the original data, such as the proportion of impervious area and terrain slope; divide the data into training set, validation set and test set to prepare for model training and evaluation. The specific process is as follows:

[0112] Determine the ratio of training set, validation set and test set according to the data volume and model requirements. The specific split ratio is 70% training set, 15% validation set and 15% test set, or 80% training set, 10% validation set and 10% test set. Before splitting, randomly shuffle the entire data set to ensure the randomness of the data and avoid the deviation caused by data sorting. According to the determined ratio, split the training set, validation set and test set from the shuffled data set.

[0113] Check the data distribution for each data set to ensure that they are similar in statistical characteristics and avoid distribution bias between the training set and the test set; for supervised learning tasks, separate the features (input variables) and labels (output variables) in the data set to ensure that the model can correctly learn the relationship between features and labels; store the split data set in an appropriate format, such as a CSV file, a database, or a specific data structure, for subsequent use, and verify the split data set to ensure that there is no data leakage, that is, the information of the training set cannot be used for the prediction of the test set;

[0114] If the data set is smaller than the preset standard, use cross-validation methods, such as k-fold cross-validation, to improve the stability and reliability of model evaluation; back up the original data set and the segmented data set to facilitate subsequent model training and evaluation. During the model training and evaluation phases, load the corresponding data set to ensure that the data can be correctly input into the model; during the model training process, update the data set as needed, such as adding new data or adjusting the data set ratio, and record the process and results of data set segmentation, including segmentation ratio, data distribution and other information, to facilitate communication between team members and subsequent model reproduction.

[0115] Deep learning model training: Use deep learning algorithms to train models and learn the patterns of waterlogging through a large amount of historical data;

[0116] Select the appropriate deep learning model architecture according to the characteristics of the problem and design the network structure, including the number of layers, number of neurons, activation function, etc. The specific process is as follows:

[0117] Identify the problem type, such as classification, regression, time series prediction or sequence labeling, understand the data type, distribution, feature dimension and sample size, etc., investigate existing solutions and model architectures for similar problems; select basic model types according to the problem type, such as convolutional neural network (CNN) for image data, recurrent neural network (RNN) or long short-term memory network (LSTM) for time series data, and select or design model architectures based on the data's sequential, periodic, spatiotemporal correlation and other characteristics; design an experimental framework, including model comparison, hyperparameter adjustment and performance evaluation, select several potentially suitable model architectures based on the problem characteristics and preliminary understanding, and test the selected model architectures, collect the basic factors and influencing factors of the test for comprehensive analysis, and determine the priorities of all selected model architectures, where the basic factors include:

[0118] Accuracy: the proportion of correct predictions made by the model; Recall: the proportion of positive classes correctly identified by the model; Mean square error: the average of the squares of the differences between the predicted value and the actual value in the regression problem; F1 score: the harmonic mean of the recall rate and the precision rate; After normalizing the obtained accuracy, recall rate, mean square error and F1 score, the product of the accuracy and recall rate is used as the radius of the bottom circle, and the ratio of the F1 score to the mean square error is used as the height, a cone model is established, and the surface area of ​​the cone model is calculated and recorded as the base evaluation value JPZ;

[0119] The influencing factors include: training time: the time required for model training; inference time: the time required for the model to predict a single sample; model size: the number of model parameters; the obtained training time, inference time and model size are normalized and entered into the following formula: The film review value YPZ is obtained, where st, tt and md are the training time, inference time and model size respectively, and the obtained film review value is used as a measure of the priority of the influencing factors on the model architecture;

[0120] Then normalize the obtained basic evaluation value and film review value and enter them into the formula: In order to obtain the comprehensive evaluation value ZPZ, this comprehensive evaluation value ZPZ is used as the evaluation criterion for measuring the priority of the model architecture; the comprehensive evaluation values ​​ZPZ obtained by all the selected model architectures are sorted according to size, and the model architecture with the largest comprehensive evaluation value ZPZ is selected as the optimal choice.

[0121] After the optimal model architecture is determined, the model's loss function, optimizer and evaluation index are configured, and the model is trained using the training set data. The training set input of the model includes urban terrain data, land use data, meteorological data and feature data. The feature data is related to waterlogging prediction through features extracted from the original data, such as the proportion of impervious area and terrain slope. The output of the model includes waterlogging prediction results: the model predicts the possibility, time, location and severity of waterlogging based on the input data; risk assessment: risk assessment based on the prediction results to assist decision-making; early warning information: model output may trigger the early warning system, providing a basis for emergency response and resource allocation; decision support: model output can be directly used to assist the decision-making process of waterlogging management, such as disaster response and resource allocation;

[0122] Optimize network weights through the back-propagation algorithm; adjust hyperparameters such as learning rate and batch size to improve model performance; use the validation set to evaluate model performance and adjust model architecture or parameters; save the trained model for subsequent use and deployment.

[0123] Dynamic parameter adjustment: Design a dynamic parameter adjustment mechanism to dynamically adjust model parameters based on real-time monitoring data;

[0124] The confirmation mechanism determines which model parameters can be dynamically adjusted based on real-time data. The confirmation mechanism process is as follows:

[0125] Determine the working principle and key parameters of the model, including learning rate, regularization coefficient, number of network layers, number of neurons, etc., and then analyze parameter sensitivity, data characteristics, real-time data impact and historical data. Specifically:

[0126] Parameter sensitivity analysis determines which parameters have the greatest impact on model performance through experimental or theoretical analysis. Sensitivity analysis can help identify parameters that have a significant impact on model output; data characteristic analysis is to analyze the characteristics of input data, including its range of variation, periodicity, seasonality, etc., to determine which model parameters may need to be adjusted based on data characteristics; real-time data analysis determines which parameters are closely related to changes in real-time data by evaluating the impact of real-time data on model predictions; historical data analysis simulates real-time data streams using historical data to analyze the performance changes of the model under different parameter settings;

[0127] Then use model diagnostic tools, such as learning curves and performance curves, to identify bottlenecks in model performance, conduct A / B tests on models with different parameter settings, compare their performance when processing real-time data, design feedback mechanisms, and dynamically adjust parameters based on the deviation between model predictions and actual observations; test the model's response to real-time data in a simulated environment, observe which parameters need to be adjusted to adapt to data changes, and apply hyperparameter optimization algorithms, such as Bayesian optimization and genetic algorithms, to find the optimal parameter adjustment strategy;

[0128] Establish a real-time monitoring system to track model performance and trigger parameter adjustments when necessary, evaluate the robustness of the model under different parameter settings, determine under what conditions parameters need to be adjusted, and apply continuous learning strategies so that the model can automatically adjust parameters over time and as data accumulates.

[0129] Analyze the relationship between real-time data and model parameters and determine adjustment rules; develop parameter adjustment strategies based on real-time data, such as threshold triggering and adaptive adjustment; implement parameter update mechanisms to ensure that the model can dynamically adjust parameters based on real-time data; design feedback loops to adjust parameters based on the deviation between model predictions and actual observations.

[0130] Real-time data integration and model calibration: Integrate IoT technology to achieve real-time data collection and online calibration of the model;

[0131] Integrate IoT devices into the monitoring system to collect data in real time; implement data stream processing to ensure that data can be transmitted and processed in real time; calibrate the model online based on real-time data and adjust model parameters; update the model regularly or as needed to adapt to new data and environmental changes; monitor model performance to ensure the effectiveness of real-time data integration and online calibration.

[0132] Waterlogging business integration: Integrate model output with waterlogging business processes to achieve real-time decision support;

[0133] Analyze the business process of waterlogging management and determine how the model output supports decision making. The specific process is as follows:

[0134] Obtain all aspects of waterlogging management, including monitoring, early warning, emergency response, resource allocation, post-disaster assessment and recovery, etc., and identify key decision points in the waterlogging management process, such as when to issue early warnings and how to deploy emergency resources; collect specific needs of waterlogging managers for model outputs, understand what information they want the model to provide to assist in decision-making, evaluate the model's capabilities, and determine what types of outputs it can provide, such as predicting the time, location, and severity of waterlogging; analyze how model outputs are combined with business processes, determine the role and importance of model outputs in the decision-making process, and develop risk assessment methods based on model outputs to help managers understand the possible consequences of different decisions;

[0135] Establish clear decision-making criteria and thresholds. When the model output reaches these thresholds, trigger specific business processes, draw a waterlogging management business process diagram, and clarify the position and role of the model output in the process; simulate different waterlogging situations, use model outputs to guide decisions, analyze the effects of decisions, and develop a feedback mechanism to feed back decision results and actual waterlogging situations to the model in order to continuously adjust and optimize model outputs, and develop update and maintenance plans for model outputs and decision support systems to ensure that the system can adapt to new business needs and technological changes.

[0136] Integrate the model into the existing water management platform; design a user interface to enable managers to easily access model outputs; develop early warning systems, risk assessment tools, etc. to assist in decision-making; conduct simulation exercises to evaluate the impact of model outputs on business processes; and continuously optimize the model and integrated system based on user feedback and business needs.

[0137] A waterlogging model creation system for urban water management, including data acquisition module, deep learning module, dynamic parameter adjustment module, real-time data integration module and waterlogging business integration module;

[0138] The data collection module is used to collect data on urban topography, land use, meteorology, etc.;

[0139] The deep learning module is used to train deep learning models;

[0140] The dynamic parameter adjustment module is used to adjust the model parameters according to the real-time monitoring data;

[0141] The real-time data integration module is used to integrate real-time monitoring data;

[0142] The waterlogging business integration module is used to embed the model output into waterlogging business.

[0143] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. Method for creating waterlogging model for urban water management, It is characterized in that The following steps are involved: Step a. Data collection and preprocessing: Collect urban topography, land use and meteorological data, and preprocess them to provide input for the model; Step b. Deep learning model training: Use deep learning algorithms to train the model and learn the patterns of waterlogging through historical data. The specific steps are as follows: Step b1: Identify the problem type and understand the data characteristics, including the data type, distribution, feature dimension, and sample size; Step b2: Research existing solutions and select deep learning model architecture based on the problem type; Step b3: Design an experimental framework for model comparison, hyperparameter adjustment, and performance evaluation; Step b4: Determine the optimal model architecture, configure the loss function, optimizer and evaluation metrics; Step b5: Use the training set data to train the model and optimize the network weights through the back propagation algorithm; Step b6: Adjust hyperparameters, including learning rate and batch size, to improve model performance; Step b7: Use the validation set to evaluate model performance and adjust the model architecture or parameters; Step b8: Save the trained model for subsequent use and deployment; The process of determining the optimal model architecture is as follows: According to the characteristics of the problem and the initial understanding, several model architectures are preliminarily screened and tested. The basic factors and influencing factors of the test are collected for comprehensive analysis to determine the priority of all selected model architectures. The basic factors include: accuracy, recall, mean square error and F1 score. After normalization, the product of accuracy and recall is used as the base circle radius, and the ratio of F1 score to mean square error is used as the height. A cone model is established, and the surface area of ​​the cone model is calculated and recorded as the base evaluation value JPZ; Influencing factors include: training time, inference time and model size, which are normalized and then put into the following formula: The film review value YPZ is obtained, where st, tt and md are the training time, inference time and model size respectively, and the obtained film review value is used as a measure of the priority of the influencing factors on the model architecture; Then normalize the obtained basic evaluation value and film review value and enter them into the formula: To obtain the comprehensive evaluation value ZPZ, and use this comprehensive evaluation value ZPZ as the evaluation standard for measuring the priority of the model architecture; Sort the comprehensive evaluation values ​​ZPZ of all selected model architectures by size, and select the model architecture with the largest comprehensive evaluation value ZPZ as the optimal choice; Step c. Dynamic parameter adjustment: Design a dynamic parameter adjustment mechanism to dynamically adjust model parameters based on real-time monitoring data. The specific steps are as follows: Determine which model parameters can be dynamically adjusted based on real-time data through a confirmation mechanism; Analyze the relationship between real-time data and model parameters and establish adjustment rules; Develop parameter adjustment strategies, including threshold triggering and adaptive adjustment mechanisms; Utilize the parameter update mechanism to enable the model to dynamically adjust parameters using real-time data; Design feedback loops to automatically adjust parameters based on deviations between model predictions and actual observations; The specific process of the confirmation mechanism is as follows: Determine key model parameters, including learning rate, regularization coefficient, number of network layers and number of neurons; Conduct parameter sensitivity analysis to identify parameters that have a significant impact on model output through experimental or theoretical analysis; Analyze data characteristics, including range, periodicity, and seasonality, to determine parameters that require dynamic adjustment; Assess the impact of real-time data on model predictions and determine parameters associated with data changes; Use historical data to simulate real-time data flow and analyze performance changes under different parameter settings; Apply model diagnostic tools to identify model performance bottlenecks; Conduct A / B testing on models with different parameter settings to compare their performance in processing real-time data; Design feedback mechanisms based on the deviation between prediction and observation to dynamically adjust parameters; Test the model’s response to real-time data in a simulated environment and identify parameters that need to be adjusted; Apply hyperparameter optimization algorithms to find tuning strategies that meet the criteria; Establish a real-time monitoring system to track model performance and trigger corresponding parameter adjustments; Evaluate the robustness of the model under different parameter settings and determine the parameter adjustment conditions; Apply continuous learning strategies to enable the model to automatically adjust parameters over time and accumulate data; Step d. Real-time data integration and model calibration: Integrate IoT technology to achieve real-time data collection and online calibration of the model. The specific process is as follows: Integrate IoT devices into monitoring systems to collect data in real time; Implement data stream processing to enable data to be transmitted and processed in real time; Calibrate the model online based on real-time data and adjust model parameters; Update the model regularly or as needed to adapt to new data and environmental changes; Monitor model performance to make real-time data integration and online calibration effective; Step e. Waterlogging business integration: Integrate the model output with the waterlogging business process to achieve real-time decision support.

2. The method for creating a waterlogging model for urban water management according to claim 1, It is characterized in that The specific process of step a is as follows: Collect urban topographic data using geographic information systems and topographic survey techniques; Collecting land use classification data, including satellite remote sensing data and urban planning data; Obtain weather data on rainfall, wind speed, and humidity from weather stations or weather services; Check the completeness of the data and remove or correct missing values ​​and outliers; Standardize the data to eliminate the impact of different dimensions and magnitudes; Extract features related to waterlogging prediction from raw data; The data is divided into training set, validation set and test set to prepare for model training and evaluation.

3. The method for creating a waterlogging model for urban water management according to claim 2, It is characterized in that The specific steps for checking the integrity of the data and removing or correcting missing values ​​and outliers in step a are as follows: Verify the integrity of the data set, including number of records, field integrity, and data format consistency; Perform logic checks to ensure that the relationships between data records meet pre-set criteria; Apply descriptive statistical analysis and graphical visualization to assess data distribution and identify outliers; For missing data, select deletion or filling strategies based on the proportion and pattern of missing values, where filling strategies include using mean, median, mode, time series filling, regression prediction, or interpolation methods; For outliers, identify them through box plots, standard deviation methods, or quantile methods, and determine their actual significance based on the business context; Correct, delete or isolate outliers, or use robust statistical methods to reduce their impact; Before determining the treatment method, evaluate the impact of missing or abnormal data on model performance, and experimentally test different treatment methods, and select a strategy that meets the standards based on the contribution of the data to the model's prediction results.

4. The method for creating a waterlogging model for urban water management according to claim 1, It is characterized in that The specific steps of dividing the data into a training set, a validation set and a test set in step a are as follows: Determine the ratio of training set, validation set and test set according to the data volume and model requirements, and choose a split ratio of 70% training set, 15% validation set and 15% test set or 80% training set, 10% validation set and 10% test set; Randomly shuffle the entire data set to avoid bias caused by data sorting; Split the training set, validation set, and test set from the shuffled dataset according to the determined ratio; Check the similarity of the statistical characteristics of the segmented data sets to ensure there is no distribution bias; For supervised learning tasks, separate features and labels to ensure that the model learns correctly; Store the segmented dataset in a CSV file or database; Verify the segmented data set to avoid data leakage; When the data set is smaller than the preset standard, the k-fold cross-validation method is used to improve the evaluation stability; Back up the original data set and the split data set; During model training, update the dataset as needed, including adding new data or adjusting the scale; Record the segmentation process and results, including proportions and distribution, for team communication and model reproduction.

5. The method for creating a waterlogging model for urban water management according to claim 1, It is characterized in that The specific process of step e is as follows: Analyze the business process of waterlogging management and determine how model outputs can support decision making; Integrate the model into existing water management platforms; Design user interfaces for managers to access model outputs; Develop early warning systems and risk assessment tools to assist decision making; Conduct simulation exercises to evaluate the impact of model outputs on business processes; Continuously optimize models and integrated systems based on user feedback and business needs.

6. A system for the method for creating a waterlogging model for urban water management as described in any one of claims 1 to 5, It is characterized in that include: Data collection module, used to collect input data of the model; Deep learning module, used to train deep learning models; Dynamic parameter adjustment module, used to adjust model parameters according to real-time monitoring data; Real-time data integration module, used to integrate real-time monitoring data; The waterlogging business integration module is used to embed the model output into the waterlogging business for output.