A data-driven real-time flood forecasting method and system

Abstract

The present invention relates to the technical field of flood forecasting, and specifically to a data-driven real-time flood forecasting method and system, the method comprising the following steps: selecting a monitoring basin range, obtaining initial hydrological information and initial rainfall data of the monitoring basin based on observation sites in the monitoring range; establishing a forecast data optimization model, processing the initial hydrological information and the initial rainfall data through the forecast data optimization model, so as to obtain a flood forecast data set for the monitoring basin; constructing a runoff prediction model based on the flood forecast data set, and using the runoff prediction model to analyze the water flow input at the outlet of the monitoring basin; introducing a neural network model to reconstruct the information of the flood forecast data set and the water flow input, and obtaining a graded forecast data set and a reconstructed monitoring basin input; combining the graded forecast data set, the reconstructed monitoring basin input and a monitoring parameter evaluation system to perform real-time analysis and flood warning on the monitoring basin range.

Description

Technical Field

[0001] The present invention relates to the technical field of flood forecasting, and in particular to a data-driven real-time flood forecasting method and system. Background Art

[0002] As a highly destructive natural disaster worldwide, flood disasters continue to pose a severe challenge to the safety of life, property and economic activities of human society. In my country, affected by the monsoon climate, the uneven distribution of rainfall has aggravated the frequency and severity of flood disasters, which not only caused huge economic losses, but also frequently triggered geological disasters, causing far-reaching impacts on the ecological environment. Traditional flood forecasting methods mainly rely on hydrological data resources and basin physical models, which can restore hydrological physical processes to a certain extent, but cannot adapt to the complex evolution of hydrological elements. In addition, the existing technology lacks effective data mining technology, making it difficult to capture the intrinsic connection between data, thus failing to achieve efficient use of data information and disaster warning.

[0003] In view of the high incidence and suddenness of rainstorm and flood events in the basin, as well as the limitations of the natural regulation capacity of the basin, it is particularly important to build an efficient and accurate flood simulation, forecasting and early warning system. In response to the above problems, it is necessary to propose a data-driven real-time flood forecasting method and system, which cleverly integrates hydrological and physical information with machine learning algorithms to build a comprehensive basin flood early warning system. Based on this, we can deeply analyze and integrate multi-dimensional data such as basin hydrology and meteorology, analyze and predict through intelligent algorithms, realize accurate simulation of basin water volume, significantly improve the basin's ability to respond to natural disasters, and minimize the losses caused by disasters. Summary of the invention

[0004] In view of the shortcomings of existing methods and the needs of practical applications, the present invention fully integrates hydrological physical information and intelligent learning algorithms, and uses intelligent algorithms to analyze, predict and verify the water volume of the basin, so as to achieve accurate simulation and forward-looking forecasting of the water volume of the basin, which can effectively improve the response speed and coping ability of the basin in the face of disasters such as floods. On the one hand, the present invention provides a data-driven real-time flood forecasting method, which includes the following steps: selecting a monitoring basin range, obtaining the initial hydrological information and initial rainfall data of the monitoring basin based on the observation stations in the monitoring range; establishing a forecast data optimization model, and processing the initial hydrological information and the initial rainfall data through the forecast data optimization model to obtain a flood forecast data set for the monitoring basin; constructing a runoff prediction model based on the flood forecast data set, and using the runoff prediction model to analyze the water flow input at the outlet of the monitoring basin; introducing a neural network model to reconstruct the information of the flood forecast data set and the water flow input, and obtaining a hierarchical forecast data set and a reconstructed monitoring basin input; combining the hierarchical forecast data set, the reconstructed monitoring basin input and the monitoring parameter evaluation system to perform real-time analysis and flood warning on the monitoring basin range. The present invention integrates technologies such as hydrological information, learning algorithms and intelligent network models to conduct real-time analysis and early warning of watersheds, provide accurate information for decision makers, and help formulate flood disaster response measures in advance to reduce the impact of disasters.

[0005] Optionally, the establishment of a forecast data optimization model, and processing the hydrological initial information and the rainfall initial data through the forecast data optimization model to obtain a flood forecast data set for the monitored basin includes: establishing a forecast data optimization model based on historical data of the monitored basin, the forecast data optimization model including a data standardization function and a data normalization function; and obtaining a flood forecast data set for the monitored basin using the data standardization function and the data normalization function, the flood forecast data set including the hydrological information and rainfall data of the monitored basin. The present invention uses data standardization and normalization functions to process the hydrological initial information and the rainfall initial data, providing high-quality data support for the construction of a subsequent runoff prediction model.

[0006] Optionally, constructing a runoff prediction model based on the flood forecast data set includes: the runoff prediction model satisfies the following relationship:

[0007]

[0008] in, express The effective runoff input at the basin outlet during the period, express Monitor the rainfall elements in the basin during the period, express The predicted rainfall in the monitored basin during the period, express The impact lag of the monitoring basin within the time period, represents the influence weight of underlying surface factors in the monitored watershed, express The present invention comprehensively considers multiple variables, can more comprehensively reflect the hydrological change process in the basin, improves the accuracy of prediction, can more accurately simulate the law of water flow movement in the basin, helps to reduce prediction errors, and improves the accuracy of flood forecasting.

[0009] Optionally, the use of the runoff prediction model to analyze the water flow input at the outlet of the monitoring basin includes: obtaining the effective runoff input at the outlet of the monitoring basin in any time period through the runoff prediction model; and obtaining the water flow input at the outlet of the monitoring basin based on the effective runoff input. The runoff prediction model of the present invention can receive and process monitoring data in real time, thereby quickly calculating the effective runoff input in any time period, so that the model can quickly respond to hydrological changes in the basin and provide timely information support for flood warning and emergency response.

[0010] Optionally, the introduction of the neural network model to reconstruct the information of the flood forecast data set and the water flow input, and obtain the graded forecast data set and the reconstructed monitoring basin input includes: the neural network model combines the time series characteristics to reconstruct the information of the flood forecast data set to obtain the graded forecast data set. The neural network model of the present invention has a strong nonlinear fitting ability and self-learning ability, can capture the complex time series characteristics and spatial distribution laws in the flood forecast data set, learn and reconstruct the hydrological information characteristics, and can more accurately simulate the hydrological process in the basin, thereby improving the accuracy of flood forecasting.

[0011] Optionally, the introduction of the neural network model to reconstruct the information of the flood forecast data set and the water flow input, and obtaining the graded forecast data set and the reconstructed monitoring basin input includes: optimizing the runoff prediction model based on the graded forecast data set, the neural network model and the water flow input, and obtaining the optimized runoff prediction model. The neural network model of the present invention can reconstruct the information of the flood forecast data set based on the time series characteristics, and then identify the key factors affecting the runoff, thereby improving the prediction accuracy of the runoff prediction model, and improving the practicality and flexibility of the data-driven real-time flood forecasting method.

[0012] Optionally, the optimized runoff prediction model satisfies the following relationship:

[0013]

[0014] in, After the neural network is optimized Monitor the flow of the basin during the period, It means that based on the neural network Time segmentation time interval, express The predicted rainfall in the monitoring basin at any time interval within the time period, represents the water balance coefficient of the monitored basin, express The effective runoff input of the monitoring basin at any time interval within the time period, express The average flow rate of the monitoring basin at any time interval within the time period, express The flow loss parameters of the basin are monitored during the period. The optimized model of the present invention reconstructs the information of the flood forecast data set and the water flow input through a neural network, which can capture the nonlinear relationship and hidden patterns in the data, so as to accurately reflect the actual hydrological process in the basin, thereby improving the accuracy of the prediction results.

[0015] Optionally, the real-time analysis and flood warning of the monitoring basin range in combination with the hierarchical forecast data set, the reconstructed monitoring basin input and the monitoring parameter evaluation system comprises: constructing a monitoring parameter evaluation system based on historical information of the monitoring basin range, the monitoring parameter evaluation system comprising a basin water storage capacity evaluation model, a monitoring parameter stability evaluation model, a monitoring parameter correlation evaluation model and a monitoring parameter error analysis model;

[0016] The watershed water storage capacity assessment model satisfies the following relationship:

[0017]

[0018] in, represents the rate of change of water storage in the monitored basin, represents the storage area of ​​the monitored basin, Indicates monitoring basin Historical traffic for a time period, Indicates the monitoring time of the monitoring basin;

[0019] The monitoring parameter stability evaluation model satisfies the following relationship:

[0020]

[0021] in, express The stable value of the monitoring parameter during the period, express The measured value of the monitoring parameter during the period, express The predicted value after monitoring parameters optimization within the time period, express The measured average value of the monitoring parameter during the period, Indicates the number of measurements of the monitoring parameter;

[0022] The monitoring parameter correlation evaluation model satisfies the following relationship:

[0023]

[0024] in, express The correlation coefficient of the monitoring parameters within the time period, express The measured value of the monitoring parameter during the period, express The predicted value after monitoring parameters optimization within the time period, express The measured average value of the monitoring parameter during the period, express The predicted average value after the monitoring parameters are optimized within the period, Indicates the number of measurements of the monitoring parameter;

[0025] The monitoring parameter error analysis model satisfies the following relationship:

[0026]

[0027] in, express The absolute error value of the monitoring parameter within the time period, express The measured value of the monitoring parameter during the period, express The predicted value after the optimization of monitoring parameters within the time period;

[0028]

[0029] in, express The relative error value of the monitoring parameters within the time period, express The measured value of the monitoring parameter during the period, express The predicted value after the monitoring parameters are optimized within the time period. The monitoring parameter evaluation system constructed by the present invention can comprehensively evaluate the hydrological conditions of the monitored basin, reflect the changes in water flow in the basin in real time, and provide accurate data support for flood warning. At the same time, combined with the hierarchical forecast data set, flood warnings of different time scales and different precisions can be achieved, improving the timeliness and accuracy of the warning method.

[0030] Optionally, the real-time analysis and flood warning of the monitoring basin range in combination with the hierarchical forecast data set, the reconstructed monitoring basin input and the monitoring parameter evaluation system includes: real-time analysis and flood warning of the monitoring basin based on the hierarchical forecast data set, the reconstructed monitoring basin input, the water storage capacity change rate, the stability value, the correlation coefficient, the absolute error value and the relative error value. The real-time analysis and flood warning system of the present invention can quickly process monitoring data and issue warning information based on the latest prediction results and evaluation indicators, which is helpful to take timely flood disaster response measures and reduce economic losses caused by disasters.

[0031] In the second aspect, in order to efficiently execute a data-driven real-time flood forecasting method provided by the present invention, the present invention also provides a data-driven real-time flood forecasting system, including a processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the data-driven real-time flood forecasting method as described in the first aspect of the present invention. The data-driven real-time flood forecasting system of the present invention has a compact structure and stable performance, and can stably execute the data-driven real-time flood forecasting method provided by the present invention, thereby improving the overall applicability and practical application capability of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flow chart of the data-driven real-time flood forecasting method of the present invention;

[0033] Figure 2 A schematic diagram of a flood forecast data optimization process in a data-driven real-time flood forecast method of the present invention;

[0034] Figure 3 A schematic diagram of a process for constructing a hierarchical forecasting data set in the data-driven real-time flood forecasting method of the present invention;

[0035] Figure 4 This is a structural diagram of the data-driven real-time flood forecasting system of the present invention. DETAILED DESCRIPTION

[0036] See also Figure 1The present invention integrates hydrological information and machine learning technology to construct an intelligent mechanism for analyzing, predicting and verifying water volume in a river basin. It not only achieves high-precision simulation of water volume in a river basin, but also has efficient disaster forecasting capabilities, significantly enhancing the response speed and response capabilities of a river basin in the face of natural disasters such as floods. The present invention provides a data-driven real-time flood forecasting method, which includes the following steps:

[0037] S1. Select the monitoring basin range, and obtain the initial hydrological information and initial rainfall data of the monitoring basin based on the observation stations in the monitoring range. The specific implementation steps and contents are as follows;

[0038] A monitoring basin is randomly determined, and the boundary of the monitoring basin is further clarified. First, the boundary of the monitoring basin is clearly defined according to the geographical coordinates, administrative divisions or natural terrain characteristics of the monitoring basin. In the embodiment, map analysis, geographic information system (GIS) technology and on-site investigation and other means can be used. Then, the monitoring basin characteristics are considered, and the natural characteristics of the monitoring basin such as topography, landform, soil type, vegetation coverage, etc. are analyzed. The influence of human activities such as reservoirs, irrigation, urbanization and other factors on the hydrological process of the monitoring basin is comprehensively considered. The above-mentioned influencing factors can provide a basis for the layout and data analysis of the basin monitoring stations.

[0039] Observation sites are selected based on the scope of the monitoring basin. Reasonable layout of information observation sites in the monitoring basin ensures that the key areas and major hydrological processes of the monitoring basin are fully covered. The layout of the sites should fully consider factors such as terrain changes, water flow convergence paths, and rainfall distribution in the monitoring basin. According to the monitoring needs and basin characteristics, different types of observation stations are installed, such as rain gauges, water level stations, flow stations, etc. Rain gauges can be used to monitor rainfall, water level stations can monitor water level changes in rivers, lakes and other water bodies, and flow stations can preliminarily measure water flow. Finally, each observation site is equipped with appropriate monitoring equipment, including but not limited to automatic rain gauges, water level meters, flow meters, etc., and ensure that the equipment has high precision, high stability and remote transmission capabilities.

[0040] Based on the observation stations in the monitoring range, the initial hydrological information and initial rainfall data of the monitored basin are obtained. In an optional embodiment, based on the observation stations within the monitoring basin, the hydrological information, meteorological data and other information of the monitored basin are collected in real time or at a predetermined period. The above data mainly cover the basic parameters of the basin's hydrological conditions, such as rainfall, water level, flow, etc. In order to more comprehensively capture the basin's environmental characteristics and its impact on the hydrological process, meteorological information such as temperature, humidity, and wind speed are also collected, which is helpful for in-depth analysis of the hydrological cycle links such as water and heat balance and evaporation and transpiration in the monitored basin, thereby further improving the prediction accuracy and foresight of basin changes.

[0041] Furthermore, as one of the core elements of hydrological monitoring, water level data records the water level heights that change over time at key monitoring stations in the basin. The relevant data points not only reflect the real-time status of water bodies such as rivers and lakes, but also provide a crucial baseline for flood warning and water resources management. Through continuous monitoring and data analysis, water level dynamics can be captured in a timely manner to predict potential hydrological disaster risks.

[0042] Flow data mainly focuses on the flow changes of rivers, channels and other water bodies in different time periods, and is an important basis for evaluating water resources, water flow speed and river conditions. Through the collection and analysis of flow data, we can deeply understand the flow characteristics of water bodies and provide scientific decision-making support for water conservancy project construction, flood scheduling and ecological environmental protection.

[0043] Water storage capacity monitoring is mainly aimed at water storage bodies such as reservoirs and lakes, and water storage capacity monitoring is an indispensable part. By regularly measuring and recording the changes in water storage capacity, the storage capacity of the water body can be accurately grasped, and reasonable water resource allocation plans can be formulated accordingly, which can also ensure the sustainable use of river basin resources.

[0044] Although water quality parameters are not directly involved in flood forecasting, they are an important reference for assessing the health of water bodies and protecting the water ecological environment. Water quality parameters include but are not limited to temperature, pH value, dissolved oxygen, turbidity, nutrient concentration and heavy metal content. Including the above parameters in the data set will help to fully understand the water quality status and provide strong support for water pollution control, water resource protection and ecological balance maintenance. At the same time, changes in water quality may also indirectly affect the hydrological cycle and provide additional auxiliary information for flood prediction.

[0045] Precipitation can mainly record the rainfall at each monitoring station or area in the basin, including but not limited to key indicators such as total rainfall, rainfall intensity, and rainfall duration; rainfall distribution requires the use of methods such as spatial interpolation to convert discrete rainfall station data into a continuous rainfall distribution map to more comprehensively reflect the rainfall conditions in the basin; rainfall types such as heavy rain and continuous rainfall are classified according to rainfall characteristics, which will help to more accurately identify different types of flood events later.

[0046] Other relevant information includes: meteorological data such as temperature, humidity, wind speed, etc. Although the above meteorological factors do not directly constitute the main input of flood forecasting, they have an indirect impact on the hydrological process in the basin; geographic information mainly includes the basin's topography, water system distribution, soil type and other geographic information, which helps to more accurately simulate the hydrological process in the basin; historical flood event records include the time, location, scale, impact range and other information of historical floods in the monitored basin, which helps to verify and calibrate subsequent prediction results.

[0047] In this embodiment, a corresponding data collection system is set up, which can not only grasp the initial hydrological information and initial rainfall data of the monitored basin in real time, but also provide solid data support and scientific basis for subsequent flood warning, water resources management and ecological protection. In addition, the observation station stores the monitored data in the system database for subsequent analysis, processing and query. The database of the embodiment should have good data management capabilities to support rapid retrieval and efficient processing of data.

[0048] Furthermore, the method for obtaining monitoring basin information and data in the present embodiment is only an optional condition of the present invention. In other embodiments, the method for obtaining basin information can be flexibly adjusted according to the actual basin monitoring situation and data collection conditions. Different basins have different hydrological characteristics and monitoring requirements due to differences in their geographical locations, climatic conditions, topography and other factors. The method for obtaining basin information can be flexibly adjusted according to actual conditions to ensure that the monitoring plan can closely fit the basin characteristics and improve the effectiveness and pertinence of data monitoring results.

[0049] S2. Establish a forecast data optimization model. Use the forecast data optimization model to process the initial hydrological information and initial rainfall data of the monitored basin to obtain the flood forecast data set of the monitored basin. The specific implementation steps and contents are as follows:

[0050] A forecast data optimization model is established based on the historical data of the monitoring basin range. The forecast data optimization model in the embodiment mainly includes a data standardization function and a data normalization function, and its specific contents are as follows:

[0051] See also Figure 2 , Schematic diagram of the optimization process of forecast data. In order to further enhance the effectiveness of the data-driven real-time flood forecasting system, the historical data within the monitoring basin are analyzed in this embodiment, based on which the comparability and practicality of the initial monitoring data can be improved. Data standardization and normalization technical means are integrated into the process, and the corresponding data processing process is set. The above process not only effectively simplifies the differences between data, but also promotes the system's ability to capture information on hydrological changes in the basin, thereby ensuring that the flood forecasting model can output more accurate and robust prediction results, laying a solid foundation for in-depth analysis and timely warning of basin disasters. Among them, the data standardization function can adjust the data scale to make it conform to a specific statistical distribution such as a standard normal distribution, thereby eliminating the deviations that may be caused by differences in dimensions or numerical ranges between different variables. In the embodiment, the mean and standard deviation of each variable are calculated, and the original data is converted into a standardized form accordingly, which can ensure that each variable contributes its information fairly during the real-time flood forecasting process.

[0052] The above data normalization function satisfies the following relationship:

[0053]

[0054] in, represents the standardized monitoring basin dataset, represents the original dataset of the monitored watershed, represents the average value of the input data, Represents the standard deviation of the input data.

[0055] The average value of the above input data satisfies the following relationship:

[0056]

[0057] in, represents the average value of the input data, represents the number of monitoring basin datasets, Represents the data of any observation station in the monitoring basin.

[0058] The standard deviation of the input data satisfies the following relationship:

[0059]

[0060] in, represents the standard deviation of the input data, represents the number of monitoring basin datasets, Represents the data of any observation station in the monitoring basin, Represents the mean of the input data.

[0061] In view of the significant difference in dimensions between the initial hydrological data of the monitored basin and the initial rainfall data, a data normalization function is designed in the embodiment to deal with the above problem. Different from the data standardization method, the normalization process focuses more on converting the monitored data to a unified numerical interval, specifically scaling it to the range of [0, 1]. This conversion step can put the original data on the same scale, thereby promoting effective comparison and collaborative calculation between data features of different magnitudes.

[0062] Data normalization can not only simplify the model's data processing complexity, but also significantly improve the training efficiency of the prediction model and accelerate the convergence speed. In addition, the consistent processing of data also enhances the model's sensitivity to data changes, which helps to explore deeper hydrological change patterns and thus improve the accuracy and reliability of flood forecasting. Data normalization is a key link in optimizing the performance of the real-time flood forecasting system and improving the prediction accuracy.

[0063] The above data normalization function satisfies the following relationship:

[0064]

[0065] in, represents the monitoring basin dataset after normalization. represents the original dataset of the monitored watershed, Represents the minimum value in the input data. Indicates the maximum value in the input data.

[0066] The flood forecast data set of the monitored basin is obtained by using the above data standardization function and data normalization function. In the embodiment, the flood forecast data set mainly includes the hydrological information and rainfall data of the monitored basin, and its specific content is as follows:

[0067] In an optional embodiment, based on step S1, it can be known that the water level data mainly records the water level heights at different time points at key hydrological sites in the monitored basin, such as river sections and in front of reservoir dams. After data standardization and normalization, the above data eliminates the differences between different sites due to different measuring equipment and units, and can more accurately capture the changing patterns of water levels in the monitored basin.

[0068] As a key indicator for measuring the dynamic flow of water bodies within a basin, flow data reveals the flow conditions of water bodies. After the data processing process, flow data has higher comparability and can maintain a consistent measurement benchmark across sites and time periods. It is conducive to a comprehensive assessment of the drainage capacity of the basin and further understanding of the potential for water flow and dispersion in the basin when it encounters rainfall or other water inputs.

[0069] Furthermore, the analysis of historical flow data can identify the regularity and trends of flow changes in the basin, which helps to accurately predict the peak flow of floods. The real-time update and precise analysis of flow data in the flood warning system provides a scientific basis for the timely formulation of flood control strategies, allocation of resources and evacuation of people, which can effectively reduce the risks and losses of flood disasters.

[0070] Precipitation, as a key indicator to measure precipitation conditions in a river basin, records the precipitation conditions at each monitoring station or monitoring area. It covers multiple dimensions such as total rainfall, rainfall intensity and rainfall duration. The above data is not only the basis for climate analysis, but also an indispensable input information for flood forecasting systems.

[0071] The processing and standardization of rainfall data is achieved through the forecast data optimization model, ensuring that rainfall data between different monitoring stations and in different time periods can be compared and analyzed under the same metric standard, thereby improving the comparability and consistency between data, and more accurately capturing the temporal and spatial distribution characteristics of rainfall, providing more solid data support for flood prediction.

[0072] In addition, rainfall data can help improve the accuracy and timeliness of flood forecasts. By analyzing rainfall data through relevant models, we can better understand the impact of rainfall processes on the hydrological conditions of the basin, predict more accurate flood flow and peak time, and provide strong support for flood prevention and disaster reduction work. Combining the spatial and temporal distribution of rainfall and spatial interpolation methods, discrete rainfall station data is converted into a continuous rainfall distribution map, which intuitively displays the spatial distribution and temporal changes of rainfall in the basin, which is conducive to understanding the impact of rainfall processes on basin hydrology.

[0073] The classification of rainfall types and characteristics can be used to classify rainfall into various types such as heavy rain and continuous rainfall according to the differences in its intensity and duration. The above classification method is not only conducive to the understanding of rainfall phenomena, but also provides a classification basis for the data-driven real-time flood forecasting system. By accurately identifying different types of rainfall events, the system can more accurately predict the possibility, scale and impact range of floods, thereby making more accurate analysis and scientific warnings.

[0074] In addition, although meteorological data is not directly used as the main input parameter for flood forecasting as system auxiliary information, it plays an indispensable role in understanding hydrological processes such as water and heat balance and evaporation and transpiration in the basin. Changes in meteorological elements such as temperature, humidity, wind speed, and wind direction directly affect the water cycle and energy exchange in the basin, and thus affect the formation and development of floods. Therefore, incorporating the above meteorological data into the consideration of flood forecasting will help improve the comprehensiveness and accuracy of forecast results and provide more comprehensive decision-making support for flood prevention and disaster reduction. Geographic information mainly includes the topography, water system distribution, soil type, etc. of the basin, which helps data-driven real-time flood forecasting methods to more accurately simulate and monitor hydrological processes in the basin.

[0075] By collecting, preprocessing and integrating the above data, a complete flood forecasting dataset can be constructed. The above dataset will be used as the input of the flood forecasting model, and can be used to train the model, simulate and predict floods, and achieve timely warning and risk assessment of flood events.

[0076] Furthermore, in the present embodiment, the processing method of the flood forecast data of the monitoring basin is only an optional condition of the present invention. In one or some other embodiments, the forecast data optimization processing method can be replaced according to the specific conditions of the monitoring basin and the data optimization standard. Different monitoring basins have their own unique hydrological characteristics, meteorological conditions and geographical environment. Timely replacement or adjustment of the forecast data optimization processing method can more flexibly adapt to actual differences, help to accurately reflect the actual conditions of the basin, and improve the accuracy and reliability of the forecast results.

[0077] S3. Construct a runoff prediction model based on the flood forecast data set, and use the runoff prediction model to analyze and monitor the water flow input at the outlet of the basin. The specific implementation steps and contents are as follows:

[0078] First, a runoff prediction model is constructed based on the flood forecast data set of the embodiment.

[0079] In this embodiment, feature selection is performed based on the water volume of the monitored basin, that is, according to the needs of the runoff prediction model, features that have a significant impact on the runoff input, such as rainfall, rainfall factors, evaporation, and soil factors, can be selected. In addition, the data set can also be divided into a training set, a validation set, and a test set for subsequent model training, validation, and testing.

[0080] The appropriate runoff prediction model is selected according to the complexity of the actual situation of the monitored basin and the characteristics of the data set. The model includes but is not limited to linear regression, nonlinear regression such as polynomial regression, machine learning models such as random forest, gradient boosting tree, neural network, etc., and distributed hydrological models based on physical processes.

[0081] According to the selected model type and using the training data, the model is constructed, which includes setting parameters of the runoff prediction model such as learning rate, number of iterations, hidden layer size, etc., and defining the relationship between input features and output targets. In the embodiment, the model is trained using a training data set, and the model parameters are adjusted using optimization algorithms such as gradient descent and Newton's method to minimize the prediction error of the model. At the same time, during the training process, the validation set can be used to monitor the performance of the model to avoid overfitting or underfitting.

[0082] The above runoff prediction model satisfies the following relationship:

[0083]

[0084] in, express The effective runoff input at the basin outlet during the period, express Monitor the rainfall elements in the basin during the period, express The predicted rainfall in the monitored basin during the period, express The impact lag of the monitoring basin within the time period, represents the influence weight of underlying surface factors in the monitored watershed, express Monitor the evaporation in the basin during the period.

[0085] The effective runoff input at the outlet of the basin in any period of time refers to the total amount of water that flows into the river system through the surface and underground pathways of the basin and finally flows out of the outlet section of the basin in any period of time. It can directly reflect the results of the combined effects of precipitation, evaporation, soil infiltration, plant interception, and depression storage in the monitored basin. The effective runoff input is a dynamic process and is affected by many factors, including climate conditions, topography, vegetation cover, soil type, etc. Therefore, when conducting runoff prediction and analysis, it is necessary to comprehensively consider the changing laws and interactions of the above factors.

[0086] Monitoring rainfall factors in a watershed at any time period comprehensively considers the influence of various characteristics of rainfall such as intensity, distribution, rainfall type, and rainfall frequency. Different factors affect the rainfall process and its impact on the watershed to varying degrees. When monitoring and analyzing rainfall factors, meteorological observation stations, radars, satellites and other observation methods can be used to obtain real-time or historical rainfall data, and mathematical models and statistical analysis methods can be used to process and analyze the data to obtain more accurate and comprehensive rainfall information.

[0087] The predicted rainfall in the monitored basin during any period of time refers to the total rainfall that falls at a certain point, area or region during a certain period of time. The rainfall is a direct indicator for assessing the intensity and total amount of rainfall.

[0088] The impact lag of monitoring a watershed in any period of time refers to the time required for water sources such as rainfall or snowmelt in the watershed to be converted into runoff and reach the outlet section of the watershed, also known as the confluence time or watershed lag. In the embodiment, the time interval is affected by many factors, including but not limited to the topography, soil type, vegetation cover, rainfall characteristics, and hydrogeological conditions of the watershed. The monitoring and evaluation of the above-mentioned watershed lag is of great significance for formulating flood prevention plans and optimizing water resources allocation. By understanding the changing laws of watershed lag, the time and scale of flood disasters can be more accurately predicted, thereby formulating more scientific and reasonable flood prevention plans.

[0089] The influence weight of the underlying surface factors in the monitoring basin refers to the relative size of the role played by different underlying surface factors including topography, vegetation, soil, geology, etc. on the changes in hydrological elements such as runoff, water level, and water quality at the outlet of the basin when evaluating the parameters such as the basin's hydrological process, water balance, and flood forecast. In the embodiment, the weight can be obtained through data analysis, model simulation, etc., and is used to quantify the contribution of each underlying surface factor to the basin's hydrological cycle. In actual application, the underlying surface conditions of different basins are different, so the influence weights of each factor will also be different. It is necessary to determine the influence weights of each underlying surface factor based on the characteristics of the monitored basin and the monitoring data requirements, so as to more accurately evaluate the characteristics of the basin's hydrological process and water balance.

[0090] Monitoring the evaporation of a watershed in any period of time refers to the total amount of water in the watershed that is converted into water vapor and escapes into the atmosphere through evaporation during a specific period of time. Based on this, the water loss and water balance in the watershed can be measured. In the embodiment, the water surface evaporator of the meteorological station can be used to continuously observe and record it; remote sensing technology can also be used to obtain detailed information on the underlying surface of the watershed, and the evaporation can be estimated in combination with meteorological data and models; isotope tracing and other methods can also be used to study the water cycle and evaporation process in the watershed.

[0091] In the embodiment, the evaporation amount in any period of time in the monitoring basin satisfies the following relationship:

[0092]

[0093] in, express Monitor the evaporation of the basin during the period. represents the net radiation of the canopy surface, It indicates the average wind speed at a height of 3 meters in the monitoring basin. express The average temperature of the monitoring basin during the period, express The average temperature of the monitoring basin during the period, represents the saturated water vapor pressure, represents the actual water vapor pressure, represents the slope of the saturated water vapor pressure curve, Represents the thermometer constant.

[0094] Monitoring the net radiation of the canopy surface in a watershed refers to the process of observing and recording the net radiation received by the vegetation canopy surface in a specific monitoring watershed. The net radiation of the canopy surface is an important component of the energy exchange between vegetation and the external environment. It reflects the comprehensive effects of absorption, reflection and emission of solar radiation and atmospheric radiation, and can provide feedback on the actual net energy input obtained by the canopy surface. It is of great significance for monitoring processes such as energy balance, photosynthesis, transpiration, and water and heat cycle within the watershed.

[0095] The average wind speed at a height of 3 meters in the monitored basin refers to the average wind speed value calculated by continuously observing the wind speed at a certain height in the basin, which is 3 meters in this embodiment, within a given time period. The above average wind speed value reflects the average rate of air flow at a specific height and time period.

[0096] The average temperature of the monitored basin during the last time period refers to the average temperature value calculated by continuous observation of temperature data at multiple observation points in the basin or in the entire basin during the last time period immediately before the current time. The average temperature value reflects the overall temperature condition in the basin during this time period. The length of the last time period at the current time needs to be determined according to specific needs and purposes. The time period can be the last hour, last day, last week or last month, etc., and needs to be corrected and determined according to the specific scope and needs of the specific monitored basin area.

[0097] Saturated water vapor pressure refers to the water vapor pressure in moist air when the water vapor in the air is in phase change equilibrium with the water or ice below it at a certain temperature. At this time, the moist air is called saturated moist air relative to the water surface or ice surface, and its water vapor pressure is the saturated water vapor pressure of the water surface or ice surface. The concept of saturated water vapor pressure is widely used in meteorological forecasting, hydrological cycle, ecological and environmental protection and other fields. In meteorological forecasting, changes in saturated water vapor pressure can reflect changes in the water vapor content in the atmosphere, and thus predict weather phenomena such as precipitation.

[0098] Actual water vapor pressure refers to the water vapor pressure of unsaturated air. The partial pressure generated by water vapor in the atmosphere can indirectly indicate the amount of water vapor in the atmosphere. The greater the water vapor pressure, the more water vapor there is in the air; conversely, the smaller the water vapor pressure, the less water vapor there is in the air. In practical applications, actual water vapor pressure is an important parameter in meteorological observation and climate research. By measuring and recording the changes in actual water vapor pressure, we can understand the temporal and spatial distribution characteristics of water vapor in the atmosphere, and then analyze its impact on weather, climate and hydrological processes.

[0099] The slope of the saturated temperature water vapor pressure curve, also known as the slope of the saturated water vapor pressure curve, is an important parameter describing the vaporization process. It reflects the rate at which water vapor pressure changes with temperature, that is, the change in water vapor pressure for every 1°C change in temperature. It is of great significance for understanding the water vapor cycle in the atmosphere, calculating evaporation, and predicting weather phenomena such as precipitation.

[0100] The calculation formula for the slope of the above saturated water vapor pressure curve is as follows:

[0101]

[0102] in, represents the slope of the saturated water vapor pressure curve, Represents temperature. The above formula describes the nonlinear relationship between saturated water vapor pressure and temperature at different temperatures, that is, the saturated water vapor pressure increases exponentially with increasing temperature.

[0103] The thermometer constant refers to the time constant of the thermometer, also known as the thermal time constant, which refers to the time required for the output temperature of the temperature sensor to rise to 63.2% of the final value when the temperature of the temperature sensor changes step by step. The time constant reflects the response speed of the temperature sensor to temperature changes.

[0104] The runoff prediction model constructed in the embodiment can effectively calculate the effective runoff input at the outlet of the basin. The above prediction model not only takes into account the influence of natural processes such as rainfall, evaporation, and infiltration on the formation of runoff, but also incorporates spatial factors such as topography, soil, and vegetation in the monitoring basin, thereby obtaining a more accurate and practical runoff prediction result.

[0105] Then, the effective runoff input at the outlet of the monitoring basin in any period of time is obtained through the above runoff prediction model; and the water flow input at the outlet of the monitoring basin is obtained based on the effective runoff input, and the specific contents are as follows:

[0106] First, the prepared data such as rainfall data, evaporation data, soil moisture data, etc. are input into the runoff prediction model. The model will calculate the effective runoff input in any time period based on the above data and built-in algorithms.

[0107] After obtaining the effective runoff input predicted by the model, historical data or an independent validation data set is needed to verify the model's prediction results. By comparing the differences between the model prediction values ​​and the actual observations, the model's prediction accuracy and reliability can be further evaluated. If the prediction error is large, the model needs to be further calibrated and adjusted. After confirming that the model's prediction results are reliable, the effective runoff input is regarded as the water flow input at the outlet of the monitoring basin.

[0108] Since the implementation of the runoff prediction model involves complex physical processes and calculation processes, a simplified pseudo code framework is provided in this embodiment to illustrate the prediction calculation process of the runoff prediction model. The specific calculation process is as follows:

[0109] Runoff prediction model function , which accepts a series of input parameters and returns the effective runoff input ( .

[0110] #The internal calculation process of the model, including multiple physical equations and algorithms

[0111] # Returns the effective runoff input

[0112] return effective_runoff_input

[0113] # Prepare input data

[0114] precipitation_data = [...] # precipitation data

[0115] evaporation_data = [...] # Evaporation data

[0116] soil_moisture_data = [...] # Soil moisture data

[0117] soil_moisture_data = [...] #Other data

[0118] # Call the model for prediction

[0119] # Output results

[0120] effective_runoff_input= , the effective runoff input at the basin outlet.

[0121] The prediction results of the runoff prediction model help us understand the changing trends and characteristics of the runoff input at the outlet of the detection basin, and can be directly applied to flood warning, water resources management, flood control scheduling and other fields, providing a scientific basis for relevant decision-making.

[0122] Furthermore, in the present embodiment, the analysis method for monitoring the water volume at the outlet of the river basin is only an optional condition of the present embodiment. In one or some other embodiments, the analysis method for the water volume of the river basin can be adjusted according to the actual situation of the monitored river basin and the target requirements of data analysis. Selecting the most suitable analysis method for the river basin can improve the accuracy of water volume monitoring and analysis.

[0123] S4. Introduce a neural network model to reconstruct the flood forecast data set and water flow input in the embodiment, and obtain a hierarchical forecast data set and a reconstructed monitoring basin input. The specific implementation steps and contents are as follows:

[0124] First, the above neural network model combines the time series characteristics to reconstruct the flood forecast data set to obtain a hierarchical forecast data set, the specific contents of which are as follows:

[0125] Extract the data features of the flood forecasting dataset in time series. Extract the features useful for flood forecasting based on the time series data, including but not limited to the processed rainfall, water level change trend, flow peak, soil saturation, etc.

[0126] In addition, the spatial characteristics of the watershed in the data are extracted, and the geospatial information is obtained from the flood forecast dataset to extract the spatial characteristics related to the occurrence of floods, such as terrain slope, water system density, land use type, etc.

[0127] The flood forecast dataset is reconstructed based on the neural network model. A suitable neural network model is selected according to the complexity of flood prediction and the characteristics of the flood forecast dataset. In an optional embodiment, the flood forecast dataset is reconstructed as follows:

[0128] The processed flood forecast dataset is input into the autoencoder model, which consists of two parts: an encoder and a decoder. The encoder is responsible for compressing the flood forecast dataset into a low-dimensional hidden vector, and the decoder recovers the original forecast data from the hidden vector.

[0129] The autoencoder model is trained by minimizing the error between the reconstructed data and the original predicted data as the loss function. During the above training process, the key parameters of the encoder and decoder need to be continuously adjusted until the preset convergence conditions or training rounds are reached.

[0130] After training, the autoencoder model is able to learn an effective low-dimensional representation of the flood forecast data, namely the latent vector, which captures the key features of the original flood forecast data.

[0131] In the flood forecasting scenario, in order to more effectively understand and classify flood disaster events, the embodiment uses the autoencoder model in the neural network model to extract and organize the key features in the flood forecast data, and represents it as a low-dimensional hidden vector, which not only captures the key information of the original data, but also reduces the data dimension, which facilitates the efficient processing and practical application of the subsequent data-driven real-time flood forecasting method.

[0132] Next, the K-means clustering unsupervised learning method is used to conduct an in-depth analysis of the hidden vectors output by the autoencoder model. The K-means clustering algorithm divides the hidden vector set into K clusters in an iterative manner. The vectors in each cluster are close to each other in the feature space, while the vectors between different clusters are relatively far away. Each cluster is represented by a cluster center, which is the mean of all vectors in the cluster. The above cluster center can reflect the typical or key features of different clusters.

[0133] In the data-driven real-time flood forecasting method, the hidden vectors are divided into different categories through K-means clustering. The above categories are mainly based on the time series characteristics of the monitored basin. At the same time, each category corresponds to flood data information in a specific time interval, including but not limited to the scale, intensity, duration and potential impact area of ​​the flood.

[0134] Based on this, not only the effective classification of flood forecast data is achieved, but also an important reference basis is provided for subsequent flood risk assessment, warning issuance and emergency response. If there is new flood forecast data in the system, its hidden vector can be extracted through the autoencoder model first, and then compared with the distance or similarity of the existing cluster center, so that the flood category to which the time interval data belongs can be quickly determined, and corresponding countermeasures can be taken accordingly. Furthermore, when new flood forecast data is received, the hidden vector of the new data is compared with the previous cluster center, and the distance or similarity between the hidden vector and each cluster center can be calculated to achieve it. Based on the comparison result, the hidden vector of the new data is divided into the cluster to which the nearest or most similar cluster center belongs, which realizes the accurate classification and rapid construction of flood forecast data in the corresponding time interval.

[0135] The flood forecast dataset is reconstructed and classified based on cluster centers. The cluster centers are used as input and reconstructed through the decoder of the autoencoder model to obtain the cluster center representation in the original data space. The reconstructed cluster centers can be used as a direct basis for flood forecast data in different time intervals.

[0136] In order to further utilize the cluster centers of different time intervals, they are reconstructed as input through the decoder part of the autoencoder model. The decoder can generate corresponding representations in the original flood forecast data space based on the input cluster centers. The above process not only realizes the mapping from low-dimensional hidden space to high-dimensional original data space, but also retains the flood characteristic information represented by the cluster centers.

[0137] The reconstructed graded forecast dataset has a clear physical meaning in the original data space. It directly corresponds to the flood forecast data in different time intervals and can be used as a direct basis in subsequent practical applications for the rapid classification, grading and warning issuance of flood forecast data.

[0138] In this embodiment, the reconstruction method of the flood forecast dataset can manage and utilize the flood forecast dataset more efficiently. When new observation data arrives, it can be compared with the reconstructed cluster center first to quickly determine the flood category and level to which it belongs, so as to take corresponding response measures. This not only improves the accuracy and timeliness of flood warnings, but also provides strong data support for flood prevention and disaster reduction work.

[0139] Based on the reconstructed hierarchical forecast dataset, new flood forecast data can be quickly and accurately classified into the existing flood data system, which facilitates subsequent flood risk assessment, warning issuance and emergency response. The above method not only improves the real-time and accuracy of flood forecasts, but also greatly simplifies the data processing and classification process.

[0140] In summary, see Figure 3 , Schematic diagram of the construction process of the hierarchical forecast dataset, where 1, 2, 3, and t represent flood forecast datasets of different time intervals, 1*, 2*, 3*, and t* represent forecast data of different time intervals after construction, and finally the hierarchical forecast dataset of the overall detection basin is obtained.

[0141] Classify and determine the attribution of new flood forecast data to provide strong support for flood prevention and disaster reduction. Reconstruct the flood forecast data set based on the neural network model, and obtain graded forecast data for different time intervals by monitoring the flood forecast data set and time interval characteristics of the basin. The above model can automatically extract key features from the original flood forecast data, and realize the graded processing and classification of the time interval of the flood forecast data through cluster analysis, providing strong support for flood prevention and disaster reduction.

[0142] Then, based on the above-mentioned graded forecast data set, neural network model and water flow input, the runoff prediction model is optimized, and the optimized runoff prediction model is obtained, the specific contents of which are as follows:

[0143] By using the neural network model to classify the flood forecast data, a hierarchical forecast data set containing flood characteristic information in different time intervals was obtained. The above data set not only realizes the effective classification of flood data, but also captures the typical patterns and key features of flood data in different time intervals through clustering centers and other methods.

[0144] Furthermore, the above-mentioned graded forecast data sets were reasonably time-tagged and classified to ensure that the data in each time interval can be accurately identified and classified. The above steps are helpful for subsequent data analysis and model prediction. Based on this, the forecast data association in the time dimension is established, and then the flood situation in a specific time interval in the future can be predicted based on historical data.

[0145] Based on the above time-stamped and graded forecast data sets, we can further obtain monitoring data related to runoff prediction in the monitoring basin. The above data mainly include key indicators such as rainfall, upstream water level, and flow. The relevant data are closely related to flood forecasting and runoff prediction. By integrating and analyzing the above data, we can have a more comprehensive understanding of the hydrological conditions in the monitoring basin and provide strong data support for the subsequent runoff prediction model.

[0146] In the process of constructing and optimizing the runoff prediction model, the hierarchical forecast data set and monitoring data can be fully utilized. The hierarchical forecast data set is used as part of the model training to enable the model to learn the characteristic differences and changing laws of flood data in different time intervals; at the same time, monitoring data can also be introduced as the input or verification basis of the model to improve the prediction accuracy and generalization ability of the model.

[0147] In order to significantly improve the accuracy and reliability of flood forecasting and runoff prediction results, and provide a more solid data foundation and decision-making support for flood prevention and disaster reduction and water resources management, a series of systematic and comprehensive optimization measures are adopted in the embodiment. The above measures are based on the hierarchical forecast data set and the runoff prediction model of step S3.

[0148] First, based on the time series-based hierarchical forecast data, we conducted an in-depth analysis of the time tags and corresponding data classification results to ensure that each type of data can accurately reflect the flood characteristics and hydrological laws of the corresponding time interval. The above steps not only provide a high-quality data foundation for subsequent model construction, but also ensure that the model can fully learn the flood change patterns and characteristics at different time scales during the training process.

[0149] Next, we carried out reasonable model construction and strategy optimization for the runoff prediction model. In terms of model construction, we fully considered the complexity and dynamics of flood forecasting, selected a neural network architecture suitable for time series data, and adjusted and optimized the model structure based on actual conditions. At the same time, we also introduced regularization technology and dropout strategy to effectively prevent model overfitting and improve the generalization ability of the model.

[0150] In terms of strategy optimization, the advantages of the hierarchical forecasting data set are fully utilized. By designing a reasonable loss function, optimization algorithm and training strategy, the model can more accurately capture the intrinsic connection and changing rules between flood forecasting and runoff forecasting. In addition, advanced technologies such as ensemble learning and transfer learning are used to further improve the prediction performance and stability of the model.

[0151] Through the above systematic and comprehensive optimization measures, an optimized runoff prediction model was successfully obtained. After the model is optimized, it can not only accurately predict the development trend of future floods and runoff changes in real time, but also provide strong data support and decision-making basis for flood prevention and water resources management.

[0152] The above optimized runoff prediction model satisfies the following relationship:

[0153]

[0154] in, After the neural network is optimized Monitor the flow of the basin during the period, It means that based on the neural network Time segmentation time interval, express The predicted rainfall in the monitoring basin at any time interval within the time period, represents the water balance coefficient of the monitored basin, express The effective runoff input of the monitoring basin at any time interval within the time period, express The average flow rate of the monitoring basin at any time interval within the time period, express Flow loss parameters of the monitored basin during the period.

[0155] The flow rate of the monitoring basin in any period after the neural network optimization refers to the prediction of the water flow value of a specific monitoring basin in any future period of time after the flood forecast data and runoff prediction model are optimized by the neural network, including time intervals such as hours, days, weeks or months. The optimized neural network model makes real-time predictions for the monitoring basin, providing strong decision-making support for flood prevention and disaster reduction and water resources management.

[0156] Based on the neural network model, the monitoring basin detection time is divided into multiple time intervals. Specifically, it refers to using the relevant mechanisms and algorithms in the neural network model to subdivide a long period of basin detection time into several identical or different continuous time periods, that is, T time intervals, so as to perform more detailed analysis, processing and prediction of the data in each time interval. The above segmentation method can play a role in a variety of application scenarios.

[0157] The water balance coefficient of the monitoring basin can reflect the water balance status in the monitoring basin. Based on the principle of water balance, that is, a basin, region or water body in any time period such as hours, days, months, years, etc., the input water volume of the monitoring basin includes precipitation, groundwater recharge, etc., minus the output water volume including evaporation, runoff, groundwater level changes, etc., and finally obtains the water storage variable of the monitoring range of the basin. The water balance coefficient of the monitoring basin is helpful to evaluate the water resources status of the basin, formulate water resources management strategies, predict flood risks, etc.

[0158] Monitoring the flow loss parameters of a basin in any period of time can be an important indicator for evaluating the water resources status and water resources management effect in the basin. The loss value is related to multiple factors such as evaporation, leakage, water withdrawal, soil erosion, etc. in the monitored basin. In practical applications, it is necessary to comprehensively consider multiple factors and data sources, and set the corresponding loss parameters for different periods according to the characteristics of the basin and monitoring needs to ensure the accuracy and reliability of the prediction results.

[0159] Furthermore, in this embodiment, the runoff prediction model after optimization by the neural network model is compared with the existing runoff prediction model, wherein NSE, RMSE, MAE and RE are set as comparison indicators. For specific comparison and analysis contents, please refer to Table 1.

[0160] Table 1 Comparative analysis of prediction models

[0161]

[0162] NSE refers to the normalized square error, which is an indicator that measures the difference between the model prediction value and the actual observation value. It is obtained by calculating the square of the prediction error and normalizing it. The smaller the NSE value, the better the model prediction performance.

[0163] RMSE refers to the root mean square error, a common method for measuring the difference between the predicted value and the true value. It is obtained by taking the square root of the average of the squares of the prediction errors. The smaller the RMSE value, the closer the model prediction value is to the actual observed value, and the better the model performance. RMSE is very sensitive to large errors, so it can well reflect the performance of the model in predicting extreme values.

[0164] MAE refers to the mean absolute error, which is another indicator to measure the accuracy of prediction. It calculates the average of the absolute errors between the predicted value and the true value. Unlike RMSE, MAE treats all errors equally and will not aggravate or mitigate their impact due to the size of the error. The smaller the MAE value, the better the prediction performance of the model.

[0165] RE refers to relative error, which can measure the degree of deviation of the predicted value from the true value. It is usually expressed as a percentage. The above RE can be the average relative error or the relative error of a single observation, which helps to understand the specific degree of deviation of the predicted value from the true value. The calculation formula of RE usually involves the absolute value of the difference between the predicted value and the true value. The smaller the value of MAE, the better the prediction performance of the model.

[0166] Based on the above comparative index information, it can be seen that the NSE value of the optimized prediction model is 0.615, which is significantly lower than 0.887 of the existing model. Since the smaller the NSE value, the better the model prediction performance, this shows that the optimized model has made significant improvements in the normalization of prediction errors, that is, the difference between its predicted values ​​and actual observed values ​​will be smaller.

[0167] Although the reduction in RMSE is not as significant as that in NSE, it still shows that the performance of the optimized prediction model has improved when predicting extreme values. RMSE is sensitive to large errors, and its reduction means that the model has better processing capabilities when dealing with observations with large errors.

[0168] The optimized prediction model also showed obvious advantages in MAE, which decreased from 12.365 to 9.023. MAE measures the average absolute error between the predicted value and the true value, indicating that the optimized prediction model has reduced the prediction error as a whole and improved the accuracy of the prediction results. The RE value decreased from 0.0681 to 0.0134, indicating that the optimized prediction model has a smaller deviation in prediction and the predicted value is closer to the actual observed value.

[0169] In summary, the optimized runoff prediction model is superior to the existing model in multiple key performance indicators, making the optimized prediction model more reliable and accurate in runoff prediction and better able to meet the needs of practical applications.

[0170] Furthermore, the method of optimizing the runoff prediction model in the present embodiment is only an optional condition of the present embodiment. In one or some other embodiments, the optimization method of the prediction model can be adjusted according to the historical information of watershed disasters and the watershed prediction objectives. By analyzing the historical information of watershed disasters, the key factors and laws affecting runoff changes can be identified. Targeted optimization of the prediction model based on the relevant factors and laws can improve the prediction accuracy and model adaptability.

[0171] S5. Combine the hierarchical forecast data set, the reconstructed monitoring basin input and the monitoring parameter evaluation system to conduct real-time analysis and flood warning for the monitoring basin. The specific implementation steps and contents are as follows:

[0172] First, a monitoring parameter evaluation system is constructed based on the historical information of the monitoring basin. The monitoring parameter evaluation system in the embodiment mainly includes: a basin water storage capacity evaluation model, a monitoring parameter stability evaluation model, a monitoring parameter correlation evaluation model and a monitoring parameter error analysis model.

[0173] In order to evaluate the water storage and regulation capacity of the monitored basin at different time scales, such as days, months, years and under different environmental conditions, the basin water storage capacity assessment model comprehensively considers factors such as the area, historical precipitation and runoff of the monitored basin, and calculates the rate of change of water storage in the monitored basin at different times and under different conditions, thereby evaluating the water storage and regulation capacity of the monitored basin, providing a scientific basis for water resources management, disaster control, and flood prevention and mitigation.

[0174] The above basin water storage capacity assessment model satisfies the following relationship:

[0175]

[0176] in, represents the rate of change of water storage in the monitored basin, represents the storage area of ​​the monitored basin, Indicates that the monitoring basin is Historical traffic within a time period, Indicates the monitoring time of the monitoring basin;

[0177] In order to evaluate the stability of the monitoring parameter runoff in the time series and identify the abnormal conditions and mutation points in the monitoring basin, the monitoring parameter stability evaluation model integrates statistical methods, such as time series analysis, trend test and other methods to process the flood forecast data set, analyze the changing trends and periodic laws of the monitoring parameters, and then evaluate the monitoring parameters, that is, the stability of the monitoring basin flow in the embodiment, to provide reference support for the water balance analysis of the monitoring basin.

[0178] The above monitoring parameter stability evaluation model satisfies the following relationship:

[0179]

[0180] in, express The stable value of the monitoring parameter during the period, express The measured value of the monitoring parameter during the period, express The predicted value after monitoring parameters optimization within the time period, express The measured average value of the monitoring parameter during the period, Indicates the number of measurements of the monitoring parameter;

[0181] In order to analyze the relationship between monitoring parameters and flood disasters, that is, the correlation between monitoring basin flow and basin flood disasters, and to reveal their intrinsic connection and degree of mutual influence, the monitoring parameter correlation evaluation model analyzes the relationship between monitoring parameters and identifies the causal relationship between key influencing factors and potential factors, which is conducive to a deep understanding of the basin hydrological process, optimizes the layout of the monitoring network, and improves the accuracy and explanatory power of the prediction model.

[0182] The above monitoring parameter correlation evaluation model satisfies the following relationship:

[0183]

[0184] in, express The correlation coefficient of the monitoring parameters within the time period, express The measured value of the monitoring parameter during the period, express The predicted value after monitoring parameters optimization within the time period, express The measured average value of the monitoring parameter during the period, express The predicted average value after the monitoring parameters are optimized within the period, Indicates the number of measurements of the monitoring parameter;

[0185] In order to evaluate the error status of monitoring parameters, that is, the possible errors in the measurement, transmission and processing of basin flow, and quantify their impact on the prediction results, the error range and uncertainty of monitoring parameters are calculated based on the monitoring parameter error analysis model.

[0186] The above monitoring parameter error analysis model satisfies the following relationship:

[0187]

[0188] in, express The absolute error value of the monitoring parameter within the time period, express The measured value of the monitoring parameter during the period, express The predicted value after the optimization of monitoring parameters within the time period;

[0189]

[0190] in, express The relative error value of the monitoring parameters within the time period, express The measured value of the monitoring parameter during the period, express The predicted value after optimizing the monitoring parameters during the time period.

[0191] Then, based on the hierarchical forecast data set, the reconstructed monitoring basin input, water storage change rate, stability value, correlation coefficient, absolute error value and relative error value, real-time analysis and flood warning of the monitoring basin are carried out. The specific contents are as follows:

[0192] In the data-driven real-time flood forecasting system, multivariate information such as graded forecast data sets, reconstructed monitoring basin input, water storage change rate, stability value, correlation coefficient, absolute error value and relative error value are used to conduct real-time analysis and flood warning of the monitoring basin.

[0193] 1. Apply the hierarchical forecast data set.

[0194] Flood forecast information within different time ranges can support decision-making needs in different time periods. The hierarchical forecast data set in the embodiment mainly includes: short-term such as hourly and daily levels; medium-term such as weekly and ten-day levels; long-term such as monthly and yearly levels of flood forecast information and historical flow results. According to the characteristics of different time intervals, time point conditions and disaster warning needs, the appropriate time interval forecast data set is selected for analysis. Short-term forecasts can be used for immediate decision-making and emergency response, while medium-term and long-term forecasts can be used to formulate disaster prevention plans and basin planning measures.

[0195] Second, the input of the monitoring basin is analyzed based on the output of the optimized runoff prediction model.

[0196] The prediction results output by the optimized runoff prediction model have better accuracy and completeness. The output prediction data not only achieves a significant improvement in accuracy, but also takes into account the comprehensiveness and timeliness of data information. It provides solid data support for the dynamic analysis of water volume in the monitoring basin, and can more accurately grasp the distribution and change trend of water flow in the basin, thereby providing a reliable data basis for water volume analysis and flood warning in the monitoring basin.

[0197] In an optional embodiment, the optimized runoff prediction model can accurately predict the basin runoff in a certain time period in the future. In flood-prone seasons or critical periods, the model can identify potential flood risks in advance and provide timely and accurate flood warning information to relevant departments, thereby effectively shortening emergency response time and reducing the dangers brought by flood disasters.

[0198] 3. Analyze the rate of change of water storage in the monitored basin.

[0199] In order to evaluate the dynamic changes in water storage in the monitored basin, determine whether there is a flood risk, and gain a comprehensive understanding of the dynamic balance and potential risks of water flow in the basin, the water storage change rate of key water bodies in the monitored basin was analyzed. The above indicators can not only intuitively reflect the storage and release rate of water flow in the basin, but are also the core parameters for evaluating changes in water flow conditions and predicting flood risks.

[0200] In an optional embodiment, the water level fluctuation of the monitored basin is continuously monitored and recorded, and the water storage capacity change rate is calculated through the basin water storage capacity assessment model. The above process not only focuses on the rapid rise and fall of water levels in the short term, but also attaches importance to the analysis of long-term trends to reveal the impact of seasonal, cyclical and even extreme climate events on the water storage capacity of the basin.

[0201] Comparing and analyzing the current rate of change of water storage in the monitored basin with historical data, long-term averages or preset thresholds can scientifically determine whether the current water flow state in the basin has deviated from the normal track, and then assess the possibility and severity of floods. Once the rate of change of water storage exceeds the safety threshold, the monitored basin is considered to be in an abnormal state, and the flood warning mechanism will be immediately triggered to issue an alarm to relevant departments and the public so that timely response measures can be taken to reduce the losses caused by flood disasters. The analysis of the rate of change of water storage in the monitored basin can effectively monitor, warn and respond to flood risks.

[0202] 4. Analyze the stability value and correlation coefficient of water volume in the monitored basin.

[0203] In order to evaluate the stability and correlation of water flow in the monitoring basin, improve the accuracy and reliability of flood warnings, and enhance the accuracy and reliability of data-driven real-time flood forecasting systems, the stability value and correlation coefficient of water flow data in the monitoring basin are analyzed in the embodiment. The stability value is a key indicator to measure the volatility of monitoring parameters in time series, and its high or low level directly reflects the strength of the stability of water flow in the basin. A high stability value indicates that the water flow state is relatively stable, which helps to establish a more reliable prediction baseline; while a low stability value may indicate the instability of the water flow state, which requires close attention and preparation for potential changes in the monitoring basin.

[0204] At the same time, the analysis of the correlation coefficient reveals the complex internal correlation and mutual influence between the monitoring parameters and flood disasters. By analyzing the correlation coefficient between the water volume in the basin and the natural disaster of floods, the possibility of floods can be quickly identified, which helps to understand the physical mechanism of flood formation and provide a basis for predicting the development trend of floods.

[0205] In addition, in order to develop more accurate and efficient monitoring and early warning strategies, we can strengthen monitoring efforts and increase data sampling frequency for key parameters closely related to flood occurrence to ensure that any abnormal signals can be captured before floods occur. At the same time, we can also continuously adjust the early warning threshold based on the comparative analysis results of historical data and real-time data to adapt to the dynamic changes in water volume in the basin and improve the accuracy and reliability of early warning signals.

[0206] 5. Analyze the absolute error and relative error values ​​of the prediction results.

[0207] In order to quantify the size of the error in the monitoring and forecasting process and evaluate the reliability of flood warnings, the absolute error value represents the absolute difference between the monitoring or forecasting results and the actual value; the relative error value represents the ratio of the above difference to the actual value. By analyzing the relevant error values, we can understand the performance of the monitoring results and forecasting system and identify potential sources of error.

[0208] The error value is used as one of the important references for flood warning results. When issuing warning signals, the error range and uncertainty information of the warning results should be given at the same time so that decision makers can make more scientific and reasonable decisions.

[0209] The above analysis results are comprehensively evaluated to provide comprehensive decision support for river basin management and flood warning. Comprehensively consider the hierarchical forecast data set, monitoring basin input, water storage change rate, stability value, correlation coefficient, absolute error value and relative error value, and judge whether there is flood risk and its severity in the basin based on the analysis results, and formulate corresponding response measures and plans.

[0210] The comprehensive analysis results can also be presented to decision makers and managers in the form of charts, reports, etc., to provide intuitive and comprehensive decision support information. At the same time, the analysis results can be released to the public and relevant institutions in real time to improve the transparency and credibility of flood warnings.

[0211] In this embodiment, a data-driven real-time flood forecasting method and a corresponding system are proposed. First, the boundary of the monitoring basin is clearly defined, and then based on the observation stations distributed in the area, the initial hydrological information and initial rainfall data are systematically collected as the basic information for disaster prediction and analysis.

[0212] Then, a forecast data optimization model was constructed. The model uses data processing algorithms to optimize the collected initial hydrological information and initial rainfall data, aiming to extract key features and eliminate redundant information, thereby generating a high-quality flood forecast data set.

[0213] Subsequently, a more accurate runoff prediction model was constructed based on the optimized flood forecast dataset. The optimized model can accurately analyze the water flow input at the outlet of the monitoring basin, thereby accurately predicting the dynamic changes of water flow in the future and providing a core reference for flood prediction.

[0214] In order to further improve the accuracy and practicality of the real-time flood forecasting method and system, the present invention also introduces a neural network model for information reconstruction. The above network model can intelligently integrate the flood forecast data set and the water flow input information, and generate a graded forecast data set and a reconstructed monitoring basin input through a complex mapping relationship, making the forecasting method more refined, hierarchical and intelligent.

[0215] Finally, by combining the graded forecast data set, the reconstructed monitoring basin input, and the monitoring parameter evaluation system, a comprehensive real-time analysis of the water volume in the monitoring basin is conducted. The above process can dynamically evaluate the flood risk in the basin and issue flood warning signals based on the evaluation and analysis results, providing decision-making basis for relevant departments and effectively reducing the impact of flood disasters.

[0216] In summary, the present invention forms a complete and efficient data-driven real-time flood forecasting system by integrating multiple key links such as data optimization, runoff prediction, neural network reconstruction and real-time analysis and evaluation, providing strong technical support for flood warning and disaster prevention and mitigation work.

[0217] See also Figure 4 In an optional embodiment, in order to efficiently execute the data-driven real-time flood forecasting method provided by the present invention, the present invention also provides a data-driven real-time flood forecasting system, in which the input device, processor, output device and memory are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the specific steps of the relevant embodiments of the data-driven real-time flood forecasting method provided by the present invention. The data-driven real-time flood forecasting system of the present invention has a complete structure, objective and stable, and can efficiently execute the data-driven real-time flood forecasting method of the present invention, thereby improving the overall applicability and practical application ability of the present invention.

[0218] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A data-driven real-time flood forecasting method, characterized in that: The steps include: Select the monitoring basin range, and obtain the initial hydrological information and initial rainfall data of the monitoring basin based on the observation stations in the monitoring range; Establishing a forecast data optimization model, and processing the hydrological initial information and the rainfall initial data through the forecast data optimization model to obtain a flood forecast data set for the monitored basin; Constructing a runoff prediction model based on the flood forecast data set, and using the runoff prediction model to analyze the water flow input at the outlet of the monitoring basin; Introducing a neural network model to reconstruct information of the flood forecast data set and the water flow input, and obtaining a hierarchical forecast data set and a reconstructed monitoring basin input; Combining the hierarchical forecast data set, the reconstructed monitoring basin input and the monitoring parameter evaluation system, real-time analysis and flood warning are carried out on the monitoring basin range; The introducing of the neural network model to reconstruct the flood forecast data set and the water flow input, and obtaining the graded forecast data set and the reconstructed monitoring basin input comprises: The neural network model combines the time series features and the autoencoder model to reconstruct the flood forecast data set to obtain a hierarchical forecast data set; Optimizing the runoff prediction model based on the graded forecast data set, the neural network model and the water flow input, and obtaining an optimized runoff prediction model; The optimized runoff prediction model satisfies the following relationship: ， in, After the neural network is optimized Monitor the flow of the basin during the period, It means that based on the neural network Time segmentation time interval, express The predicted rainfall in the monitoring basin at any time interval within the time period, represents the water balance coefficient of the monitored basin, express The effective runoff input of the monitoring basin at any time interval within the time period, express The average flow rate of the monitoring basin at any time interval within the time period, express Monitor the flow loss parameters of the basin during the period; The real-time analysis and flood warning of the monitoring basin range by combining the hierarchical forecast data set, the reconstructed monitoring basin input and the monitoring parameter evaluation system includes: Constructing a monitoring parameter evaluation system based on historical information of the monitoring basin, wherein the monitoring parameter evaluation system includes a basin water storage capacity evaluation model, a monitoring parameter stability evaluation model, a monitoring parameter correlation evaluation model, and a monitoring parameter error analysis model; The watershed water storage capacity assessment model satisfies the following relationship: ， in, represents the rate of change of water storage in the monitored basin, represents the storage area of ​​the monitored basin, Indicates monitoring basin Historical traffic for a time period, Indicates the monitoring time of the monitoring basin; The monitoring parameter stability evaluation model satisfies the following relationship: ， in, express The stable value of the monitoring parameter during the period, express The measured value of the monitoring parameter during the period, express The predicted value after the monitoring parameters are optimized within the time period, express The measured average value of the monitoring parameter during the period, Indicates the number of measurements of the monitoring parameter; The monitoring parameter correlation evaluation model satisfies the following relationship: ， in, express The correlation coefficient of the monitoring parameters within the time period, express The measured value of the monitoring parameter during the period, express The predicted value after the monitoring parameters are optimized within the time period, express The measured average value of the monitoring parameter during the period, express The predicted average value after the monitoring parameters are optimized within the period, Indicates the number of measurements of the monitoring parameter; The monitoring parameter error analysis model satisfies the following relationship: ， in, express The absolute error value of the monitoring parameter within the time period, express The measured value of the monitoring parameter during the period, express The predicted value after the optimization of monitoring parameters within the time period; ， in, express The relative error value of the monitoring parameters within the time period, express The measured value of the monitoring parameter during the period, express The predicted value after the optimization of monitoring parameters within the time period; The real-time analysis and flood warning of the monitoring basin range by combining the hierarchical forecast data set, the reconstructed monitoring basin input and the monitoring parameter evaluation system includes: Based on the graded forecast data set, the reconstructed monitoring basin input, the water storage capacity change rate, the stability value, the correlation coefficient, the absolute error value and the relative error value, the monitoring basin is analyzed in real time and flood warning is carried out.

2. The data-driven real-time flood forecasting method according to claim 1, characterized in that: The step of establishing a forecast data optimization model and processing the initial hydrological information and the initial rainfall data through the forecast data optimization model to obtain a flood forecast data set for the monitoring basin includes: Establishing a forecast data optimization model based on historical data of the monitored river basin, wherein the forecast data optimization model includes a data standardization function and a data normalization function; The data standardization function and the data normalization function are used to obtain a flood forecast data set for the monitored basin, wherein the flood forecast data set includes hydrological information and rainfall data for the monitored basin.

3. The data-driven real-time flood forecasting method according to claim 1, characterized in that: The constructing of a runoff prediction model based on the flood forecast data set comprises: The runoff prediction model satisfies the following relationship: ， in, express The effective runoff input at the basin outlet during the period, express Monitor the rainfall elements in the basin during the period, express The predicted rainfall in the monitored basin during the period, express The impact lag of the monitoring basin within the time period, represents the influence weight of underlying surface factors in the monitored watershed, express Monitor the evaporation in the basin during the period.

4. The data-driven real-time flood forecasting method according to claim 3 is characterized in that: The use of the runoff prediction model to analyze and monitor the water flow input at the outlet of the basin includes: The effective runoff input at the outlet of the monitoring basin in any period of time is obtained by using the runoff prediction model; The water flow input at the outlet of the monitored watershed is obtained based on the effective runoff input.

5. A data-driven real-time flood forecasting system, characterized in that: The system includes a processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the data-driven real-time flood forecasting method as described in any one of claims 1-4.

Images