Reservoir inflow rain flood coupling prediction method and system and electronic equipment

Abstract

The embodiment of the application provides a reservoir inflow rain flood coupling prediction method, system and electronic equipment, and belongs to the technical field of reservoir inflow prediction. The method comprises the following steps: extracting a rain catchment area of a reservoir automatically delimited according to DEM; predicting a precipitation forecast amount of the rain catchment area according to a constructed fusion prediction mode; constructing a flood prediction model of the reservoir inflow amount; obtaining a reservoir evapotranspiration prediction amount, inputting the reservoir evapotranspiration prediction amount and the precipitation forecast amount into the flood prediction model for prediction, and obtaining a reservoir inflow prediction amount. The prediction method of the application combines meteorological and hydrological real-time observation data and other monitoring data, timely captures a precipitation process, improves the accuracy of areal rainfall and precipitation prediction results, and reduces precipitation prediction errors; provides fine reservoir area areal rainfall prediction and reservoir inflow increment prediction information services for flood control departments and reservoir power stations, provides decision support for making water resource management plans in advance, and improves economic benefits to a certain extent.

Description

Technical Field

[0001] This invention relates to the field of reservoir inflow forecasting technology, specifically to a rain-flood coupled forecasting method for reservoir inflow, a rain-flood coupled forecasting system for reservoir inflow, an electronic device, and a computer-readable storage medium. Background Technology

[0002] Rainfall in Jiangxi Province and other parts of China is expected to be above average from June to August, with a prolonged rainy season, concentrated rainfall events, and locally heavy downpours. Major rivers are at risk of significant regional flooding, posing a serious threat to the economic and social development of the surrounding areas. Therefore, rational water resource management is essential, and accurate reservoir inflow forecasts can provide effective data-driven decision support for water resource management. Current reservoir inflow forecasting technology consists of two parts: precipitation forecasting and flood forecasting.

[0003] Precipitation forecasts are primarily based on forecasts from global and regional models, as well as localized corrections using applied techniques, including the following:

[0004] 1. Dynamic multi-model ensemble based on previous forecast errors. The models and their weights for subsequent forecasts are determined by sliding the forecast error of a previous period.

[0005] 2. Frequency matching method. This method can significantly improve the systematic biases of rainfall and rain area in precipitation forecasts of the AREM numerical weather prediction model, especially in the light rainfall period. Localized corrections significantly improve the quality of qualitative precipitation forecasts of "rain or no rain" in numerical weather prediction.

[0006] 3. Statistical downscaling method. It requires less computation and is easy to implement.

[0007] Existing precipitation forecasting methods still have at least the following shortcomings:

[0008] Climate simulation has a relatively large systematic error, lacks integrated forecasting research on short-term and medium-term conditions, and the accuracy of element forecast results is also insufficient. In terms of predicting the inflow of water into reservoir areas, traditional hydrological models have numerous parameters, and the manual parameter adjustment process often requires a lot of experience, which is extremely labor-intensive. Summary of the Invention

[0009] The purpose of this invention is to provide a method, system, and electronic device for coupled rainwater and flood forecasting of reservoir inflow, so as to at least solve the problems of large errors, lack of integrated short-term and medium-term forecasting research, and insufficient accuracy of forecast results in the above-mentioned precipitation forecasting methods.

[0010] To achieve the above objectives, embodiments of the present invention provide a method for coupled stormwater and flood forecasting of reservoir inflow, the method comprising:

[0011] Extract the catchment area of ​​the reservoir automatically delineated based on the DEM;

[0012] The precipitation forecast for the catchment area is predicted based on the constructed fusion forecast model;

[0013] Construct a flood forecasting model for reservoir inflow;

[0014] The predicted evapotranspiration of the reservoir is obtained, and the predicted evapotranspiration and precipitation are input into the flood forecasting model for prediction, so as to obtain the predicted inflow of water into the reservoir.

[0015] Optionally, the flood forecasting model is a neural network model or the Xin'anjiang model.

[0016] Optionally, the method further includes: constructing a fused forecast model, including:

[0017] The selected forecast models are downscaled to obtain high-resolution forecast models; the forecast models include one or more of the following: QPF precipitation model, rapid update assimilation model, regional numerical model, EC, GFS and JMA global model;

[0018] Multiple high-resolution forecast models are deployed for different future time periods;

[0019] Obtain the overlapping time periods of multiple high-resolution forecast models existing in various future time periods;

[0020] Calculate the fusion weights of overlapping high-resolution forecast models for each cross-period;

[0021] The high-resolution forecast models that overlap in the corresponding cross-periods are fused according to the fusion weights to obtain the fused forecast model.

[0022] Optionally, the fusion weights of overlapping high-resolution forecast models in each cross-period are calculated, including:

[0023] Obtain the grid-point precipitation forecasts for each high-resolution forecast model during this cross-period.

[0024] Calculate the error between the grid point precipitation forecasts of each high-resolution forecast model and the actual precipitation during the cross-period;

[0025] The constructed convolutional neural network is used to process the error to obtain the fusion weights of the overlapping high-resolution forecast models in the cross-period.

[0026] Optionally, the precipitation forecast for the catchment area is predicted based on the constructed fusion forecast model, including:

[0027] The real-time rainfall field is obtained, which is the real-time rainfall amount generated after quality control processing of rainfall data monitored by meteorological stations, rain gauges or satellite radar;

[0028] The areal rainfall in the catchment area is predicted based on the fusion forecast model.

[0029] The areal rainfall is localized and corrected based on the actual rainfall field to obtain the precipitation forecast for the catchment area.

[0030] Optionally, the areal rainfall can be localized and corrected based on the actual rainfall field, including:

[0031] Obtain the cumulative probability of actual precipitation and the cumulative probability of precipitation forecast over a past time period;

[0032] The precipitation amount is adjusted based on the cumulative probability of actual precipitation and the cumulative probability of precipitation forecast.

[0033] Optionally, extract the catchment areas of the reservoirs automatically delineated based on the DEM, including:

[0034] The watershed topography of the reservoir is preprocessed based on the DEM;

[0035] The water flow direction and flow rate of the pre-processed watershed topography were calculated using the D8 algorithm.

[0036] The river network system is extracted based on the set threshold and the direction of water flow;

[0037] The extracted river network is interactively connected based on the flow rate to obtain a watershed, which is used to characterize the catchment area of ​​the reservoir.

[0038] Optionally, the preprocessing includes depression filling.

[0039] Optionally, the method further includes reservoir water level forecasting: calculating the new water level of the reservoir based on the forecasted inflow and reservoir parameters, including reservoir capacity and minimum water level.

[0040] This invention also provides a reservoir inflow coupled rainfall-flood forecasting system for implementing the above-mentioned reservoir inflow coupled rainfall-flood forecasting method. The system includes:

[0041] The extraction module is used to extract the catchment areas of reservoirs automatically delineated based on the DEM;

[0042] The precipitation forecast module is used to predict the precipitation forecast amount of the catchment area based on the constructed fusion forecast model;

[0043] The model building module is used to build flood forecasting models for reservoir inflow.

[0044] The inflow forecast module is used to obtain the reservoir evapotranspiration forecast. The reservoir evapotranspiration forecast and precipitation forecast are input into the flood forecast model for prediction to obtain the reservoir inflow forecast.

[0045] This invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described method for coupled rainwater and flood forecasting of reservoir inflow.

[0046] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for coupled rainwater and flood forecasting of reservoir inflow.

[0047] Through the above technical solutions, the forecasting method of the present invention combines real-time meteorological and hydrological observation data and other monitoring data to capture precipitation processes in a timely manner, improve the accuracy of areal rainfall and precipitation forecast results, and reduce precipitation forecast errors; it provides information services such as refined areal rainfall forecasts and reservoir inflow increment forecasts for flood control departments and reservoir power stations, and provides decision support for the formulation of water resource management plans in advance, thereby improving economic benefits to a certain extent.

[0048] Secondly, the fusion forecasting model enables integrated short-term and medium-term precipitation forecasts, which can improve the accuracy of areal rainfall and precipitation forecast results.

[0049] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0050] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0051] Figure 1 This is a flowchart of a method for coupled rainwater and flood forecasting of reservoir inflow provided by one embodiment of the present invention;

[0052] Figure 2 This is a flowchart of a method for delineating rainwater catchment areas provided in an optional embodiment of the present invention;

[0053] Figure 3 This is a flowchart of a method for constructing a fused forecast model provided by an optional embodiment of the present invention;

[0054] Figure 4 This is a flowchart of a localized correction calculation method provided by an optional embodiment of the present invention;

[0055] Figure 5This is a block diagram of a reservoir inflow coupled forecasting system provided by an optional embodiment of the present invention. Detailed Implementation

[0056] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0057] Figure 1 This is a flowchart of a method for coupled rainwater and flood forecasting of reservoir inflow provided by one embodiment of the present invention, as shown below. Figure 1 As shown, a method for coupled rainfall and flood forecasting of reservoir inflow is provided, the method comprising:

[0058] Step S1: Extract the rainwater catchment area of ​​the reservoir automatically delineated based on the digital elevation model (DEM).

[0059] In this embodiment, the catchment area of ​​the reservoir is the sum of the catchment areas of all rivers that flow into the reservoir, and the catchment area is the size of the collected rainwater surface per unit length and width.

[0060] like Figure 2 As shown, as a further optimization of this embodiment, the delineation of the rainwater catchment area includes the following steps:

[0061] Step S101: Preprocess the watershed topography of the reservoir based on the DEM.

[0062] In the river network extraction process described below, errors caused by data input errors, interpolation errors during DEM generation, and DEM resolution result in some depressions on the DEM surface, i.e., depressions. Therefore, before calculating the flow direction, the original DEM data should be preprocessed to fill the depressions to obtain a DEM without depressions. The depression filling process involves calculating the depression areas in the DEM data using the flow direction data, calculating their depths, and filling the depressions according to a set filling threshold.

[0063] Step S102: Calculate the water flow direction and flow rate of the preprocessed watershed topography according to the D8 algorithm.

[0064] The D8 algorithm uses numerical values ​​to represent the flow direction of each unit. The range of the numbers is 1 to 255, where 1 represents due east, 2 represents southeast, 4 represents due south, 8 represents southwest, 16 represents due west, 32 represents northwest, 64 represents due north, and 128 represents northeast.

[0065] The water flow direction is as follows: assuming rainwater falls on a cell in the terrain, the water will flow to the lowest cell in the surrounding 8 cells. If multiple cells have the same maximum downward direction, the range of adjacent cells will be expanded until the steepest downward direction is found.

[0066] The flow rate is calculated as follows: the cumulative flow is calculated by using a regular grid to represent the digital elevation model. Each point has one unit of water volume. Then, based on the water flow direction data of the grids within the area, the value of the water volume flowing through each grid is calculated to obtain the cumulative flow of the area.

[0067] Step S103: Extract the river network system based on the set threshold and the direction of water flow.

[0068] If a grid point in a DEM belongs to the same river system, it must be supported by a certain upstream catchment area. Based on the obtained runoff accumulation data and considering factors such as climate and topography of the study area, a threshold (i.e., runoff capacity threshold) is determined. When the runoff accumulation at a grid point exceeds this threshold, the grid point is considered to belong to a certain river system. The interconnection of these grid points forms a river network, and the size of this threshold determines the accuracy and detail of the river network extraction.

[0069] Step S104: Based on the flow rate, the extracted river network is interactively connected to obtain a watershed, which is used to characterize the catchment area of ​​the reservoir.

[0070] Watersheds can be nested; for example, a large watershed can contain several smaller watersheds, meaning a large watershed contains several smaller watersheds.

[0071] Step S2: Predict the rainfall forecast for the catchment area based on the constructed fusion forecast model.

[0072] In this embodiment, the fusion forecast model can achieve integrated short-term and medium-term precipitation forecasts. As a further optimization of this embodiment, such as... Figure 3 As shown, the method further includes: constructing a fused forecast model, including:

[0073] Step S201: Downscale the selected forecast model to obtain a high-resolution forecast model. The forecast model includes one or more of the following: Quantitative Precipitation Forecast (QPF) model, Rapid Update Assimilation Model, Regional Numerical Model, European Medium-Range Weather Forecast (EC), Global Forecast System (GFS), or Japan Meteorological Agency Numerical Forecast Model (JMA).

[0074] In this embodiment, the scale of the selected forecast model is usually around 12.5 km, which has relatively low resolution and lacks detailed regional climate information, making it difficult to make reasonable predictions about regional climate. Downscaling can reduce the scale of the forecast model to 1 km, improve the resolution of the forecast model, and thus enable regional climate prediction.

[0075] Downscaling is preferably performed using the DeepSD model based on deep learning. DeepSD is composed of multiple layers of SRCNN stacked together. SRCNN downscaling utilizes convolutional neural networks to extract high-dimensional features and reconstruct the data, thereby achieving downscaling. In this method, to achieve a more accurate transformation from large-scale climate information to regional scale, regional and topographic factors are also considered. The specific implementation process is as follows: a three-layer convolutional neural network is constructed. The first two layers are used to extract high-dimensional spatial features of topography and the original data (i.e., to find features in large-scale climate model output variables and topographic information that are highly correlated with small-scale meteorological elements). The third layer of the convolutional neural network performs data reconstruction (i.e., to construct and transform the nonlinear relationship between high-dimensional spatial features and small-scale meteorological elements).

[0076] The deep convolutional neural network model used in the downscaling process can improve the accuracy of downscaling by utilizing convolution and subsampling, thereby reducing the amount of computation and achieving downscaling more efficiently and accurately.

[0077] Step S202: Deploy multiple high-resolution forecast models for multiple future time periods.

[0078] In this embodiment, the distribution of high-resolution forecast models for each future time period is as follows: QPF is used for the 0-6 hour forecast, the Fast Update Assimilation Model is used for the 2-24 hour forecast, the Regional Numerical Model is used for the 6-72 hour forecast, and EC, GFS, and JMA are used for the 8-hour to 7-day forecast.

[0079] Step S203: Obtain the overlapping time periods of multiple high-resolution forecast models existing in each future time period.

[0080] Step S204: Calculate the fusion weights of the overlapping high-resolution forecast models in each cross-period.

[0081] In this embodiment, there are overlapping time periods in each future time period, and these overlapping time periods have overlapping portions of multiple high-resolution forecast models. Specifically, during the overlapping time period of 2h-6h, there is an overlap between the QPF and the Fast Update Assimilation Model; during the overlapping time period of 6h-8h, there is an overlap between the Fast Update Assimilation Model and the Regional Numerical Model; during the overlapping time period of 8h-24h, there is an overlap between the Fast Update Assimilation Model, the Regional Numerical Model, EC, GFS, and JMA; and during the overlapping time period of 24h-72h, there is an overlap between the Regional Numerical Model, EC, GFS, and JMA. Therefore, when making forecasts, it is necessary to perform fusion processing on the high-resolution forecast models in the overlapping portions.

[0082] Step S205: Based on the fusion weight, the high-resolution forecast models that overlap at the corresponding cross-periods are fused to obtain a fused forecast model, which is used to predict the precipitation forecast for the next 0-7 days.

[0083] The fusion forecasting model enables integrated short- and medium-term precipitation forecasts, improving the accuracy of areal rainfall and precipitation forecast results.

[0084] When selecting weights for the fusion of traditional forecast models, the methods often involve manually adjusting them based on past experience or setting fixed weights that will not be changed. These methods can lead to inappropriate weight settings and increased errors.

[0085] As a further optimization of this embodiment, the fusion weights of the overlapping high-resolution forecast models in each cross-period are calculated, including:

[0086] Step a1: Obtain the grid precipitation forecast of each high-resolution forecast model during the cross-period. A grid is a grid obtained by dividing the unit area of ​​the catchment area into grids. Grid precipitation is the precipitation on each grid. The areal rainfall forecast of each high-resolution forecast model during the cross-period is the recent forecast data, such as the forecast data for the next 1-3 months.

[0087] Step a2: Calculate the error between the grid precipitation forecast of each high-resolution forecast model and the actual precipitation during the crossover period. The actual precipitation is based on the rainfall data monitored by meteorological stations, rain gauges or satellite radar during the crossover period.

[0088] Step a3: Use the constructed convolutional neural network to process the error and obtain the fusion weights of the overlapping high-resolution forecast models in the cross-period.

[0089] Therefore, the fusion weights of the present invention can automatically adjust the weights based on recent data, thereby improving the forecast accuracy of the fusion forecast model.

[0090] As a further optimization of this embodiment, such as Figure 4 As shown, the precipitation forecast for the catchment area is predicted based on the constructed fusion forecast model, including:

[0091] Step b1: Obtain the real-time rainfall field, which is the real-time rainfall amount generated after quality control processing of rainfall data monitored by meteorological stations, rain gauges or satellite radar;

[0092] Step b2: Predict the areal rainfall in the catchment area based on the fused forecast model obtained in step S205;

[0093] Step b3: Based on the actual rainfall field, perform localized correction on the areal rainfall to obtain the rainfall forecast for the catchment area.

[0094] Since the physical models and parameterization schemes in regional numerical models are only estimates of atmospheric conditions and cannot fully reflect the actual conditions, the aforementioned models such as QPF, Fast Update Assimilation, Regional Numerical Models, EC, GFS, or JMA all have certain errors in their forecasts. To reduce these errors, localized corrections need to be made to the fused forecast models.

[0095] In this embodiment, localization correction is performed based on the actual rainfall field surface area, including:

[0096] Step c1: Obtain the cumulative probability of actual precipitation and the cumulative probability of precipitation forecast over a past period; for example, the distribution of the cumulative probability of actual precipitation over the past 1-3 months at different durations (1 / 3 / 6 / 12 / 24h), and the distribution of the cumulative probability of precipitation forecast over the past 1-3 months at different durations (1 / 3 / 6 / 12 / 24h). The distribution of the cumulative probability of actual precipitation and the cumulative probability of precipitation forecast are represented by a cumulative distribution function.

[0097] Step c2: Adjust the surface rainfall based on the cumulative probability of actual precipitation and the cumulative probability of precipitation forecast.

[0098] In this embodiment, the aforementioned localization correction is specifically completed in two steps: weather correction and precipitation correction. Weather correction is performed first, followed by precipitation correction. Weather correction is achieved using a dynamically optimal classification threshold GBDT model. Factors such as time, precipitation, relative humidity, and total cloud cover are selected, and a GBDT gradient boosting decision tree model is trained to obtain the optimal probability threshold for weather classification. The classification results include rain grid points and sunny grid points.

[0099] After the weather correction, when the classification result is a rain grid point, the precipitation is corrected for that rain grid point, i.e., a probability matching precipitation correction function is constructed:

[0100] F(x) = D[N(x)] - x;

[0101] Where x is the precipitation forecast from the fusion forecast model, F(x) is the precipitation correction function related to the precipitation forecast, N(x) is the fitting function of the cumulative probability of precipitation forecast, and D(y) is the inverse function of the fitting function of the cumulative probability of actual precipitation.

[0102] The inverse function D(y) is calculated as follows: Define the cumulative probability of precipitation forecast pi(p1,p2,p3,...,tn) and the cumulative probability of actual precipitation qi(q1,q2,q3,...,qn) for different durations of the precipitation forecast ti(t1,t2,t3,...,tn), where F(ti)=pi, (x<=ti). First, construct the cumulative probability curve of precipitation forecast using coordinates (p1,t1), (p2,t2), (p2,t2),..., (pn,tn). Then, calculate the precipitation thresholds g1,g2,...,gn corresponding to the cumulative probability of actual precipitation on the cumulative probability curve of precipitation forecast using interpolation. Construct the correction curve of precipitation forecast and precipitation threshold using coordinates (g1,t1), (g2,t2), (g2,t2),..., (gn,tn). This correction curve is the inverse function D(y).

[0103] After constructing the probability matching precipitation correction function, the precipitation correction can be completed simply by adding the correction function F(x) to the forecast precipitation.

[0104] In this embodiment, localized correction can also employ an Encoder-Decoder neural network model. The correction model Seq2Seq is constructed using a convolutional neural network (CNN) and a long short-term memory network (LSTM). The core of the Seq2Seq model is CNN and LSTM. The model consists of two parts: an Encoder and a Decoder. The Encoder is used to parse historical information, while the Decoder is used to compile historical information to generate correction results.

[0105] Step S3: Construct a flood forecasting model for reservoir inflow.

[0106] In this embodiment, the flood forecasting model is a neural network model or the Xin'anjiang model. When the reservoir has a large amount of historical flow data and the data of each basin is relatively abundant, the neural network model can be directly used for modeling; while when the reservoir flow data is relatively scarce, the Xin'anjiang model is used.

[0107] Step S4: Obtain the reservoir evaporation forecast. Input the reservoir evaporation forecast and precipitation forecast into the flood forecast model for prediction to obtain the reservoir inflow forecast.

[0108] In this embodiment, the Xin'anjiang model is used as the flood forecasting model. The Xin'anjiang model mainly includes four parts: evapotranspiration calculation, runoff generation calculation, water source division, and confluence calculation.

[0109] In the actual measurement process, the Xin'anjiang model takes measured rainfall and measured evapotranspiration as inputs and outputs the basin flow. The measured evapotranspiration is obtained by measuring the conversion coefficient between water surface evaporation and basin evapotranspiration capacity through an evaporator. The measured rainfall is the actual precipitation measured in the basin. In other words, the actual inflow of water into the reservoir can be calculated based on the measured precipitation.

[0110] In the forecasting process, the predicted precipitation and evapotranspiration are used as input values ​​for the Xin'anjiang model to predict the inflow of water into the reservoir.

[0111] The forecasting method of this invention combines real-time meteorological and hydrological observation data and other monitoring information to capture precipitation processes in a timely manner, thereby improving the accuracy of areal rainfall and precipitation forecast results and reducing precipitation forecast errors. It provides information services such as refined areal rainfall forecasts and reservoir inflow increment forecasts for flood control departments and reservoir power stations, and provides decision support for the formulation of water resource management plans in advance, thereby improving economic benefits to a certain extent.

[0112] As a further optimization of this embodiment, the method also includes constructing a Long Short-Term Memory (LSTM) network model, using daily meteorological data and reservoir basin attribute data as inputs to the LSTM network model to predict the daily flow of the reservoir. The daily meteorological data includes data such as precipitation and temperature, and the reservoir basin attribute data includes data such as slope. Slope is one of the important influencing factors in the formation of the basin's flow.

[0113] As a further optimization of this embodiment, the method also includes water level prediction, which calculates the new water level of the reservoir based on the forecast inflow and reservoir parameters, including reservoir capacity and minimum water level.

[0114] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention.

[0115] Figure 5 This is a block diagram of a reservoir inflow coupled forecasting system provided by an optional embodiment of the present invention, such as... Figure 5 As shown, the system is used to implement the above-mentioned reservoir inflow coupled stormwater forecasting method, and the system includes:

[0116] The extraction module is used to extract the catchment areas of reservoirs automatically delineated based on the DEM;

[0117] The precipitation forecast module is used to predict the precipitation forecast amount of the catchment area based on the constructed fusion forecast model;

[0118] The model building module is used to build flood forecasting models for reservoir inflow.

[0119] The inflow forecast module is used to obtain the reservoir evapotranspiration forecast. The reservoir evapotranspiration forecast and precipitation forecast are input into the flood forecast model for prediction to obtain the reservoir inflow forecast.

[0120] The forecasting method of this invention combines real-time meteorological and hydrological observation data with other monitoring information to promptly capture precipitation processes, improving the accuracy of areal rainfall and precipitation forecast results, and reducing precipitation forecast errors. It provides flood control departments and reservoir power stations with refined information services such as reservoir area areal rainfall forecasts and reservoir inflow increment forecasts. Furthermore, it provides decision support for the formulation of water resource management plans in advance, thereby enhancing economic benefits to a certain extent.

[0121] Secondly, the fusion forecasting model enables integrated short-term and medium-term precipitation forecasts, which can improve the accuracy of areal rainfall and precipitation forecast results.

[0122] This invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described method for coupled rainwater and flood forecasting of reservoir inflow.

[0123] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for coupled rainwater and flood forecasting of reservoir inflow.

[0124] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not describe the various possible combinations separately.

[0125] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0126] Furthermore, various different implementations of the present invention can be combined arbitrarily, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed in the present invention.

Claims

1. A method for coupled rainfall and flood forecasting of reservoir inflow, characterized in that, The method includes: Extract the catchment area of ​​the reservoir automatically delineated based on the DEM; The precipitation forecast for the catchment area is predicted based on the constructed fusion forecast model. The prediction includes: acquiring the actual rainfall field, predicting the areal rainfall of the catchment area based on the fusion forecast model, and performing localized corrections on the areal rainfall based on the actual rainfall field to obtain the precipitation forecast for the catchment area. The localized correction includes: first performing a weather-related correction, and then performing a precipitation correction. The weather-related correction is performed using a dynamically optimal classification threshold GBDT model, selecting time, precipitation, relative humidity, and total cloud cover as factors, and training a GBDT gradient boosting decision tree model. The optimal probability threshold is obtained for rain / sunshine classification. After rain / sunshine correction, when the classification result is a rain grid point, the precipitation is corrected for that rain grid point. A probability-matching precipitation correction function F(x)=D[N(x)]-x is constructed, where x is the precipitation forecast from the fused forecast model, F(x) is the precipitation correction function related to the precipitation forecast, N(x) is the fitting function of the cumulative probability of the precipitation forecast, and D(y) is the inverse function of the fitting function of the cumulative probability of the actual precipitation. The correction function F(x) is added to the forecast precipitation to complete the precipitation correction. The classification result includes rain grid points. The fused forecast model is constructed as follows: Selected forecast models are downscaled to obtain corresponding high-resolution forecast models; these models include one or more of the following: QPF precipitation model, Fast Update Assimilation model, regional numerical model, EC, GFS, or JMA global model; multiple high-resolution forecast models are deployed across multiple future time periods; the overlapping time periods of the multiple high-resolution forecast models in each future time period are obtained; the fusion weights of the overlapping high-resolution forecast models in each overlapping time period are calculated; the overlapping high-resolution forecast models in the corresponding overlapping time periods are fused according to the fusion weights to obtain the fused forecast model; the fusion weights of the overlapping high-resolution forecast models in each overlapping time period are calculated, including: obtaining the grid-point precipitation forecasts of each high-resolution forecast model in the overlapping time period; calculating the error between the grid-point precipitation forecasts of each high-resolution forecast model in the overlapping time period and the actual rainfall in the overlapping time period; and processing the error using a constructed convolutional neural network to obtain the fusion weights of the overlapping high-resolution forecast models in the overlapping time period. Construct a flood forecasting model for reservoir inflow; The predicted evapotranspiration of the reservoir is obtained, and the predicted evapotranspiration and precipitation are input into the flood forecasting model for prediction, so as to obtain the predicted inflow of water into the reservoir.

2. The method according to claim 1, characterized in that, The flood forecasting model is either a neural network model or the Xin'anjiang model.

3. The method according to claim 1, characterized in that, The real-time rainfall field is the actual rainfall amount generated after quality control processing of rainfall data monitored by meteorological stations, rain gauges, or satellite radar.

4. The method according to claim 1, characterized in that, Extract the catchment areas of reservoirs automatically delineated based on the DEM, including: The watershed topography of the reservoir is preprocessed based on the DEM; The water flow direction and flow rate of the pre-processed watershed topography were calculated using the D8 algorithm. The river network system is extracted based on the set threshold and the direction of water flow; The extracted river network is interactively connected based on the flow rate to obtain a watershed, which is used to characterize the catchment area of ​​the reservoir.

5. A reservoir inflow coupled rainfall-flood forecasting system, used to implement the reservoir inflow coupled rainfall-flood forecasting method according to any one of claims 1-4, characterized in that, The system includes: The extraction module is used to extract the catchment areas of reservoirs automatically delineated based on the DEM; The precipitation forecast module is used to predict the precipitation forecast amount for the catchment area based on the constructed fusion forecast model. Predicting the precipitation forecast amount for the catchment area includes: acquiring the actual rainfall field, predicting the areal rainfall of the catchment area based on the fusion forecast model, and performing localized correction on the areal rainfall based on the actual rainfall field to obtain the precipitation forecast amount for the catchment area. The localized correction includes: first performing a weather-related correction, and then performing a precipitation correction. The weather-related correction is performed using a dynamically optimal classification threshold GBDT model, selecting time, precipitation, relative humidity, and total cloud cover as elements, and training the GBDT gradient. The decision tree model is improved to obtain the optimal probability threshold for rain / sunshine classification. After rain / sunshine correction, when the classification result is a rain grid point, the precipitation is corrected for that rain grid point. A probability-matching precipitation correction function F(x) = D[N(x)] - x is constructed, where x is the precipitation forecast from the fused forecast model, F(x) is the precipitation correction function related to the precipitation forecast, N(x) is the fitting function of the cumulative probability of the precipitation forecast, and D(y) is the inverse function of the fitting function of the cumulative probability of the actual precipitation. The correction function F(x) is added to the forecast precipitation to complete the precipitation correction. Classification result The forecast model includes both rain and clear grid points. The fused forecast model is constructed as follows: Selected forecast models are downscaled to obtain corresponding high-resolution forecast models; these models include one or more of the following: QPF precipitation model, Rapid Update Assimilation model, Regional Numerical Model, EC, GFS, or JMA global model; Multiple high-resolution forecast models are deployed across multiple future time periods; the overlapping time periods of the multiple high-resolution forecast models in each future time period are obtained; the fusion weights of the overlapping high-resolution forecast models in each overlapping time period are calculated; the overlapping high-resolution forecast models in the corresponding overlapping time periods are fused according to the fusion weights to obtain the fused forecast model; the fusion weights of the overlapping high-resolution forecast models in each overlapping time period are calculated, including: obtaining the grid point precipitation forecasts of each high-resolution forecast model in the overlapping time period; calculating the error between the grid point precipitation forecasts of each high-resolution forecast model in the overlapping time period and the actual rainfall in the overlapping time period; and processing the error using a constructed convolutional neural network to obtain the fusion weights of the overlapping high-resolution forecast models in the overlapping time period. The model building module is used to build flood forecasting models for reservoir inflow. The inflow forecast module is used to obtain the reservoir evapotranspiration forecast. The reservoir evapotranspiration forecast and precipitation forecast are input into the flood forecast model for prediction to obtain the reservoir inflow forecast.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the rain-flood coupling forecasting method for reservoir inflow as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the stormwater-flood coupling forecasting method for reservoir inflow as described in any one of claims 1-4.

Images