Similar rainfall process searching method considering drainage basin rainfall center transfer path

Abstract

The invention belongs to the technical field of hydrological forecasting, and particularly provides a similar rainfall process searching method considering a drainage basin rainfall center transfer path, and the method comprises the steps: collecting drainage basin rainfall condition data; calculating a central moving path per 3h rainfall in a historical rainfall process; classifying historical rainfall processes according to the rainfall center moving path; according to future rainfall information, identifying 3h-by-3h rainfall center transfer paths, and performing classification; and performing similarity analysis on the 3h-by-3h area rainfall sequences of the forecast and historical rainfall processes, and further selecting the rainfall process with the highest similarity from the historical processes of the same category as the historical similar process of the rainfall according to the time sequence similarity. According to the method, the spatial distribution of the rainfall process and the dynamic change process of the rainfall level along with time are fully considered, and the accuracy of similar rainfall search is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of hydrological forecasting, and in particular relates to a method for searching similar rainfall processes taking into account the transfer path of rainfall centers in a watershed. Background Art

[0002] Flood forecasting is a key component of flood prevention and disaster reduction, as well as the efficient use of water resources. Accurate flood forecasting can support scientific flood control and scheduling decisions, effectively reduce loss of life and property caused by floods, and improve water resource utilization in river basins. Current flood forecasting methods mostly rely on hydrological models. Traditional physical hydrological models based on hydrological forecasting models rely on complex parameter calibration and real-time data-driven methods, resulting in problems such as high parameter calibration difficulty and insufficient dynamic parameter adaptability. Similar rainfall and flood forecasting methods utilize machine learning, big data, and other technologies to explore the rainfall and flood patterns of historical rainfall and flood events. By matching real-time rainfall processes with historical processes, flood forecast results can be quickly generated, providing new insights for flood forecasting.

[0003] However, the spatial distribution and rainfall amount of a rainfall process change dynamically over time. Existing similarity forecasting methods mostly focus on establishing a mapping between the rainfall characteristic index set and the flood characteristic index set of a rainfall flood process. The quantitative representation of the spatiotemporal distribution characteristics such as the spatial differentiation characteristics of rainfall and the movement path of rain belts is insufficient, resulting in biased screening of similar events, especially in complex scenarios with dynamic displacement of the rainfall center or frequent local severe convection. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for searching similar rainfall processes taking into account the transfer path of the rainfall center in the basin, which fully considers the spatial distribution of the rainfall process and the dynamic change process of the rainfall level over time, thereby improving the accuracy of similar rainfall search.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: A method for finding similar rainfall processes considering the transfer path of rainfall centers in a watershed includes the following steps: S1. Collection of water and rainfall data in the basin; S2. Calculation of the moving path of the rainfall center every 3 hours during the historical rainfall process; S3. Classify historical rainfall processes according to the movement path of the rainfall center; S4. Based on the future rainfall information, identify the 3-hour rainfall center transfer path and classify it; S5. Perform similarity analysis on the 3-hour surface rainfall series of the forecast and historical rainfall processes, and further select the rainfall process with the highest similarity from the same category of historical processes based on the similarity of the time series as the historical similar process of this rainfall.

[0006] In the preferred solution, in step S1, the water and rainfall data of the basin include hourly flow data of the basin control station, hourly basin surface rainfall data and site rainfall data, historical water and rainfall data of the basin and rainfall site information within the basin, and the historical water and rainfall data of the basin include compiled data of historical flood events in the basin.

[0007] In a preferred embodiment, step S1 includes the following sub-steps: S11. Process outliers and missing values ​​in the flow data series of the basin control station. If the basin data at a certain time point is abnormal or missing, linear interpolation is performed using the flow data of two adjacent time points. If the basin control station is a reservoir, the inflow flow data is smoothed. S12. Unify the time series length of rainfall data at each rainfall station in the basin, process abnormal values ​​and missing values ​​in the rainfall data series of the stations, and if the rainfall data of a station is abnormal or missing, replace or supplement it with the average rainfall data of the three nearest neighboring stations; S13. The collected historical flood data include rainfall start time, rainfall end time, flood rise time, flood recede time, flood peak flow, and peak time.

[0008] In a preferred solution, in step S11, a Savitzky-Golay filter is used to smooth the incoming flow data.

[0009] In a preferred embodiment, step S2 includes the following sub-steps: S21, extracting rainfall data from each rainfall station within the rainfall start time and rainfall end time period during a rainfall flood process, and calculating the cumulative rainfall of each station every 3 hours; S22. Calculate the mean 3-hour rainfall of each rainfall station and its three nearest stations, take the central station corresponding to the one with the largest mean rainfall as the location of the 3-hour rainfall center, and connect the rainfall center stations in chronological order to form the movement path of the rainfall center of this process.

[0010] In a preferred embodiment, step S3 includes the following sub-steps: S31. Calculate the path characteristic indicators for each 3-hour rainfall center path of each rainfall process, including the ratio Rc of the number of sites of the identified rainfall center to the total number of sites, the path fractal dimension FD, and the path direction entropy. H ; S32, based on the three path characteristic indices calculated in step S31, calculate the discrimination index to determine whether a rainfall process belongs to a basin-type complex rainfall process. S The calculation method is: ; Where, 、 、 Respectively represent the ratio of the total number of sites Rc, the path fractal dimension FD and the path direction entropy H The weights assigned to the three indicators; when When , the rainfall process is classified as Class 0, which means that the rainfall center moving path covers a large area and the moving path is complex; S33. For other rainfall processes that are not classified as complex paths, a hierarchical clustering algorithm is used to cluster all rainfall center paths.

[0011] In a preferred solution, in step S31, the calculation formulas for various path characteristic indicators are as follows: 1) The ratio Rc of the number of stations with identified rainfall centers to the total number of stations reflects the mobile spatial coverage of rainfall centers. The calculation formula is: ; in, is the number of stations identified as rainfall centers, is the number of all rainfall stations in the basin; 2) Path fractal dimension FD reflects the complexity of the movement path of the rainfall center. Its calculation formula is: ; in, l is the actual total length of the rainfall center track, d is the straight-line distance between the starting and ending points of the trajectory; 3) Path direction entropy H , reflecting the change in the direction of the rainfall center path, and its calculation formula is: ; Among them, K is the number of direction bins, is the frequency that the trajectory movement direction falls into the kth interval. The larger the H value is, the more disordered the direction change is.

[0012] In a preferred solution, in step S33, when clustering all rainfall center paths using the hierarchical clustering method, the DTW distance is used to characterize the difference between the rainfall center paths of two different rainfall processes, and the calculation method is as follows: Assumptions and Represents the rainfall center coordinate sequence of two rainfall processes, where each point and They are all two-dimensional vectors, representing the coordinates of the rainfall center point; Calculate the Euclidean distance between each pair of points in the two sequences. The calculation formula is: ; in, For rainfall process X No. i The coordinates of the rainfall center are two-dimensional coordinate points. is the first dimension coordinate value, is the second dimension coordinate value; For rainfall process Y No. j The coordinates of the rainfall center, is the first dimension coordinate value, is the second dimension coordinate value; for and The Euclidean distance between After the above calculation, we get Matrix D ,in According to the following recursive formula: ; in, That is the sequence and sequence The DTW distance between them.

[0013] In a preferred embodiment, step S4 includes the following sub-steps: S41. Calculate a 10-day 3-hourly forecast surface rainfall data sequence based on the 10-day 3-hourly forecast rainfall product, and interpolate from the grid to the rainfall station to obtain a 10-day 3-hourly forecast rainfall data sequence at each rainfall station location; S42, splicing the 3-hourly measured surface rainfall data for 10 days before the forecast day and the 3-hourly forecast surface rainfall data for the next 10 days to form a continuous surface rainfall sequence, and splicing the 3-hourly measured rainfall data for each rain gauge station for 10 days before the forecast day and the 3-hourly forecast rainfall data for each rain gauge station for the next 10 days obtained by interpolation to form a continuous station rainfall data sequence; S43. Set a surface rainfall interval threshold T. If the interval between two rainfall periods exceeds the threshold T, they are considered to be two independent rainfall processes. Based on this rule, identify a complete rainfall process in the basin around the forecast date and record the rainfall start and end times. S44, obtaining rainfall data from each rainfall station within the time period, and calculating the rainfall center path of this rainfall process according to the method described in step S2; S45: Classify the real-time rainfall events according to the method described in step S3.

[0014] In a preferred embodiment, step S5 includes the following sub-steps: S51. Calculate the DTW distance between the surface rainfall series of the historical rainfall flood process with the same rainfall center path as the forecast rainfall process and the surface rainfall series of the forecast rainfall process; S52. Select the historical rainfall process with the smallest DTW distance to the forecast rainfall process as the similar rainfall process for the current forecast process.

[0015] This paper proposes a method for finding similar rainfall processes that considers the path of rainfall centers within a watershed. The search for similar historical rainfall processes occurs in two steps. First, based on the paths of rainfall centers within each rainfall process, a complexity index representing the path of rainfall centers is calculated to classify historical rainfall processes within the watershed. Second, based on the similarity of surface rainfall time series within the watershed, the most similar historical rainfall process is found within the same category. The method's key innovation lies in fully considering the spatial distribution of rainfall processes and the dynamic changes in rainfall levels over time, resulting in more scientific and accurate matching results for similar rainfall processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below with reference to the accompanying drawings and examples: Figure 1 This is a flow chart of a method for searching similar rainfall processes considering the transfer path of the rainfall center in a watershed according to the present invention; Figure 2 This is a diagram of rainfall flow process in part of Qujiang River Basin; Figure 3 This is a 3-hour moving path diagram of the rainfall center during a rainstorm in the Qujiang River Basin. Figure 4 It is a hierarchical clustering dendrogram of the historical rainfall process; Figure 5 It is the rainfall center path of the forecast rainfall process and the rainfall center path of the same type of historical rainfall process; Figure 6 is the final result of searching for similar historical rainfall processes; Figure 7 It is a comparison chart between the forecast surface rainfall series and the surface rainfall series of similar historical rainfall processes. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] like Figure 1As shown in the figure, a method for searching similar rainfall processes considering the transfer path of the rainfall center in the basin is proposed. Based on the collected historical water and rainfall data of the basin, the method is divided into two parts to search for similar rainfall processes starting from the transfer of the rainfall center over time and the time distribution of the average rainfall in the basin. The spatial distribution of the rainfall process and the dynamic change of the rainfall level over time are fully considered to construct a basin-like rainfall and flood forecasting method considering the spatiotemporal distribution of rainfall in the basin.

[0019] The specific steps include: S1. Collection of water and rainfall data in the basin.

[0020] The water and rainfall data of the river basin include hourly flow data of the river basin control station, hourly river basin surface rainfall data and station rainfall data, historical water and rainfall data of the river basin and rainfall station information within the river basin. The historical water and rainfall data of the river basin include the compilation data of historical flood events in the river basin.

[0021] The following sub-steps are included: S1. Process outliers and missing values ​​in the flow data series of the basin control station. If the basin data at a certain time point is abnormal or missing, linear interpolation is performed using the flow data of two adjacent time points. If the basin control station is a reservoir, the Savitzky-Golay filter is used to smooth the inflow flow data. S12. Unify the time series length of rainfall data at each rainfall station in the basin, process abnormal values ​​and missing values ​​in the rainfall data series of the stations, and if the rainfall data of a station is abnormal or missing, replace or supplement it with the average rainfall data of the three nearest neighboring stations; S13. The collected historical flood data include rainfall start time, rainfall end time, flood rise time, flood recede time, flood peak flow, and peak time.

[0022] In this example, the Qujiang River Basin in the upper reaches of the Yangtze River is taken as an example. Hourly flow data, station rainfall data, and surface rainfall data are collected from 2015 to 2024. A total of 81 flood processes are also collected from 2015 to 2024. Figure 2 This is a diagram of some of the rainfall flow processes.

[0023] S2. Calculation of the 3-hour moving path of the rainfall center during historical rainfall processes: Statistically analyze the 3-hour rainfall center position and rainfall amount during all previous rainstorms and floods in the basin to obtain the 3-hour moving path of the rainfall center during historical rainfall processes.

[0024] The following sub-steps are included: S21. Extract rainfall data from each rainfall station within the rainfall start time and rainfall end time period during a rainfall flood process, and calculate the cumulative rainfall of each station every 3 hours.

[0025] S22. Calculate the mean 3-hour rainfall of each rainfall station and its three nearest stations, take the central station corresponding to the one with the largest mean rainfall as the location of the 3-hour rainfall center, and connect the rainfall center stations in chronological order to form the movement path of the rainfall center of this process.

[0026] Figure 3 This is a 3-hour movement path map of the rainfall center during a rainfall flood process in the Qujiang River Basin.

[0027] S3. Classify historical rainfall processes based on the movement path of the rainfall center.

[0028] The following sub-steps are included: S31. Calculate the path characteristic indicators for each 3-hour rainfall center path of each rainfall process, including the ratio Rc of the number of sites of the identified rainfall center to the total number of sites, the path fractal dimension FD, and the path direction entropy. H .

[0029] The calculation formulas for each path characteristic index are as follows: 1) The ratio Rc of the number of stations with identified rainfall centers to the total number of stations reflects the mobile spatial coverage of rainfall centers. The calculation formula is: ; in, is the number of stations identified as rainfall centers, is the total number of rainfall stations in the basin.

[0030] 2) Path fractal dimension FD reflects the complexity of the movement path of the rainfall center. Its calculation formula is: ; in, l is the actual total length of the rainfall center track, d is the straight-line distance between the starting and ending points of the trajectory.

[0031] 3) Path direction entropy H , reflecting the change in the direction of the rainfall center path, and its calculation formula is: ; Among them, K is the number of direction bins, usually 8, representing 8 azimuths. is the frequency that the trajectory movement direction falls into the kth interval. The larger the H value is, the more disordered the direction change is.

[0032] S32, based on the three path characteristic indices calculated in step S31, calculate the discrimination index to determine whether a rainfall process belongs to a basin-type complex rainfall process. S The calculation method is: ; Where, 、 、 Respectively represent the ratio of the total number of sites Rc, the path fractal dimension FD and the path direction entropy H The weights assigned to the three indicators.

[0033] In this embodiment, w 1=0.5, w 2=0.2, w 3=0.3.

[0034] when When , the rainfall process is classified as Class 0, which means that the rainfall center moving path covers a large area and the moving path is complex. It is believed that the rainfall process covers a small area and has a relatively simple movement path, and its movement path will be further classified.

[0035] S33. For other rainfall processes that are not classified as complex paths, a hierarchical clustering algorithm is used to cluster all rainfall center paths.

[0036] When clustering all rainfall center paths using the hierarchical clustering method, the DTW distance is used to characterize the differences between the rainfall center paths of two different rainfall processes: The DTW distance between the rainfall center paths of each two different rainfall processes is calculated as follows: Assumptions and Represents the rainfall center coordinate sequence of two rainfall processes, where each point and They are all two-dimensional vectors, representing the coordinates of the rainfall center point; Calculate the Euclidean distance between each pair of points in the two sequences. The calculation formula is: ; in, For rainfall process X No. i The coordinates of the rainfall center are two-dimensional coordinate points. is the first dimension coordinate value, is the second dimension coordinate value; For rainfall process Y No. j The coordinates of the rainfall center, is the first dimension coordinate value, is the second dimension coordinate value; for and The Euclidean distance between .

[0037] After the above calculation, we get Matrix D ,in According to the following recursive formula: ; in, That is the sequence and sequence The DTW distance between them.

[0038] Figure 4 This is a hierarchical clustering dendrogram for the historical rainfall process.

[0039] S4. During the real-time forecast process, based on the rainfall information for the next 10 days, identify the 3-hour rainfall center transfer path and classify it.

[0040] The following sub-steps are included: S41. Calculate the surface rainfall data sequence for the next 10 days, which is based on the 3-hourly forecast rainfall product for the next 10 days, and interpolate from the grid to the rainfall station to obtain the 3-hourly forecast rainfall data sequence for the next 10 days at each rainfall station location.

[0041] In this embodiment, the grid rainfall product released at 8:00 am on July 12, 2024 is obtained. The product covers the Qujiang River Basin, has a forecast period of 240 hours, and a time scale of 3 hours.

[0042] S42. Splice the 3-hourly measured surface rainfall data for 10 days before the forecast day with the 3-hourly forecast surface rainfall data for the next 10 days to form a continuous surface rainfall sequence. Splice the 3-hourly measured rainfall data for each rain gauge station for 10 days before the forecast day with the 3-hourly forecast rainfall data for each rain gauge station for the next 10 days obtained by interpolation to form a continuous station rainfall data sequence.

[0043] S43. Set the surface rainfall interval time threshold T. If the interval time between two rainfall periods exceeds the threshold T, they are considered to be two independent rainfall processes. According to this rule, identify a complete rainfall process in the basin before and after the forecast day, and record the rainfall start time and rainfall end time.

[0044] In this example, the rainfall interval threshold is set to 24 hours. It is calculated that there will be a rainfall process in the Qujiang River Basin between July 12 and July 22. The information of this rainfall process is shown in Table 1. This rainfall will start at 2:00 on July 13 and last until 19:00 on July 16, with a cumulative rainfall of 82.6 mm.

[0045]

[0046] S44: Obtain the rainfall data of each rainfall station within the time period, and calculate the rainfall center path of this rainfall process according to the method described in step S2.

[0047] S45. Classify the real-time rainfall events according to the method described in step S3. The rainfall process information of the same category is shown in the following table: The rainfall process information of the same category is shown in Table 2, Figure 5 From the rainfall center path of this predicted rainfall process and the rainfall center path of historical rainfall processes of the same category, it can be seen that the rainfall centers of rainfall processes classified into this category are more concentrated in the northwest of the basin.

[0048]

[0049] S5. Perform similarity analysis on the 3-hour surface rainfall series of the forecast and historical rainfall processes, and further select the rainfall process with the highest similarity from the same category of historical processes based on the similarity of the time series as the historical similar process of this rainfall.

[0050] The following sub-steps are included: S51. Calculate the DTW distance between the surface rainfall series of the historical rainfall flood process similar to the rainfall center path of the forecast rainfall and the surface rainfall series of the forecast rainfall process.

[0051] S52. Select the historical rainfall process with the smallest DTW distance to the forecast rainfall process as the similar rainfall process to this forecast process. The most similar historical rainfall process is the rainfall process from 3:00 on September 7, 2019 to 15:00 on September 10, 2019.

[0052] Figure 6 Find the results for the final historical similar rainfall process, Figure 7 This is a comparison chart of the predicted surface rainfall series and the surface rainfall series of similar historical processes. It can be seen that the two rainfall processes have strong similarities in terms of both the spatial distribution of cumulative rainfall and the temporal distribution trend of the surface rainfall series.

[0053] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for finding similar rainfall processes considering the transfer path of rainfall centers in a watershed, characterized by: The following steps are involved: S1. Collection of water and rainfall data in the basin; S2. Calculation of the moving path of the rainfall center every 3 hours during the historical rainfall process; S3. Classify historical rainfall processes according to the movement path of the rainfall center; S4. Based on the future rainfall information, identify the 3-hour rainfall center transfer path and classify it; S5. Perform similarity analysis on the 3-hour surface rainfall series of the forecast and historical rainfall processes, and further select the rainfall process with the highest similarity from the same category of historical processes based on the similarity of the time series as the historical similar process of this rainfall.

2. The method for searching similar rainfall processes considering the transfer path of the rainfall center in a watershed according to claim 1 is characterized in that: In step S1, the water and rainfall data of the basin include hourly flow data of the basin control station, hourly basin surface rainfall data and site rainfall data, historical water and rainfall data of the basin and rainfall site information within the basin. The historical water and rainfall data of the basin include compiled data of historical flood events in the basin.

3. The method for searching similar rainfall processes considering the transfer path of the rainfall center in a watershed according to claim 2 is characterized in that: The step S1 includes the following sub-steps: S11. Process outliers and missing values ​​in the flow data series of the basin control station. If the basin data at a certain time point is abnormal or missing, linear interpolation is performed using the flow data of two adjacent time points. If the basin control station is a reservoir, the inflow flow data is smoothed. S12. Unify the time series length of rainfall data at each rainfall station in the basin, process abnormal values ​​and missing values ​​in the rainfall data series of the stations, and if the rainfall data of a station is abnormal or missing, replace or supplement it with the average rainfall data of the three nearest neighboring stations; S13. The collected historical flood data include rainfall start time, rainfall end time, flood rise time, flood recede time, flood peak flow, and peak time.

4. The method for searching similar rainfall processes considering the transfer path of the rainfall center in a watershed according to claim 3 is characterized in that: In step S11, a Savitzky-Golay filter is used to smooth the incoming flow data.

5. The method for searching similar rainfall processes considering the transfer path of the rainfall center in a watershed according to claim 1 is characterized in that: The step S2 includes the following sub-steps: S21, extracting rainfall data from each rainfall station within the rainfall start time and rainfall end time period during a rainfall flood process, and calculating the cumulative rainfall of each station every 3 hours; S22. Calculate the mean 3-hour rainfall of each rainfall station and its three nearest stations, take the central station corresponding to the one with the largest mean rainfall as the location of the 3-hour rainfall center, and connect the rainfall center stations in chronological order to form the movement path of the rainfall center of this process.

6. The method for searching similar rainfall processes considering the transfer path of the rainfall center in a watershed according to claim 1 is characterized in that: The step S3 includes the following sub-steps: S31. Calculate the path characteristic indicators for each 3-hour rainfall center path of each rainfall process, including the ratio Rc of the number of sites of the identified rainfall center to the total number of sites, the path fractal dimension FD, and the path direction entropy. H ; S32, based on the three path characteristic indices calculated in step S31, calculate the discrimination index to determine whether a rainfall process belongs to a basin-type complex rainfall process. S The calculation method is: ； Where, 、 、 Respectively represent the ratio of the total number of sites Rc, the path fractal dimension FD and the path direction entropy H The weights assigned to the three indicators; when When , the rainfall process is classified as Class 0, which means that the rainfall center moving path covers a large area and the moving path is complex; S33. For other rainfall processes that are not classified as complex paths, a hierarchical clustering algorithm is used to cluster all rainfall center paths.

7. A similar rainfall process search method considering the transfer path of the rainfall center in the basin according to claim 6, characterized in that: In step S31, the calculation formulas for each path characteristic index are as follows: 1) The ratio Rc of the number of stations with identified rainfall centers to the total number of stations reflects the mobile spatial coverage of rainfall centers. The calculation formula is: ； in, is the number of stations identified as rainfall centers, is the number of all rainfall stations in the basin; 2) Path fractal dimension FD reflects the complexity of the movement path of the rainfall center. Its calculation formula is: ； in, l is the actual total length of the rainfall center track, d is the straight-line distance between the starting and ending points of the trajectory; 3) Path direction entropy H , reflecting the change in the direction of the rainfall center path, and its calculation formula is: ； Among them, K is the number of direction bins, is the frequency that the trajectory movement direction falls into the kth interval. The larger the H value is, the more disordered the direction change is.

8. A similar rainfall process search method considering the transfer path of the rainfall center in a watershed according to claim 6, characterized in that: In step S33, when clustering all rainfall center paths using the hierarchical clustering method, the DTW distance is used to characterize the difference between the rainfall center paths of two different rainfall processes. The calculation method is as follows: Assumptions and Represents the rainfall center coordinate sequence of two rainfall processes, where each point and They are all two-dimensional vectors, representing the coordinates of the rainfall center point; Calculate the Euclidean distance between each pair of points in the two sequences. The calculation formula is: ； in, For rainfall process X No. i The coordinates of the rainfall center are two-dimensional coordinate points. is the first dimension coordinate value, is the second dimension coordinate value; For rainfall process Y No. j The coordinates of the rainfall center, is the first dimension coordinate value, is the second dimension coordinate value; for and The Euclidean distance between After the above calculation, we get Matrix D ,in According to the following recursive formula: ； in, That is the sequence and sequence The DTW distance between them.

9. The method for searching similar rainfall processes considering the transfer path of the rainfall center in a watershed according to claim 1 is characterized in that: The step S4 includes the following sub-steps: S41. Calculate a 10-day 3-hourly forecast surface rainfall data sequence based on the 10-day 3-hourly forecast rainfall product, and interpolate from the grid to the rainfall station to obtain a 10-day 3-hourly forecast rainfall data sequence at each rainfall station location; S42, splicing the 3-hourly measured surface rainfall data for 10 days before the forecast day and the 3-hourly forecast surface rainfall data for the next 10 days to form a continuous surface rainfall sequence, and splicing the 3-hourly measured rainfall data for each rain gauge station for 10 days before the forecast day and the 3-hourly forecast rainfall data for each rain gauge station for the next 10 days obtained by interpolation to form a continuous station rainfall data sequence; S43. Set a surface rainfall interval threshold T. If the interval between two rainfall periods exceeds the threshold T, they are considered to be two independent rainfall processes. Based on this rule, identify a complete rainfall process in the basin around the forecast date and record the rainfall start and end times. S44, obtaining rainfall data from each rainfall station within the time period, and calculating the rainfall center path of this rainfall process according to the method described in step S2; S45: Classify the real-time rainfall events according to the method described in step S3.

10. The method for searching similar rainfall processes considering the transfer path of the rainfall center in a watershed according to claim 1, characterized in that: The step S5 includes the following sub-steps: S51. Calculate the DTW distance between the surface rainfall series of the historical rainfall flood process with the same rainfall center path as the forecast rainfall process and the surface rainfall series of the forecast rainfall process; S52. Select the historical rainfall process with the smallest DTW distance to the forecast rainfall process as the similar rainfall process for the current forecast process.

Images