A method and system for time-scale conversion of parameters in a high-frequency real-time flood forecasting model.

Abstract

This invention discloses a method and system for time-scale conversion of parameters in high-frequency real-time flood forecasting models. Addressing the technical shortcomings of hydrological model parameters in minute-level, sub-hourly, and hourly high-frequency forecasting scenarios, such as significant time-scale effects, low efficiency of multi-scale independent calibration, and poor accuracy of direct parameter transfer, this invention constructs multi-time-scale hydrological and meteorological parameter sample sets, establishes a parameter time-scale sensitivity function to screen sensitive parameters, structurally identifies parameter scale variation patterns, constructs quantitative conversion relationships that consider both hydrological physical meaning and value boundaries, forms an expandable parameter conversion knowledge base, and achieves cross-scale and cross-basin adaptive parameter migration, ultimately completing high-frequency real-time flood forecasting and accuracy verification. This invention can quickly achieve multi-scale parameter conversion at 15-minute, 30-minute, and 1-hour scales, with the converted forecast NSE consistently above 0.9, significantly reducing parameter calibration workload and improving the deployment efficiency and simulation accuracy of high-frequency flood forecasts.

Description

Technical Field

[0001] This invention belongs to the field of hydrology, water resources and flood control and disaster reduction engineering technology, specifically relating to a method and system for converting time scale parameters of hydrological models for high-frequency real-time flood forecasting. Background Technology

[0002] In existing real-time flood forecasting operations, as the frequency of meteorological and hydrological data acquisition continues to increase, forecasting systems are gradually expanding from daily and hourly scales to minute and sub-hourly scales. At different forecast time scales, the parameters of conceptual hydrological models, semi-distributed or distributed hydrological models do not remain constant, but exhibit significant time-scale effects as the input and output time resolution changes. Existing research only provides conversion methods for parameters of water resource prediction models at monthly and annual scales, lacking identification and quantification conversion techniques for parameters of high-frequency real-time flood forecasting models at hourly or even minute scales.

[0003] To address the challenges of applying models at different time scales, two approaches are typically employed: one is to conduct independent parameter calibration at each time scale; the other is to directly transplant parameters from existing time scales, perform empirical conversions, or make simple proportional adjustments. The former is labor-intensive, heavily reliant on data, exhibits poor stability during high-frequency periods, and struggles to support rapid deployment of forecast models; the latter lacks rigorous quantitative evidence, fails to reflect the true patterns of parameter changes over time, and is prone to parameter distortion, model mechanism corruption, and a significant decrease in forecast accuracy.

[0004] The time-scale responses of different model parameters are inconsistent: some parameters are highly sensitive, while others change less; some parameters are suitable for linear transformation, while others are constrained by hydrological runoff generation mechanisms and require nonlinear transformations that preserve physical meaning and value range. Existing technologies cannot construct a unified analytical framework for multiple models and multiple time scales, nor can they identify the parameters that need to be transformed and the corresponding transformation relationships.

[0005] Therefore, the existing technology has the following shortcomings: First, it lacks a sensitivity identification method for the time scale effect of hydrological model parameters; second, it lacks a structured representation and identification method for the time scale variation law of parameters; third, it lacks a quantitative conversion relationship of the time scale of high-frequency real-time flood forecasting model parameters that takes into account the hydrological and physical meaning, the reasonable range of parameter values, and the function of model modules; fourth, it lacks a mechanism for adaptive matching and extended application of parameter time scales between different high-frequency forecasting models. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for time scale conversion of parameters in high-frequency real-time flood forecasting models, thereby solving the aforementioned technical problems existing in the prior art.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following solution:

[0008] A method for time-scale transformation of parameters in a high-frequency real-time flood forecasting model includes the following steps:

[0009] Step S1: Collect rainfall, evapotranspiration, and flow data at multiple time scales to construct a multi-time scale model parameter sample set;

[0010] Step S2: Based on the multi-timescale model parameter sample set, construct the parameter timescale sensitivity function, calculate and screen out the sensitive parameters that respond significantly to changes in timescale;

[0011] Step S3: Based on the sensitive parameters obtained from the screening, perform standardization processing, construct a parameter-time scale correspondence dataset, and structurally identify the change patterns of the sensitive parameters;

[0012] Step S4: Based on the identified change patterns, combined with the hydrological and physical significance, parameter value boundaries, and calculation module functions, construct the parameter time scale quantitative conversion relationship;

[0013] Step S5: Structure and store the model type, calculation module, parameter attributes, sensitivity, change pattern, quantization conversion relationship and quality evaluation index to establish a parameter conversion relationship knowledge base;

[0014] Step S6: Based on the similarity of computing module functions, the consistency of parameter physical meaning, the matching of parameter value ranges, and the proximity of time scale sensitivity, adaptively match the transformation relationship from the parameter transformation relationship knowledge base to realize the transfer application of parameters across time scales, watersheds, or models;

[0015] Step S7: Input the target scale parameters obtained from the migration into the flood forecasting model, carry out high-frequency real-time flood forecasting, and verify the results using parameter error index and flow simulation accuracy index.

[0016] Further optimization, step S1.1: Set a unified time scale set ΔT = for high-frequency real-time flood forecasting. The timescale set ΔT includes at least two timescales selected from 5 minutes, 10 minutes, 15 minutes, 30 minutes, 1 hour, 3 hours, 6 hours, and 24 hours; wherein The source timescale.

[0017] Step S1.2: Collect raw rainfall data P, evapotranspiration data E, and flow data Q for the target watershed at various time scales.

[0018] Step S1.3: Process the raw data of rainfall, evapotranspiration, and flow rate according to each single time scale. A unified scale conversion is performed, specifically as follows:

[0019] Accumulate rainfall data over time periods according to the target scale: ;

[0020] Evapotranspiration data are time-division converted according to the target scale: ;

[0021] Perform water balance conversion on the target scale based on the flow data: ;

[0022] Where n is a single time scale The number of original data time periods included.

[0023] Step S1.4: For each time scale The parameters of the hydrological model were calibrated to obtain the parameter set. :

[0024] ,in Let N be the calibration value of the i-th parameter at the j-th time scale, where N is the total number of parameters and M is the total number of time scales.

[0025] Step S1.5: Calculate the values ​​of each individual time scale ΔT j The corresponding P(ΔT) j ), E(ΔT) j ), Q(ΔT) j ), and the set of model parameters calibrated under this single time scale. One-to-one matching is performed to form a multi-timescale model parameter sample set D;

[0026] .

[0027] Further optimization, step S2 specifically includes the following sub-steps:

[0028] Step S2.1: Based on the multi-timescale parameter samples obtained in Step S1, calculate the parameter timescale sensitivity:

[0029] ; (1)

[0030] in, For parameters, time-scale sensitivity; Let be the change in the i-th model parameter relative to the source time scale at the j-th single time scale; For the i-th model parameter at the source time scale The parameter values ​​below; Let j be the j-th single time scale in the set of time scales; Let be the change in the j-th single time scale relative to the source time scale. The source timescale.

[0031] Step S2.2: Calculate the average parameter sensitivity under multi-scale combination. :

[0032] ; (2)

[0033] Where M is the total number of time scales in the set ΔT, and N is the total number of model parameters.

[0034] Step S2.3: Parameters with sensitivity values ​​higher than the average sensitivity value are identified as sensitive parameters, and the set of sensitive parameters is output.

[0035] Further optimization, step S3 specifically includes the following sub-steps:

[0036] Step S3.1: Standardize the changes in the sensitive parameters and the changes over the time scale.

[0037] Step S3.2: Construct a dataset F showing the variation of sensitive parameters over time:

[0038] ; (3)

[0039] Where F is the dataset showing the correspondence between parameter changes and changes over time; This represents the relative change of the i-th parameter relative to the source time scale at the j-th time scale. It represents the relative change in time scale of the j-th time scale relative to the source time scale.

[0040] Step S3.3: Based on the aforementioned correspondence dataset By using a dual-axis line graph or data matrix identification chart, the direction, magnitude, monotonicity, inflection point, and degree of nonlinearity of the sensitive parameters are identified, and the change patterns are classified into linear and nonlinear change types.

[0041] Further optimization is achieved in step S4. Based on the parameter time-scale sensitivity function, the variation pattern obtained in step S3, and the physical meaning of the parameters themselves, and fully considering the hydrological runoff generation, confluence mechanism, and characteristics of the calculation module to which the parameters belong, a machine learning or deep learning model is used to construct a parameter and time-scale quantification transformation mechanism. This establishes the quantification transformation relationship between model parameters and time-scale changes, specifically:

[0042] Step S4.1: Construct a linear quantization transformation relationship for parameters of the linear change type:

[0043] ; (4)

[0044] Step S4.2: For parameters exhibiting nonlinear attenuation, storage / discharge, outflow recursion, or value constraints within an interval, a nonlinear transformation model that maintains consistency between the physical meaning and the value boundaries is used for expression. The transformation relationship is expressed as follows:

[0045] when >1, and < hour: ; (5-1)

[0046] When 0 < <1, and < hour: ; (5-2)

[0047] in, For the target time scale Next, the i-th model parameter; Source timescale The baseline value of the i-th model parameter; As the source timescale, For the target time scale; k i a i The transformation coefficient for the i-th parameter is obtained by fitting the parameter-timescale dataset constructed in step S3.

[0048] Further optimization, step S5 specifically includes the following sub-steps:

[0049] Step S5.1: Construct a hierarchical index structure for the knowledge base. The hierarchical index structure includes, from top to bottom, the following layers: model type layer, calculation module layer, parameter physical meaning layer, parameter value range layer, sensitivity level layer, change pattern tag layer, conversion formula layer, conversion coefficient layer, and quality evaluation index layer.

[0050] Among them, the model type is used to distinguish between conceptual hydrological models, semi-distributed hydrological models, and distributed hydrological models; the calculation module is used to identify functional units such as evapotranspiration, runoff generation, water source delineation, slope runoff, and river runoff; the change law label is used to identify types such as linear change, nonlinear change, storage and discharge recursion, and proportional constraint.

[0051] Step S5.2: The multi-timescale model parameter sample set constructed in step S1 is classified and stored according to model type and calculation module.

[0052] Step S5.3: Mark the parameter time scale sensitivity and sensitive parameter screening results obtained in step S2 according to the sensitivity level and associate them with the corresponding parameter entries;

[0053] Step S5.4: Bind the parameter change pattern type identified in step S3 to the corresponding parameter entry in the form of a label.

[0054] Step S5.5: The parameter time scale quantization transformation relationship and transformation coefficients constructed in step S4 are associated with the corresponding change law labels in the form of callable functions.

[0055] Step S5.6: Calculate the average relative error of the parameters obtained in step S7. The Nash efficiency coefficient (NSE) quality evaluation index is added as a reliability indicator of the transformation relationship and stored separately.

[0056] Step S5.7: Establish a multi-dimensional index based on model type, calculation module, parameter name, time scale, and change pattern type, supporting dynamic updates, retrieval, and version management, forming an scalable parameter transformation relationship knowledge base. When a new watershed, model, or time scale sample is added, the knowledge base automatically updates the sensitivity statistics, transformation relationship coefficients, and quality evaluation information based on the new sample.

[0057] Further optimization, step S6 specifically includes the following sub-steps:

[0058] Step S6.1: Determine the set of parameters to be transformed for the target scene and the target time scale. and source timescale Extract the model type, computation module, physical meaning of parameters, value range and sensitivity features of the target scene.

[0059] Step S6.2: Based on the similarity of calculation module functions, the consistency of parameter physical meaning, the matching of parameter value ranges, and the similarity of time scale sensitivity, retrieve matching items from the parameter transformation relationship knowledge base, specifically as follows:

[0060] 1) Calculation module functional similarity: Match calculation modules that have the same hydrological processes (such as runoff generation and confluence) as the target scenario;

[0061] 2) Consistency of physical meaning of parameters: Match parameter entries with physical meanings consistent with the target parameters (such as the decline coefficient and outflow coefficient);

[0062] 3) Parameter value range matching: Match parameter entries whose overlap with the target parameter value range is greater than or equal to a preset threshold;

[0063] 4) Time scale sensitivity proximity: Match parameter items whose sensitivity level difference is ≤ preset level.

[0064] Step S6.3: Based on the change pattern type corresponding to the matching item, call the parameter time scale quantization transformation relationship and transformation coefficient k constructed in step S4. i ai .

[0065] Step S6.4: Set the parameter values ​​of the i-th model parameter at the source time scale. Source timescale and target time scale Substituting into the conversion formula, the target time scale parameter values ​​are calculated. .

[0066] Step S6.5: Calculate the... Perform hydrological physical boundary verification to ensure that the values ​​are within the preset reasonable hydrological range. Parameters that exceed the boundary are truncated or corrected according to physical constraints.

[0067] Step S6.6: Output the validated target parameters to complete the adaptive parameter transfer across time scales, watersheds, or models.

[0068] Further optimization, step S7 specifically includes the following sub-steps:

[0069] Step S7.1: Obtain the measured rainfall, evapotranspiration, and flow data sequences of the target watershed within the forecast period, and extract the measured flow Q at time t. o,t .

[0070] Step S7.2: Calculate the target time scale parameters obtained in step S6. Input a high-frequency flood forecasting model and run it to obtain the simulated flow rate Q at time t. c,t .

[0071] Step S7.3: Verify the accuracy of parameter transformation using the parameter average relative error index.

[0072] ; (6)

[0073] Where N is the number of model parameters, Let be the baseline value of the i-th model parameter at the source time scale. For the target time scale The transformed value of the i-th model parameter.

[0074] Step S7.4: Verify the accuracy of the flow simulation using the Nash efficiency coefficient:

[0075] ; (7)

[0076] in, t represents the arithmetic mean of the measured flow rate; L represents the total number of samples in the flow rate series; and t represents the time series number.

[0077] Step S7.5: Set the parameter error threshold With flow simulation accuracy threshold When satisfied ,and If the parameter transformation and forecast results are deemed valid, it indicates that the constructed parameter time scale quantization transformation relationship can achieve good parameter adaptive matching under the target time scale and has the ability to be extended to multiple models; if not, it is fed back to step S4 to adjust the transformation coefficients or step S5 to update the knowledge base.

[0078] Step S7.6: Output high-frequency real-time flood forecast results, parameter conversion accuracy and flow simulation accuracy reports to complete the forecast and verification process.

[0079] A time-scale conversion system for parameters of a high-frequency real-time flood forecasting model, used to implement the above method, includes:

[0080] The data preprocessing module is used to perform step S1, which involves collecting and organizing hydrological and meteorological data at multiple time scales and constructing a multi-time scale model parameter sample set.

[0081] The sensitivity analysis module is used to perform step S2, calculate the time-scale sensitivity of parameters, and filter sensitive parameters;

[0082] The pattern identification module is used to execute step S3, construct a parameter-time scale relationship dataset and identify the type of change pattern;

[0083] The quantization conversion module is used to execute step S4 and construct the parameter time scale quantization conversion relationship based on the type of change pattern.

[0084] The knowledge base storage module is used to execute step S5, which stores the model type, calculation module, parameter attributes, sensitivity, change law, quantization conversion relationship and quality evaluation index in a structured manner to establish a parameter conversion relationship knowledge base; and supports index retrieval and dynamic updates by calculation module, parameter, time scale and formula type.

[0085] The adaptive matching module is used to execute step S6. Based on the similarity of the calculation module functions, the consistency of the physical meaning of the parameters, the matching of the parameter value range, and the proximity of the time scale sensitivity, it adaptively matches the transformation relationship from the parameter transformation relationship knowledge base to realize the transfer application of parameters across time scales, watersheds, or models.

[0086] The forecasting application module is used to execute step S7, input the target scale parameters obtained from the migration into the flood forecasting model, carry out high-frequency real-time flood forecasting, and verify the results using parameter error index and flow simulation accuracy index.

[0087] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for time-scale conversion of parameters in a high-frequency real-time flood forecasting model.

[0088] Compared with the prior art, the present invention has the following advantages:

[0089] 1. The method described in this invention screens sensitive parameters through parameter time scale sensitivity analysis, and constructs a suitable quantitative conversion relationship by combining the parameter change pattern type, avoiding blind conversion and improving parameter conversion accuracy; at the same time, it adopts dual verification of parameter error and flow simulation accuracy to ensure the reliability of conversion results.

[0090] 2. The method described in this invention constructs a structured parameter transformation relationship knowledge base, supports multi-dimensional index retrieval and dynamic updates, realizes parameter migration and reuse across time scales, watersheds and models, reduces parameter calibration costs and improves high-frequency forecast efficiency.

[0091] 3. The method described in this invention forms a complete closed loop: sample construction → sensitivity analysis → pattern identification → quantification conversion → knowledge base establishment → adaptive transfer → accuracy verification. The steps described are reproducible and meet the requirements of hydrological forecasting mechanism and patent feasibility.

[0092] 4. The method described in this invention adopts two transformation methods, linear and piecewise nonlinear, for parameters with different variation patterns, adapting to the parameter characteristics of different hydrological models and different watersheds, and covering various high-frequency real-time flood forecasting scenarios. Attached Figure Description

[0093] Figure 1 This is a flowchart illustrating the overall process of a method for converting time scale parameters in a high-frequency real-time flood forecasting model according to the present invention.

[0094] Figure 2 A graph showing the variation of sensitive parameters in the Changshan River Basin over time.

[0095] Figure 3 The calculated and measured flow rates for the 15-minute, 30-minute, and 1-hour timescale model parameters of the Kaihua watershed are shown in the figure. Detailed Implementation

[0096] The technical solution of the present invention will be described in detail below with reference to the embodiments, but the scope of protection of the present invention is not limited to the embodiments described.

[0097] This embodiment selects two representative watersheds: the Changshan and Kaihua watersheds in the upper reaches of the Qiantang River, for the time-scale transformation of model parameters for high-frequency real-time flood forecasting. Historical flood data at 15-minute, 30-minute, and 1-hour timescales in the Changshan watershed are used to calibrate the corresponding time-scale model parameters. Then, the sensitivity of the parameters to time scales is analyzed, and a quantitative transformation relationship between different time scales and model parameters is constructed. Through this transformation relationship, the sensitive parameters of the Xin'anjiang model at the calibrated 1-hour timescale in the Kaihua watershed are transferred and quantified to the 30-minute and 15-minute timescales. Finally, the goodness-of-fit evaluation index between the model calculation results and the measured results in the Kaihua watershed at different time scales is examined.

[0098] Example 1:

[0099] In this embodiment, the implementation process of the time scale transformation method for high-frequency real-time flood forecasting model parameters is as follows: Figure 1 As shown, the specific steps include the following:

[0100] Step S1: Construction of multi-timescale sample sets

[0101] Data from 20 floods in the Changshan River Basin from 2018 to 2025 were compiled, including rainfall, evaporation, and flow data at three time scales: 1 hour, 30 minutes, and 15 minutes. Taking the distributed Xin'anjiang model based on natural river system division as an example, the SCE-UA parameter optimization algorithm was used, with the objective function calibration parameters being runoff depth relative error ≤15%, flood peak relative error ≤10%, NSE ≥0.8, and peak occurrence time difference ≤3 hours, forming a parameter-time scale corresponding sample set, as shown in Table 1.

[0102] Table 1. Excellence values ​​of model parameters at different time scales in the Changshan watershed.

[0103]

[0104] Statistical indices were applied to calculate parameters for 20 floods at different time scales, yielding an average NSE of 0.879 for the 1-hour scale, 0.896 for the 30-minute scale, and 0.898 for the 15-minute scale. The average NSE for all three time scales is greater than 0.85, indicating high forecast accuracy. This demonstrates that the obtained parameter values ​​at these three time scales can be used to construct quantitative transformation relationships.

[0105] Step S2: Calculation of parameter time-scale sensitivity

[0106] Calculate the ratio of the relative change of each model parameter between different time scales to the relative change of time scale according to formula (1) to obtain the sensitivity value of the parameter on each scale combination; then calculate the average of all scale combinations according to formula (2) to obtain the average sensitivity value of the parameter, as shown in Table 2. Parameters with an average sensitivity higher than the overall mean of the model parameters are selected as sensitive parameters.

[0107] Table 2. Calculated values ​​of model parameters' sensitivity over time scales

[0108]

[0109] After calculation, parameters with an average timescale sensitivity greater than 0.1 were selected as sensitive parameters for timescale transformation, resulting in four sensitive parameters: KI, KG, CS, and KE.

[0110] Step S3: Identification of Change Patterns

[0111] Construct a dataset F based on the variation law of sensitive parameters with time scale according to equation (3), and determine whether the parameter is linear or nonlinear based on the results of the dual-coordinate broken line graph and data matrix identification.

[0112] In this embodiment, the changes of the four sensitive parameters over time are as follows: Figure 2 As shown in the figure, (a), (b), (c), and (d) represent the changes of four sensitive parameters, KI, KG, CS, and KE, over time, respectively. Figure 2 It was determined that the three parameters KI, KG, and CS change non-linearly, while the parameter KE changes linearly.

[0113] Step S4: Quantitative Transformation Relationship Construction

[0114] For parameters in the linear approximation module, such as confluence parameters and empirical coefficient parameters with scale scaling characteristics, Equation (4) is used for modeling; for storage and discharge recursion parameters, outflow reduction parameters, or proportional coefficients with values ​​between 0 and 1, the nonlinear transformation relationship shown in the equation is used for modeling, and the transformation coefficients of Equation (5) are fitted and solved according to the random forest model. After modeling, the parameter boundary preservation, transformation stability, and simulation applicability are verified through mutual conversion experiments between different time scales. The quantization transformation formula of sensitive parameters with time scale is as follows:

[0115] (1) Formula for KI quantization conversion over time scale:

[0116] ;

[0117] In the formula, express Model parameter values ​​at different time scales The unit is minutes, T represents the number of minutes at the corresponding time scale, and KI represents the model parameter values ​​for the watershed to be converted at a 1-hour time scale.

[0118] (2) Formula for quantization conversion of KG over time scale:

[0119] ;

[0120] In the formula, express The model parameter values ​​at the time scale, where KG represents the model parameter values ​​of the watershed to be converted at the 1-hour time scale.

[0121] (3) CS quantization conversion formula with time scale:

[0122] ;

[0123] In the formula, express The model parameter values ​​at the time scale, where CS represents the model parameter values ​​of the watershed to be converted at a 1-hour time scale.

[0124] (4) KE quantization conversion formula with time scale:

[0125]

[0126] In the formula, express Model parameter values ​​at a given time scale, where KE represents model parameter values ​​known at a 1-hour time scale.

[0127] Step S5: Establishing a Parameter Conversion Knowledge Base

[0128] After completing the quantification and transformation relationships of the four sensitive parameters KI, KG, CS, and KE, a parameter transformation relationship knowledge base is established. The knowledge base uses transformation relationship records as the basic unit, and each record includes at least: model type, watershed name, calculation module, parameter name, parameter symbol, parameter physical meaning, parameter value range, source time scale, target time scale range, sensitivity value, sensitivity level, change law label, transformation relationship formula, relationship coefficient, boundary constraints, and quality evaluation index.

[0129] In this embodiment, KI corresponds to the soil outflow module, KG corresponds to the groundwater outflow module, CS corresponds to the surface runoff recession module, and KE corresponds to the Muskingan channel confluence module. The knowledge base labels KI and KG as "outflow ratio type—nonlinear polynomial transformation," CS as "recession coefficient type—logarithmic transformation," and KE as "confluence propagation time type—scale conversion." For example, the formula field in the KI record is... The applicable time scale range is 15 minutes to 60 minutes, and the value boundary is 0 < <1. Quality assessment information includes Changshan watershed modeling NSE.

[0130] After the knowledge base is established, module function indexes, parameter physical meaning indexes, time scale indexes, and formula type indexes are set for each transformation relationship. When a new watershed or a new model needs parameter time scale transformation, the system can directly retrieve the transformation relationship based on the index. After the new sample is validated, the new formula coefficients, NSE results, and applicable conditions can also be updated to the knowledge base, thereby improving the reliability of subsequent parameter migration.

[0131] Step S6: Adaptive Matching and Extended Applications

[0132] When it is necessary to migrate the parameters of the distributed Xin'anjiang model at a 1-hour time scale in the Kaihua River Basin to 30-minute and 15-minute time scales, first read the module to which the parameter to be converted belongs, its physical meaning, value range, sensitivity level, and source time scale; then retrieve the conversion relationship with the same or similar calculation module and physical meaning from the knowledge base.

[0133] Based on the existing 1-hour timescale sensitive parameters in the Kaihua watershed, the corresponding 15-minute and 30-minute timescale sensitive parameters were calculated, as shown in Table 3.

[0134] Furthermore, for hydrological models with similar structures, if their parameters have the same hydrological function as the parameters already in the knowledge base, such as representing soil outflow, groundwater outflow, surface runoff recession, or river confluence propagation time, then the corresponding conversion relationship can be extended and applied to the model.

[0135] Table 3. Results of the transformation of the best values ​​of sensitive parameters at different time scales in the Kaihua watershed.

[0136]

[0137] Step S7: Result Verification:

[0138] The representative flood event in the Kaihua River Basin on June 8, 2025, was used for verification. Figure 3 This is a comparison of rainfall-discharge processes at different time scales for the 20250608th flood in the Kaihua River Basin. Figure 3 Comparison of rainfall-flow processes at three timescales (15 min, 30 min, and 60 min) in (a), (b), and (c) respectively, where P represents the measured rainfall process, Q represents the measured flow process, and Qc represents the model-simulated flow process. Figure 3 It is evident that the peak location, flood volume, and process curve morphology of the simulated and measured flow processes at various time scales are highly consistent, indicating that the parameter time scale conversion method proposed in this invention can effectively maintain the accuracy of flood simulation at different time scales and meet the application requirements of high-frequency real-time flood forecasting.

[0139] Select the forecast simulation index shown in equation (7) The conversion results were verified. The calculated NSE values ​​were 0.913 for the 1-hour scale, 0.907 for the 30-minute scale, and 0.902 for the 15-minute scale. The NSE values ​​were all greater than 0.9 at the three time scales, indicating that the constructed parameter time scale conversion method can maintain the accuracy of flood process simulation well and can directly obtain the corresponding time scale parameters required by the high-frequency forecast model. The specific statistical results are shown in Table 4.

[0140] Table 4. Statistics on the simulation accuracy of time-scale transformation parameters in the Kaihua watershed.

[0141]

[0142] Example 2:

[0143] In this embodiment, a high-frequency real-time flood forecasting model parameter time-scale conversion system based on the above method is described. This system can be deployed on a server, workstation, or flood forecasting operational system. It includes a data preprocessing module, a sensitivity analysis module, a pattern identification module, a quantization conversion module, a knowledge base storage module, an adaptive matching module, and a forecasting application module. The modules interact with each other through parameter sample data, conversion relationship records, and a forecasting model interface.

[0144] The data preprocessing module is used to read rainfall (P), evaporation (E), flow rate (Q), and model parameters for the Changshan and Kaihua watersheds. Sample records were generated according to time scales of 1 hour, 30 minutes, and 15 minutes. For rainfall and evaporation data, they are accumulated or converted according to the target time scale; for flow data, they are averaged or converted to water volume consistency according to the target time scale; for model parameters, the correspondence between parameter symbols, parameter names, calculation modules, and parameter value ranges is established.

[0145] The sensitivity analysis module is used to calculate the time-scale sensitivity of each parameter and obtain the average sensitivity according to formula (2). This module identifies KI, KG, CS and KE as sensitive parameters and outputs them to the pattern identification module.

[0146] The pattern identification module is used to construct a dataset F showing the changing patterns of sensitive parameters over time. Based on the curve shape and matrix identification results of the changes in sensitive parameters over time, this module identifies the direction, magnitude, and degree of nonlinearity of KI, KG, CS, and KE, and outputs a label of "linear transformation" or "nonlinear transformation".

[0147] The quantization conversion module is used to call the corresponding formula based on the labels output by the pattern recognition module. For KI and KG, the formulas are called respectively. and For CS, call For KE, call .

[0148] The knowledge base storage module is used to write the above four types of transformation relationships and their applicable conditions into the parameter transformation relationship knowledge base. Each knowledge base record includes model type, calculation module, physical meaning of parameters, parameter value range, source time scale, target time scale range, transformation formula, formula coefficients, boundary constraints, and NSE quality evaluation information. This module also provides an index for searching by calculation module, parameter symbol, time scale, and formula type.

[0149] The adaptive matching module is used to retrieve transformation relationships from the knowledge base and call the quantization transformation module to generate target time scale parameters when the Kaihua watershed or other similar structural models lack target time scale calibration parameters.

[0150] The forecasting application module is used to input the converted 15-minute, 30-minute, or 1-hour timescale parameters into the distributed Xin'anjiang model to conduct rolling flood forecasts and output simulated flow processes and NSE evaluation results at the target timescale. Thus, the modules in the device can not only perform parameter timescale conversions within a single watershed, but also extend their applications across watersheds, timescales, and structurally similar models through a knowledge base and adaptive matching.

[0151] In this embodiment, the data preprocessing module, sensitivity analysis module, pattern identification module, quantization conversion module, knowledge base storage module, adaptive matching module, and forecasting application module can be implemented by independent software program modules, or they can be integrated into the same flood forecasting operational system. The modules can interact with each other through databases, configuration files, interface programs, or in-memory data objects. The division of modules in this embodiment is only for illustrating the functional structure of the invention and does not constitute a limitation on the physical structure of the device.

[0152] Example 3:

[0153] An electronic device includes a processor, a memory, and a computer program stored in the memory and executable by the processor. When the processor executes the computer program, it implements the steps in the above-mentioned method for converting the time scale of parameters in a high-frequency real-time flood forecasting model, including constructing a multi-time scale sample set, screening sensitive parameters, identifying change patterns, constructing quantitative conversion relationships, establishing a parameter conversion relationship knowledge base, adaptive matching, and verifying high-frequency real-time flood forecasting results.

[0154] The memory may include random access memory, read-only memory, hard disk, solid-state drive, portable storage medium, or cloud storage unit. The processor may be a central processing unit, graphics processing unit, digital signal processor, application-specific integrated circuit, field-programmable gate array, or other processing unit with data computing capabilities. The electronic device may be a single computer, server, cloud computing node, or a distributed computing system composed of multiple computing devices.

[0155] Example 4:

[0156] A computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the aforementioned method for time-scale conversion of parameters in a high-frequency real-time flood forecasting model. The computer-readable storage medium may be a disk, optical disk, read-only memory, random access memory, flash memory, portable hard drive, solid-state drive, or other media capable of storing program code.

[0157] It should be noted that the above-described device embodiments and method embodiments belong to the same inventive concept. For details not described in the device embodiments, please refer to the corresponding descriptions in the foregoing method embodiments. The above modules can be implemented in software, hardware, or a combination of both. Any application that can achieve the parameter time-scale sensitivity analysis, change pattern identification, quantization conversion, knowledge base matching, and high-frequency real-time flood forecasting functions described in this invention should fall within the protection scope of this invention.

Claims

1. A method for time-scale conversion of parameters in a high-frequency real-time flood forecasting model, characterized in that, Includes the following steps: Step S1: Collect rainfall, evapotranspiration, and flow data at multiple time scales to construct a multi-time scale model parameter sample set; Step S2: Based on the multi-timescale model parameter sample set, construct the parameter timescale sensitivity function, calculate and screen out the sensitive parameters that respond significantly to changes in timescale; Step S3: Based on the sensitive parameters obtained from the screening, perform standardization processing, construct a parameter-time scale correspondence dataset, and structurally identify the change patterns of the sensitive parameters; Step S4: Based on the identified change patterns, combined with the hydrological and physical significance, parameter value boundaries, and calculation module functions, construct the parameter time scale quantitative conversion relationship; Step S5: Structure and store the model type, calculation module, parameter attributes, sensitivity, change pattern, quantization conversion relationship and quality evaluation index to establish a parameter conversion relationship knowledge base; Step S6: Based on the similarity of computing module functions, the consistency of parameter physical meaning, the matching of parameter value ranges, and the proximity of time scale sensitivity, adaptively match the transformation relationship from the parameter transformation relationship knowledge base to realize the transfer application of parameters across time scales, watersheds, or models; Step S7: Input the target scale parameters obtained from the migration into the flood forecasting model, carry out high-frequency real-time flood forecasting, and verify the results using parameter error index and flow simulation accuracy index.

2. The method according to claim 1, characterized in that, Step S1, constructing the multi-timescale model parameter sample set, specifically includes the following sub-steps: Step S1.1: Define a unified timescale set ΔT for high-frequency real-time flood forecasting. The timescale set ΔT includes at least two timescales selected from 5 minutes, 10 minutes, 15 minutes, 30 minutes, 1 hour, 3 hours, 6 hours, and 24 hours; wherein Source timescale; Step S1.2: Collect raw rainfall data P, evapotranspiration data E, and flow data Q for the target watershed at various time scales; Step S1.3: Process the raw data of rainfall, evapotranspiration, and flow rate according to each single time scale. A unified scale conversion is performed, specifically as follows: Accumulate rainfall data over time periods according to the target scale: ; Evapotranspiration data are time-division converted according to the target scale: ; Perform water balance conversion on the target scale based on the flow data: ; Where n is a single time scale The number of original data time periods included; Step S1.4: For each time scale The parameters of the hydrological model were calibrated to obtain the parameter set. : ,in Let N be the calibration value of the i-th parameter at the j-th time scale, where N is the total number of parameters and M is the total number of time scales. Step S1.5: Calculate the values ​​of each individual time scale ΔT j The corresponding P(ΔT) j ), E(ΔT) j ), Q(ΔT) j ), and the set of model parameters calibrated under this single time scale. One-to-one matching is performed to form a multi-timescale model parameter sample set D; 。 3. The method according to claim 2, characterized in that, Step S2 specifically includes the following sub-steps: Step S2.1: Based on the multi-timescale parameter samples obtained in Step S1, calculate the parameter timescale sensitivity: ； in, For parameter time scale sensitivity; Let be the change in the i-th model parameter relative to the source time scale at the j-th single time scale, i.e. ; For the i-th model parameter at the source time scale The parameter values ​​below; Let j be the j-th single time scale in the set of time scales; Let be the change in the j-th single time scale relative to the source time scale. ; Step S2.2: Calculate the average parameter sensitivity under multi-scale combination. : ； Where M is the total number of time scales in the set ΔT, and N is the total number of model parameters; Step S2.3: Determine the parameters with sensitivity values ​​higher than the average sensitivity as sensitive parameters, and output the set of sensitive parameters.

4. The method according to claim 3, characterized in that, Step S3 specifically includes the following sub-steps: Step S3.1: Standardize the changes in the sensitive parameters and the changes over the time scale, respectively; Step S3.2: Construct a dataset showing the correspondence between parameter changes over time: ； Where F is the dataset showing the correspondence between parameter changes and changes over time; This represents the relative change of the i-th parameter relative to the source time scale at the j-th time scale. The relative change in time scale of the j-th time scale relative to the source time scale; Step S3.3: Use a dual-coordinate line graph or data matrix to identify the direction, magnitude, monotonicity, inflection point, and degree of nonlinearity of the sensitive parameters, and classify the change patterns into two categories: linear change and nonlinear change.

5. The method according to claim 4, characterized in that, Step S4 specifically includes the following sub-steps: Step S4.1: Construct a linear quantization transformation relationship for parameters of the linear change type: ； Step S4.2: Construct piecewise nonlinear transformation relationships for parameters of nonlinear variation type: when >1, and < hour: ; When 0 < <1, and < hour: ; in, For the target time scale The i-th model parameter; Source timescale The baseline value of the i-th model parameter; As the source timescale, For the target time scale; k i a i The transformation coefficient for the i-th parameter is obtained by fitting the parameter-timescale dataset constructed in step S3.

6. The method according to claim 5, characterized in that, Step S5 specifically includes the following sub-steps: Step S5.1: Construct a hierarchical index structure for the knowledge base. The hierarchical index structure includes, from top to bottom, the following layers: model type layer, calculation module layer, parameter physical meaning layer, parameter value range layer, sensitivity level layer, change pattern tag layer, conversion formula layer, conversion coefficient layer, and quality evaluation index layer. Step S5.2: Classify and store the multi-timescale model parameter sample set constructed in step S1 according to model type and calculation module; Step S5.3: Mark the parameter time scale sensitivity and sensitive parameter screening results obtained in step S2 according to the sensitivity level and associate them with the corresponding parameter entries; Step S5.4: Bind the parameter change pattern type identified in step S3 to the corresponding parameter entry in the form of a label; Step S5.5: Quantize the parameter time scale transformation relationship and transformation coefficients constructed in step S4 and associate them with the corresponding change pattern labels in the form of callable functions; Step S5.6: Calculate the average relative error of the parameters obtained in step S7. The Nash efficiency coefficient (NSE) quality evaluation index is added as a reliability indicator of the transformation relationship and stored separately. Step S5.7: Establish a multi-dimensional index based on model type, calculation module, parameter name, time scale, and change pattern type, supporting dynamic updates, retrieval, and version management, forming an scalable parameter transformation relationship knowledge base.

7. The method according to claim 6, characterized in that, Step S6 specifically includes the following sub-steps: Step S6.1: Determine the set of parameters to be transformed for the target scene and the target time scale. and source timescale Extract the model type, computation module, physical meaning of parameters, value range and sensitivity features of the target scene; Step S6.2: Based on the similarity of calculation module functions, the consistency of parameter physical meaning, the matching of parameter value ranges, and the similarity of time scale sensitivity, retrieve matching items from the parameter transformation relationship knowledge base; Step S6.3: Based on the change pattern type corresponding to the matching item, call the parameter time scale quantization transformation relationship and transformation coefficient k constructed in step S4. i a i ; Step S6.4: Set the parameter values ​​of the i-th model parameter at the source time scale. Source timescale and target time scale Substituting into the conversion formula, the target time scale parameter values ​​are calculated. ; Step S6.5: Calculate the... Perform hydrological physical boundary verification to ensure that the values ​​are within the preset reasonable hydrological range. Parameters that exceed the boundary are truncated or corrected according to physical constraints. Step S6.6: Output the validated target parameters to complete the adaptive parameter transfer across time scales, watersheds, or models.

8. The method according to claim 7, characterized in that, Step S7 specifically includes the following sub-steps: Step S7.1: Obtain the measured rainfall, evapotranspiration, and flow data sequences of the target watershed within the forecast period, and extract the measured flow Q at time t. o,t ; Step S7.2: Calculate the target time scale parameters obtained in step S6. Input a high-frequency flood forecasting model and run it to obtain the simulated flow rate Q at time t. c,t ; Step S7.3: Verify the accuracy of parameter transformation using the parameter average relative error index. ； Where N is the number of model parameters, Let i be the value of the i-th reference parameter at the source time scale. For the target time scale The i-th transformation parameter; Step S7.4: Verify the accuracy of the flow simulation using the Nash efficiency coefficient. ； in, t represents the arithmetic mean of the measured flow rates; L represents the total number of samples in the flow rate series; t represents the time series sequence number. Step S7.5: Set the parameter error threshold With flow simulation accuracy threshold When satisfied ,and If the parameter conversion and forecast results are deemed valid, then the process is fed back to step S4 to adjust the conversion coefficients or step S5 to update the knowledge base. Step S7.6: Output high-frequency real-time flood forecast results, parameter conversion accuracy and flow simulation accuracy reports to complete the forecast and verification process.

9. A time-scale conversion system for parameters of a high-frequency real-time flood forecasting model, used to implement the method described in any one of claims 1-8, characterized in that, include: The data preprocessing module is used to perform step S1, which involves collecting and organizing hydrological and meteorological data at multiple time scales and constructing a multi-time scale model parameter sample set. The sensitivity analysis module is used to perform step S2, calculate the time-scale sensitivity of parameters, and filter sensitive parameters; The pattern identification module is used to execute step S3, construct a parameter-time scale relationship dataset and identify the type of change pattern; The quantization conversion module is used to execute step S4 and construct the parameter time scale quantization conversion relationship based on the type of change pattern. The knowledge base storage module is used to execute step S5, which stores the model type, calculation module, parameter attributes, sensitivity, change law, quantization conversion relationship and quality evaluation index in a structured manner to establish a parameter conversion relationship knowledge base; and supports index retrieval and dynamic updates by calculation module, parameter, time scale and formula type. The adaptive matching module is used to execute step S6. Based on the similarity of the calculation module functions, the consistency of the physical meaning of the parameters, the matching of the parameter value range, and the proximity of the time scale sensitivity, it adaptively matches the transformation relationship from the parameter transformation relationship knowledge base to realize the transfer application of parameters across time scales, watersheds, or models. The forecasting application module is used to execute step S7, input the target scale parameters obtained from the migration into the flood forecasting model, carry out high-frequency real-time flood forecasting, and verify the results using parameter error index and flow simulation accuracy index.

10. 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 time scale conversion method for high-frequency real-time flood forecasting model parameters as described in any one of claims 1-8.

Images