A non-invasive correction method for flood forecasting models

Abstract

A non-invasive correction method for flood forecasting models includes the following steps: S10: obtaining the measured rainfall, measured flow, and flow calculation process for the time period to be corrected; S20: analyzing the deterministic coefficient of the calculated flow process; S30: analyzing and correcting the length of the reserved freedom period; S40: calculating the comprehensive correction characteristic process; S50: performing a dynamic input + feedback output comprehensive correction to obtain a corrected deterministic coefficient; S60: determining whether the corrected deterministic coefficient has improved. The present invention provides a non-invasive correction method for flood forecasting models. This method, based on a dynamic input + feedback output comprehensive correction mode, can rapidly and in real time correct the errors of the forecast model for various error scenarios without intrusively modifying the original forecast model, thereby ensuring the long-term stability and reliability of the forecast model.

Description

Technical Field

[0001] The present invention belongs to the technical field of flood forecast model correction methods, and in particular relates to a non-invasive correction method for flood forecast models. Background Art

[0002] Flood forecasting models, as the core of flood control and scheduling systems, have a direct impact on the scientific and effective nature of flood prevention decisions. However, forecast model accuracy is affected by factors such as data quality, model status, and model architecture. Especially in super-standard flood scenarios, complex hydrological nonlinear response mechanisms can still lead to forecast bias. If forecast models lack the ability to correct errors in real time, they will not only fail to fully extract effective information from real-time data, but may also cause error propagation and error amplification, leading to a sharp increase in the risk of model inaccuracy during flood seasons.

[0003] Existing real-time correction methods for flood forecast models can be mainly divided into three categories: model input correction, model state correction, and model output correction. Model input correction is generally based on rainfall sequence correction, but rainfall correction has non-negative constraints, and the degree of freedom period reserved during correction is subjective, which may lead to instability in the correction effect. Model state correction generally corrects the internal state variables of the forecast model. The forecast model is disassembled according to the selected corrected state variables, and the correction method is coupled to the corresponding model module. The degree of customization is high, and it is necessary to fully understand the technical details and operating mechanism of the original forecast model. Model output correction generally adopts an autoregressive correction method, which mainly relies on the autocorrelation of flow errors. It will lose the forecast period of the forecast model, and the correction effect is poor in the period near the flood peak. There may be a problem of error correction mutation.

[0004] Therefore, based on the above situation, this application develops a non-invasive correction method for flood forecasting models. Summary of the Invention

[0005] In response to the above-mentioned deficiencies in the existing technology, the present invention provides a non-invasive correction method for flood forecasting models. Through the non-invasive correction method based on the dynamic input + feedback output comprehensive correction mode, the error of the forecast model can be quickly corrected in real time for various error scenarios without intruding or modifying the original forecast model, thereby ensuring the stability and reliability of the long-term operation of the forecast model.

[0006] The present invention is solved by the following technical solutions.

[0007] A non-intrusive correction method for a flood forecast model includes the following steps:

[0008] S10: Obtaining the measured rainfall, measured flow and calculated flow process for the period to be corrected;

[0009] S20: Analyze and calculate the flow process certainty coefficient ;

[0010] S30: Analyze and modify the length of the reserved freedom period ;

[0011] S40: Calculation of comprehensive correction feature process ;

[0012] S50: Dynamic input + feedback output comprehensive correction to obtain the corrected certainty coefficient ;

[0013] S60: Determine the corrected certainty coefficient Whether to improve:

[0014] like , then update the correction status and set the dynamic correction coefficient , then proceed to steps S40, S50, and S60;

[0015] The update and correction status includes:

[0016] (1) Update the rainfall process to be corrected ; is the revised rainfall process;

[0017] (2) Update the calculation flow error process ;

[0018] (3) Update the calculation process certainty coefficient ;

[0019] like , then determine the dynamic correction coefficient Whether it exceeds the corresponding maximum limit ,like , then end the process; if , then set the dynamic correction coefficient , then proceed to steps S40, S50, and S60.

[0020] In a preferred embodiment, in S20, the flow process certainty coefficient is analyzed and calculated. ,as follows:

[0021]

[0022]

[0023]

[0024]

[0025] in Calculate the function for the flood forecast model, To calculate the flow error process; 、 Respectively The calculated flow and measured flow of each period, For the measured flow process, For the measured flow process The mean value, unit ; For the measured rainfall process The length of the period, in units .

[0026] In a preferred embodiment, in S30, the length of the reserved freedom period is analyzed and corrected. ,as follows:

[0027]

[0028]

[0029]

[0030] in is the confluence duration of the flood forecast model, The peak time period length calculation function, unit ; For the The rainfall in this period is And the rest The rainfall in each period is The rainfall process, unit ; Considering the influence of the confluence of the previous period, it is generally set is a larger value.

[0031] In a preferred embodiment, in S40, the comprehensive correction feature process is calculated ,as follows:

[0032]

[0033]

[0034]

[0035]

[0036]

[0037] in is the length of the comprehensive correction period, unit , ; Comprehensive modified characteristic matrix for bidirectional reserved degrees of freedom; for Calculated by the least squares method The rainfall error value needs to be corrected during the period, unit ; For the Time period The unit flow feedback process of the rainfall forecast model, , is the dynamic correction coefficient, is the corresponding maximum limit; is the rainfall process to be corrected, unit .

[0038] In a preferred embodiment, in S50, dynamic input + feedback output comprehensive correction is performed to obtain the corrected deterministic coefficient ,as follows:

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045] in It is a comprehensive correction result of dynamic input + feedback output. and They are respectively the dynamic input and feedback output flow correction process considering the actual rainfall non-negative constraint, and the unit .

[0046] In a preferred embodiment, in S60, the update correction state is as follows:

[0047] (1) Update the rainfall process to be corrected , is the revised rainfall process;

[0048] (2) Update the calculation flow error process ;

[0049] (3) Update the calculation process certainty coefficient .

[0050] Compared with the existing technology, the present invention has the following beneficial effects: it provides a non-invasive correction method for flood forecasting models. Through the non-invasive correction method based on the dynamic input + feedback output comprehensive correction mode, it can quickly realize real-time correction of the forecast model errors for various error scenarios without intruding and modifying the original forecast model, thereby ensuring the stability and reliability of the long-term operation of the forecast model. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Schematic diagram of the process of the non-invasive correction method of the present invention.

[0052] Figure 2 For the settings in the embodiment of the present invention The overall correction effect diagram.

[0053] Figure 3 For the settings in the embodiment of the present invention The overall correction effect diagram. DETAILED DESCRIPTION

[0054] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] In the following embodiments, the same or similar numbers throughout represent the same or similar components or elements with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limiting the present invention.

[0056] To achieve non-invasive correction of flood forecast models, a black-box system identification is performed on the forecast model. Dynamic analysis methods are used to study the correlation mechanism between external inputs and system feedback. While ensuring the integrity of the forecast model's internal structure, the model's external inputs and model outputs are directly corrected. The technical solution in this application considers factors such as rainfall non-negativity constraints, model forecast period, and correction stability. Through a comprehensive correction model of dynamic input + feedback output, a non-invasive correction method for flood forecast models is proposed.

[0057] The method in the present application obtains the measured rainfall, measured flow and calculated flow process of the forecast model in the period to be corrected and analyzes the deterministic coefficient of the calculated flow process; analyzes and corrects the length of the reserved freedom period by setting the characteristic rainfall process; combines the latest correction status, constructs a comprehensive correction characteristic matrix of bidirectional reserved freedom through dynamic correction coefficients, and recalculates the comprehensive correction characteristic process based on the least squares method; cyclically corrects and updates the correction status through a correction mode of dynamic input + feedback synthesis, and finally realizes the overall non-invasive correction of the flood forecast model.

[0058] See also Figures 1 to 3 , an embodiment of the present application is as follows.

[0059] The prediction results of a reservoir flood forecast model developed by a third-party unit for floods in 2024 are selected, such as Figure 2 As shown by the dotted line, it can be seen that there are large errors between the calculation of the flood forecast model and the measured inflow in terms of flood peak, flood volume and overall process, and error correction is needed to ensure the accuracy of the flood forecast model.

[0060] The specific steps of the non-invasive correction method for flood forecasting models provided by the present invention are as follows:

[0061] Step 1: Obtain the measured rainfall, measured flow and calculated flow process for a total of 120 periods during the reservoir flood period;

[0062] Step 2: Analyze and calculate the certainty coefficient of the flow process :

[0063]

[0064]

[0065]

[0066]

[0067] in Calculate the function for the flood forecast model, To calculate the flow error process; 、 Respectively The calculated flow and measured flow of each period, For the measured flow process, For the measured flow process The mean value, unit ; For the measured rainfall process The length of the period, in units In this embodiment, , the analysis can obtain the certainty coefficient of the original flood forecast model for calculating the flow process ;

[0068] Step 3: Analyze and modify the length of the reserved freedom period :

[0069]

[0070]

[0071]

[0072] in is the confluence duration of the flood forecast model, The peak time period length calculation function, unit ; For the The rainfall in this period is And the rest The rainfall in each period is The rainfall process, unit ; Considering the influence of the confluence of the previous period, it is generally set is a larger value. In this embodiment, , the analysis can be used to obtain the length of the modified reserved freedom period ;

[0073] Step 4: Calculate the comprehensive correction feature process :

[0074]

[0075]

[0076]

[0077]

[0078]

[0079] in is the length of the comprehensive correction period, unit , ; Comprehensive modified characteristic matrix for bidirectional reserved degrees of freedom; for Calculated by the least squares method The rainfall error value needs to be corrected during the period, unit ; For the Time period The unit flow feedback process of the rainfall forecast model, , is the dynamic correction coefficient, is the corresponding maximum limit; is the rainfall process to be corrected, unit In this embodiment , ;

[0080] Step 5: Comprehensively correct the dynamic input and feedback output to obtain the corrected deterministic coefficient :

[0081]

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] in It is a comprehensive correction result of dynamic input + feedback output. and They are respectively the dynamic input and feedback output flow correction process considering the actual rainfall non-negative constraint, and the unit .

[0088] Step 6: Determine the corrected certainty coefficient Whether to improve:

[0089] like , then update the correction status and set the dynamic correction coefficient , then proceed to steps four, five, and six;

[0090] Among them, the update correction status is as follows:

[0091] (1) Update the rainfall process to be corrected , is the revised rainfall process;

[0092] (2) Update the calculation flow error process ;

[0093] (3) Update the calculation process certainty coefficient ;

[0094] like , then determine the dynamic correction coefficient Whether it exceeds the corresponding maximum limit ,like , then end the process; if , then set the dynamic correction coefficient , then proceed to steps four, five, and six;

[0095] By using a non-invasive correction method for flood forecasting models provided by the present invention, and The overall correction effects are as follows: Figure 2 、 Figure 3As shown by the dashed lines in the figure. The corrected flood peak errors were increased from -39.01% to -1.36% and -1.69% respectively, the flood volume errors were increased from -44.27% to -16.62% and -6.81% respectively, and the coefficient of certainty was increased from 0.387 to 0.726 and 0.906 respectively. The results show that the correction method provided by the present invention can improve the calculation accuracy of the flood forecast model without intruding or modifying the original flood forecast model. The more information in the correction period, the better the correction effect. After correction The calculated flow rate during the period is basically consistent with the measured flow rate process, and The subsequent flow calculation process of the time period also has a good correction effect.

[0096] From the above description, it can be seen that the non-invasive correction method for flood forecast models in this application is based on a comprehensive correction mode of dynamic input + feedback output, which can not only eliminate the limitation of the model input correction method subject to the non-negative constraint of rainfall, but also avoid the influence of the model output correction method on the forecast period of the forecast model, and the feedback output correction does not depend on the autocorrelation of the flow error, and there will be no problem of error correction mutation. By reserving the correction freedom period in both directions, the influence of the special treatment of the calculated flow in the confluence period by some flood forecast models can be effectively avoided without affecting the overall correction effect. The method does not introduce subjective coefficients and has strong versatility. It can be used as an independent correction method for flood forecast models. Without intrusive modification of the original forecast model, it can quickly realize the real-time correction of the forecast model error for various error scenarios, effectively avoiding the dependence of the traditional intrusive correction method on the internal structure of the forecast model.

[0097] The protection scope of the present invention includes but is not limited to the above embodiments. The protection scope of the present invention is based on the claims. Any replacement, deformation, and improvement of the technology that can be easily thought of by those skilled in the art fall within the protection scope of the present invention.

Claims

1. A non-invasive correction method for flood forecasting models, characterized in that: The following steps are involved: S10: Obtain the measured rainfall, measured flow and calculated flow process for the period to be corrected and perform dynamic input; S20: Analyze and calculate the flow process certainty coefficient ; S30: Analyze and modify the length of the reserved freedom period ; S40: Calculation of comprehensive correction feature process ; S50: Dynamic input + feedback output comprehensive correction to obtain the corrected certainty coefficient , ； ； ； ； ； ； in It is a comprehensive correction result of dynamic input + feedback output. and They are respectively the dynamic input and feedback output flow correction process considering the actual rainfall non-negative constraint, and the unit ; S60: Determine the corrected certainty coefficient Whether to improve: like , then update the correction status and set the dynamic correction coefficient , then proceed to steps S40, S50, and S60; like , then determine the dynamic correction coefficient Whether it exceeds the corresponding maximum limit ,like , then end the process; if , then set the dynamic correction coefficient , then proceed to steps S40, S50, and S60; In S20, analyze and calculate the flow process certainty coefficient Middle: Calculation function through flood forecasting model , Calculate flow error process , calculated flow and measured flow in multiple time periods, and the length of the measured rainfall process period to analyze and calculate the flow process certainty coefficient ; In S30, the length of the reserved freedom period is analyzed and corrected. Middle: Confluence duration via flood forecasting models , Peak time period length calculation function , rainfall process, and the influence of the previous period on the flow of the previous period to analyze and modify the length of the reserved freedom period .

2. A non-invasive correction method for flood forecasting models according to claim 1, characterized in that: In S40, the comprehensive correction feature process is calculated Medium: By comprehensively correcting the time period length , comprehensive modified characteristic matrix of two-way reserved degrees of freedom , the rainfall error value that needs to be corrected during the period, the unit flow feedback process of the forecast model, and the rainfall process to be corrected, to calculate the comprehensive correction characteristic process .

3. The non-invasive correction method for flood forecasting model according to claim 2, characterized in that: In S40, the rainfall error value that needs to be corrected for the time period is calculated using the least squares method.

4. The non-invasive correction method for flood forecasting model according to claim 3, characterized in that: In S50, the corrected deterministic coefficient is obtained by comprehensive correction of dynamic input + feedback output, considering the dynamic input and feedback output flow correction process of the actual rainfall non-negative constraint. .

5. The non-invasive correction method for flood forecasting model according to claim 4, characterized in that: In S60, updating the correction state includes: updating the rainfall process to be corrected , The corrected rainfall process.

6. The non-invasive correction method for flood forecasting model according to claim 5, characterized in that: In S60, updating the correction state includes: updating the flow error calculation process ; It is a comprehensive correction result of dynamic input + feedback output; This is the measured flow process.

7. The non-invasive correction method for flood forecasting model according to claim 6, characterized in that: In S60, updating the correction state includes: updating the calculation flow process certainty coefficient .

Images