A hydrological prediction model uncertainty correction method, device and electronic equipment

Abstract

This application provides a method, apparatus, and electronic device for correcting uncertainty in a hydrological prediction model. The method includes: when predicting the outlet flow of a target watershed using a target hydrological prediction model, subjecting the target hydrological prediction model to a preset number of perturbations to obtain a prediction result set; calculating the residual values ​​of each prediction result based on a preset water conservation equation; determining physical consistency weighting coefficients based on the residual values; performing weighted statistics on the prediction result set based on the physical consistency weighting coefficients corresponding to each prediction result in the prediction result set to construct a weighted probability distribution function; and determining the confidence interval for the outlet flow of the target watershed based on the weighted probability distribution function and a preset quantile. This application integrates the physical constraint of water conservation into the uncertainty quantification process, weakening the impact of physically unreasonable prediction results on uncertainty assessment and improving the reliability and engineering applicability of the hydrological prediction model uncertainty assessment.

Description

Technical Field

[0001] This application relates to the field of watershed hydrological model evaluation technology, and in particular to a method, device and electronic equipment for correcting uncertainty in hydrological prediction models. Background Technology

[0002] With the application of deep learning in the field of hydrology, hydrological prediction models based on neural networks and artificial intelligence can quickly predict flow or groundwater conditions. However, these models are mostly black-box structures, and their uncertainty assessment often relies on traditional sampling or Bayesian approximation methods, resulting in a lack of physical interpretation of confidence intervals, which in turn reduces the reliability and engineering applicability of hydrological models. Summary of the Invention

[0003] This application provides a method, apparatus, and electronic device for correcting uncertainty in a hydrological prediction model, in order to address the problem of how to improve the certainty of the confidence interval of the hydrological prediction model output.

[0004] In a first aspect, embodiments of this application provide a method for correcting the uncertainty of a hydrological prediction model, the method comprising: When predicting the outflow of a target watershed using a target hydrological prediction model, the target hydrological prediction model is subjected to a preset number of perturbations. Obtain the prediction results output by the target hydrological prediction model after each disturbance processing to obtain a prediction result set; Based on the preset water conservation equation, the residual values ​​of each prediction result in the prediction result set are calculated respectively; For each prediction result, a physical consistency weight coefficient is determined based on the residual value corresponding to the prediction result. Based on the physical consistency weight coefficients corresponding to each prediction result in the prediction result set, a weighted probability distribution function is constructed by performing weighted statistics on the prediction result set. The confidence interval for the outflow of the target watershed is determined based on the weighted probability distribution function and the preset quantile.

[0005] Secondly, embodiments of this application also provide a hydrological prediction model uncertainty correction device, the device comprising: The processing module is used to perform a preset number of perturbations on the target hydrological prediction model when predicting the outflow of the target watershed using the target hydrological prediction model. The first acquisition module is used to acquire the prediction results output by the target hydrological prediction model after each disturbance processing, and obtain a prediction result set. The first calculation module is used to calculate the residual value of each prediction result in the prediction result set based on the preset water conservation equation. The first determining module is used to determine the physical consistency weight coefficient corresponding to each prediction result based on the residual value corresponding to the prediction result. The construction module is used to perform weighted statistics on the prediction result set based on the physical consistency weight coefficients corresponding to each prediction result in the prediction result set, and construct a weighted probability distribution function. The second determining module is used to determine the confidence interval of the outlet flow of the target watershed based on the weighted probability distribution function and the preset quantile.

[0006] Thirdly, embodiments of this application also provide an electronic device, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the above-described hydrological prediction model uncertainty correction method.

[0007] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described hydrological prediction model uncertainty correction method.

[0008] The embodiments of this application include at least the following technical effects: The technical solution of this application embodiment generates a prediction result set by perturbing the target hydrological prediction model, calculates the residual value of each prediction result based on the water conservation equation, determines the physical consistency weight coefficient based on the residual value, constructs a weighted probability distribution function by weighted statistics, and determines the outlet flow confidence interval according to the preset quantile. This application can integrate the physical constraints of water conservation into the entire uncertainty quantification process without changing the model structure, weakens the impact of physically unreasonable prediction results on uncertainty assessment, and makes the final confidence interval have both statistical rationality and physical consistency, improves the credibility and engineering applicability of the hydrological prediction model uncertainty assessment, and makes the confidence interval of the outlet flow prediction conform to the hydrological and physical laws of the watershed, providing more reliable and physically interpretable uncertainty quantification results for hydrological forecasting. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0010] Figure 1 This is a flowchart illustrating the hydrological prediction model uncertainty correction method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the hydrological prediction model uncertainty correction device provided in the embodiments of this application; Figure 3 A block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0013] In the various embodiments of this application, it should be understood that the sequence number of each process described below does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0014] like Figure 1 As shown in the embodiment of this application, a method for correcting uncertainty in a hydrological prediction model is provided. The method includes: Step 101: When predicting the outflow of the target watershed using the target hydrological prediction model, the target hydrological prediction model is subjected to a preset number of perturbations.

[0015] The target hydrological prediction model is a pre-trained hydrological AI model used to predict the outflow of a watershed. Specifically, it is a black-box hydrological forecasting model based on deep learning or neural networks combined with artificial intelligence. It can output predicted outflow of the watershed based on input data such as rainfall, temperature, soil moisture, and evapotranspiration. The target watershed is the watershed whose outflow is to be predicted.

[0016] The hydrological prediction model uncertainty correction method provided in this application can introduce the water conservation constraint from the hydrological physical laws into the uncertainty quantification without changing the structure and training process of the target hydrological prediction model. Compared with traditional methods that only focus on the statistical distribution of prediction results and do not judge whether the sample conforms to the balance between changes in watershed rainfall, evapotranspiration, and water storage, resulting in a narrow confidence interval, this application can weaken the impact of physically unreasonable prediction results on uncertainty assessment by integrating the water conservation physical constraint into the entire uncertainty quantification process. This makes the final confidence interval have both statistical rationality and physical consistency, improving the credibility and engineering applicability of the hydrological prediction model uncertainty assessment.

[0017] Specifically, in this embodiment of the application, when predicting the outlet flow of a target watershed using a target hydrological prediction model, the target hydrological prediction model is subjected to a preset number of perturbations. These perturbations include both input perturbations and parameter perturbations, which can be used individually or in combination. To ensure the stability of the samples obtained after perturbation and to represent the overall fluctuation characteristics of the model, the preset number is generally required to be no less than 100 perturbations.

[0018] Specifically, the input perturbation involves superimposing random noise within the actual observation error range onto the model's actual input data, such as rainfall, temperature, soil moisture, and evapotranspiration. The parameter perturbation involves small-scale random sampling of the internal weights, regularization coefficients, and hidden layer parameters of the already trained hydrological prediction model.

[0019] Step 102: Obtain the prediction results output by the target hydrological prediction model after each disturbance processing to obtain a prediction result set.

[0020] After each perturbation, the predicted watershed outlet discharge value output by the target hydrological prediction model is obtained, i.e., the prediction result. The prediction results corresponding to all perturbations are then integrated to obtain the prediction result set. The distribution of the output of this prediction result set under slight fluctuations in the input and parameters of the target hydrological prediction model reflects the statistical uncertainty of the model itself; at this point, all samples are not distinguished by importance.

[0021] For example, after 150 perturbation processes, a prediction result set including 150 export flow prediction values ​​is obtained.

[0022] Step 103: Based on the preset water conservation equation, calculate the residual value of each prediction result in the prediction result set.

[0023] The pre-set water conservation equation is derived from the fundamental hydrological physical law of watershed water balance. Specifically, the pre-set water conservation equation can be expressed as the sum of watershed rainfall, outflow, evapotranspiration, watershed water storage change, and residual values. By substituting each predicted result into the pre-set water conservation equation, the deviation between each predicted result and the physical balance is calculated, yielding the residual value. The residual value represents the degree to which the current predicted result deviates from the water balance and serves as the basis for judging physical consistency.

[0024] This application's embodiments convert hydrophysical constraints into calculable numerical indicators, enabling a quantitative scoring of the physical rationality of each prediction result.

[0025] Step 104: For each prediction result, determine the physical consistency weight coefficient corresponding to the prediction result based on the residual value corresponding to the prediction result.

[0026] After obtaining the residual value for each prediction result, a physical consistency weighting coefficient is calculated based on the residual value. The residual value is negatively correlated with the physical consistency weighting coefficient; the larger the residual value, the lower the physical consistency weighting coefficient. This reduces the impact of physically unreasonable prediction results on statistics and increases the impact of physically reasonable prediction results on statistics.

[0027] The physical consistency weight coefficient can be calculated exponentially to ensure that the weight decreases rapidly when the residual increases slightly, thus significantly suppressing predictions that clearly violate water conservation. To avoid excessive penalty leading to the loss of interval boundaries, a minimum weight lower limit is set, typically 0.05, to retain a small number of extreme predictions to ensure complete coverage of the confidence interval. For example, if the residual value of one prediction is very small, close to water balance, its weight will be close to 1; if the residual value of another prediction is very large, clearly not conforming to physical laws, its weight will rapidly decrease to near the lower limit of 0.05.

[0028] The embodiments of this application incorporate physical rationality into statistical weights through residual values, thereby reducing the impact of unreasonable prediction results on the overall distribution without removing samples, thus preserving statistical information while strengthening physical constraints.

[0029] Step 105: Based on the physical consistency weight coefficients corresponding to each prediction result in the prediction result set, perform weighted statistics on the prediction result set to construct a weighted probability distribution function.

[0030] After obtaining the physical consistency weights of all prediction results, the prediction result set is re-statistically analyzed using methods such as weighted mean and weighted variance to construct a weighted probability distribution function. Weighted statistics allow physically reasonable prediction results to have a higher contribution, while the contribution of physically unreasonable prediction results is reduced. The final distribution is a corrected distribution that incorporates hydrophysical laws.

[0031] Step 106: Determine the confidence interval of the outlet flow of the target watershed based on the weighted probability distribution function and the preset quantile.

[0032] Based on the weighted probability distribution, commonly used preset quantiles in engineering are selected, typically 2.5% and 97.5%, corresponding to the lower and upper limits of the confidence interval, respectively, thus obtaining the final outlet flow confidence interval after physical consistency correction. This correction interval automatically avoids extreme values ​​that do not conform to water conservation, and the interval width changes reasonably with rainfall intensity, being wider during heavy rain and narrower during dry periods, consistent with the actual hydrological response patterns.

[0033] In this embodiment, a set of prediction results is generated by perturbing the target hydrological prediction model. The residual values ​​of each prediction result are calculated based on the water conservation equation, and the physical consistency weight coefficient is determined according to the residual values. A weighted probability distribution function is constructed by weighted statistics, and the confidence interval of the outlet flow is determined according to the preset quantile. This application can integrate the physical constraints of water conservation into the entire process of uncertainty quantification without changing the model structure, thereby weakening the impact of physically unreasonable prediction results on uncertainty assessment. The final confidence interval has both statistical rationality and physical consistency, improving the credibility and engineering applicability of the uncertainty assessment of the hydrological prediction model. The confidence interval of the outlet flow prediction conforms to the hydrological and physical laws of the watershed, providing more reliable and physically interpretable uncertainty quantification results for hydrological forecasting.

[0034] In an optional embodiment of this application, the target hydrological prediction model is subjected to a preset number of perturbation processes, including: The target hydrological prediction model is subjected to a first preset number of variable perturbation processes, wherein the variable perturbation process is to superimpose random noise less than a preset noise threshold onto at least one input variable in the input variable set corresponding to the target hydrological prediction model. The target hydrological prediction model is subjected to a second preset number of model perturbation processes; wherein, the model perturbation process is to adjust the model parameters corresponding to the target hydrological prediction model to a value less than a preset amplitude threshold. Wherein, the sum of the first preset quantity and the second preset quantity is equal to the preset quantity.

[0035] This application's embodiments perturb the target hydrological prediction model using two methods: input variable perturbation and model parameter perturbation. These two perturbation methods complement each other, avoiding the sample distribution bias problem caused by a single perturbation method.

[0036] When performing the first preset number of variable perturbations, specifically, driving variables such as rainfall, temperature, soil moisture, and evapotranspiration are selected from the set of input variables of the target hydrological prediction model. Random noise controlled within a preset noise threshold is superimposed on these input variables. This preset noise threshold is determined based on the error range of the actual observation equipment and the observation accuracy of the watershed data, ensuring that the noise amplitude always remains within the range that conforms to the actual observation conditions and that no extreme input values ​​exceeding physical meaning occur. In actual execution, noise can be applied to a single input variable or multiple input variables simultaneously to simulate the real scenario of multiple sources of observation errors acting together. For example, when predicting the outlet flow of a watershed, rainfall and soil moisture are selected as input variables, and random noise conforming to conventional observation errors is superimposed on their observed values. The first preset number of independent perturbations are repeated. Each perturbation keeps the model structure and parameters unchanged, only changing the numerical state of the input variables, thereby obtaining a prediction result driven by input uncertainty.

[0037] When performing the second preset number of model perturbations, the parameters of the already trained model within the target hydrological prediction model are adjusted. The adjustment range is controlled within a preset threshold to ensure that the parameters only fluctuate slightly without changing the overall learning ability and basic mapping relationship of the model. The adjustable parameters include key parameters that constitute the internal calculation logic of the model, such as the hidden layer weights, regularization coefficients, and bias terms. Each perturbation only makes small random adjustments to the parameters without changing the configuration of the network layers, activation functions, model structure, etc. For example, the weights of the fully connected layers of a deep learning hydrological AI model are slightly and randomly adjusted. Without destroying the hydrological patterns already learned by the model, the prediction drift caused by parameter uncertainty is stimulated. The second preset number of independent perturbations are repeated to obtain the prediction results driven by the uncertainty of the model's internal parameters.

[0038] The above-described implementation scheme of this application, by performing variable perturbation processing and model perturbation processing on the target hydrological prediction model respectively, fully stimulates and covers the prediction uncertainty caused by the observation error of input variables and the fluctuation of model parameters. Under the premise of controlling the noise amplitude and parameter adjustment amplitude, it ensures that all perturbations are within a reasonable physical range and effective model state, avoids interference from invalid or non-physical perturbation samples, and provides a statistically sufficient prediction result set that fits the real uncertainty source for subsequent physical consistency correction.

[0039] In an optional embodiment of this application, based on a preset water conservation equation, the residual values ​​of each prediction result in the prediction result set are calculated, including: Obtain the set of input variables corresponding to the prediction results; the set of input variables includes the rainfall, evapotranspiration and water storage changes of the target watershed; Based on the fact that the rainfall is equal to the sum of the evapotranspiration, the prediction result, the change in water storage, and the residual value, the residual value corresponding to the prediction result is determined.

[0040] In this embodiment of the application, when calculating the residual value of the prediction result, the watershed water conservation is used as the evaluation criterion. Each predicted outflow result in the prediction result set is substituted into the water conservation equation, and the residual value is obtained by calculating the deviation between the predicted value and the physical equilibrium state.

[0041] Specifically, an input variable set matching the current prediction results is obtained. This set includes rainfall, evapotranspiration, and changes in water storage in the target watershed at the corresponding prediction time. These three variables are components of the watershed's water balance and, together with the predicted outflow, form a complete water exchange closed loop. It is important to note that the acquisition process must ensure a complete correspondence between the time and spatial scope and the prediction results to avoid distortion in residual calculations due to spatiotemporal mismatches. Furthermore, to reduce calculation biases caused by observational noise or local outliers, time smoothing or moving averages can be applied to evapotranspiration and changes in water storage, making the variables involved in the calculation more reflective of the overall water volume change trend of the watershed, rather than simply data fluctuations.

[0042] After obtaining the set of input variables, the water conservation equation is applied as follows:

[0043] in, For rainfall, For the predicted result, i.e., export flow, Evaporation amount This represents the change in water storage. These are the residual values. And all these variables correspond to the same time point.

[0044] Substitute the corresponding values ​​one by one and perform balance calculations to obtain the residual value corresponding to the current prediction result. The residual value is the degree to which the current predicted outlet flow deviates from the actual water balance state of the basin. The larger the absolute value of the residual, the more obvious the conflict between the prediction result and the hydrophysical law, and the lower the physical consistency. Conversely, the smaller the residual value, the more the prediction result conforms to the water conservation constraint and has a higher degree of physical reliability.

[0045] The above-described implementation scheme of this application, by matching the rainfall, evapotranspiration and water storage changes at the corresponding time of the prediction results, and solving the residual values ​​of each prediction result according to the preset water conservation equation, transforms the hydrophysical constraints into quantifiable physical deviation indicators, providing an objective and reliable quantitative basis for subsequent physical consistency weight calculation.

[0046] In an optional embodiment of this application, determining the physical consistency weight coefficient corresponding to the prediction result based on the residual value corresponding to the prediction result includes: Obtain the target penalty coefficient that matches the target watershed; The physical consistency weight coefficient corresponding to the prediction result is calculated by taking the opposite of the product of the absolute value of the residual and the target penalty coefficient as the exponent of the natural exponential function.

[0047] The process of obtaining the target penalty coefficient that matches the target watershed includes: Based on the correspondence between watershed type and basic penalty coefficient range, the target basic penalty coefficient range corresponding to the target watershed type is determined according to the target watershed type. Obtain the historical penalty coefficient, actual observation coverage, and confidence interval width corresponding to the target watershed; Based on the actual observation coverage and confidence interval width, the historical penalty coefficient is adjusted within the target basic penalty coefficient interval to obtain the target penalty coefficient.

[0048] In this embodiment of the application, when determining the physical consistency weight coefficient based on the residual value, the target penalty coefficient that matches the target watershed is first obtained.

[0049] Specifically, based on the pre-defined correspondence between watershed types and basic penalty coefficient intervals, the corresponding basic penalty coefficient interval is determined by combining the watershed type of the target watershed. Different watersheds exhibit significant differences in scale, climate, and underlying surface conditions, resulting in varying requirements for physical constraint strength. Therefore, it is necessary to first define a reasonable value range according to the type to avoid the penalty coefficient being too high or too low, which could lead to the failure of physical constraints or the loss of statistical information. After determining the basic penalty coefficient interval, operational indicators such as historical penalty coefficients, actual observation coverage, and confidence interval width corresponding to the target watershed are further retrieved. Using the historical penalty coefficient as an initial benchmark, dynamic adjustments are made based on whether the actual observation coverage meets the standard and whether the confidence interval width conforms to the hydrological response law. When the observation coverage is too low, the penalty intensity is appropriately weakened; when the confidence interval is too wide or physical anomalies occur, the penalty intensity is appropriately strengthened. The process is iteratively optimized within the basic interval to ultimately obtain the target penalty coefficient that adapts to the current watershed and current forecast conditions. In this embodiment, the penalty coefficient is not a fixed value but a key parameter that adaptively changes with watershed characteristics and actual operational effects.

[0050] Furthermore, after obtaining the target penalty coefficient, the weight coefficient is calculated according to the preset exponential function rule. The absolute value of the residual is multiplied by the target penalty coefficient, and the negative number is used as the exponent of the natural exponential function for calculation. This yields the physical consistency weight coefficient corresponding to the current prediction result. The calculation formula is as follows:

[0051] in, This is the physical consistency weighting coefficient. The target penalty coefficient, This represents the residual value.

[0052] It should be noted that the exponential decay form ensures that the weights decrease rapidly when the residuals increase slightly, thereby suppressing physically unreasonable samples that clearly violate the water conservation law. Furthermore, to avoid excessive penalty, a lower limit can be set for the weights, such as 0.05, to retain a small number of extreme value samples for interval boundary calculations.

[0053] The above-described implementation scheme of this application adaptively determines the penalty coefficient by matching the characteristics of the target watershed, and constructs a natural exponential function with the absolute value of the residual and the penalty coefficient to calculate the physical consistency weight coefficient. It can achieve differentiated weighting based on the physical rationality of the prediction results, suppress the influence of physically unreasonable samples on the statistical distribution, provide a weighting basis that conforms to physical laws for subsequent weighted statistics and confidence interval construction, and improve the physical consistency and watershed adaptability of the uncertainty quantification process.

[0054] In an optional embodiment of this application, a weighted probability distribution function is constructed by performing weighted statistics on the prediction result set based on the physical consistency weight coefficients corresponding to each prediction result in the prediction result set, including: Calculate the weighted mean and variance based on the physical consistency weight coefficients corresponding to each prediction result in the prediction result set. The weighted probability distribution function is constructed based on the weighted mean and the variance.

[0055] In this embodiment of the application, when constructing the weighted probability distribution function, the weighted mean and variance are calculated based on the physical consistency weight coefficients corresponding to each prediction result. The calculation formula is as follows:

[0056]

[0057] in, For weighted average, The number of predicted results in the prediction result set. For the i-th prediction result, Let be the physical consistency weighting coefficient for the i-th prediction result. Let Variance be the variance.

[0058] The calculation of weighted mean and variance means that the prediction results are no longer given equal statistical contributions. Instead, they are statistically analyzed according to the weight of physical consistency. Prediction results with higher physical consistency have a higher weight in the calculation of mean and variance, while prediction results with lower physical consistency have a significantly reduced statistical impact.

[0059] After obtaining the weighted mean and variance that reflect physical constraints, a corresponding weighted probability distribution function is constructed based on these two statistics. This distribution function is no longer the statistical distribution output by the original model, but a corrected distribution that incorporates the physical rules of water conservation. Its overall shape, central tendency and dispersion are both subject to the dual constraints of statistical laws and hydrological physical laws, which can effectively avoid problems such as distribution shift, peak anomalies and unreasonable intervals that occur in traditional statistical distributions due to the inclusion of samples with physical imbalances.

[0060] The above-described implementation scheme of this application calculates the weighted mean and variance of the prediction result set based on the physical consistency weight coefficient, and constructs a weighted probability distribution function accordingly. This can integrate physical rationality into the sample statistical process, effectively weaken the statistical influence of physically unreasonable samples, strengthen the distribution contribution of prediction results that conform to the water conservation constraint, and form a weighted probability distribution that combines statistical robustness and physical consistency. This provides a reliable distribution basis that conforms to hydrological and physical laws for the subsequent determination of the confidence interval of outlet flow, and improves the physical rationality and statistical credibility of uncertainty quantification.

[0061] In an optional embodiment of this application, determining the confidence interval of the outlet flow of the target watershed based on the weighted probability distribution function and the preset quantile includes: Based on the weighted probability distribution function and the preset quantiles, determine the upper and lower bounds of the boundary; The confidence interval for the outflow is determined based on the upper and lower bounds of the boundary.

[0062] In determining the confidence interval of the outlet flow of the target watershed, this application embodiment transforms the statistical distribution after physical consistency correction into a directly usable confidence interval of the outlet flow based on the weighted probability distribution function that has incorporated the physical constraint of water conservation. This makes the final interval no longer rely solely on traditional statistical sampling, but simultaneously satisfy both statistical confidence requirements and hydrophysical laws.

[0063] Specifically, firstly, based on the weighted probability distribution function that has already undergone physical correction, the upper and lower boundary limits are located according to the preset quantiles. Here, the preset quantiles correspond to the conventional confidence levels in hydrological forecasting and uncertainty assessment. Their values ​​are determined based on engineering decisions, risk prevention and control, and forecast accuracy requirements. The flow values ​​corresponding to the quantiles are extracted from the weighted distribution to ensure that the obtained boundaries are the results after physical rationality weighting.

[0064] Furthermore, after determining the upper and lower boundary limits, these are combined to form a complete confidence interval for the outlet flow. This interval is the final output of the entire physical consistency correction method, presenting the reliable range of flow prediction after physical constraints. Compared with the traditional uncorrected interval, this corrected interval exhibits reasonable response characteristics depending on the watershed hydrological conditions. When rainfall replenishment is sufficient, the interval shows reasonable width variations, and when water balance is stable, the interval is more compact, perfectly matching the actual hydrological process. For example, in a watershed flood peak forecast scenario, the traditional statistical interval may be too narrow and exceed physical possibilities, while the interval determined based on the weighted distribution will automatically adjust to a reasonable range that conforms to water conservation, while maintaining effective coverage of the observed values.

[0065] The above-described implementation scheme of this application determines the upper and lower limits of the confidence interval based on the weighted probability distribution function and the preset quantile. It uses the statistical distribution characteristics corrected for physical consistency to output the outlet flow confidence interval that simultaneously meets the statistical confidence requirements and the physical law of water conservation. This effectively avoids the defects of traditional intervals being physically unreasonable and lacking physical interpretation, and improves the reliability and engineering applicability of the quantification results of hydrological prediction uncertainty.

[0066] In an optional embodiment of this application, after calculating the residual values ​​of each prediction result in the prediction result set based on a preset water conservation equation, the method further includes: Obtain the average rainfall of the target watershed within a preset time period; Calculate the average of the residuals of each prediction result in the prediction result set to obtain the residual mean; The ratio of the difference between the average rainfall and the mean residual to the average rainfall is determined as the physical consistency index.

[0067] In this application embodiment, after obtaining the residuals of each prediction result based on the water conservation calculation, a standardized quantitative index is constructed based on the mean of the residuals and the average rainfall of the watershed. This index transforms the overall physical rationality of the prediction result set into a physical consistency index that can be intuitively interpreted and compared horizontally, providing an intuitive credibility criterion for the uncertainty correction results.

[0068] Specifically, after calculating the residual values ​​of all prediction results in the prediction result set based on the water conservation equation, in order to further quantitatively evaluate the physical compliance of the prediction results from an overall perspective, the calculation process of the physical consistency index is executed. First, the average rainfall of the target watershed within a preset time period is obtained. This average rainfall serves as the benchmark value for watershed water input, representing the overall water supply level of the watershed within that time period. Selecting the average value within the preset time period can effectively eliminate the impact of instantaneous rainfall fluctuations, enabling the index to reflect the overall physical equilibrium state rather than local instantaneous deviations.

[0069] Then, the residual values ​​corresponding to all prediction results in the prediction result set are averaged to obtain the mean residual value that can represent the degree of physical deviation of the whole set of samples. This reflects the average level of deviation of all prediction samples from the law of water conservation. The larger the mean residual value, the more significant the conflict between the overall prediction result and the water balance, and the worse the physical consistency. Conversely, the smaller the mean residual value, the closer the overall prediction result is to the hydrological physical law.

[0070] Finally, the difference between the average rainfall and the mean residual is divided by the average rainfall, and this ratio is determined as the physical consistency index corresponding to the current forecast result set. The calculation formula is as follows:

[0071] in, This is a physical consistency index. This represents the average rainfall. This represents the mean of the residuals.

[0072] The physical consistency index is normalized using average rainfall, enabling stable comparisons across different watersheds, rainfall conditions, and forecast periods. The index ranges from 0 to 1; values ​​closer to 1 indicate better overall water conservation and higher physical reliability in the forecast set. For example, within a specific watershed forecast period, a smaller mean residual will result in a higher physical consistency index, reflecting good physical rationality of the forecast results.

[0073] It should be noted that, in addition to the overall physical consistency index, a local physical consistency index can also be calculated for each time step to identify periods of physical imbalance. This index provides an intuitive basis for engineering applications, such as quickly identifying outliers that may be caused by physical imbalances during flood peak prediction.

[0074] The above-described implementation scheme of this application calculates the physical consistency index by means of average rainfall and residual mean. This index can normalize and quantitatively characterize the overall physical equilibrium state of the prediction result set, reflecting the degree to which the hydrological prediction results conform to the law of water conservation. It provides a quantifiable and comparable evaluation basis for the physical credibility of the model, effectively improves the uncertainty quantification system, and enhances the interpretability and engineering application value of the hydrological prediction results.

[0075] The above describes the hydrological prediction model uncertainty correction method provided in the embodiments of this application. The hydrological prediction model uncertainty correction device provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0076] like Figure 2 As shown, this embodiment of the invention also provides a hydrological prediction model uncertainty correction device, the device comprising: Processing module 201 is used to perform a preset number of perturbations on the target hydrological prediction model when predicting the outflow of the target watershed using the target hydrological prediction model. The first acquisition module 202 is used to acquire the prediction results output by the target hydrological prediction model after each disturbance processing, and obtain a prediction result set; The first calculation module 203 is used to calculate the residual value of each prediction result in the prediction result set based on the preset water conservation equation. The first determining module 204 is used to determine the physical consistency weight coefficient corresponding to each prediction result based on the residual value corresponding to the prediction result. The construction module 205 is used to perform weighted statistics on the prediction result set according to the physical consistency weight coefficients corresponding to each prediction result in the prediction result set, and construct a weighted probability distribution function. The second determining module 206 is used to determine the confidence interval of the outlet flow of the target watershed based on the weighted probability distribution function and the preset quantile.

[0077] Optionally, the processing module includes: The first processing submodule is used to perform a first preset number of variable perturbation processes on the target hydrological prediction model, wherein the variable perturbation process is to superimpose random noise less than a preset noise threshold onto at least one input variable in the input variable set corresponding to the target hydrological prediction model. The second processing submodule is used to perform a second preset number of model perturbation processes on the target hydrological prediction model; wherein, the model perturbation process is to adjust the model parameters corresponding to the target hydrological prediction model to a value less than a preset amplitude threshold. Wherein, the sum of the first preset quantity and the second preset quantity is equal to the preset quantity.

[0078] Optionally, the first computing module includes: The first acquisition submodule is used to acquire the set of input variables corresponding to the prediction result; the set of input variables includes the rainfall, evapotranspiration and water storage change of the target watershed; The first determining submodule is used to determine the residual value corresponding to the prediction result based on the fact that the rainfall is equal to the sum of the evapotranspiration, the prediction result, the change in water storage, and the residual value.

[0079] Optionally, the first determining module includes: The second acquisition submodule is used to acquire the target penalty coefficient that matches the target watershed. The first calculation submodule is used to take the opposite of the product of the absolute value of the residual and the target penalty coefficient as the exponent of the natural exponential function, and calculate the physical consistency weight coefficient corresponding to the prediction result.

[0080] Optionally, the building modules include: The second calculation submodule is used to calculate the weighted mean and variance based on the physical consistency weight coefficients corresponding to each prediction result in the prediction result set. A submodule is constructed to construct the weighted probability distribution function based on the weighted mean and the variance.

[0081] Optionally, the second determining module includes: The second determining submodule is used to determine the upper boundary and lower boundary based on the weighted probability distribution function and the preset quantile; The third determining submodule is used to determine the confidence interval of the outflow based on the upper boundary limit and the lower boundary limit.

[0082] Optionally, after calculating the residual values ​​of each prediction result in the prediction result set based on the preset water conservation equation, the device further includes: The second acquisition module is used to acquire the average rainfall of the target watershed within a preset time period; The second calculation module is used to calculate the average value of the residuals of each prediction result in the prediction result set, and obtain the residual mean. The third determining module is used to determine the physical consistency index as the ratio of the difference between the average rainfall and the mean residual to the average rainfall.

[0083] Optionally, the second acquisition submodule includes: The first determining unit is used to determine the target basic penalty coefficient range corresponding to the target watershed type based on the correspondence between the watershed type and the basic penalty coefficient range, according to the target watershed type corresponding to the target watershed. The acquisition unit is used to acquire the historical penalty coefficient, actual observation coverage, and confidence interval width corresponding to the target watershed. An adjustment unit is used to adjust the historical penalty coefficient within the target basic penalty coefficient range based on the actual observation coverage and the confidence interval width, so as to obtain the target penalty coefficient.

[0084] The hydrological prediction model uncertainty correction device provided in this application, and the technical solution of this application embodiment, generates a prediction result set by perturbing the target hydrological prediction model, calculates the residual value of each prediction result based on the water conservation equation, determines the physical consistency weight coefficient based on the residual value, constructs a weighted probability distribution function by weighted statistics, and determines the outlet flow confidence interval according to the preset quantile. This application can integrate the physical constraints of water conservation into the entire uncertainty quantification process without changing the model structure, weakens the impact of physically unreasonable prediction results on uncertainty assessment, and makes the final confidence interval have both statistical rationality and physical consistency, improves the credibility and engineering applicability of hydrological prediction model uncertainty assessment, and makes the confidence interval of outlet flow prediction conform to the hydrological and physical laws of the watershed, providing more reliable and physically interpretable uncertainty quantification results for hydrological forecasting.

[0085] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0086] This application also provides an electronic device, including: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the various processes of the above-described hydrological prediction model uncertainty correction method embodiment and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0087] For example, Figure 3 A schematic diagram of the physical structure of an electronic device is shown. (For example...) Figure 3As shown, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other through the communications bus 340. The processor 310 can call logic instructions in the memory 330. The processor 310 is used to perform the following steps: when predicting the outlet flow of a target watershed using a target hydrological prediction model, the target hydrological prediction model is subjected to a preset number of perturbation processes; the prediction results output by the target hydrological prediction model after each perturbation process are obtained to obtain a prediction result set; based on a preset water conservation equation, the residual values ​​of each prediction result in the prediction result set are calculated respectively; for each prediction result, the physical consistency weight coefficient corresponding to the prediction result is determined according to the residual value corresponding to the prediction result; the prediction result set is weighted statistically based on the physical consistency weight coefficients corresponding to each prediction result in the prediction result set to construct a weighted probability distribution function; and the confidence interval of the outlet flow of the target watershed is determined according to the weighted probability distribution function and a preset quantile. The processor 310 can also execute other schemes in the embodiments of this application, which will not be further described here.

[0088] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0089] This application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described hydrological prediction model uncertainty correction method embodiment and achieves the same technical effect. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0090] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0091] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0092] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0093] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0094] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0095] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0096] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0097] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0098] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0099] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A hydrologic forecast model uncertainty correction method, characterized by, The method includes: When predicting the outflow of a target watershed using a target hydrological prediction model, the target hydrological prediction model is subjected to a preset number of perturbations. Obtain the prediction results output by the target hydrological prediction model after each disturbance processing to obtain a prediction result set; Based on the preset water conservation equation, the residual values ​​of each prediction result in the prediction result set are calculated respectively; For each prediction result, a physical consistency weight coefficient is determined based on the residual value corresponding to the prediction result. Based on the physical consistency weight coefficients corresponding to each prediction result in the prediction result set, a weighted probability distribution function is constructed by performing weighted statistics on the prediction result set. The confidence interval for the outflow of the target watershed is determined based on the weighted probability distribution function and the preset quantile.

2. The method for correcting uncertainty in a hydrological prediction model according to claim 1, characterized in that, The target hydrological prediction model is subjected to a predetermined number of perturbation processes, including: The target hydrological prediction model is subjected to a first preset number of variable perturbation processes, wherein the variable perturbation process is to superimpose random noise less than a preset noise threshold onto at least one input variable in the input variable set corresponding to the target hydrological prediction model. The target hydrological prediction model is subjected to a second preset number of model perturbation processes; wherein, the model perturbation process is to adjust the model parameters corresponding to the target hydrological prediction model to a value less than a preset amplitude threshold. Wherein, the sum of the first preset quantity and the second preset quantity is equal to the preset quantity.

3. The method for correcting uncertainty in a hydrological prediction model according to claim 1, characterized in that, Based on the preset water conservation equation, the residual values ​​of each prediction result in the prediction result set are calculated, including: Obtain the set of input variables corresponding to the prediction results; the set of input variables includes the rainfall, evapotranspiration and water storage changes of the target watershed; Based on the fact that the rainfall is equal to the sum of the evapotranspiration, the prediction result, the change in water storage, and the residual value, the residual value corresponding to the prediction result is determined.

4. The method for correcting uncertainty in a hydrological prediction model according to claim 1, characterized in that, Based on the residual value corresponding to the prediction result, the physical consistency weight coefficient corresponding to the prediction result is determined, including: Obtain the target penalty coefficient that matches the target watershed; The physical consistency weight coefficient corresponding to the prediction result is calculated by taking the opposite of the product of the absolute value of the residual and the target penalty coefficient as the exponent of the natural exponential function.

5. The method for correcting uncertainty in a hydrological prediction model according to claim 1, characterized in that, Based on the physical consistency weight coefficients corresponding to each prediction result in the prediction result set, a weighted statistical method is constructed to gather the prediction result set and include: Calculate the weighted mean and variance based on the physical consistency weight coefficients corresponding to each prediction result in the prediction result set. The weighted probability distribution function is constructed based on the weighted mean and the variance.

6. The method for correcting uncertainty in a hydrological prediction model according to claim 1, characterized in that, Based on the weighted probability distribution function and the preset quantile, the confidence interval for the outlet flow of the target watershed is determined, including: Based on the weighted probability distribution function and the preset quantiles, determine the upper and lower bounds of the boundary; The confidence interval for the outflow is determined based on the upper and lower bounds of the boundary.

7. The method for correcting uncertainty in a hydrological prediction model according to claim 1, characterized in that, After calculating the residual values ​​of each prediction result in the prediction result set based on the preset water conservation equation, the method further includes: Obtain the average rainfall of the target watershed within a preset time period; Calculate the average value of the residuals of each prediction result in the prediction result set to obtain the residual mean; The ratio of the difference between the average rainfall and the mean residual to the average rainfall is determined as the physical consistency index.

8. The method according to claim 4, characterized in that, Obtaining the target penalty coefficient matching the target watershed includes: Based on the correspondence between watershed type and basic penalty coefficient range, the target basic penalty coefficient range corresponding to the target watershed type is determined according to the target watershed type. Obtain the historical penalty coefficient, actual observation coverage, and confidence interval width corresponding to the target watershed; Based on the actual observation coverage and confidence interval width, the historical penalty coefficient is adjusted within the target basic penalty coefficient interval to obtain the target penalty coefficient.

9. A device for correcting uncertainty in a hydrological prediction model, characterized in that, include: The processing module is used to perform a preset number of perturbations on the target hydrological prediction model when predicting the outflow of the target watershed using the target hydrological prediction model. The first acquisition module is used to acquire the prediction results output by the target hydrological prediction model after each disturbance processing, and obtain a prediction result set. The first calculation module is used to calculate the residual value of each prediction result in the prediction result set based on the preset water conservation equation. The first determining module is used to determine the physical consistency weight coefficient corresponding to each prediction result based on the residual value corresponding to the prediction result. The construction module is used to perform weighted statistics on the prediction result set based on the physical consistency weight coefficients corresponding to each prediction result in the prediction result set, and construct a weighted probability distribution function. The second determining module is used to determine the confidence interval of the outlet flow of the target watershed based on the weighted probability distribution function and the preset quantile.

10. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the hydrological prediction model uncertainty correction method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the hydrological prediction model uncertainty correction method as described in any one of claims 1 to 8.

Images