Full-scale runoff prediction method and system based on error cascade correction

Abstract

The invention provides a full-scale runoff prediction method and system based on error cascade correction, and the method comprises the steps: carrying out the runoff prediction of different time scales for a calculation section through employing an existing method, and carrying out the analysis and calculation of a multi-scale prediction error through combining with a historical observation runoff sequence; respectively carrying out distribution law analysis and transfer law analysis by adopting a Gaussian mixture model and a state space model according to different time scale errors such as fine, medium and coarse time scale errors; then, a cascade correction model is constructed, error rolling correction from fine scale to medium scale and from medium scale to coarse scale is achieved, and correction values are obtained; and applying the correction values at all levels to initial forecasting to obtain a corrected full-scale runoff prediction result. Through cascade correction for runoff prediction error characteristics of different scales, error correction can be realized more scientifically, efficiently and specifically, so that the full-scale runoff prediction precision is further improved.

Description

Technical Field

[0001] This invention relates to the field of hydrological and water resources forecasting technology, and in particular to a full-scale runoff prediction method and system based on error cascade correction. Background Technology

[0002] Hydrological forecasting is a core technical support for water resource management and flood and drought disaster prevention. Its accuracy directly affects the effectiveness of water project scheduling decisions and the rationality of water resource allocation. Against the backdrop of increasingly scarce water resources and frequent extreme hydrological events, forecasting errors directly translate into decision-making risks. For reservoir scheduling, underestimating peak flood flow may lead to insufficient flood control capacity, triggering dam failure risks; while overestimating it may result in inappropriate water release, reducing beneficial effects. In inter-basin water resource allocation, forecasting errors can mislead water allocation plans, causing inter-regional water-related conflicts.

[0003] However, forecasting error, as an inherent phenomenon, permeates the entire process of runoff prediction. Runoff forecasting error stems from various factors, including model simplification, parameter uncertainty, input data noise, and spatiotemporal scale effects, and is an objective reality that cannot be eliminated in hydrological forecasting. It is noteworthy that error is not static but exhibits significant dynamic accumulation and amplification characteristics as the forecast period extends. In traditional forecasting models, small initial deviations are continuously propagated and amplified through iterative calculations of model state variables as the forecast step progresses, forming an "error cascade" effect. This effect is particularly prominent in distributed hydrological models or coupled meteorological-hydrological models, where errors from upstream units propagate downstream and overlap with locally generated errors, potentially leading to forecast results at the end of the forecast period that deviate significantly from reality. Existing correction techniques mostly target a single forecast period or a specific node for independent correction, lacking explicit description and intervention of the propagation law of errors along spatiotemporal paths, thus failing to effectively curb the cascading amplification trend of errors.

[0004] Traditional error correction methods often focus on local corrections to the final forecast results, such as post-processing correction or real-time correction. While these methods can improve instantaneous forecast accuracy to some extent, their effects are often limited to the short term. Furthermore, they fail to consider the propagation and feedback of errors in different scheduling stages (such as forecasting, decision-making, and execution), and thus fail to systematically address the dynamic propagation and cumulative effects of errors in the forecast chain. Autoregressive models, Kalman filters, and their variants, on the other hand, primarily rely on statistical analysis of the residuals between the forecast and observation sequences, achieving passive, "post-hoc" correction. When scheduling decisions are based on a "point correction" result without error propagation analysis, the seemingly precise short-term improvements may lead to larger systemic problems in the medium to long term—a risk of "correction shortsightedness." Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art by providing a full-scale runoff prediction method and system based on error cascade correction. This method can obtain a clearer picture of the evolution of forecast uncertainty over the lead time, thereby enabling the formulation of more forward-looking and adaptive scheduling schemes and systematically reducing the decision-making risks caused by error accumulation.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a full-scale runoff prediction method based on error cascade correction, comprising: S1. Select the calculation section and obtain the historical runoff sequence, denoted as... ; S2. Use existing methods to conduct runoff prediction and generate forecast runoff sequences at different time scales, i.e., initial forecasts; Among them, the hourly forecasts for the next 1 to 10 days are recorded as the fine-scale predicted runoff sequence. ; Daily forecasts for the next 1-4 weeks are recorded as mesoscale predicted runoff sequences. ; Monthly forecasts for the next 1 to 6 months are recorded as coarse-scale runoff sequences. ; If needed, further encryption can be used to generate ten-day forecasts of runoff for the next 1-3 months and quarterly forecasts of runoff for the next 1-12 months. To ensure that forecasts at different scales are aligned in time, i.e., starting from the same initial moment; S3. Calculate the multi-scale forecast error; S4. Error distribution pattern analysis; S5. Error propagation law analysis; S6. Construct a cascaded correction model; S7: Correction Calculation and Output: Apply the correction values ​​at each level to the initial forecast to obtain and output the corrected full-scale runoff forecast results.

[0007] Furthermore, S3 specifically involves calculating the differences between observed and predicted values ​​for the runoff sequences at different scales (fine, medium, and coarse), as follows: Fine-scale error: ; Mesoscale error: Aggregate hourly observations into a daily average or take a fixed time each day as the daily forecast value. ; Coarse-scale error: Aggregation yields monthly observations Monthly forecast value , ; in, hourly series Prediction error of time; hourly series The observed value at time; hourly series Forecast value for the time; Total hours; For the daily sequence number Forecast error for days; For the daily sequence number The observed value of the day; taken as the first day. The average value of different observation periods over a day or the observed value of a certain period; For the daily sequence number Forecast value for the day; Similarly, Corresponding to the number The average value for different forecast periods or the forecast value for a certain period; D is the total number of days; or ,in, For the first The total number of observation or forecast periods per day ; For the monthly sequence number Monthly forecast error; Total number of months; For the monthly sequence number Monthly observations; For the monthly sequence number Forecast value for the month; , Simultaneously take as the first The average of daily or several days' observations and forecasts for a month, or the average of a single day's observations and forecasts. If the prediction results of existing methods For the first The monthly average, then Let's take it as the first. The average of daily observations for the month.

[0008] Furthermore, S4 specifically includes: The historical error sequence is fitted using a Gaussian mixture model, specifically as follows: ; in, As a scale; , , , As weight; It follows a Gaussian distribution.

[0009] Furthermore, S5 specifically involves: analyzing the error propagation law using a state-space model, as follows: S501. Calculate the error status using the following formula: ; in, For the first The set of all hour-scale error states contained in the sky is the mean or variance; For the first Solar-scale error status; It can be a linear or nonlinear transfer function; This is the set of parameters for the model; This is process noise; S502, Calculate the mapping relationship, that is, through the formula Establish a mapping relationship between error values ​​and error states; in, This refers to the diurnal scale error calculated based on actual observations. For measurement functions; For measuring noise.

[0010] Furthermore, S6 specifically includes: S601, Model Input: Real-time fine-scale error, i.e., in the early stages of forecasting, based on the actual observations available from the previous few hours or days. Calculate real-time error ; The initial forecasts at the mesoscale and coarse scales are as follows: , ; S602. Correction process: Correction is performed using Bayesian update, specifically as follows: S6021. Using the error state calculation formula, based on the current hourly error state... Predicting the prior distribution of errors on future day-scale scales Once the actual observation data for the first day is obtained, the daily error will be calculated in real time. As The mapping values ​​are used to update the posterior distribution of the daily-scale error state using Bayes' theorem. Thus, the daily-scale error correction amount is obtained. Corrected mesoscale forecast: ;in, This is the corrected forecast value; S601, Correction from mesoscale to coarse scale: Establish the error state and mapping relationship from mesoscale to coarse-scale using the pre-corrected mesoscale forecast sequence. Error status of data updated a few days ago on a monthly scale ; Calculate the error correction amount on a monthly scale Correction of coarse-scale forecasts: .

[0011] Furthermore, a full-scale runoff prediction system based on error cascade correction includes at least one processor; and a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor to implement the full-scale runoff prediction method based on error cascade correction.

[0012] The beneficial effects of this invention are as follows: First, existing methods are used to conduct runoff forecasting at different time scales for the calculation cross section, and multi-scale forecast errors are calculated by combining historical runoff observation sequences. Second, for errors at different time scales such as fine, medium, and coarse, a Gaussian mixture model and a state-space model are used to analyze the distribution patterns and identify the transfer functions. Then, a cascaded correction model is constructed to achieve rolling error correction from fine scale to medium scale and from medium scale to coarse scale, obtaining the correction values. Finally, the correction values ​​at each level are applied to the initial forecast to obtain the corrected full-scale runoff prediction results. By cascaded correction targeting the characteristics of runoff prediction errors at different scales, error correction can be achieved more scientifically, efficiently, and specifically, thereby further improving the accuracy of full-scale runoff prediction. Attached Figure Description

[0013] Figure 1 This is a flowchart of a full-scale runoff prediction method based on error cascade correction; Figure 2 This is a schematic diagram of the forecasting process of a reservoir at different time scales in the example; Figure 3 This is a schematic diagram of the 1-3 day forecast process for a certain reservoir in the example; Figure 4 This is a schematic diagram of the 1-7 day forecast process for a certain reservoir in the example; Figure 5 This is a schematic diagram of the 1-10 day forecast process for a certain reservoir in the example. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0015] Please see Figure 1 A full-scale runoff prediction method based on error cascade correction includes: S1. Select the calculation section and obtain the historical runoff sequence, denoted as... ; S2. Use existing methods to conduct runoff prediction and generate forecast runoff sequences at different time scales, i.e., initial forecasts; Among them, the hourly forecasts for the next 1 to 10 days are recorded as the fine-scale predicted runoff sequence. ; Daily forecasts for the next 1-4 weeks are recorded as mesoscale predicted runoff sequences. ; Monthly forecasts for the next 1 to 6 months are recorded as coarse-scale runoff sequences. ; If needed, further encryption can be used to generate ten-day forecasts of runoff for the next 1-3 months and quarterly forecasts of runoff for the next 1-12 months. To ensure that forecasts at different scales are aligned in time, i.e., starting from the same initial moment; S3. Calculate the multi-scale forecast error; S4. Error distribution pattern analysis; S5. Error propagation law analysis; S6. Construct a cascaded correction model; S7: Correction Calculation and Output: Apply the correction values ​​at each level to the initial forecast to obtain and output the corrected full-scale runoff forecast results.

[0016] Specifically, S3 involves calculating the differences between observed and predicted values ​​for the runoff sequences at different scales (fine, medium, and coarse), as follows: Fine-scale error: ; Mesoscale error: Aggregate hourly observations into a daily average or take a fixed time each day as the daily forecast value. ; Coarse-scale error: Aggregation yields monthly observations Monthly forecast value , ; in, hourly series Prediction error of time; hourly series The observed value at time; hourly series Forecast value for the time; Total hours; For the daily sequence number Forecast error for days; For the daily sequence number The observed value of the day; taken as the first day. The average value of different observation periods over a day or the observed value of a certain period; For the daily sequence number Forecast value for the day; Similarly, Corresponding to the number The average value for different forecast periods or the forecast value for a certain period; D is the total number of days; or ,in, For the first The total number of observation or forecast periods per day ; For the monthly sequence number Monthly forecast error; Total number of months; For the monthly sequence number Monthly observations; For the monthly sequence number Forecast value for the month; , Simultaneously take as the first The average of daily or several days' observations and forecasts for a month, or the average of a single day's observations and forecasts. If the prediction results of existing methods For the first The monthly average, then Let's take it as the first. The average of daily observations for the month.

[0017] Specifically, S4 is: The historical error sequence is fitted using a Gaussian mixture model, specifically as follows: ; in, As a scale; , , , As weight; It follows a Gaussian distribution.

[0018] Specifically, S5 involves using a state-space model to analyze the error propagation pattern, as follows: S501. Calculate the error status using the following formula: ; in, For the first The set of all hour-scale error states contained in the sky is the mean or variance; For the first Solar-scale error status; It can be a linear or nonlinear transfer function; This is the set of parameters for the model; This is process noise; S502, Calculate the mapping relationship, that is, through the formula Establish a mapping relationship between error values ​​and error states; in, This refers to the diurnal scale error calculated based on actual observations. For measurement functions; For measuring noise.

[0019] Specifically, S6 is: S601, Model Input: Real-time fine-scale error, i.e., in the early stages of forecasting, based on the actual observations available from the previous few hours or days. Calculate real-time error ; The initial forecasts at the mesoscale and coarse scales are as follows: , ; S602. Correction process: Correction is performed using Bayesian update, specifically as follows: S6021. Using the error state calculation formula, based on the current hourly error state... Predicting the prior distribution of errors on future day-scale scales Once the actual observation data for the first day is obtained, the daily error will be calculated in real time. As The mapping values ​​are used to update the posterior distribution of the daily-scale error state using Bayes' theorem. Thus, the daily-scale error correction amount is obtained. Corrected mesoscale forecast: ;in, This is the corrected forecast value; S601, Correction from mesoscale to coarse scale: Establish the error state and mapping relationship from mesoscale to coarse-scale using the pre-corrected mesoscale forecast sequence. Error status of data updated a few days ago on a monthly scale ; Calculate the error correction amount on a monthly scale Correction of coarse-scale forecasts: .

[0020] A full-scale runoff prediction system based on error cascade correction includes at least one processor; and a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor to implement the full-scale runoff prediction method based on error cascade correction.

[0021] Taking a reservoir in the upper reaches of the Yangtze River as an example, we will conduct rolling forecasts of inflow at different time scales to verify the feasibility and effectiveness of the method of the present invention.

[0022] As a core water project connecting the upper and lower reaches of the Yangtze River, a certain reservoir plays a vital role in the accurate forecasting of its runoff. Accurate runoff forecasting is crucial for the coordinated scientific management of the upstream reservoir group and for flood control safety in the middle and lower reaches. Improving runoff forecasting accuracy at different time scales, such as daily, ten-day, and monthly, has always been a goal pursued by technical professionals. This paper uses the short-, medium-, and long-term inflow runoff forecasts of a reservoir during the flood season of a certain year as an example to verify the feasibility and effectiveness of the method of this invention. Figure 2 A comparison diagram of the process lines of the short-to-medium-term (1-10 days) progressive forecast for the reservoir and the actual situation was drawn. Figures 3 to 5 Forecast process lines at different time scales were plotted during the rolling forecast process. The shorter lead time forecast was obtained by considering the deviation correction between the actual data and the forecast data based on the longer lead time forecast.

[0023] Tables 1 and 2 analyze the comparison between the long-term (monthly) ten-day runoff forecast and the actual flow of the reservoir. It can be found that using the method of the present invention for multi-scale runoff rolling forecast, based on the actual development process, and dynamically correcting the forecast error for different time scales can effectively improve the accuracy of the forecast.

[0024] Table 1. Initial Long-Term Forecast Results for a Certain Reservoir time Forecast flow (natural) Natural flow tendency value Real-time natural flow Forecast flow (actual) Actual traffic tendency Actual traffic volume Late September 27000～33000 29000 31600 25000～30000 27000 28500 early October 23000～28000 25000 28500 20000～25000 22000 25000 mid-october 16000～21000 18000 21500 15000～20000 17000 19000 Late October 12000～17000 14000 14500 11000～16000 13000 12000 Table 2. Long-term ten-day correction forecast results for a certain reservoir time Forecast flow (natural) Natural flow tendency value Real-time natural flow Forecast flow (actual) Actual traffic tendency Actual traffic volume Late September (some live footage has already been released) 29000～33000 31500 31600 27000～30000 28500 28500 early October 25000～30000 27000 28500 23000～28000 25000 25000 mid-october 19000～24000 21000 21500 16000～21000 18000 19000 Late October 12000～17000 15000 14500 11000～16000 13000 12000 Depend on Figures 2 to 5 As shown in Tables 1 and 2, the method of this invention employs a Gaussian mixture model and a state-space model for distribution pattern analysis and transfer function identification, and constructs a cascaded correction model to quickly achieve rolling error correction from fine-scale to mesoscale and from mesoscale to coarse-scale. Through cascaded correction targeting the error characteristics of runoff prediction at different scales, the accuracy of runoff prediction across all scales is improved, demonstrating the feasibility and effectiveness of this method. Therefore, this method exhibits superior application results in runoff prediction across all scales.

[0025] Based on the above analysis, it can be seen that the method of the present invention is scientific, practical and operable, and can effectively improve the accuracy of runoff prediction at different time scales. By cascading correction of the error characteristics of runoff prediction at different scales, error correction can be achieved more scientifically, efficiently and in a more targeted manner, so as to further improve the accuracy of runoff prediction at all scales and solve the problems of insufficient accuracy and large uncertainty in runoff forecast at different time scales.

[0026] In summary, this invention has advantages such as practicality and strong operability, and can quickly realize full-scale runoff prediction and obtain high-precision runoff prediction results, providing a more scientific and efficient new method for watershed water resource allocation and water project scheduling.

[0027] The embodiments described above are merely illustrative of implementation methods of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be defined by the appended claims.

Claims

1. A full-scale runoff prediction method based on error-cascaded correction, characterized by, include: S1, select the calculation section, obtain the historical observed runoff sequence, denoted as ; S2. Use existing methods to conduct runoff prediction and generate forecast runoff sequences at different time scales, i.e., initial forecasts; Among them, the hourly forecasts for the next 1 to 10 days are recorded as the fine-scale predicted runoff sequence. ; Daily forecasts for the next 1-4 weeks are recorded as mesoscale predicted runoff sequences. ; Monthly forecasts for the next 1 to 6 months are recorded as coarse-scale runoff sequences. ; If needed, further encryption can be used to generate ten-day forecasts of runoff for the next 1-3 months and quarterly forecasts of runoff for the next 1-12 months. To ensure that forecasts at different scales are aligned in time, i.e., starting from the same initial moment; S3. Calculate the multi-scale forecast error; S4. Error distribution pattern analysis; S5. Error propagation law analysis; S6. Construct a cascaded correction model; S7: Correction Calculation and Output: Apply the correction values ​​at each level to the initial forecast to obtain and output the corrected full-scale runoff forecast results.

2. The full-scale runoff prediction method based on error cascade correction according to claim 1, characterized in that, Specifically, S3 involves calculating the differences between observed and predicted values ​​for the runoff sequences at different scales (fine, medium, and coarse), as follows: Fine-scale error: ; Mesoscale error: Aggregate hourly observations into a daily average or take a fixed time each day as the daily forecast value. ; Coarse-scale error: Aggregation yields monthly observations Monthly forecast value , ; in, hourly series Prediction error of time; hourly series The observed value at time; hourly series Forecast value for the time; Total hours; For the daily sequence number Forecast error for days; For the daily sequence number The observed value of the day; taken as the first day. The average value of different observation periods over a day or the observed value of a certain period; For the daily sequence number Forecast value for the day; Similarly, Corresponding to the number The average value for different forecast periods or the forecast value for a certain period; D is the total number of days; or ,in, For the first The total number of observation or forecast periods per day ; For the monthly sequence number Monthly forecast error; Total number of months; For the monthly sequence number Monthly observations; For the monthly sequence number Forecast value for the month; , Simultaneously take as the first The average of daily or several days' observations and forecasts for a month, or the average of a single day's observations and forecasts. If the prediction results of existing methods For the first The monthly average, then Let's take it as the first. The average of daily observations for the month.

3. The full-scale runoff prediction method based on error cascade correction according to claim 2, characterized in that, Specifically, S4 is: The historical error sequence is fitted using a Gaussian mixture model, specifically as follows: ； in, As a scale; , , , As weight; It follows a Gaussian distribution.

4. The full-scale runoff prediction method based on error cascade correction according to claim 3, characterized in that, Specifically, S5 involves using a state-space model to analyze the error propagation pattern, as follows: S501. Calculate the error status using the following formula: ； in, For the first The set of all hour-scale error states contained in the sky is the mean or variance; For the first Solar-scale error status; It can be a linear or nonlinear transfer function; This is the set of parameters for the model; This is process noise; S502, Calculate the mapping relationship, that is, through the formula Establish a mapping relationship between error values ​​and error states; in, This refers to the diurnal scale error calculated based on actual observations. For measurement functions; For measuring noise.

5. The full-scale runoff prediction method based on error cascade correction according to claim 4, characterized in that, Specifically, S6 is: S601, Model Input: Real-time fine-scale error, i.e., in the early stages of forecasting, based on the actual observations available from the previous few hours or days. Calculate real-time error ; The initial forecasts at the mesoscale and coarse scales are as follows: , ; S602. Correction process: Correction is performed using Bayesian update, specifically as follows: S6021. Using the error state calculation formula, based on the current hourly error state... Predicting the prior distribution of errors on future day-scale scales ; Once the actual observation data for the first day is obtained, the daily error will be calculated in real time. As The mapping values ​​are used to update the posterior distribution of the daily-scale error state using Bayes' theorem. Thus, the daily-scale error correction amount is obtained. Corrected mesoscale forecast: ;in, This is the corrected forecast value; S601, Correction from mesoscale to coarse scale: Establish the error state and mapping relationship from mesoscale to coarse-scale using the pre-corrected mesoscale forecast sequence. Error status of data updated a few days ago on a monthly scale ; Calculate the error correction amount at the monthly scale Correction of coarse-scale forecasts: .

6. A full-scale runoff prediction system based on error cascade correction, characterized in that: It includes at least one processor; and a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor to implement the full-scale runoff prediction method based on error cascade correction as described in any one of claims 1 to 5.

Images