SSP scene reservoir group flood control scheduling method considering hydrological forecast uncertainty

Abstract

The invention discloses an SSP scene reservoir group flood control scheduling method considering hydrological forecast uncertainty, and the method comprises the steps: obtaining SSP climate scene meteorological data to drive an SWAT model to generate daily runoff, and deducing and designing a flood hydrograph; a flow error sequence is calculated based on historical simulation and measured data, optimal distribution is optimized through fitting of a multi-probability distribution model, and a 95% confidence interval is generated through Monte Carlo simulation to correct design flood; and constructing an optimization model with the goal of minimizing the highest water level of flood regulation, bringing in various constraints, solving an optimal scheduling process by adopting a discrete differential dynamic programming algorithm, and finally refining an adaptive scheduling rule. The method solves the problems that prediction uncertainty is not considered and solving efficiency is low in the prior art, flood control safety and dispatching adaptability of the reservoir group are improved, and the method is suitable for flood control dispatching of the reservoir group under climate change.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water conservancy and water resources management, and particularly relates to a SSP scenario reservoir group flood control scheduling method considering hydrological prediction uncertainty. BACKGROUND

[0002] The SSP scenario shows that the uncertainty and risk of future extreme flood events have significantly increased, and the traditional design flood estimation method based on historical sequences and reservoir scheduling strategy are facing severe challenges. Although existing research attempts to couple climate models, such as SSP, with hydrological models, such as SWAT, to predict future floods, it generally ignores the systematic error of hydrological prediction - directly using the original prediction value to drive the scheduling optimization, which leads to insufficient flood control safety when the actual flood deviates from the prediction. At the same time, the high-dimensional complexity of multi-stage flood control optimization of cascade reservoirs makes conventional intelligent algorithms, such as dynamic programming algorithm and chaos frog leap algorithm, have low solving efficiency and unstable convergence.

[0003] Therefore, the present application provides a SSP scenario reservoir group flood control scheduling method considering hydrological prediction uncertainty. SUMMARY

[0004] The present application aims to overcome the above-mentioned deficiencies and provide a SSP scenario reservoir group flood control scheduling method considering hydrological prediction uncertainty to solve the problems raised in the background art.

[0005] The present application proposes a SSP scenario reservoir group flood control scheduling method considering hydrological prediction uncertainty, comprising the following steps: Step one, obtain meteorological data under multiple SSP climate scenarios, drive the SWAT hydrological model, simulate the daily runoff process of the target basin in the future period; based on the simulated daily runoff data, use the P-III type frequency distribution curve, combine the moment method for parameter estimation, and through the same frequency amplification method, obtain the design flood hydrograph of different return periods under each SSP scenario; Step two, based on the historical period SWAT model simulation runoff data and the same period measured runoff data, calculate the daily scale flow error sequence; use multiple candidate probability distribution models to fit the daily scale flow error sequence, and obtain the optimal probability distribution model of the error according to the AIC information criterion minimization principle; Step three, select the target return period design flood hydrograph as the reference; based on the optimal probability distribution model, use the Monte Carlo random simulation method to generate the 95% confidence interval range of the daily flow value of the reference design flood hydrograph; calculate the mean value of the daily flow in the confidence interval to form the input flood hydrograph considering the prediction uncertainty; Step four, establishing a flood control optimization scheduling model with the minimization of the highest water level of the reservoir group as a single objective function, and incorporating reservoir capacity constraints, water level constraints, downstream discharge constraints, water balance constraints, flood control safety constraints, and scheduling period end water level constraints; Step five, inputting the flood process line considering forecast uncertainty into the flood control optimization scheduling model constructed in step four, applying a discrete differential dynamic programming algorithm to solve, and outputting the optimal water level control process and optimal downstream discharge process of the reservoir group; Step six, based on the optimal water level control process and optimal downstream discharge process, analyzing and refining the flood control adaptive scheduling rules of the reservoir group under the target SSP scenario and the return period.

[0006] Preferably, the plurality of SSP climate scenarios in step one include SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5.

[0007] Preferably, the candidate probability distribution model in step two includes symmetric distribution, thick-tailed distribution, and asymmetric distribution, wherein the symmetric distribution includes normal distribution and Logistic distribution, the thick-tailed distribution includes t distribution and Cauchy distribution, and the asymmetric distribution includes Johnson SU distribution and skewed normal distribution.

[0008] Preferably, the determination method of the 95% confidence interval range in step three is: for each daily flow value of the reference design flood process line, a large number of random error samples based on the optimal probability distribution model are generated by Monte Carlo simulation, and the possible value set of the daily flow is obtained by superposition, and the 2.5% quantile of the set is taken as the lower bound of the daily confidence interval, and the 97.5% quantile is taken as the upper bound of the daily confidence interval.

[0009] Preferably, the flood process line considering forecast uncertainty in step five is a flood process line formed by the arithmetic mean of the daily flow values within the 95% confidence interval range.

[0010] Preferably, the flood control safety constraint in step five includes that the highest flood level of the reservoir does not exceed the check flood level, and the downstream discharge does not exceed the downstream river channel safety discharge.

[0011] Preferably, the discrete differential dynamic programming algorithm in step five solves by discretizing the reservoir state variables and decision variables, and using the Bellman optimality principle for reverse recursion and forward optimization, wherein the state variables are reservoir capacity or water level, and the decision variables are downstream discharge or reservoir outflow.

[0012] Preferably, the flood control adaptive scheduling rule in step six is obtained by analyzing the relationship between the optimal water level control process and the incoming flood process, the time period, and fitting the obtained reservoir scheduling water level control rule or gate opening rule.

[0013] Preferably, the input data of the SWAT model in step two includes DEM, soil type data and land use data, wherein the resolution of DEM is 30m, the soil type data adopts HWSD, the land use data adopts CNLUCC2015, and the NSE and R2 of the model calibration period and the verification period are both higher than 0.7.

[0014] Preferably, the time period division of the discrete differential dynamic programming algorithm in step five is 360 time periods, corresponding to a typical flood period of 15 days, the time period step is 1 hour, the number of state variable discrete grids is 20, the number of control variable discrete grids is 30, the hard constraint penalty coefficient is 1000, and the soft constraint penalty coefficient is 100.

[0015] The present application has the following beneficial effects: 1. The present application corrects the design flood by quantifying the forecast uncertainty, reduces the highest water level for flood regulation by 0-5.5 meters, and optimizes the discharge by 4.6%-35.2%. 2. The present application adopts the DDDP algorithm, which shortens the solution time and has high convergence accuracy, and is suitable for multi-stage scheduling. 3. The present application generates dynamic scheduling rules based on the SSP scenario, adjusts the pre-discharge water level threshold according to the flood frequency, and reserves the reservoir capacity for extreme floods. 4. The present application takes into account the multi-dimensional constraints such as water level and discharge, and optimizes the flood regulation benefit while ensuring the safe discharge of the downstream river. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 Figure 1 is a diagram of the SWAT model daily scale calibration and verification results of the Qingjiang River Basin; Figure 2 Figure 2 is a scatter plot of the annual average runoff under four future scenarios in the Qingjiang River Basin; Figure 3 Figure 3 is a diagram of the error interval of the design flood in the reference period of the Qingjiang River Basin; Figure 4 Figure 4 is a diagram of the design flood error interval under the future scenario of Shuibuya; Figure 5 Figure 5 is a diagram of the design flood error interval under the future scenario of Geheyan; Figure 6 Figure 6 is a comparison diagram of the adaptive scheduling process and the original scheduling process of Shuibuya in the reference period; Figure 7 Figure 7 is a comparison diagram of the adaptive scheduling process and the original scheduling process of Geheyan in the reference period. Figure 8 The comparison chart of adaptive scheduling process and original scheduling process of Shuibuya Reservoir is SSP1-2.6. Figure 9 The comparison chart of adaptive scheduling process and original scheduling process of Geheyan Reservoir is SSP1-2.6. Figure 10 The comparison chart of adaptive scheduling process and original scheduling process of Shuibuya Reservoir is SSP2-4.5. Figure 11 The comparison chart of adaptive scheduling process and original scheduling process of Geheyan Reservoir is SSP2-4.5. Figure 12 The comparison chart of adaptive scheduling process and original scheduling process of Shuibuya Reservoir is SSP3-7.0. Figure 13 The comparison chart of adaptive scheduling process and original scheduling process of Geheyan Reservoir is SSP3-7.0. Figure 14 The comparison chart of adaptive scheduling process and original scheduling process of Shuibuya Reservoir is SSP5-8.5. Figure 15 The comparison chart of adaptive scheduling process and original scheduling process of Geheyan Reservoir is SSP5-8.5. DETAILED DESCRIPTION

[0017] The application will be further described below in combination with the drawings and examples: Reference is made to the drawings shown: Figures 1-15 1. Implementation environment and data preparation 1.1 Research area Taking Shuibuya and Geheyan Reservoirs of Qingjiang cascade as implementation objects, the reservoir capacities are 45.8 billion m³ and 33.4 billion m³ respectively.

[0018] 1.2 Climate data Download daily scale precipitation and temperature data under SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5 scenarios from 2015 to 2100 from the CMIP6 climate model data set website, and spatially downscale to the basin grid, i.e. 0.50°x0.50°.

[0019] 1.3 Hydrological data SWAT model input: DEM is 30m, soil type is HWSD, and land use is CNLUCC 2015; 1992-2004 is the calibration period; 2005-2014 is the verification period, and the NSE and R 2 of daily scale simulation are higher than 0.7, and the simulation results of the SWAT model are shown in Figure 1 . ​

[0020] 2. Design flood routing 2.1 Runoff simulation The daily runoff under each SSP scenario was extracted from the SWAT model output, and the sequence from 2015 to 2100 was used as the design basis. The scatter plot of the annual average runoff under the four future scenarios in the Qingjiang River Basin is shown in Fig. 2. Figure 2

[0021] 2.2 Frequency analysis Firstly, the P-III distribution curve was selected, and the parameter estimation was completed combined with the moment method. The peak flow and different duration flood under the baseline period and the four future climate scenarios were calculated, and the design flood hydrograph was obtained by using the same frequency amplification method. 3. Forecast error quantification and correction 3.1 Error sequence construction The forecast error is often used in hydrological forecasting to characterize the uncertainty of the forecast. The forecast error value is defined as the difference between the actual flow and the simulated flow, and the calculation formula is as follows:

[0022] In the formula: e t The error value of runoff forecast at time t , is represented by Q obs,t The measured runoff value at time period t , is represented by Q sim,t The simulated runoff value at time period t , is represented by

[0023] Based on the historical measured runoff data from the initial stage of the reservoir construction to 2015 and the SWAT model simulation runoff data at the same period, the daily runoff sequence in the flood season was extracted, and the error sequence was calculated according to formula 3-1.

[0024] 3.2 Distribution fitting optimization The error sequence was fitted by using the probability distribution fitting method. The model was selected based on the Akaike information criterion, abbreviated as AIC, and the simulation results are shown in Table 1.

[0025] Table 1 Comparison of AIC values of different distribution models for fitting of cascade reservoirs

[0026] The results show that the AIC value of Johnson SU distribution is the lowest, indicating that it has the best fitting effect compared with other candidate distributions.

[0027] 3.3 Monte Carlo correction of design flood ​​​Therefore, the Johnson SU distribution is used to fit the error sequence of Qingjiang cascade reservoirs, and 10,000 error samples are randomly generated to obtain the 95% confidence interval of Shuibu'ya error sequence [-2079.99, 622.41] and Geheyan error sequence [-3571.15, 898.94]. The error range of the design flood of the two reservoirs is shown in Figure 3 、 Figure 4 、 Figure 5 .

[0028] 4. Solution of flood control optimization model 4.1 Model construction Objective function:

[0029] Constraint conditions: 1) Water balance equation constraint

[0030] In the formula: 、 is the initial and final reservoir storage at t period, is the average inflow of the reservoir at t period, is the average discharge of the reservoir at t period, is the time length.

[0031] 2) Reservoir water level constraint

[0032] In the formula: is the minimum water level allowed during reservoir regulation; is the initial water level of the reservoir at t, is the maximum water level allowed by the reservoir.

[0033] 3) Reservoir discharge capacity constraint

[0034] In the formula: is the average discharge at t period; is the maximum discharge.

[0035] 4) Reservoir downstream flood control safety constraint

[0036] In the formula: is the safe discharge of the downstream river channel, is the discharge capacity when the reservoir water level is z. 5) Final water level constraint

[0037] Where: is the final water level of the i-th reservoir operation period; is the flood limit water level of the i-th reservoir. The final water level is generally controlled at the flood limit water level. If there is still a large flood in the later period of the forecast, the target water level can be controlled below the flood limit water level.

[0038] 4.2 DDDP Algorithm Implementation In the implementation of discrete differential dynamic programming (DDDP) algorithm, the key parameters are set as follows: 1) Time period division: The total number of scheduling time periods T = 360, corresponding to a typical flood cycle of 15 days, i.e. 15 days × 24 hours, and the time period step length Δ t = 1 hour. This setting balances calculation accuracy and timeliness to meet the short-term flood control and scheduling needs of cascade reservoirs.

[0039] 2) Iteration control: Optimize the number of iterations This value ensures that the objective function converges to the optimal solution, that is, the optimization cost is minimized.

[0040] 3) State space discretization: state variables, i.e., the number of discrete grids for reservoir capacity / water level Ns =20, control variables, such as the number of discrete grids for the discharge flow Nc= 30.

[0041] 4) Constraint Penalty Mechanism: The model evaluates optimization results using an optimization cost. This cost is composed of the sum of the highest flood control levels of the cascade reservoirs and the penalty for violating any constraint. Constraints are categorized as hard and soft. Constraints other than the final water level constraint are considered hard constraints. Violating a hard constraint incurs a significant penalty, while violating the final water level constraint incurs a relatively small penalty. The penalty coefficient for hard constraints is λ = 1000, and the penalty coefficient for soft constraints is λ = 100.

[0042] 5. Scheduling Strategy Refinement 5.1 Analysis of optimization results Comparison results of conventional and optimized operation of the Qingjiang cascade reservoirs under the baseline period and four future scenarios are as follows: Figures 6-15 As shown, the results indicate that the optimal solution for the four frequency floods in the baseline period and the SSP1-2.6, SSP2-4.5, and SSP3-7.0 scenarios and the 1% frequency flood in the SSP5-8.5 scenario was obtained by constructing a flood control scheduling model. Under the optimized scheduling, the water level of the Shuibuya Reservoir dropped in the range of 0.0-5.0m, and the discharge flow dropped in the range of 4.6%~35.2%. The water level of the Geheyan Reservoir dropped in the range of 1.1-5.5m, and the discharge flow dropped in the range of -8.9%~16.4%. When the inflow flood is large, the discharge flow will be smaller than the existing scheduling rules. When the inflow flood is small, a larger discharge flow than the existing scheduling rules will be adopted under the premise of ensuring the safety of the downstream. 5.2 Schedule policy generation According to the comparison of the benchmark period and the future scenario scheduling results, it is shown that the adaptive scheduling can effectively guarantee the safe operation of the reservoir and reduce the flood control pressure of the downstream by implementing pre-discharge regulation before the flood peak arrives. The regulation strategy dynamically responds to climate change by optimizing the pre-discharge water level threshold, and the specific pre-discharge water level threshold is shown in Table 2 and Table 3.

[0043] Table 2 Pre-discharge minimum water level of Shuibuya Unit: m

[0044] Table 3 Pre-discharge minimum water level of Geheyan Unit: m

[0045] As shown in Table 2 and Table 3, the flood frequency is significantly negatively correlated with the pre-discharge minimum water level, which is mainly affected by the upper limit of the reservoir pre-discharge capacity and the flood size. Based on this, the pre-discharge water level corresponding to the 1% frequency is set as the benchmark water level for flood control scheduling under the future climate scenario, such as 382.2 meters for SSP1-2.6 scenario of Shuibuya. The advantage is that it can meet the reservoir capacity regulation demand of regular flood and reserve emergency buffer space for extreme flood.

[0046] The above embodiments are only preferred technical solutions of the present application, and should not be regarded as a limitation of the present application. The protection scope of the present application should be based on the technical solutions recited in the claims, including equivalent replacement solutions of the technical features recited in the claims. That is, within this scope, equivalent replacement improvements are also within the protection scope of the present application.

Claims

1. A SSP scenario reservoir group flood control scheduling method considering hydrological forecast uncertainty, characterized in that, The method comprises the following steps: Step one, obtaining meteorological data under multiple SSP climate scenarios, driving the SWAT hydrological model to simulate the daily runoff process of the target basin in the future period; based on the simulated daily runoff data, using P-III type frequency distribution curve, combining with the moment method to estimate the parameters, and through the same frequency amplification method, the design flood hydrograph of different return periods under each SSP scenario is obtained; Step two, based on the historical period SWAT model simulation runoff data and the same period measured runoff data, the daily scale flow error sequence is calculated; a plurality of candidate probability distribution models are used to fit the daily scale flow error sequence, and the optimal probability distribution model of the error is obtained according to the AIC information criterion minimization principle; Step three, selecting the target return period design flood hydrograph as the benchmark; based on the optimal probability distribution model, the Monte Carlo random simulation method is used to generate the 95% confidence interval range of the daily flow value of the benchmark design flood hydrograph; The mean value of the daily flow in the confidence interval is calculated to form the input flood hydrograph considering the prediction uncertainty; Step four, establishing a flood control optimization scheduling model with the minimum of the highest water level of the reservoir group as the single objective function, and incorporating the reservoir capacity constraint, water level constraint, discharge constraint, water balance constraint, flood control safety constraint and scheduling period end water level constraint; Step five, inputting the input flood hydrograph considering the prediction uncertainty into the flood control optimization scheduling model constructed in step four, and applying discrete differential dynamic programming algorithm to solve, outputting the optimal water level control process and optimal discharge process of the reservoir group; Step six, based on the optimal water level control process and optimal discharge process, the flood control adaptive scheduling rules of the reservoir group under the target SSP scenario and return period are analyzed and refined. 2.The SSP scenario reservoir group flood control scheduling method considering hydrological forecast uncertainty according to claim 1, characterized in that, The multiple SSP climate scenarios in step one include SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.

5. 3.The SSP scenario reservoir group flood control scheduling method considering hydrological forecast uncertainty according to claim 1, characterized in that, The candidate probability distribution models in step two include symmetric distribution, thick-tailed distribution and asymmetric distribution, wherein the symmetric distribution includes normal distribution and Logistic distribution, the thick-tailed distribution includes t distribution and Cauchy distribution, and the asymmetric distribution includes Johnson SU distribution and skew normal distribution. 4.The SSP flood control scheduling method for reservoir groups considering hydrological forecast uncertainty according to claim 1, characterized in that, The determination method of the 95% confidence interval range in step three is as follows: for each daily flow value of the benchmark design flood hydrograph, a large number of random error samples based on the optimal probability distribution model are generated by Monte Carlo simulation, and the possible value set of the daily flow is obtained by superposition, and the 2.5% quantile of the set is taken as the lower limit of the daily confidence interval, and the 97.5% quantile is taken as the upper limit of the daily confidence interval.

5. The SSP scenario-based reservoir group flood control scheduling method considering hydrological forecast uncertainty according to claim 1, characterized in that, The input flood hydrograph considering the prediction uncertainty in step five is the arithmetic mean of the daily flow values in the 95% confidence interval range.

6. The SSP scenario reservoir group flood control scheduling method considering hydrological forecast uncertainty according to claim 1, characterized in that, The flood control safety constraint in step five includes that the highest flood level of the reservoir does not exceed the check flood level, and the discharge does not exceed the safe discharge of the downstream river channel.

7. The SSP scenario reservoir group flood control operation method considering hydrological forecast uncertainty according to claim 5 is characterized in that: The discrete differential dynamic programming algorithm in the step five is solved by discretizing the reservoir state variable and decision variable, and using the Bellman optimality principle for reverse recursion and forward optimization, wherein the state variable is reservoir capacity or water level, and the decision variable is discharge or outflow. 8.The SSP scenario reservoir group flood control scheduling method considering hydrological forecast uncertainty according to claim 1, wherein, The flood control adaptive scheduling rule in the step six is obtained by analyzing the relationship between the optimal water level control process and the incoming flood process and the time period, and fitting the obtained reservoir scheduling water level control rule or gate opening rule. 9.The SSP scenario reservoir group flood control scheduling method considering hydrological forecast uncertainty according to claim 1, wherein, The input data of the SWAT model in the step two includes DEM, soil type data and land use data, wherein the resolution of the DEM is 30m, the soil type data adopts HWSD, and the land use data adopts CNLUCC 2015; the NSE and R2 of the model in the calibration period and the verification period are both higher than 0.

7.

10. The SSP scenario reservoir group flood control scheduling method considering hydrological forecast uncertainty according to claim 7, characterized in that, The time period division of the discrete differential dynamic programming algorithm in the step five is 360 time periods, corresponding to a 15-day typical flood period, the time period step is 1 hour, the state variable discrete grid number is 20, the control variable discrete grid number is 30, the hard constraint penalty coefficient is 1000, and the soft constraint penalty coefficient is 100.

Images