Professional Documents
Culture Documents
Flood Routing
Flood Routing
Abstract: Floods spreading through reservoir spillways are usually generalized into a first-order nonlinear ordinary differential equation
in engineering. With the classical RungeKuttaFehlberg RKF method, ignoring the local stability of the numerical solution, a Runge
Downloaded from ascelibrary.org by Technische Universiteit Delft on 10/17/17. Copyright ASCE. For personal use only; all rights reserved.
KuttaFehlbergReverse RKFR method is advanced and employed to route reservoir floods. In the improved method the step length will
further be rectified following the RKF using a local stability parameter, which is based on the fact that as the local truncation error is small
enough, the local stability might be approximately characterized by the relative error between the initial dependent variable associated
with every step and the feedback that may be calculated through a RungeKutta RK method. The method presented is first verified with
two numerical examples, and the results are more accurate and stable with comparable computed speed compared with the RKF method.
Subsequently, the RKFR is coupled with the reservoir flood routing program, and tested with a flood control decision support system of
the HuanrenHuilong reservoir group in Hunjiang river basin in Northeast China. The testing results demonstrate that it is more scientific
and reliable than the RKF and RK methods.
DOI: 10.1061/ASCE1084-0699200914:2197
CE Database subject headings: Reservoirs; Flood routing; Numerical analysis; China; Water quality.
1 qt = f VV 1b
Ph.D. Graduate Student, School of Civil and Hydraulic Engineering,
Dalian Univ. of Technology, Dalian 116024, China corresponding For a reservoir with gates, Eq. 1b would be discontinuous
author. E-mail: xgli828@yahoo.com.cn because of the fluctuation of opening or the amount of outflow
2
Professor, School of Civil and Hydraulic Engineering, Dalian Univ. equipment according to the demand of downstream protective
of Technology, Dalian 116024, China. targets. Since the discontinuous may be treated as the serially
3
Ph.D. Graduate Student, School of Civil and Hydraulic Engineering,
connected continuous, the following will mainly be concentrated
Dalian Univ. of Technology, Dalian 116024, China.
Note. Discussion open until July 1, 2009. Separate discussions must
in the continuous case of Eq. 1b: Substituting for q from
be submitted for individual papers. The manuscript for this technical note Eq. 1b into Eq. 1a yields
was submitted for review and possible publication on January 4, 2007;
approved on May 6, 2008. This technical note is part of the Journal of
Hydrologic Engineering, Vol. 14, No. 2, February 1, 2009. ASCE, dVt
= Qt f VV = f 0V,t 2
ISSN 1084-0699/2009/2-197202/$25.00. dt
ment systems, and consequently, dangerous situations often dete-
riorated Yan et al. 2007. Actually in the RK solution, the time t t
k2 = f 0 tn+1 ,Vn+1 k1 3c
step t selected empirically induces stability and accuracy to de- 2 2
cline for the reservoir whose inflow rating or volume-discharge
curve is markedly nonlinear, especially for a small volume reser-
voir with large floods such as downstream in a reservoir group.
k3 = f 0 tn+1
t
2
t
,Vn+1 k2
2
3d
8. Calculate Vn as the fourth-order feedback value V4n 1. Select the parameters: tmin, tmax, smin, smax, tmin, and
by introducing tn + tn, V5n+1 and tn into Eqs. 3a3e tmax upon the accurate request of flood routing, and gener-
instead of tn+1, Vn+1, and tn, respectively; ally tmax = tint;
9. Vn / Vn as the estimated value sn+1;
Define absV4n 2. Define m = 0 and the initial step length t0 = tmax;
10. Compare tn with tmin. If tn = tmin, then go to Step 12; 3. Define n = m, and calculate Vn+1 and tn+1 associated with
11. Compare sn+1 with smax. If sn+1 smax, then tn tn+1 until tn+1 tm + tint through the previous RKFR adaptive
= tn / 2, and go back to Step 2; procedure with the initial volume Vm and the initial step
12. Define tn + tn, V5n+1 and tn as tn+1, Vn+1 and tn+1, respec- length tm;
tively; 4. Compare tn+1 and tm + tint. If tn+1 = tm + tint, then define
13. Compare tn+1 with tmin and sn+1 with smin. If tn+1 Vm+1 = Vn+1; If tn+1 tm + tint, then calculate Vn+1 as Vm+1 via
tmin and sn+1 smin, then tn+1 = 2tn+1; and the fifth-order RKF method with the initial volume Vn and
14. Define n = n + 1, and go back to Step 2. t = tm + tint tn;
5. Define tm+1 = tn+1, tm+1 = tm + tint, m = m + 1 and go back to
Step 3 until the temporal hydrograph of Vm is properly ob-
Reservoir Flood Routing RKFR Adaptive Solution tained; and
Procedure 6. Determine qm from Vm through Eq. 1b.
To amend precision of the operative reservoir flood routing RK
solution with the rapid development of a reservoir group decision
support system, the RKFR method presented is inserted to replace Solution Validation
the RK method. Given the initial volume V0, the reservoir
discharge qm at time level tm can be calculated along with the Compared with the RKF, two different types of numerical ex-
following steps noticed that the time interval tint = tm tm1 amples are employed for confirming the performance of the im-
would in general be invariable for facilitating decision making: proved RKFR method. Subsequently, a case reservoir group is
introduced to test the reservoir flood routing RKFR adaptive nu- error and nothing of local stability. Seeking another parameter to
merical solution. cooperate with tn seems to be indispensable for improvement of
RKF.
RKFR Method Verification Referring to Table 1 again, sn 16,801.7 is extremely large
as tn is almost nonfunctional for rn. The reason should be that
The first example is concerned with the equation the errors V5n+1 and Vrn+1 induce a heavy increase of the errors
dV 2V V4n and Vrn. This shows that decreasing sn might be one
= t + 15/2 + 4a effective and efficient way of attenuating rn when tn is small
dt t+1 enough.
of the initial value condition See Table 2 for results from the proposed RKFR with addi-
tional parameters: smin = 0.01 and smin = 0.05 relative to the
V00 = 1 4b above-employed RKF. Compared with Table 1, rn in Table 2 is
and the corresponding analytical solution is about two orders less in magnitude but has approximately three
times as many steps 11/ 4. It is clear that the new method
V = t + 12 3 t + 13/2 +
2 1
3 4c greatly improves precision with a smaller temporal price, which is
what should be expected and also fits the theoretical basis: RKFR
which is a concave function and V will tend to infinity as t tends tests both local truncation error and local stability using respec-
to infinity. Table 1 shows part of the calculations with the RKF tive parameters at every step, while RKF only concerns the
method of parameters: tmin = 0.01, tmax = 0.05, tmin = 0.25, and former.
tmax = 16 in detail. Considering the second example
As Table 1 indicates, tn 0.00631 is much less than tmin
0.01 at the first step, yet rn 0.12804 is much larger than tn dV
0.00631, which implies that an attempt to decrease rn by re- = 2tv2 5a
dt
ducing tn unboundedly is sometimes fruitless. That illustrates
the flaw of RKF: tn represents a majority of the local truncation of the initial value condition
V00 = 1 5b and more accurate and stable for a one-order nonlinear ordinary
differential equation with an initial value compared with the RKF
the corresponding analytical solution is and RK.
1
V= 5c Reservoir Flood Routing RKFR Adaptive Solution
t2 + 1
Confirmation
which is a convex function and V tends to zero as t tends to
infinity. Table 3 details part of the calculations based on the RKF As shown in Fig. 3, Huanren and Huilong reservoirs are the first
method with the same parameters as Example 1. and second projects in the cascade development of Hunjiang river
Concisely, the two lists sn and rn of Table 3 are included in basin of Northeast China, respectively. The reservoir group is the
Fig. 1. prototype of the national flood control decision support system of
It is obvious that sn magnifies crests of rn in synchrony, China, and selected as the study area for testing different solu-
which would be responsible for the RKFR method. When local tions. Briefly, the reservoir flood routing RKF solution takes out
truncation error is regulated to satisfy certain precision demand, the part where local stability checks for the RKFR solution, while
rn is mainly derived from the state of local stability. Thus, sn the RK solution leaves both the local truncation error and the
is called a local stability evaluation parameter in the new RKFR local stability aside.
method as tn is treated as the local truncation error evaluation
parameter in the classical RKF.
Results of the RKFR method are displayed in Table 4 with the
additional parameters: smin = 0.01 and smax = 0.05.
Fig. 2 consists of the lists rn of Tables 3 and 4. rn
from Table 4 about RKFR is obviously much less than from
Table 3 about RKF with comparable computing steps 18/ 15.
That again demonstrates the necessity of a stability check in solv-
ing the initial value problem of a one-order nonlinear ordinary
differential equation, and the rationality of the definition sn+1
Vn / Vn of the local stability evaluation parameter
= abcV4n
in the RKFR method.
The above two examples from concave and convex perspec- Fig. 4. Relative error curves of Huilong reservoir discharge predicted
tives, respectively, demonstrate that the RKFR method is operable by different flood routing solutions on 20-year flood