ASSESSING AND OPTIMIZING THE OPERATION OF THE HOABINH RESERVOIR IN VIETNAM BY MULTI-OBJECTIVE OPTIMAL CONTROL TECHNIQUES doc

613.3 Hoabinh reservoir: Historical inow minus release blueline, water level light blue line, normal water level redline over the period 1995-2004.. 713.11 ddp results: release strategie

POLITECNICO DI MILANODipartimento di Elettronica e InformazioneRESEARCH DOCTORAL PROGRAM IN INFORMATION

TECHNOLOGY

ASSESSING AND OPTIMIZING THE OPERATION OF THEHOABINH RESERVOIR IN VIETNAM BY MULTI-OBJECTIVE

OPTIMAL CONTROL TECHNIQUES

Doctoral Dissertation of:

Xuan Quach

Advisor:

Prof Rodolfo Soncini-Sessa

Tutor:

Prof Carlo Piccardi

The Chair of the Doctoral Program:

Prof Carlo Fiorini

Co-supervisor:

Dr Andrea Castelletti

Dr Francesca Pianosi

2011 - XXIV

1 The Red River Basin and the Hoabinh reservoir 7

1.1 Physical and socio-economic system 7

1.2 Important issues of the system 8

1.2.1 Flood 9

1.2.2 Water shortages 11

1.2.3 Electricity 13

1.2.4 Navigation and water pollution 13

1.2.5 River bed erosion 14

1.2.6 Climate change 14

1.2.7 Transboundary issues 15

1.3 Institutional and normative framework for Hoabinh oper-ation 16

1.4 Objectives of the study 19

1.5 Conceptualization: actions, criteria and design indicators 21 2 Modelling the Hoabinh water system 27 2.1 The Time Step 27

2.2 Upstream catchments 29

2.2.1 Data analysis 29

2.3 The Hoabinh reservoir 32

2.3.1 Characteristic curves of Hoabinh reservoir 34

2.3.2 Rating curve of Hoabinh reservoir 36

2.3.3 Mass balance equation of the reservoir 39

2.3.4 Minimum instantaneous discharge function 42

2.3.5 Maximum instantaneous discharge function 43

2.3.6 Validation of the reservoir model 44

2.3.7 Time constant of the reservoir 46

2.3.8 Model of the hydropower plant 48

2.4 The downstream river network 51

2.4.1 Background 52

2.4.2 The data 52

2.4.3 The model structure 53

2.4.4 Model identication 55

3 Improvement potential to Hoabinh reservoir operation 59 3.1 Historical operation of the Hoabinh reservoir 59

3.1.1 Water supply 61

3.1.2 Flood control 61

3.1.3 Hydropower production 61

3.2 The ideal operation of the Hoabinh reservoir 65

3.2.1 Solution of DDP problem 67

3.2.2 Discretization 67

3.2.3 Results and discussion 68

3.3 Summary of the main results 75

4 Optimization by Stochastic Dynamic Programming 77 4.1 Review of application of SDP to reservoir operation 77

4.2 Formulation of the SDP problem 78

4.3 The catchment model 81

4.4 SDP results under `white inow' assumption 83

4.5 SDP results under `not-white inow' assumption 87

4.6 Summary of the main results 90

5 Optimization by Policy Search 93 5.1 Implicit Stochastic Optimization (ISO) 93

5.1.1 Discussion of ISO results and its limits 95

5.2 Evolutionary Multi-objective optimization (EMO) 101

5.2.1 Non-dominated Sorting Genetic Algorithm (NSGA-II) 103

5.2.2 The initialization procedure 104

5.2.3 Measure of Pareto frontier quality: the Hypervol-ume Indicator (HD) 104

5.2.4 Application results and discussion 105

5.3 EMO with hydrometeorological information 108

5.4 Summary of the main results 113

6 Comparison and discussion 115 6.1 Comparison of the reservoir optimization methods 115

6.2 Signicance of the results for the Hoabinh management 118

6.3 Further research 120

4

List of Figures

1.1 The Red River Basin 81.2 Energy production of Hoabinh reservoir from 1989 to 2008 131.3 Percentage of monthly ow rates from China sub-catchment

at Laichau (black), Tabu (gray) and Hoabinh (light gray)

on the Da River 161.4 Historical water level of Hoabinh reservoir (1995-2004) 211.5 Yearly pattern of water demand at Sontay 242.1 The scheme of the Hoabinh water system and of its model 272.2 Minimum annual ows of the Da River at Hoabinh dam(blue), of the Thao River at Yenbai (green), and of the LoRiver at Vuquang (red) 322.3 Mean annual ows of the Da River at Hoabinh dam (blue),

of the Thao River at Yenbai (green), and of the Lo River

at Vuquang (red) 332.4 Maximum annual ows of the Da River at Hoabinh dam(blue), of the Thao River at Yenbai (green), and of the LoRiver at Vuquang (red) 332.5 The dam of the Hoabinh reservoir 342.6 Water level vs storage (a); measured (solid) and esti-mated (cross) water level of Hoabinh reservoir (b) 352.7 Hoabinh reservoir: surface area vs storage 362.8 Downstream water level of Hoabinh reservoir: observed(blue) and estimated by piecewise linear (green) and bypolynomial of order 3 (red) over the period 1995-2004 372.9 Release through penstocks vs hydraulic head (a); Releasethrough penstocks vs storage (b) 372.10 Hoabinh reservoir: rating curve of bottom gates 382.11 Hoabinh reservoir: spillway rating curve 392.12 Hoabinh reservoir: daily unitary surface evaporation overone year 402.13 Hoabinh reservoir: instantaneous maximum and minimumreleases 442.14 Validation scheme of the Hoabinh reservoir model 45

List of Figures

2.15 Observed (blue) and simulated (green) release; observed(purple), simulated (black) and dead (red) storage overthe period 1995-2004 452.16 Linearizing the storage-discharge relation in correspon-dence with various storage values ˇs Note the sharp in-crease in the slope when ˇs is greater than the maximumoperative storage ¯s 472.17 Validation scheme of the Hoabinh hydropower plant model 492.18 Measured (solid line) and estimated (dotted) releases throughthe turbines at the beginning (a) and at the end (b) of theperiod 1995-2004 502.19 Observed vs estimated downstream water level of theHoabinh dam over the period 1995-2004 502.20 Historical annual mean of step energy production (thickline) and corresponding rate of release through the tur-bines (thin line) over the period 1995-2004 512.21 Cross-correlation at dierent lag values between observedtime series at dierent stations in the RRB basin over theperiod (1989-2004) 542.22 Scatter plot of measured and modeled water level at Hanoi(a), and ow at Sontay (b) 573.1 Yearly pattern of the Hoa Binh level with historical oper-ation over the horizon 1995-2004 603.2 Historical inow (dotted line) and release (solid line) ofHoabinh reservoir over year 1996 613.3 Hoabinh reservoir: Historical inow minus release (blueline), water level (light blue line), normal water level (redline) over the period 1995-2004 623.4 Flow at Sontay over the period 1995-2004: historical value(vertical axis) and hypothetical value (horizontal axis)without regulation of Hoabinh reservoir 633.5 Total water decit (a) and mean decit step cost (b) ineach year over the period 1995-2004 633.6 Water level at Hanoi in 1996: historical value (solid line)and hypothetical value without the Hoabinh reservoir (dot-ted line) 643.7 Maximum water level in Hanoi (a), and mean ood stepcost (b) in each year over the period 1995-2004 643.8 Total hydropower production (a) and mean of minus hy-dropower step cost (b) in each year over the period 1995-

2004 656

List of Figures

3.9 Performances of some ddp-policies with dierent tion grids of control: U14 (cyan), U20 (green) and U41(blue) that dominate the historical one (red cross) overthe evaluation horizon 1995-2004 (a); zoom the box in theleft panel (b); The diameter of the circles is proportional

discretiza-to the hydropower production 683.10 The Hoabinh water level produced by ddp-25 over theevaluation horizon 1995-2004 713.11 ddp results: release strategies of Hoabinh reservoir forreducing water decit ddp-2 (a,b), for avoiding ood atHanoi ddp-3 (c,d), and for compromising among the twoddp-4 (e,f) in 1998 and 2002 (release (blue), inow (red),water level (light blue), and normal water level (dottedred)) 723.12 Water level at Hanoi when the Hoabinh release is perma-nently equal to zero (horizontal) and produced by ddp-4policy (vertical) 733.13 Result of ddp-4 Top panel: water level (blue) and deadwater level (dotted red); middle panel: release of Hoabinhreservoir (blue), inow to Hoabinh (red), ow of Thao(pink), and ow of Lo (yellow); bottom panel: water level

at Hanoi (blue), water level at Hanoi without Hoabinh lease (black), and ooding threshold (dotted red) in Au-gust, 2002 743.14 Hoabinh water level in 2004 under dierent policies (ddp-

re-1 (blue), ddp-2 (green), ddp-3 (red), and ddp-25 (lightblue)) when the optimization horizon ends in 2005 (a) and

in 2004 (b) 75

4.1 Autocorrelation of Da (a), Thao (c) and Lo (e) river ows;spatial correlation of river ows between Da-Thao (b), Da-

Lo (d) and Lo-Thao (f) 824.2 The periodic time patterns of µ (a) and σ (b) of the model(4.7a,b) for Da catchment, together with the samples fromwhich they were estimated (calibration period) 834.3 Flow distribution of Da (a), Thao (c) and Lo (e); auto-correlogram of yt (b), ηY B

t (d) and ηV Q

t (f) 844.4 Water levels at Hanoi with historical (red) and ddp-4(blue) and sdp-ar0-4 (green) policies 86

List of Figures

4.5 The eects induced by ddp-4 (blue) and sdp-ar0-4 (green)during the ood event of 1996: water levels in Hoabinhreservoir (top panel); Hoabinh releases and Da (red), Thao(violet) and Lo (yellow) inows (middle panel); water lev-els at Hanoi and the ood threshold at 9.5 m (red dot-ted)(bottom panel) 874.6 The inow (red line); the release (continuous lines) andstorage (dotted lines) produced by the ddp-4 (blue) andsdp-ar0-4 (green) 884.7 Event of January-February 1999: the eects induced bythe ddp-26 (blue) and sdp-ar0-26 (green) policies: waterlevels in Hoabinh reservoir (top panel); Hoabinh releasesand Da (red), Thao (violet) and Lo (yellow) inows (mid-dle panel); supply decits (bottom panel) 894.8 Auto-correlation of ξt+1 in equation (4.8) 904.9 The eects induced by ddp-4 (blue), sdp-ar0-4 (green),and sdp-ar1-4 (light blue) during the ood event of 1996:water level and the maximum level (red dotted) in Hoabinhreservoir (top panel); Hoabinh releases and Da (red), Thao(violet) and Lo (yellow) inows (middle panel); water lev-els at Hanoi and the ood threshold (red dotted) at 9.5 m(bottom panel) 914.10 Performances of ddp policies (blue), sdp-ar0 (brown),sdp-ar1 (cyan) policies that dominate the historical one(red cross) The diameter of the circles is proportional tothe hydropower production 925.1 Procedure of Implicit Stochastic Optimization (iso) 945.2 Relationship: between time (day of the year), inow andrelease decision (a) by ddp-15; between time, inow andinterpolated release decision (b) by iso-a (blue), and iso-

b (green) (calibration dataset) 965.3 Trajectories of ddp-15 (blue), iso-a (green), and iso-b(red), inow (light blue) over the evaluation horizon 1995-

2004 975.4 Release decision of ddp (blue), water demand at Sontay(green), and interpolated release decision of iso-c (red)over the evaluation horizon 1995-2004 988

List of Figures

5.5 Release decision by ddp-15 (blue), iso-c `interpolated'(red) and iso-c `simulated' (green) over the calibrationhorizon 1958-1977 (a) A blow up of the box in the leftpanel (b): the red and light blue dotted lines are the stor-age trajectories by ddp-15 and iso-c 995.6 Interpolated release decision of iso-c with dierent values

of storage while other inputs assume their historical values(t = 6500) 995.7 Release decision by ddp-15 (blue), iso-c `interpolated'(red) and iso-c `simulated' (green) over the calibrationhorizon 1958-1977 (a) A blow up of the box in the leftpanel (b): the red and light blue dotted lines are the stor-age trajectories by ddp-15 and iso-c+ 1005.8 Procedure of Evolutionary Multi-Objective algorithm (emo)1025.9 The hypervolume is the surface area of the grey regioncomprised between the reference point H and the points

in the approximate Pareto frontier (black) divided by thesurface area of the rectangle with vertices in H and U(dashed lines) 1055.10 Performances of ddp policies (blue), emo policies of Ta-ble 5.3 (pink), and the historical one (red cross) Thediameter of the circles is proportional to the hydropowerproduction 1075.11 Performance of historical (blue), ddp-25 (green), and emo-

6 (brown) policy over the evaluation horizon 1995-2004 1095.12 Water level at Hanoi produced by: historical (red), ddp-

25 (blue), and emo-6 (pink) policy in July and August ofsome years in the period 1995-2004 The red dotted line

is the ood threshold of 9.5 m 1105.13 Procedure of Input Variable Selection ivs 1115.14 Location of hydrological and meteorological stations of theRed River Basin 1126.1 Performances (in terms of design indicators) of several op-erating policies designed by dierent optimization meth-ods The diameter of the circles is proportional to thehydropower production 1166.2 Performances of several operating policies designed by dif-ferent optimization methods in terms of physical indica-tors The diameter of the circles is proportional to thehydropower production (evaluation horizon 1995-2004) 119

List of Tables

1.1 Historical ood events at some stations in the Red RiverBasin 111.2 Water level vs discharge at Sontay cross-section over 3decades 142.1 The main variables of the model 282.2 Available data in Red River system 312.3 General eciency of hydropower generation of the Hoabinhhydropower plant 492.4 Annual energy production of history and by model (TWh/y) 492.5 Performance of the downstream river network models overthe period 1995-2004 573.1 Results of ddp: average value of the step-costs (hyd: hy-dropower production; def: water decit; o: ood control)over the evaluation horizon (1995-2004) with dierent dis-cretization grids of the control (`na' means not available) 694.1 Performances of sdp-ar0 policies over the validation hori-zon 1995-2004, compared with the performances of thecorresponding ddp policies 854.2 Performances of sdp-ar1 policies, compared with the per-formances of the corresponding sdp-ar0 policies 915.1 Results of iso 965.2 Quality of Pareto frontiers (hypervolume value HD) ob-tained by emo under dierent input vector and number

ν of neurons in the decision rule, population size P andnumber of generations G 1055.3 Results of emo: average value of step costs over the val-idation horizon 1995-2004 of policies found by emo with

It= |t, st, at, qtY B, qtV Q|and ν=7 107

List of Tables

5.4 Quality of Pareto frontiers (hypervolume value HD)

ob-tained by emo with exogenous information under

dier-ent input vector and number ν of neurons in the decision

rule, population size P and number of generations G 112

5.5 Result of emo with exogenous information: average values

of step costs over the validation horizon 1995-2004 of

poli-cies found by EMO with It= |t, s, at, qY B

6.1 Performances of dierent operating policies (and Utopia

point of ddp) in terms of design indicators over the

eval-uation period 1995-2004 116

6.2 Performances of dierent operating policies (and Utopia

point of ddp) in terms of physical indicators over the

eval-uation period 1995-2004 119

12

List of TablesAcknowledgments

I am thankful to my supervisor Rodolfo Soncini-Sessa for the tunity he gave me and for his continuous teaching and consulting duringfulllment of this thesis

oppor-I am greatly indebted to my co-supervisors Andrea Catelletti, EnricoWeber, and Francesca Pianosi, who taught and assisted me during thewhole research period in both dealing with informatics problems and de-veloping optimization schemes

I greatly appreciate my tutor Prof Carlo Piccardi, who took care of

my study plan

I gratefully acknowledge other members of the research group on ning and management of water systems at Dei, especially Simone Bizzi,Stefano Galelli, and Daniele Anghileri

plan-My gratitude goes to the Institute of Water Resources Planning where

I used to work and to all my colleagues from the Institute, in particular,Nam Le, Manh Nguyen, Phung Nguyen, Van Anh Truong, and Nam Vufor their help in collecting data

My largest acknowledgment is for my parents, my husband, and mydaughter, who provide me with strength and motivation to complete thisthesis and to face any other challenge in my life

The purpose of this study was to investigate the operation of the Hoabinhreservoir in the Red River Basin of Vietnam, and assess the room for itsimprovement by application of system analysis and optimal control tech-niques The study aimed at establishing a foundation for further research

on inter-reservoir regulation of the Basin Finally, this study provided

a testing ground for developing and comparing dierent reservoir mization methods

opti-Chapter 1 and 2 provide a general and mathematical description of thesocio-economic and physical system of the Red River Basin, including thethree main objectives of hydropower production, ood control, and wa-ter supply Conceptual and data-driven modeling tools were used to thispurpose In Chapter 3, the potential of the current infrastructure wasassessed by application of Deterministic Dynamic Programming, whichprovides the upper bound of system performance under the ideal assump-tion of perfect information about future inows to the system Resultsproved that, in fact, there exists a large room for improvement of thehistorical regulation, although the existing storage capacity is not fullyadequate for ood control In Chapter 4 and 5 several dierent optimalcontrol methods were applied to design new operating policies, includingthe standard Stochastic Dynamic Programming and Implicit StochasticOptimization, and a novel method that combines Evolutionary Multi-Objective optimization and Articial Neural Networks Results of thiswork showed that there are a lot of operating policies that prove Pareto-dominant over the historical one, that is, they can improve all three ob-jectives simultaneously However, while the improvement is rather signif-icant with respect to hydropower production and water supply, it is muchmore limited in terms of ood control To overcome this limitation, thevalue of hydrometeorological information for anticipatory managementwas explored by a novel Input Variable Selection technique combinedwith Evolutionary Multi-Objective optimization The last Chapter ofthe thesis compares the results obtained by the dierent optimizationtechniques in terms of their eectiveness and signicance to the Hoabinhwater management It also discusses the limits of the study and topicsfor future research

In many fast-developing countries water is a key renewable resource tocomplement carbon-emitting energy production and support food secu-rity in the face of demand pressure from fast-growing industrial pro-duction and urbanization In this context, water resources managementneeds proper strategies to eciently balance their multiple purposes.Starting in the late Eighties, Vietnam has undertaken a comprehensivereform (Doi Moi) of liberalization of economic production and exchange,which has been the key driver of its explosive economic and demographicdevelopment in the last two decades [Toan et al., 2010] The rapid growthresulted in an increased energy demand, which has been growing at anannual rate of nearly 15% in the last ten years; but also boosted inter-nal migration from rural areas to the main cities, which are sprawlinguncontrolled [Hoang et al., 2010] Water resources play a central role inthis development: hydropower is the primary renewable energy resource

in the country and, despite the considerably increasing importance of theindustrial and service sectors, irrigated agriculture is still the main eco-nomic drive [Nguyen et al., 2002] and a primary source to ensure foodsecurity in the face of demand pressure Unfortunately, water is alsoresponsible for most of the worst natural disasters occurred in the coun-try in recent years Severe oods are plaguing Hanoi every year duringthe heavy rain monsoon season with increasing damage in the unusuallyoverdeveloped river urban area

To cope with this heterogeneous and fast-evolving context, water sources development and management needs to be reconsidered to im-prove resilience of economy, society and environment in the entire Viet-nam Increased water storage at the river basin level is certainly a majorcomponent of vulnerability reduction strategies, however the optimal re-operation of the available storing capacity is an economically interestingand potentially eective alternative, or simply complementary option, toinfrastructure development

re-The Hoabinh reservoir in the Red River Basin, Vietnam is a typicalcase Mitigation of ood damage is one of the highest priorities of thereservoir operation, but hydropower production is also very signicant aswell as water supply to civil, industrial and agricultural uses Due to themulti-purpose character of the Hoabinh reservoir, conicts and disputes

List of Tables

about its operation have been ongoing since its construction Researches

on nding solutions to improve the Hoabinh operation and mitigate theconicts are strongly encouraged

In the literature, we found only two works on the operation of theHoaBinh Ngo et al [2008] use traditional scenarios analysis to compar-atively assess three alternative operating policies on ood control andhydropower production focusing on the ood season only Built on theseresults, Ngo et al [2007] explore the reservoir re-operation by param-eterization and subsequent optimization of the operating rules throughthe Shued Complex Evolution algorithm Their study revealed thatthere exists room for improvement of the current management, at least

in terms of hydropower production and ood mitigation However, otherproblems such as water shortages and environmental pollution need to

be considered, and the analysis should be extended to cover the entireyear not only the ood season

The scope of this research is twofold First is to better understand thefunctioning of the Hoabinh water system, its structural potential, theneed for capacity expansion and the space for improving its operation.With respect to the state of the art on the Hoabinh water system, thisthesis takes a step forward by: (i) enlarging the tradeo analysis to thewater supply sector; (ii) enlarging the optimization horizon to the entireyear thus allowing for inter seasonal water transfer; (iii) exploiting moredata availability to introduce a clear distinction between the dataset usedfor optimization and the one used for validation of the optimized poli-cies, which allows for a fair and statistically sound comparison with thehistorical operation (iv) exploiting the value of exogenous information(e.g meteorological data) to design anticipatory operating policies Theknowledge base built in this research will constitute an important contri-bution for future research on the inter-reservoir management includingother main reservoirs under construction in the Basin The researchuses system analysis and optimal control techniques, including novel op-timization methods The research thus has also the theoretical purpose

of developing and testing novel optimal control techniques by application

to the complex Hoabinh case study

6

1 The Red River Basin and the Hoabinh reservoir

This Chapter presents a brief introduction to the Red River Basin andthe Hoabinh reservoir, discussing the main natural and socio-economicconditions, the specic objectives of the study and the indicators thatwill be used to evaluate and design new operating policies of the Hoabinhreservoir in the following Chapters

1.1 Physical and socio-economic system

The Red River Basin is the second largest of Vietnam It is locatedbetween 20o00N and 25o30N, and 100o00E and 107o10E The totalarea of the basin is approximately 169,000 km2, of which 81,240 km2

(48%) in China's terriory, 86,600 km2 (51.35%) in Vietnam, and the rest

in Laos Administratively, the Red River basin covers 26 provinces andcities in the Northern region of Vietnam, with a total population of about

30 million in 2009

The Red River, the main River downstream from Viettri, is conuence

of three upstream tributaries: Da, Thao and Lo River (Fig 1.1) Allthese tributaries originate from China Even though the catchment areas

of the Da and Thao River basins are almost the same, the Da Rivercontributes 42%, while the Thao River contributes only 19% of total

ow to the Red River The Lo River basin is the smallest one, but itscontribution is 25.4%

The whole basin is characterized by two distinguished seasons: rainyseason from May to October and dry season from November to April

of the following year Annual rainfall varies from from 1,200 to 4,800mm/year in Vietnam part, and about 80% of rainfall occurs in the rainyseason The ood (high ow) season is from June to October, and low

ow season is longer, from November to the next May Because of unevenrainfall, ows through the basin are unevenly distributed in time, causing

oods and water-logging in the rainy season and water shortages in thedry season

Among water sectors, irrigation is the biggest user, accounting for90% of total used water The agricultural land occupies approximately

1 The Red River Basin and the Hoabinh reservoir

Figure 1.1: The Red River Basin

1,874,100 ha and forestry land occupies 2,570,775 ha Potential areafor future agriculture and forestry development is estimated at about3,919,500 ha, little bit decrease due to the urbanization and industrial-ization but the number is still high Other water sectors use less amountbut highly depend upon the surface water resources of the basin.Several reservoirs have been built and operated since 1970s Thacbareservoir (Fig 1.1) is located on Chay River, a tributary of the Lo,starting its regulation in 1971 The main objective of this reservoir ishydropower generation However, it increases the ow in the dry seasonfrom about 100 to 200 m3/s The Tuyenquang reservoir (Fig 1.1) onthe Gam River (belonged to the Lo) and the SonLa reservoir (Fig 1.1),upstream of Hoabinh reservoir on the Da River, are under construction

Up to now, Hoabinh reservoir is the biggest one of Vietnam So far ithas been playing an important role in preventing and controlling ood,generating hydro electricity, and supplying water to irrigation, domesticuse, industry, and other water users of the Red River Basin However,due to rapid growth of population, quick development of the economy,and climate change, the basin has been facing many problems such assevere oods, water shortages, water pollution, and etc These problemswill be described in detail in the next Section

1.2 Important issues of the system

Presenting the important issues of the rrb does not mean that this thesiswill propose solutions to all the problems, but by listing them here the8

1.2 Important issues of the system

authors hope that it is helpful to clearly understand the basin situation,and establish an excellent background for further works Let's start withthe rst issue of ood

1.2.1 Flood

In recent years, big oods have frequently happened in Vietnam in eral and in the Red River Basin in particular Some historical events are

gen-in deep memories of people

In 1971, from August 12nd to 21st, due to combined eects of thetropical converged range, Western low grooves combined with Pacichigh pressure, very heavy rain happened in the whole basin In thisperiod, average rainfall was 255 mm in the Red River Basin and was 200

mm in the Red River Delta On the Da River, at Hoabinh station, the

ow was 14,800 m3/s On the Thao River, the ow at Yenbai was 10,530

m3/s, and on the Lo River at Vuquang the ow reached 14,000 m3/s.The ood of the Lo River was the biggest one in history

The ood of the Red River in 1971 was the combination of the biggest

ood of the Lo River, the second biggest ood of the Thao River, and thefourth biggest ood of the Da River These caused the historical ood

in the Red River Basin, the biggest one since 1902, with the maximum

ow at Sontay of 37,800 m3/s The ood peak in Hanoi was 14.13 m(equivalent to 14.8 m if converted1), 1.63 m and 2.63 m above the thirdand the second alarming levels, respectively The water level above thethird alarming lasted in eight consecutive days As a result, dike spillingand dike breaking happened in several places On August 19th, dikesbroke in the downstream of the Lo and Da River, in the Northern Hanoi

to the left bank, and at the Southern Hanoi to the right bank of theRed River At 4 o'clock on August 20th, Vancoc dam was opened todivert ood into the Day River (Fig 1.1) At the same time, the dike

of the Thao River was broken At one o'clock the dike of the Da Riverwas broken At 17 o'clock on 22nd, the ood detention area at the righthand side of the Thao River was activated In the very following day,dikes breaks happened every where, one after the other, causing a deepinundation in the entire delta

According to the price of 1971, total property damage to the Statesectors under management of the Central Government was more thanVND 44 billion In addition, 100000 dead, other losses to local people,and ood eects such as epidemics and interrupted production could not

be counted out

1 `converted' here means that if dikes had not broken, the water level could have reached 14.8 m

1 The Red River Basin and the Hoabinh reservoir

Another bad event was the ood happened in 1996 when Hoabinhreservoir have been operated for several years From 9thto 20th August,

1996, it kept raining in the Red River Basin During this period, thebasin average rainfall was 432 mm; that gure of the Da, Thao and

Lo river basin was 380 mm, 317 mm, and 349 mm, respectively Therainfall in the Northern delta was about 300 mm In some rain centers2,the gures were from 300 to 400 mm The ow at Hoabinh station onthe Da River was measured at 21,000 m3/s (equivalent to 22,700 m3/s

if converted) These gures at Vuquang on the Lo River and at Yenbai

on the Thao River were 3,920 m3/s and 6,420 m3/s, respectively As

a result, the third biggest ood occurred in the Red River, which wasthe combination of the biggest ood on the Da River and relatively big

oods on the other two The ood peak measured at HaNoi was 12.43

m (equivalent to 13.30 m if converted) at 21 o'clock 19th, a 0.93 m abovethe third alarming level, and lasted for 6 days The peak ow at Hanoiwas 14,800 m3/s and at Sontay was 27,400 m3/s In the mean time,unfortunately, the Typhoon No.4 landed in the south of Red River Delta,causing sea level to rise from 1 to 2 m This rising hampered the RedRiver drainage ood to the sea As a consequence, the high water levelslasted for many days, seriously threatening the entire dike system in theNorthern delta Most of the local dikes were spilled or broken Therewere four people died, 61 missed and 161 injured; houses, schools, clinicsand hospitals were collapsed; rice and sub-crops were ooded; irrigation,transport, and energy works were damaged

In general, the Red River oods are combinations of oods from threeupstream tributaries: Da, Thao and Lo River If they are formed by one

of three tributaries, their characteristics are somehow similar to oods

of the mountainous area They have several peaks, high uctuation, fastrising and fast decreasing If they are combination of more than onetributary, often the case, they are fast rising, big and long duration.Flood of the Da River is the main source causing big oods of the RedRiver On average, it contributes 49% of 8 day ood water (max is 69%)

at Sontay If only the ood peak is considered, there are 69% of oodpeaks of the Da River constituting the ood peaks of the Red River.The Lo River ood is the second biggest source of the Red River

ood One interesting note is that oods of the Lo and Da Rivers oftencoincided, and there are about 34% of chance to form the big ood ofthe Red River

There are some reasons that make the downstream ood of the Red

2 some places are called rain centers because of their so high rainfall compared to the rest

10

1.2 Important issues of the system

Table 1.1: Historical ood events at some stations in the Red River

of simultaneous occurrence was rather high (72%) Third, the weatherhas been changed in a very disadvantage way that caused so heavy,intensive and long-duration rain in the ood season Fourth, by nature,the ood water level in the river is often 5 to 7 meters higher thanthe elevation of the delta, and the elevation of the Delta is lower thanthe msl, specically, 58.4% area of the Red River Delta is 2.0 m belowthe msl Many coastal provinces have more than 80% of their landthat is 2.0 m below the msl [Nguyen, 2010a] As a result, once dykebreaks or heavy rain happens inside the delta, it is dicult to drain thiswater Other possible man-made reasons maybe deforestation, and themismanagement of the reservoir operation

1.2.2 Water shortages

In the Red River Basin, water shortages often happened in the dry season(from November to the next May), and they have become more serious.The minimum water level in the Red River at Hanoi station has beengetting lower and lower It was 2 m, 1.94 m, and 1.46 m in 1999, 2004 and

2006, respectively Especially, it reached 0.66 m in the end of March, and0.4 m in the beginning of April in 2010, the latter being the historical

1 The Red River Basin and the Hoabinh reservoir

event in the last 100 years3 Minimum discharge at Sontay in March

2002 was only 380 m3/s, while total water demand this time was about

500 m3/s, leading to a lack of more than 10 million m3/day

The rst reason, like the case of ood, by nature, is the uneven tribution of the rainfall Total rainfall in the dry season (7 months)accounts for only 24% The lowest ow occurs in February or Marchwhen the irrigation demands are tentional at most The second reason

dis-is due to the abnormal of the weather, specically the LaNi˜na menon which has happened every four or ve years The third reason

phenon-is erosion of the river since Hoabinh was put in operation Forest exploited and chopped down is the man-made reason that reduces theground water table These both lower the water level along the RedRiver creating diculties in withdrawing water to the Delta

over-The last but not least is the low inow from China According to theNational Center for meteorology and hydrology, the inows from China

in all three tributaries are at the lowest levels in history Minimum ows

in November 2009 and March 2010 on the Da River at Laichau are 23

m3/s and 56 m3/s, respectively, while the corresponding lowest levels

in the past were 332 m3/s and 103 m3/s These gures on the ThaoRiver at Laocai were 141 m3/s and 130 m3/s in comparison with thelowest ones of 161 m3/s and 97 m3/s in the past The inows to allreservoirs in the basin also assumed minimum values in history This

is the main reason leading to the lower water levels compared to thedesigned ones of some big reservoirs In 2009, maximum water level ofHoabinh, TuyenQuang and ThacBa were 116.44 m, 107.6 m and 53.4

m, lower than designed levels 0.56 m, 12.4 m, and 4.5 m, respectively.Especially, in 2010, up to March 31st, water levels at almost reservoirswere lower in comparison with the same period of previous year, forinstance the dierences were 1.28 m, 3.93 m, and 12.98 m at Hoabinh,Thacba and Tuyenquang, respectively4 As a consequence, releases todownstream were lower than usual Clearly, if all the reservoirs werefull before the dry season, the downstream release could be signicantlyimproved

3 information from 01110200.html by News Investment and Development Company, last visited 30 th

http://www.tinmoi.vn/Giua-long-thu-do-bo-tung-tang-gam-co-April, 2011

4 information from http://www.boxitvn.net/bai/2802 by Vietnam Bauxit, last ited: 30 th April, 2011

vis-12

1.2 Important issues of the system

Figure 1.2: Energy production of Hoabinh reservoir from 1989 to 2008

1.2.3 Electricity

Hydropower has been the main energy source of Vietnam; it contributedfrom 35 to 40% of total energy consumed Hoabinh hydropower planthas been the biggest one that annually supplied about 8 billion kwhsince 1994, accounting for 35% energy of the northern part, and 15% ofthe country In 20 years of operation, the yearly energy production hasincreased from 1.3 TWh in 1989 to 10.136 billion kWh in 2008 (Fig 1.2)

It was low from 1989 to 1994 because the hydoropower plan was notcompletely constructed, not fully operated

The electric demand of Vietnam has been increased about 15 to 20

% per year Vietnam has started to import electricity from China since

2005 In 2011, it is estimated that the electricity decit is about 4.67TWh, and it is also forecasted that hydropower electricity continuescontributing about 33% of total energy production of Vietnam [Toan

et al., 2010] Therefore Hoabinh hydropower plant will keep playing themost important role in supplying hydropower to the electric system of thecountry Thus, balancing amongst objectives will be a sharp challengefacing the decision maker in optimising reservoir operation

1.2.4 Navigation and water pollution

Downstream ow has been decreased leading to the interrupted tion, and pollution of water sources in the basin The water level requiredfor local navigation on the Red River is 2.15 m at Hanoi station Waterlevel data from 1956 to 2004 shows that on average the annual number

naviga-1 The Red River Basin and the Hoabinh reservoir

Table 1.2: Water level vs discharge at Sontay cross-section over 3

Water pollution in the river network of the basin is a parallel nomenon with the interruption of transportation It is caused by thedeclined water level of the Red River Less water diversion to NhueRiver (see Fig 1.1) has led to environmental damages in Nhue RiverBasin because of lack of water for dilution In November, 2003, sheswere mass-died in downstream of Nhue-Day River, brought about disap-pearance of aquaculture and shing villages here

phe-1.2.5 River bed erosion

Before 1989 the cross-section of river downstream was getting smaller due

to sedimentation With the same discharge, the water level was increasedfrom 20 to 30 cm in comparision with the previous decade However,since 1990, the sediment has stayed in the reservoir; downstream riverbed has been eroding The water level in the downstream has been lower

in the dry season even with higher release regulated by Hoabinh Withthe same discharge of 1000 m3/s, water level at Sontay has decreased0.44 m from 5.69 m to 5.25 m (Table 1.2)

1.2.6 Climate change

Observed warming over several decades has been linked to changes inthe large-scale hydrological cycle such as: increasing atmospheric watervapour content; changing precipitation patterns, intensity and extremes;reduced snow cover and widespread melting of ice; and changes in soilmoisture and runo Precipitation changes show substantial spatial andinter-decadal variability Over the 20th century, precipitation has mostlyincreased over land in high Northern latitudes, while decreases have dom-inated from 10°S to 30°N since the 1970s The frequency of heavy pre-cipitation events (or proportion of total rainfall from heavy falls) has14

1.2 Important issues of the system

increased over most areas Globally, the area of land classied as verydry has more than doubled since the 1970s There have been signicantdecreases in water storage in mountain glaciers and Northern Hemispheresnow cover Shifts in the amplitude and timing of runo in glacier- andsnowmelt-fed rivers, and in ice-related phenomena in rivers and lakes,have been observed (high condence)[Wu et al., 2008]

Increased precipitation intensity and variability are projected to crease the risks of ooding and drought in many areas The frequency

in-of heavy precipitation events (or proportion in-of total rainfall from heavyfalls) will be very likely to increase over most areas during the 21st cen-tury, with consequences for the risk of rain-generated oods At thesame time, the proportion of land surface in extreme drought at anyone time is projected to increase, in addition to a tendency for drying

in continental interiors during summer, especially in the sub-tropics, lowand mid-latitudes

Instability in rainfall would cause more severe oods in rainy seasonand droughts in dry season Increase in frequency and intensity of ty-phoons, storms would cause high oods and inundation, ash oods,landslide and erosion Increasing water shortage and growing water de-mand threaten water supply, water use conicts, less energy generationdue to drought Unstable ow regimes may lead to conict in water reg-ulation at hydropower stations, more energy consumption due to hightemperature and humidity [Wu et al., 2008]

1.2.7 Transboundary issues

All tributaries of the Red River Basin are international ones shared byChina, Laos and Vietnam Detailed information of ows from Chineseparts are inaccessible, except for some general information listed here

In the Chinese part of the Red River Basin, up to 2005, there havebeen about 8 medium-scale reservoirs with total storage of 198 millioncubic meters, and 698 small-scale reservoirs with total storage of 347million cubic meters covering for 4,933 ha, and 27,886 ha, respectively

In addition, approximate 1,346 irrigation pumps, 1,868 water intercepts,and 1,573 temporary structures have been responsible for total 89,132

ha irrigated land There exists the LucThuyHa hydropower with aninstalled capacity of 57,500 kw and other 843 small hydropower plantswith total capacity of 99,400 kw Total water supplied by the irrigationstructures at frequency of 75% (i.e there are 75 out of 100 years in whichwater demands are met) is about 1.3 billion m3, accounting for 3% totalwater of basin (total water amout of the whole basin at frequency of 75%

is 40.6 billion m3 per year)[Le et al., 2007]

1 The Red River Basin and the Hoabinh reservoir

Figure 1.3: Percentage of monthly ow rates from China sub-catchment

at Laichau (black), Tabu (gray) and Hoabinh (light gray) onthe Da River

As mentioned in the section 1.2.2, the low ow from China has beenkept decreasing Unfortunately, it is estimated that the ow from Chi-nese part constitutes more than 50% (Fig 1.3) of total ow of the DaBasin at Hoabinh station in the dry season [Truong and Vu, 2010] IfChina overexploited all the uspstream water resources, the low ow left

to Vietnam could be zero It follows that the inow to Hoabinh reservoircould be reduced one half This may pose challenges to Hoabinh reser-voir operation when the locally-originated inow is not enough to feedthe reservoir as designed

Many issues have arisen from the socio-economic development in ferent countries within the basin, including the reservoir building andwaste discharging from upstream area of China These eects have be-come more severe to the downstream areas of Vietnam Thus, the inte-grated water management in the Red River Basin has to be considered as

dif-a trdif-ansbounddif-ary issue, requiring the joint eorts of the three countries.Unfortunately, this topic up to now is not yet appropriately addressed[Nguyen et al., 2007a]

1.3 Institutional and normative framework for Hoabinh operation

This Section explores the current management of the Hoabinh reservoir.The main controllable infrastructure in the RRB, the Hoabinh reservoir,

is operated by the Hoabinh Power Company (Operator from now on)under the Ministry of Industry and Trade During the ood seasonthe operation is supervised by the Ministry of Agriculture and Rural16

1.3 Institutional and normative framework for Hoabinh operation

Development, and it must comply with the guideline reported in theDecision No.80/2007/QD-TTg issued on July 1st, 2007 by the PrimeMinister of Vietnam The main rules in this Decision are briefed here.Notice that the Hoabinh reservoir is working in collaboration with theother two existing reservoirs of the Lo tributary: ThacBa on the ChayRiver, and Tuyenquang on the Gam River (see Figure 1.1)

To ensure absolute safety for Hoabinh hydropower plant, and to tively face all the uncertainties, the reservoir water level is not allowed

ac-to be over the maximum water level 122 m in any oods with returnperiod smaller than or equal to 10,000 years (at Sontay base5)

To ensure the safety for the Northern Delta, the operation of theHoabinh reservoir in collaboration with ThacBa and TuyenQuang reser-voirs should keep the Red River water levels at Hanoi station not ex-ceeding 13.1 m and 13.4 m if oods happened having return periods of

150 years and 250 years (at Sontay base), respectively

To ensure both safety and ood control, while improving the eciency

of hydropower generation, the ood season (dened by the Decision:from June 15thto September 15th) is divided into three periods as follows:the early ood period is from June 15th to July 15th, the main oodperiod is from July 16th to August 25th, and the late ood period isfrom August 26th to September 15th

The pre-ood water level of the reservoir during the early ood periodmust be controlled below 98 m From July 1st, if ood control is notactivated, water level is gradually regulated to close to the pre-oodwater level of the main ood period, so that on July 16th the waterlevel is around 90 m to 94 m On the contrary, if the Red River waterlevel at Hanoi is forecasted to exceed 10.5 m in 24 hours, the storage partbetween level 98 m and 102 m is used to reduce ood in the downstream,and keep water level of Red River at Hanoi not to exceed 10.5 m If it ispredicted that the water level at Hanoi may exceed 11.5 m in 24 hours,the volume between 102 m and 105 m is used to control the ood, andkeep water level at Hanoi not to be over 11.5 m After the ood peak,when the Red River water levels at Hanoi station is below 11 m, theOperator release water, gradually bring the water level of the reservoirclosing to the pre-ood level of the early ood period

In the main ood period, pursuant to the weather forecast of theNational Centre of Meteorology and Hydrology-Ministry of Natural Re-sources and Environment, if ooding is likely to appear with return pe-riod of more than 100 years (at Sontay base), the reservoir water level

5 The magnitude of oods are dened by considering their ows at Sontay station because downstream of this station the Red River branches

1 The Red River Basin and the Hoabinh reservoir

must be immediately decreased to 90 m If the water level at Hanoi isforecasted to exceed 11.5 m in 24 hours, the volume between 90 (or 94m) and 98 m is used to store ood water in order to hold the water level

at Hanoi station not to exceed 11.5 m When it is already over 11.5 mand forecast shows that in 24 hours the water level will raise rapidly, theOperator continues to store water in Hoabinh reservoir, in collaborationwith Tuyenquang and Thacba reservoir, to prevent water level of theRed River at Hanoi from exceeding 13.1 m, and water level of Hoabinhreservoir from exceeding 115 m In the case the water level at Hanoi ispredicted to exceed 13.1 m in 24 hours, the coordination in regulation

of three reservoirs keeps continuing, to hold water level of the Red River

at Hanoi not exceeding the 13.4 m, and water level of Hoabinh reservoirnot exceeding the 117 m After the ood peak, when the water level atHanoi is below 11.5 m, Operator releases water to hold the water level

of the lake at the pre-ood level of the main ood period After August

16th, based on weather conditions identied by the National Center ofMeteorology and Hydrology - Ministry of Natural Resources and Envi-ronment, if the main ood seems to end soon, the Center Flood ControlCommittee allows the Operator gradually raising the lake water level butnot exceeding 106 m

During the late food period, pursuant to the forecast the NationalCentral of Meteorology and Hydrology - Ministry Natural Resources andEnvironment, if the ood season seems to end soon, it is allowed togradually rise up the water level to the normal water level (+117 m).When the lake water level is 117 m, if the late ood occurs, release is setequal to inow, holding water level of the lake not exceeding 117.3 m.When the water level at the Hoabinh reservoir is already at 117 m,but ood of the Da River is predicted to continue, the lake water levelmay exceed the 117.3 m, safety modes should be activated: the bottomgates and the spillways are gradually or consecutively open in order thatwhen the water level reaches the 117.3 m, all the bottom gates, spillways,water intakes, valves are fully opened

For safety reason, the open of sluice gates should be done in a specialway The rst six or the last six bottom gates are opened or closed,respectively, following the 6 hour-rule, i.e opening or closing of the nextone will be done in 6 hours after the very previous one, or maximal 4bottom gates can be opened and closed within a day In other words,roughly, the absolute dierence between releases of two consecutive daysmust be smaller than or equal to 7332 m3/s (1833 × 4)

18

1.4 Objectives of the study

1.4 Objectives of the study

Figure 1.4 shows the yearly pattern of the Hoabinh level from 1995 (thedate when the reservoir lling can be considered completed) to 2004, im-plying that the Hoa Binh reservoir was operated according to a seasonalstrategy From January to June the reservoir release ranges from 500

to 2000 m3/s, which is generally enough to support the water supply atSontay In this period, the reservoir release is generally higher than thenatural ow of the Da River and, correspondingly, the Hoa Binh leveldecreases of about 25-30 m in six months The decrease in the Hoa Binhlevel is favorable for ood control as the reservoir reaches its minimumlevel just by the beginning of June, in anticipation of the oods that mayoccur in July and especially August From September to October, as thethreat of oods diminishes, the reservoir is relled and by the beginning

of November the full capacity is reached again

As all reservoirs in the world, Hoabinh reservoir has been facing theproblem of conicting interests among stakeholders In the ood seasonthere is a conict between ood control on the one hand and hydropowerand irrigation on the other In this season the reservoir level is kept low inorder to create the reserve volume necessary for ood control From thehydropower and irrigation perspective, the reservoir should be relled asquickly as possible in order to increase the hydraulic head and the storagebefore the dry season; on the other hand, if the reservoir is lled in tooearly, late oods that occasionally occur in September and October maycause a disaster for downstream

In the dry season, especially from January to March, when water quirements are intensive, if water is released to fully meet all the require-ments, the reservoir may become empty in the following months of April,May, and June As a consequence, power generation may be halted if thewater level in the reservoir falls below 75 m The possible total damagewould be very large because this situation does not only destroy the hy-dropower plant's equipments, environmental and navigation downstreambut also aect user's daily lives and productive activities in the basin.The goal of this study is to assess the room for operation improvement

re-of the Hoabinh reservoir and to design operating policies that can reducethe conict among dierent operation objectives

To the authors' knowledge, the only study about the Hoa Binh voir in the international literature are Ngo et al [2007] and Ngo et al.[2008] Ngo et al [2008] presents a what-if analysis comparing the im-pacts on ood control and hydropower production of three alternativeregulation policies They use Mike-11 hydraulic model to simulate thedierent policies over the ood season in the years from 1963 to 1996

reser-1 The Red River Basin and the Hoabinh reservoir

This study is the basis for further developments presented in Ngo et al.[2007], where the parameters of the regulation rules (e.g threshold waterlevel values at which the release decision is changed) are optimized bythe Shued Complex Evolution optimization algorithm and the tradeobetween ood control and hydropower production is analyzed

The assumption that the reservoir operation in the ood season can

be optimized separately from the rest of the year is questionable FromFigure 1.4 it can be noticed that while on the 1st of November, whenthe transition from the wet to the dry season takes place, the Hoa Binhreservoir is always at full capacity, at the dry-to-wet transition (1st ofJune), the Hoa Binh level varies between 77.5 and 96.2 m depending onthe year, meaning that occasionally water is transferred from one season

to the other, in order to maintain the hydraulic head as high as possible

It follows that while it is possible to simulate and optimize the systemmanagement over one year starting from the 1st of November with thestorage at full capacity, disconnecting the dry and wet season on the 1st

of June would unnecessarily limit the potential for optimizing the storagevalue at the transition

Following these considerations, the specic objectives of this thesis are

to advance the state-of-the-art along some directions:

ˆ enlarging the tradeo analysis to the water supply sector;

ˆ enlarging the optimization horizon to the entire year thus allowingfor inter seasonal water transfer;

ˆ exploiting more data availability to introduce a clear distinctionbetween the dataset used for optimization and the one used forvalidation of the optimized policies, which allows for a fair andstatistically sound comparison with the historical operation

ˆ exploiting the value of exogenous information (e.g meteorologicaldata) to design anticipatory operating policies

This study also aims at creating a reference for guiding collaborationbetween the dierent institutions (Ministry of Industry and Trade andMinistry of Agriculture and Rural Development) that currently managethe system, and for further study about inter-reservoir regulation in theRed River Basin, where other reservoirs are under-construction

To achieve the above objectives, several optimization algorithms will

be applied The research will thus also constitute a testing ground fordierent and novel reservoir optimization techniques

20

1.5 Conceptualization: actions, criteria and design indicators

Figure 1.4: Historical water level of Hoabinh reservoir (1995-2004)

1.5 Conceptualization: actions, criteria and

design indicators

In this thesis, the action that will be considered to improve water sources management is a change in the operating policy of the Hoabinhreservoir The three main criteria are ood mitigation, hydropower gen-eration and water supply Three design indicators in correspondencewith the three criteria will be dened and used in the control problem.Denition of such indicators is discussed below

re-Social and economic interests in the Red River Basin are modeledthrough indicators that quantify the evaluation criteria the relevant stake-holders adopt in judging and comparing alternative operating policies.The formulation and subsequent identication of these indicators shouldtake into consideration some fundamental properties and concepts: (i)indicators are supposed to accurately reproduce the stakeholders view-points and should thus reect their perception of the problem; (ii) theymust meet some technical requirements imposed by the control algorithmadopted to design the operating policies Precisely, the design indicatorsmust be formulated as the integral over a reference time horizon of im-mediate costs that should be, in turn, easily computable from the systemmodel output without adding to much to the problem complexity Tobalance delity and computational complexity, immediate costs are for-mulated as simple physical relationships including empirical parameters

tted to the stakeholder risk perception

1 The Red River Basin and the Hoabinh reservoir

Hydropower production

The Vietnamese electricity market is regulated by the Government andthe energy sold at a xed rate decided on the basis of the average energyproduction cost and the current economic development strategy Elec-tricity prices change depending upon the energy destination (industrial

or domestic use) and the total energy consumed but not within the day

or the week In economic terms, given the xed cost of hydropower eration, maximizing the energy production is equivalent to maximize theassociated revenue Yet, the fast-growing national energy demand [Toan

gen-et al., 2010] and the recently increasing frequency of power shortages

in the last three months of the dry season, from April to June, makethe smaller energy available in this period much more valuable than inothers To account for this seasonal variability, in formulating the im-mediate cost, the daily energy production Pt+1 [GWh] (see Eq (2.18))

is ltered by a time-varying coecient αt[GWh−1], expressing the valuegiven to one GWh on day t, i.e

Based on the analysis of the energy decit and consequent import fromChina in the dierent season, αtis assumed equal to 2 from April to Juneand 1 in the other months Being the indicators formulated as costs, theproduction in (1.1) is changed in sign

Water supply

Wet-rice agriculture is key to national food security but also the mostimportant segment of the Vietnamese economy [FAOSTAT, 2003] Theoptimal climatic conditions and plentiful water resources of this tropi-cal monsoonal region enabled an intensive rice production in the RedRiver Delta (rrd), composed of 31 irrigation schemes servicing around

850 000 ha of irrigated agriculture [Turral and Chien, 2002] and formingthe second largest rice production area in the country after the MekongDelta The maximization of the net crop return (including variable and

xed costs) is the economic indicator traditionally adopted by the agriculture sector (e.g see Kipkorir et al., 2001) However, both cropprice and yield dynamics do require sophisticated models, which arenot easily identiable from conventional observational data and wouldconsiderably add to the problem computational burden In addition,the extensive use of pumping stations in the rrd distribution network[George et al., 2003] implies substantial energy costs in operating theirrigation scheme that, however, are hardly estimable due to the lack of22

wet-1.5 Conceptualization: actions, criteria and design indicators

data [Harris, 2006] For these reasons, the average annual water decitcan be adopted as a proxy of the annual crop yield and the disaggre-gated daily decit the corresponding immediate cost This is a provablyreasonable hypothesis under the assumption that the considered operat-ing policies will not move to much away from the current average watersupply [Soncini-Sessa et al., 2007a] Further, to make the surrogationmore reliable, the annual decit is not linearly reallocated on a dailybasis, but modulated by a time-varying coecient βt that accounts forthe combined varietal phenological stages and climate conditions andthe associated time-varying risk of stress (e.g Kulshreshtha and Klein,1989) Finally, farmers are not insensitive to the magnitude of the dailydecit since, the integral eect of water shortages being the same, severalsmall decits might be more acceptable than one single severe shortagethat might strongly aect crop production (e.g see Draper and Lund,

2004 and references therein) A behavioral coecient n is thus used tocharacterize farmers' risk aversion: n = 1 means no risk aversion, whilefor n → ∞ the aversion is maximum and the indicator is equivalent to amin-max formulation [Soncini-Sessa et al., 2007b] Correspondingly, theimmediate cost for the water supply is formulated as a power function:

t+1 [m3/s] are the daily water demand (given as shown

in Fig 1.5) and supply at SonTay (see Fig 1.1), and βt [(m3/s)−2] isequal to 2 from January to March, when the diverted ow from the RedRiver is the only source for the submersion of paddy elds for winter-spring rice crop, and 1 in the rest of the year when the submersion forthe summer-autumn crop is also supported by rainfall The power n is

xed equal to 2, which ensure that vulnerability is a minimum according

to Hashimoto and Loucks [1982]

The yearly pattern of the water demand was estimated as part ofthe Integrated Red-Thaibinh River Basin Planning project (2002-2006)[Nguyen et al., 2007b] The whole basin was divided into seven sites,and computation of water demand in each site takes into account allwater use sectors, including industry, domestic use, irrigation, livestock,aquaculture, navigation and environment Water demand for industrywas estimated by assuming a specic water demand of 1000 m3/1000usd for food industry, 400 m3/1000 usd for light industry, and 200

m3/1000 usd for heavy industry Water demand for domestic use isabout 150 liters per person per day for the urban area, and 60 litersper person per day for the rural area Water demand for livestock is

1 The Red River Basin and the Hoabinh reservoir

Figure 1.5: Yearly pattern of water demand at Sontay

computed based on the water criteria per head per day, depending onthe size of the animal Water requirement for aquaculture is 20000 m3

per ha per month Water demand for irrigation is computed using theCROPWAT [Allen et al., 1998] software There is no ocial criteria forenvironmental water requirement In the project, it is set from 10 to

30 % of total other demands And water requirement for navigation istransformed from water level required at some certain points along theRed River

Flood mitigation

Hanoi and its unusually overdeveloped river urban area (rua) are tected by a system of two series of dykes for a total length of 2700 km.Floods mainly occur in July and August and inundations produce enor-mous damage every time dykes break [Hansson and Ekenberg, 2002],

pro-as regularly happened nearly once per decade in the lpro-ast century Inprinciple, an accurate modelling of ood inundations and the associateddamage requires to combine a 2D model of the oodplain to estimatethe ooded surface area (e.g Hoang et al., 2007) and a record of past

ood recovery costs and associated river ow rates to interpolate thecorresponding damage (e.g de Kort and Booij, 2007) Because of theregularly disruptive eects of the ood routing process following a dykebreaching on the RUA morphology, any ood routing model should berecalibrated after every ood event Further, the fast uncontrolled ur-ban development in the rua is quickly changing the size and shape ofthe oodplain, thus making totally incomparable damages registered in24

1.5 Conceptualization: actions, criteria and design indicators

dierent years Damages can thus not be included as an indicator in ourdecision model Nevertheless, it is observable [Vorogushyn et al., 2010]that high and persisting ood water levels in Hanoi correspond to highrisk of dike break, and consequently high potential damage An indi-rect way of accounting for ood damage is thus to penalize operatingpolicies that produce river water levels higher than some appropriatelyselected threshold Once again, economic relevance and risk perceptionare implicitly accounted for using some empirical coecients: the higherdamage potential of oods in August on the summer-autumn crop [Le

et al., 2007] is given a higher weight, while the increased stakeholders'risk aversion to extreme ood is modelled by using a power law Theresulting immediate cost has the following form:

t+1is the water level [cm] at Hanoi station, ¯h (=950 cm) is the 1stalarm ood level [Hansson and Ekenberg, 2002], δt[cm−2] is the seasonalcoecient (equals 2 in August and 1 otherwise), and m is the coecientreecting risk aversion here assumed equal to 2 The rational is that

ood risk comes from either the overtopping of the levees or the leveebreaches The latter are more probable in August because the meanwater level is 8.29 m (against 4.53 m in the rest of the year) and thusthe soil volume of the levee saturated with water is larger Water levelexcesses are thus given more weight in August Further, the total force

on the levee, which is the driver of collapse, increases with the square ofthe water level, which motivates the choice of power 2 in Eq (1.3)

2 Modelling the Hoabinh water system

The three main components of the Hoabinh water system are the stream catchment, the reservoir, and the downstream river network Theglobal model, therefore, includes three elements: the models of the up-stream catchments (Sec 2.2) to predict inow to the system, the model ofthe Hoabinh reservoir (Sec 2.3), including the model of the hydropowerplant, and the model of the downstream river network (Sec 2.4) Actu-ally, two downstream ow routing models are needed, one for computingthe decit ow at Sontay and the other for computing the cost of ood

up-at Hanoi For the reader's convenience, a list of main variables used inthe global model is given in Table 2.1 The choice of the modelling timestep is also discussed in the next paragraph

2.1 The Time Step

Decision step is the time between one decision and the next It is served that the decision step is not actually constant, because in criticalsituations, when for example there is a high risk of ooding, the Operatortakes decisions with a step that is much shorter than in normal condi-tions However, for the study purpose, the decision step can be assumed

ob-Figure 2.1: The scheme of the Hoabinh water system and of its model

2 Modelling the Hoabinh water system

Table 2.1: The main variables of the model

No Name Meaning

1 r The release from the Hoabinh reservoir

2 rt The release through the turbines of the Hoabinh power plant

3 q t,max The maximum turbine capacity of the Hoabinh power plant

4 rt,max The maximum possible releases through the turbines of the Hoabinh power plant

5 rb,max The maximum possible release through the bottom gates of the Hoabinh reservoir

6 r s,max The maximum possible release through the spillways of the Hoabinh reservoir

7 hup The water level in the Hoabinh reservoir

8 hdo The water level downstream of the Hoabinh reservoir

9 H The hydraulic head of the Hoabinh reservoir

10 η Turbine eciency

11 a The inow to the Hoabinh reservoir

12 s The storage of the Hoabinh reservoir

13 S The surface of the Hoabinh reservoir

14 e The unitary evaporation of the Hoabinh reservoir

15 E The evaporation of the Hoabinh reservoir

16 q mef The environmental ow required from the Hoabinh reservoir

17 h Y B , q Y B The observed water level and discharge at the Yenbai station

18 h V Q , q V Q The observed water level and discharge at the Vuquang station

19 h ST , q ST The observed water level and discharge at the Sontay station

20 h HN , q HN The observed water level and discharge at the Hanoi station

21 w The nominal water demand at Sontay for all downstream users

22 ¯h The ood threshold at Hanoi (1 of 3 alarming levels at 9.5, 10.5, 11.5 m)

23 ghyd Step cost of hydropower production

24 gdef Step cost of water decit

25 gf lo Step cost of ood

as constant and equal to a day without any signicant loss of accuracy.The choice of decision step clearly inuences the choice of modelling step,because the latter must be equal to the rst or a submultiple of it Sincethe decision step is equal to one day, the modelling step will be 1/k with

k is an integer number To choose the value of k, one must consider:

ˆ the sampling time of the hydrological measurements used for modelcalibration and validation;

ˆ that the computing time needed to solve the control problems and

to evaluate the alternative operating policies increase linearly withk;

ˆ that the discrete-time description of the system should not produce

an excessive loss of information (the system is time continuous).Most of the time series in our possession have a daily time step, so it isreasonable to assume k =1 This assumption also has the advantage ofreducing the computing time to a minimum To verify that the selectedvalue of k does not bring about an excessive loss of information, the suf-

cient conditions is that the modelling time step be shorter (by aboutone order of magnitude) than the smallest time constant of the model(Shannon's Sampling Theorem, Shannon, 1949) The time constant is

a measure of the speed at which the output of an asymptotically stablesystem converges to the equilibrium output, when the input is kept con-stant It can be properly dened only for linear time-invariant systems,28