On-line flow forecasting for automatic operation of a flood reservoir

Transcription

1 Hydrological forecasting - Prévisions hydrologiques (Proceedings of the Oxford Symposium, April 1980; Actes du Colloque d'oxford, avril 1980): IAHS-AISH Publ. no On-line flow forecasting for automatic operation of a flood reservoir A. VAN DER BEKEN, G. L. VANDEWIELE, J. IBERGHIEN, J. MARIËN, J. MARIVOE and M. MEERSSEMAN Brussels, Belgium Abstract his paper presents a project concerned with a flood reservoir. he study aims to implement a fully automatic control system based on real time measurements and model calculations. Forecasts are required both of flow into the reservoir and at the downstream control section where uncontrolled tributary flows occur. he processing of existing data and the collection of additional field data are discussed. Quantitative temperature and precipitation forecasts are proposed. Different schemes for using rainfall-runoff and flood routing models are given. hese models include the probability limits for their expected forecast values and real time corrections by using an ARMA model. he control policy will be based upon a heuristic approach using a simulation model. Prévision des débits en temps réel en vue d'une gestion automatisée d'un réservoir de crue Résumé. L'article concerne un projet de réservoir de crue. L'étude a comme objectif la réalisation d'un système automatique de gestion du réservoir basé sur les mesures et les calculs par modèles en temps réel. Dans ce but des prévisions de débits sont indispensables aussi bien à l'entrée du réservoir qu'à la section critique d'aval où des affluents aux apports non-controlés viennent grossir les eaux de la rivière. Le traitement des données existantes et la collecte de données supplémentaires sur le terrain sont discutés. Des prévisions quantitatives de température et de précipitation sont proposées. Plusieurs schémas de calcul basés sur des modèles de 'relation pluie-débit' et de 'propagation des crues' sont présentés. ous ces modèles fournissent des prévisions des valeurs et de leurs limites de probabilités. En outre ont été prévues des corrections en temps réel grâce à l'utilisation des modèles ARMA. La gestion sera basée sur un modèle de simulation selon la méthode heuristique. INRODUCION he Dijle River basin has a total area of 3462 km 2 and forms part of the Scheldt River basin ( km 2 ) (Fig. 1). Its main tributary, however, is the Demer River with a basin area of 2200 km 2 (Coen, 1978). Fairly large cities are drained by the two rivers, such as Diest and Aarschot by the Demer and Wavre and Leuven by the Dijle River. After their confluence, the Dijle Demer River crosses the city of Mechelen where tidal effects are observed. he largest part of the river basin is covered by a loamy soil of thickness between 0.5 and 1.0 m or more. he ertiary deposits which underlie the soil are often sandy with a large water bearing capacity. he relief of the basin varies between 10 and 160 m. he general slope index as defined by Roche (1963) is for the Dijle basin and for the Demer basin. Although the overall land use in the basin is still mainly agricultural (arable land), the region has been increasingly urbanized, especially around the cities. wo main highways (E5 and E40) have been built across the basin in the last 10 years. he new university city Louvain-la-Neuve is situated in the Dijle basin. Moreover, this basin is adjacent to the Brussels area with its eastwards expanding urbanization. he climatological effect of this area shown by heavier summer thunderstorms may eventually reach part of the Dijle basin. Historically, the Dijle Demer rivers have always been subjected to flooding. he flooding becomes extremely heavy when spring tides in the Scheldt River coincide with high discharges of its tributaries. Urbanization increases this flood hazard. 557

2 558 A. van der Bekenef al. FIGURE 1. he Dijle River basin as part of the Scheldt River basin. herefore, flood relief programmes for both the Dijle and the Demer have been planned. his paper deals with part of this programme for the Dijle River. HE DIJLE FLOOD RESERVOIR For the Dijle basin upstream of the city of Leuven with an area of 867 km 2, the 100-year return flood has been estimated at 120 m 3 /s. he permissible flow at Leuven however, is only 60 m 3 /s and this has a return period of 25 years. A (single purpose) flood reservoir of 2.8 x 10 s m 3 has been proposed at Neerijse, about 10 km upstream of Leuven, with a basin area of 738 km 2 (Fig. 2). Downstream from this reservoir two tributaries, the River Voer with a basin area of 52 km 2 and the River Molenbeek with a basin of 42 km 2, have estimated 100-year return floods of 22 and 18 m 3 /s respectively. Hence, the control operation objective of the flood reservoir is the limitation of the total downstream flow at Leuven (control section) to the critical level of 60 m 3 /s by releasing appropriate flows at the outlet of the reservoir. he construction of the outlet of this reservoir will be a movable cylindrical crested weir. It will be on the river bottom when the reservoir is not in use (Bogaerts, 1977). HE SUDY PROGRAMME he goal of the study, sponsored by the Ministries of Agriculture and the Flemish Region, Department of Agricultural Hydraulics, is the future implementation of a fully automatic operation system of the weir by using real time forecasts for the different parts of the flow system. It has been estimated that a non-automatic

3 On-line flow forecasting 559 FIGURE 2. he projected flood reservoir with existing measurement stations. operation system would require a larger flood reservoir which is technically impossible in this region. On the other hand, the automatic system should be as simple as possible in order to improve reliability. he study programme, presently underway, can be divided as follows: (1) processing of existing data; (2) collection of additional field data; (3) formulation and off-line cahbration of several models for: (a) temperature and precipitation forecasts, (b) rainfall-runoff relationships, (c) flood routing, (d) operational strategy of the flow releases from the reservoir; (4) design of inteligent measurement stations for future use in the real time system; (5) design of the on-line data transmission system and process-computer configuration; (6) extensive simulation and efficiency study and formulation of follow-up requirements, including rules for manual operation. PROCESSING OF EXISING DAA Since this study is aimed at the off-line calibration of several models, existing data should ideally have the folowing characteristics:

4 560 A. van der Beken et al. (1) data collected in the basin under study, (2) similar time base, (3) exact synchronization of all data, (4) available over the same periods, preferably for several years, (5) preferably in digital form. It is most likely that existing data will not meet all these requirements. For instance, short-duration meteorological data (10 min or hourly) were only available at three stations in the neighbourhood of the basin (see Fig. 2). Moreover, exact synchronization of all stations is certainly not guaranteed. Digitizing and handling of data, analysis by correlation and interpolation, critical use of height discharge curves are some of the many tedious aspects of the general tasks of processing existing data; these tasks must be executed accurately. COLLECION OF ADDIIONAL FIELD DAA Since many existing data do not fulfil all requirements, additional data have to be collected during the study period. his will allow us (1) to improve the off-line calibration of the models at a later stage; (2) to assess the value of the measuring stations and to choose the location of the future permanent stations for the real time system. his field work requires ideally two persons for installation, control, maintenance and for velocity or discharge measurements. For reasons of quick installation, three portable pneumatic bulb type recorders have been installed on the rivers Voer and Molenbeek (see Fig. 2). At the other places, O-limnigraphic recorders are used. A pluviographic recorder (Hellmann-type) is also in use. All recorders are synchronized as reported by Van der Beken et al. (1980) in the poster session of this Symposium. FORMULAION OF HE ON-LINE CALCULAIONS he consecutive on-line calculations and tasks to be performed by the process computer in the future real time system are given in Fig. 3. he double outlined boxes represent the models where calculations take place. he incoming observations are on-line measurements which must be accepted or rejected on the basis of welldefined criteria. For the time being, only temperature, precipitation and water level measurements are planned. According to the availability of the treated observations, different models can be chosen. his is represented in Fig. 4 which is a more detailed version of Fig. 3. he forecasts of the flows are not known exactly due to measurement errors and the imperfection of the forecasting models (especially the precipitation forecasting model). o meet this difficulty, we use probabilistic models where the predicted value has also been given probability limits. However, the inflow into the reservoir is assumed to be less important since the reservoir acts as a buffer. herefore, probability limits are only required for the uncontrolled flows of the downstream tributaries Voer and Molenbeek. Quantitative temperature and precipitation forecasts Since temperature is used in the rainfaë-runoff models for evaluating thermal losses as well as freezing and thawing processes, a temperature forecast is required. As the importance of the temperature model is assumed to be rather marginal, the very simple model illustrated in Fig. 5 has been formulated.

5 observations peteorological and hydrological models ) forecast of the inflow ï in the reservoir forecast with probability limits of the uncontrolled flow at the control section flood-routing model between the reservoir and the control section regulation rules t optimal weir position FIGURE 3. computer. alarms { statistics! On-line flow forecasting 561 Scheme of the consecutive on-line calculations and tasks of the process observations; acceptance or rejection of observations calculation of missing ± observations calculation of areal precipitations calculation of flows from levels always available treated observations i forecasting models for temperature and precipitation forecasted temperature and precipitation not-always I available treated observations choice of the models used rainfall-runoff models and flood-routing models ± forecasts of the inflow in the reservoir forecasts with probability limits of the uncontrolled flows FIGURE 4. On-line calculations from the observations to the forecasts. A quantitative precipitation forecast is required in the rainfall-runoff and flood routing models for the downstream tributaries. However, practical models for operational hydrology are not available at the present time. Moreover, within the limits of our study programme and in view of the final goal, the decision was made not to search for advanced numerical models or to use radar observations.

6 562 A. van der Bekenef al. 'FORECAS *I FIGURE 5. NOW Model for temperature forecast. A statistical study was based on an analysis of a 12 year period of 2-h data. he influence of a number of variables, the values of which can be measured in real time, was considered. he significance of their influencing future precipitation during 2 and 6 h was determined. he recent change in atmospheric pressure and recent precipitation proved to be significant. he atmospheric pressure itself, temperature and wind direction seem unimportant. Research is continued in the direction of 24 h forecasts. Rainfall-runoff and flood routing models We consider as inputs precipitation series \P(t)} and temperature series {(t)}. he output of the rainfall runoff model is the flow series {A(f)} with appropriate probability distributions and probability limits. he rainfall runoff model distinguishes between direct runoff and delayed runoff while thermal losses are accounted for by a temperature relationship. For the transport of the direct runoff, a linear model is assumed. For estimating the delayed runoff, however, on-line measurements will be used. he flood routing model is also a linear model. Details of these models and their off-line calibration are given by Marivoet and Vandewiele (1980) at this Symposium. hese authors also present the use of ARMA models for real time correction of the forecasted flows. In the case of a forecast without flood routing, the scheme in Fig. 6 summarizes the flow chart of the calculations. In this scheme, Q(t), is the on-line measurement of flow and/? (t) = O(t) A(t) forms the residual. From a past series of residuals, we can extrapolate a series of future residuals by using an auto regressive moving average (ARMA) model. he notation is as follows: the superscript v means a forecast ; the underlining means a stochastic variable for which we can describe the uncertainty with a probability distribution and confidence limits; the accent means an improved forecast by using the ARMA model. When flood routing is also incorporated, the necessary calculations are shown in Fig. 7, where the subscript 'up' means 'upper basin', 'do' means 'downstream' and 'si' means 'lateral inflow'. Similar schemes are required when some on-line measurements fail. he optimal control policy of the flood reservoir he regulation of the flow releases at the reservoir must be optimized in real time in the sense that the expected maximum positive difference between the actual flow at

7 On-line flow forecasting 563 P(t), (t),(t), t 1 rainfall -runoff «r (t) model * r Q(t) * V A v (t) r A(t). w R(t) ARMA i(t) X-< t) FIGURE 6. Scheme of calculations in the case of a forecast without flood routing. Q(t)!P(t), (O, (t), (t) (rainfall-runoff model J^J 'A (t) Q (t)j' (t) 'up up Flood t routing model X(t)=X si (t) + X^o(t) R(t)=<J(t)-A(t) t f K(t) X'(t) FIGURE 7. Scheme of calculations when flood routing is incorporated. the control section and the critical flow (60 m 3 /s) for the forecasted horizon is minimized. he problem is complicated since we have to take into account the stochastic nature of the forecasted flows. o the best of our knowledge, such a complex, real time, nonlinear, dynamic optimization problem cannot be solved by explicit methods of operations research. herefore, a heuristic method based on simulation is developed. Some experimental results obtained with this simulation model prove that optimal control is better

8 564 A. van der Bekenefa/. achieved with a forecast of the inflow into the reservoir than without this forecast. Further details are given in the paper by Mariën (1980) at this Symposium. CONCLUSIONS he case study presented in this paper is the first project of on-line flow forecasting in Belgium. he planned flood reservoir is a single purpose project which certainly facilitates the model formulations, the measurement system and the control policy. Nevertheless, the many tasks of the study, which are not all discussed in this paper, require a teamwork where statistics, operations research methods, electronics and information theory as well as hydrology must play a well-balanced role in order to propose the most efficient system at the present stage of our knowledge. Acknowledgements. his study is supported by the Department of Agricultural Hydraulics, Ministry of Agriculture and Ministry of the Flemish Region. he authors would like to thank F. Bouttefeux, P. Demeester and K. O. for their initiative and support. hey also wish to thank J. P. Bogaerts from N. V. Seges, Consultants, who planned the flood reservoir, for fruitful discussions. Many institutions were helpful in providing data and comments, particularly the Royal Meteorological Institute, the Ministry of Public Works, the Ministry of ransport, the Ministry of Public Health and the University of Louvain-la-Neuve. Special thanks are due to the National Water Supply Agency for providing measurement facilities. REFERENCES Bogaerts, J. P. (1977) Studie van de hoge afvoeren op de Dijle (A flood study of the River Dijle). Symposium on Flood Studies (Brussels, November 1977). Coen, I. (1978) Neerslag- afvoerrelaties voor Schelde en bijrivieren (Rainfall-runoff relationships for the River Scheldt and its tributaries). ijdschrift der Openbare Werken van BelgiëNr3, Mariën, J. (1980) Regulation of a flood reservoir with the use of on-line forecasts. In Hydrological Forecasting (Proceedings of the Oxford Symposium, April 1980), pp : IAHS Publ.no Marivoet, J. and Vandewiele, G. L. (1980) A real time rainfall-runoff model. In Hydrological Forecasting (Proceedings of the Oxford Symposium, April 1980) pp : IAHS Publ.no Roche, M. (1963) Hydrologie de Surface: Gauthiers-Villars, Paris, France. Van der Beken, A., Marivoet, J. and iberghien, J. (1980) Instruments and their improvement for hydrological forecasting. Poster paper presented at the Symposium on Hydrological Forecasting, Oxford, April 1980.