I. D. Cluckie, J. Yuan, R. Norreys, G. Shepherd, J. Tyson

Introduction

Large urban drainage systems (UDSs) consist of many thousands of pipe sections and manholes. These systems generally transport combined flows, that is both sewage and storm water, and they are designed to operate within specific load levels. Since these UDSs are physically large, it is uncommon for spare capacity to be available in one area of the system at the same time as Combined Sewer Overflows (CSOs) are operating elsewhere. Such behavior means that part of the capital invested in the UDS is being wasted and, conversely, if the available storage volume can be utilized by the application of an RTC system, benefits must accrue from the reduction of flood damage to property and the environment.

Many control systems use flow measurements and levels to control storage allocation and as such are reactive systems. The conversion of the RTCs to ones using predictive algorithms will provide a more flexible, efficient and cost effective control philosophy. The small additional cost in establishing the predictive RTC strategy will assist in making the overall system more productive in that the frequency and volume of CSOs will be reduced and, providing that the particular drainage system has some storage capacity available, operational performance will be improved.

Since the last symposium on hydrological applications of weather radars, held at Hannover in 1992, significant developments in RTC systems have been made (Schilling, 1994). This paper attempts to highlight some of the important aspects of RTC development and their dependency on radar hydrology.

WHICH RTC SYSTEM?

To be an RTC system there must be at least one adjustable regulating device operated under rules defined in the system strategy. This device could be a pump, a gate, a penstock or an adjustable weir. There are many types of RTC systems, each being defined by its Control Scope, Controller, Control Mode and Control Target. For example, if a UDS is globally controlled by an automatic controller in a predictive mode to maintain flow quality, then the RTC system can be classified as a Global-Automatic-Predictive-Quality control system. In order that such systems may operate successfully, it is essential that high quality rainfall measurements must be achieved by the use of hydrological radars.

The Control Scope or extent of the control system may be local in which case it is probably non optimal, regional which would include a number of interactive branches of the network, or system wide. Local control is only suitable for problems that can be trapped and removed rather than partially cured or transferred elsewhere. To achieve regional control objectives, it is necessary to co-ordinate control of the inter-related branches within the scope of the RTC. For example, without co-ordination the emptying of one tank upstream may cause the flow to be restored in a tank located downstream. The efficiency of the RTC will depend on the effectiveness of the co-ordination process. It is this co-ordinated approach to control that differentiates between regional and local control systems. The global control strategy is formed by co-ordinating the operation of all the control devices and may be achieved by operating the regional control systems under hierarchical procedures.

The Control Mode of an RTC system is the way that the regulating device reacts to the system status. There are three possible modes: Passive, Reactive and Predictive. The control settings for the regulating devices are constant for Passive mode, and are adjustable in Reactive/Predictive modes. In the Reactive mode, the methods used for the adjustment of the control settings are based on real-time measurements from the UDS. In the Predictive mode, real-time data are used as input to predictive algorithms to provide estimates of the future state of the UDS. Since all dynamic control systems have a real possibility of component failure, all RTC systems have to be able to function in a fail-safe manner. The automatic detection of such failures and the necessary actions to return the regulating device to passive control must be part of any system.

THE SPRINT PROJECT AND THE BOLTON RTC SYSTEM

In 1989, a project to protect rivers from pollution and to minimize the risk of urban flooding was launched using a pilot study of the Bolton catchment. The project was named SPRINT, an acronym for Specific PRojects for Intra-community INovation and Technology transfer. The project was partly funded by the European Community and includes a consortium of water-related companies and local authorities, the UK, Denmark and Spain. The UK representatives are North West Water (NWW) Ltd and the Water Research Center (WRc). An RTC system has been constructed for the Bolton UDS as part of the project, the structure of which is shown in Figure 1.

Fig. 1 - The Bolton RTC system

The main control objectives were to optimally use the storage tanks, minimize CSO operations and eliminate the problem of surface flooding. At the present time the RTC system is a Global_Supervisory_Reactive_Quantity control system. A system operator and a telemetry engineer have been employed to run and maintain the system during the current testing phase. The master control station was equipped with an IBM compatible PC to interrogate the telemetry outstations and an RISC workstation to run a simulation model developed using the MOUSE program from the Danish Hydraulic Institute. A total number of 29 flow sensors and 9 raingages have been installed in the system although difficulty has been experienced in maintaining a sufficient raingage density originally stated as being 1 gage per 7 km2. If the simulation model is found to be capable of predicting flows for 1-2 hour ahead, the RTC system will be modified to operate in Predictive mode. Flow prediction is necessary to enable the UDS to be prepared to accept inflow caused by the predicted storm event. Due to the inadequate quality of the rainfall data available from the raingage network, it has been identified that data from the MARS urban hydrological C-band radar (Cluckie, et. al., 1995) needs to be incorporated into the RTC system. Current work also includes the development of novel modeling techniques to enable the behavior of complex pipe networks to be simulated in real-time.

THE EVALUATION OF RTC POTENTIAL USING WEATHER RADAR DATA

It is possible to evaluate the RTC potential of an existing UDS using radar rainfall data. This evaluation process is demonstrated in a case study carried out for the Fylde UDS. The Fyld area is now better known for its bathing waters and the pleasure beaches. The area is relatively flat at an average altitude of about 15 meters above sea level. The location of the area under study is shown in Figure 2.

The major objectives of the current rehabilitation projects for the Fylde system are to protect the Bathing water for coastal communities and to ensure that the discharges of storm water do not prejudice compliance with the requirements of the EC Bathing Water Directive (Council of the European Communities, 1976).

The catchment area drained by the system is approximately 17 km2 and this is shown schematically in Figure 3. There are two main storage facilities, an off-line tank at Moor Park with a capacity of 11300 m3 and the Warren Drive (on-line) tank sewer with a capacity 5420 m3. A pumping station at the Anchorsholme outfall is responsible for lifting the screened sewage into the Irish sea.

Fig. 2 - The location of the Fylde

There are emergency overflows at the Moor Park tank and at the Anchorsholme pumping station and CSOs at six other locations. Regulators in the form of pumps and penstocks have been installed at the following locations:

Fig. 3 - The schematic of the UDS used in the assessment

The penstocks and pumps are automatically activated using local water levels in a reactive mode. The existing RTC system can therefore be defined as a Local-Automatic-Reactive-Quantity control system. The objective of the study was to assess the potential of a Global-Supervisory-Predictive-Quantity control system.

Assessment Methodology

There are many ways of quantifying the potential of RTC for a UDS (Green, 1991). The method adopted for this assessment was to evaluate the improvement in storage utilization when an RTC strategy was applied. The evaluation was carried out using the rainfall data collected by the radar and a simulation model of the UDS built using a commercial simulation package, WASSP (Department of the Environment/National Water Council, 1983). The improvement in storage utilization can be quantified by an Improvement Factor (IF), defined as:

where d EX is the total CSO volume in m3 spilled during the event under existing control, dRTC is the total CSO volume spilled during the same storm under the intended RTC, and SEX is the utilized storage volume in m3 during the same storm under the existing control. It is assumed that the improvement in storage utilization is equal to the reduction in CSO volume expressed as a percentage of SEX. To be a valid indicator of the RTC potential, the process of reducing the CSO volume should not cause any excessive flooding or other damage compared with the existing control strategy.

Determination Of Rainfall Events

All enclosed UDSs have a finite storage capacity and therefore the maximum reduction of CSO spill volume that can be achieved is reached when the system is operating at full capacity. The Improvement Factor, IF, is therefore a function of storm scale which, in turn is a function of Storm Duration and Return Period. Furthermore, actual storms will have variable spatial severity and will produce different degrees of storage utilization within the UDS. To obtain an objective relationship between the Improvement Factor and Storm Scale, the rainfall events used are assumed to have a spatially uniform scale distribution, that is every point within the storm area has an equal storm duration and return period. The temporal distribution of the rainfall was intact (i.e. as measured) and the spatial distribution synthetic being based on the spatial rainfall patterns of widespread frontal storms. The catchment used is covered by 6 radar data grids each 2 km by 2 km, and each event contains 6 time series rainfall data sets derived from the Hameldon Hill weather radar located some 50 km from the Fylde coast. a total of 9 storms were constructed, the details of which are shown in Table 1. An example of one such semi synthetic storm is included in Figure 2.

Table 1 Selected Rainfall Events

RTC Simulation Results

The performance of the existing RTC system is shown in Figure 4. It can be seen that a large proportion of the flow had been spilled at Anchorsholme pumping station whilst the available storage at Warren Drive tank had not been activated because of inappropriate control settings for this particular event. In order to improve the performance of this UDS a Global-Predictive RTC system is desirable.

By using a calibrated simulation model, the storage utilization and overflow status can be predicted for a rainfall data set corresponding to a particular storm event. Regulatory operations can then be carried out on the basis of the simulation result. It can be seen from Figure 4 that if the automatic operation of Regulator 2 is modified to enable the storage at Warren Drive to be used, the overflow at the pumping station will be reduced. Most of the available storage at Moor Park tank had been used during the event and so no adjustments were made to the operation of Regulator 6. The performance of the Global-Predictive control strategy was simulated using the above operational strategy together with the 9 distributed storms and as such was a single step operation. If multiple step adjustment of the regulators had been used, the improvement in the storage utilization might have been even greater and therefore the results shown in Figure 5 indicate the minimum potential of a Global-Predictive RTC strategy. These results show that the benefits gained from the implementation of a predictive RTC system decrease with the increase in storm scale since the storage volume in the UDS is tending to become more completely utilized and there is less scope for the RTC system to make any meaningful decisions. However, for moderate storms, Improvement Factors between 20% and 50% are achievable. In the case of the selected UDS, a 1-hour 1 year event can be treated as a moderate storm and with a more complex predictive operational strategy an even more substantial improvement in the storage utilization could be made.

Fig.4 - Storage capacity at Warren Drive Tank sewer having not been used

MODELLING REQUIREMENTS FOR PREDICTIVE REAL-TIME CONTROL

A predictive control system requires several different modeling elements to be inter-linked in order to predict the effects of a control strategy upon the entire catchment. The catchment might include elements such as the drainage system with controllable elements, the waste water treatment plant with flow constraints and receiving waters which have legislated quality requirements.

Fig.5 - The improvement of storage utilization as a function of storm scale

Figure 6 illustrates the decision algorithm for a predictive control system which relies upon data inputs from measurements made within the system to provide the initial conditions and radar rainfall data to enable the hydrological predictions to be made. Once the hydrological data has been generated, a control loop is entered requiring that the complete drainage system is controlled in such a way as to ensure that the predicted rainfall causes minimal impact upon other operational criteria such as treatment cost, flooding, final effluent quality or operational cost. In order to identify the optimum strategy for a predicted event, the full range of available options have to be examined and their relative performances assessed. The chain of elements making up the algorithm could include:

Short term rainfall forecast model;

Rainfall runoff model;

UDS flow simulation model;

Sewer quality model;

Receiving water quality model.

Fig.6 - Decision algorithm for predictive control

These model elements have to be capable of running very fast since they interact with one another requiring multiple passes of the complete algorithm to be computed before an optimal strategy can be selected. The selection and integration of these model elements is critical to the successful introduction of such schemes. To this end, several of these elements are being developed at Salford (Abes, 1995, Griffith, 1995, Norreys 1991), Han and Cluckie (Sao Paulo, 1995) discusses the problems associated with the development of the rainfall data. This paper will concentrate on the development of techniques used to simulate flow in a UDS.

Modeling Methodologies For RTC

It is a basic requirement of any RTC system that the models are computationally fast since any decision in respect to the selection of a control strategy has to be made within one control time step, generally between 2 to 15 minutes. In order to select the strategy, each of the models have to be run consecutively several times within the duration of the control loop and despite the development of more powerful small computers, the computational speed of each module is of prime importance. The simulation models developed for design purposes have to be able to represent the detailed behavior of the feature being modeled. The models required for RTC work can be empirical, gray or black box models since the physical characteristics of the system are not to be altered. There are many modeling techniques available for creating suitable empirical models. These include the use of linear transfer function models, PRTF models (Han, 1991), CPTF models (Yuan, 1994) or non-linear neural network models. In general, empirical models are more applicable to RTC work than mechanistic models since they are faster running and since they posses fewer degrees of freedom, simpler to calibrate and adjust or to modify and recalibrate. Flow surveys will have to be conducted as part of the RTC scheme and the data produced are used to calibrate and prove whatever type of model has been selected for the simulation of the UDS.

Real-Time Urban Drainage System Modeling

The UDS model uses the telemetered data from the physical system together with the forecast hydrological data to simulate the influence of the selected control strategy on the overall system. The outputs from this model are used to calculate notional operating costs, CSO volumes, flooding impact and the loading of receiving waters. For this model to meet these requirements it must be capable of simulating the changes that take place as a result of a control action such as a pump starting or a penstock opening. Although dynamic models have been used for more than 20 years, the rapid simulation of a large UDS has not yet been achieved. The conventional model uses a mechanistic approach solving variations of the unsteady flow equations and pipe flow equations using finite difference techniques. Unfortunately, in order to maintain stability, the time step has to be small which significantly increases the model run time, or the model has to use simplified hydraulic relationships which ignore some of the physical behavior of the flow but allows larger time steps. Some attempts have been made to use empirical models to simulate UDSs (Beck, 1977) but these have attempted to use a single empirical model for the whole system and although these have been partially successful, they are unable to model changes within the system caused by the operation of a control device.

At Salford a hybrid approach has been adopted which combines the speed of the empirical model with the ability to simulate complex and non-linear aspects of the dynamic/mechanistic models. The UDS is divided into different subsystems which can either be represented by a dynamic model or an empirical model is constructed from field data to represent the response of that part of the system. This approach is shown in Figure 7. Computer software called ëthe knowledge acquisition moduleí (KAM) has been developed to divide an existing WALLRUS (Department of the Environment/National Water Council, 1992) or MOUSE model of a complete UDS into various subsystems, dynamically model them off-line and then use the results of the simulation to develop one of a variety of empirical models. The KAM then reconnects these simplified models to provide a representation of the entire system.

Fig.7 - Developing empirical models of the urban drainage system

Research is still being undertaken to establish which empirical modeling techniques provide the most suitable models for RTC, but some success has been achieved with this approach (Norreys & Cluckie 1994). The success of this technique is shown in Figure 8 where the performance of two different modeling techniques are compared.

SUMMARY

A Global-Supervisory-Predictive RTC strategy should be adopted at the planning stage of an RTC system. The examination of the behavior of the RTC system currently operating in control of the Bolton UDS revealed that a successful predictive control strategy requires the application of rainfall data that is best obtained from a hydrological radar system.

Fig.8 - The performance of the empirical UDS model compared with a dynamic model

Such RTC systems will also need the development of real-time adaptive and accurate network simulation models. The spatial and temporal quality of the radar rainfall data makes it very suitable for the evaluation of the potential for RTC systems and when combined with other real-time data from the drainage system, the hydrological radar will provide the basis for a robust control capability for the system manager.

REFERENCES

Abes B. 1995,"Real-rime Control of Urban Drainage Systems Using Weather Radar", PhD Thesis, Department of Civil Engineering, University of Salford.

Beck M.B. , 1977, "The Identification and Adaptive Prediction of Urban Sewer Flows",. International Journal of Control, Vol. 25 No 3. pp. 425-440.

Cluckie, I. D., et. al., 1995, ëOperational Utilization of High Resolution Hydrological C-band Radar", III International Symposium On Hydrological Applications Of Weather Radars, August 20-23, 1995, Sao Paulo, Brazil.

Council of the European Communities, 1976, "Directive Concerning the Quality of Bathing Water. (76/160/EEC)". Official Journal of the European Community No. L31.

Department of the Environment/National Water Council (UK), 1983, The Wallingford Procedure, DOE/NWC Standing Technical committee Report No.28, Vol. 1 to 5.

Department of the Environment/National Water Council (UK), 1992, The Wallingford Procedure, The Wallingford Software, WALLRUS.

Griffith R.J. , 1995, "Weather Radar and Strategic Urban Drainage in Real-time", PhD thesis, Department of Civil Engineering, University of Salford.

Green, M. J., 1991, Benefits of Real-Time Control for Urban Systems. WRc Reports, UC 1045, pp71.

Han D., 1991, "Radar Information Processing and Real-Time Flow Forecasting" PhD Thesis, Department of Civil Engineering, University of Salford.

Norreys R.J., 1991, "Water Quality River Impact Model (RIM) for River Basin Management",. PhD Thesis, Department of Civil Engineering, University of Salford.

Norreys R.J. & Cluckie I.D., 1994, Neural Network Modeling of Urban Drainage Systems. Report for North West Water Ltd, Dept. Civil Engineering, University of Salford.

Schilling, W., 1994, "Smart Sewer Systems: Improved performance by Real-Time Control". European Water Pollution Control, Vol.4, Part 5. p24-31.

Yuan, J., 1994, Hydrological Modeling with Weather Radar Data in Urban Drainage Systems. PhD Thesis., Department of Civil Engineering, University of Salford., pp265.