Msc Results

Master of science abstracts
Links to Moodle and library

Optimising the Apure river basin reservoir system using a multi-model approach

Avila Torres -

The present work deals with the optimization of reservoir operation in the Apure River Basin in Venezuela. Apure River Basin is one of the main tributary basins of the Orinoco River in Venezuela. In Apure River Basin there are three reservoirs whose main purpose of hydropower production and now there are plans to use their releases to increase the navigation period at a downstream reach.

The main objective of this study is to develop an approach which would allow construction of a control strategy for the operation of the hydropower multireservoir system. Under the control strategy we understand the time sequence of decisions (releases from the reservoirs) that should be taken to accomplish the operational objectives. This research presents a work where classical dynamic programming has been used to optimize the performance of a multipurpose multi reservoir system. The energy production of the multi reservoir system is affected by the discharges and net head at the reservoirs. The discharges from the reservoirs affect positively the downstream water levels. Therefore, they affect the navigation along the downstream reaches. To get adequate water levels for navigation an increase on the releases from the reservoirs is necessary during the dry periods in the subbasin. However, the increase of releases from the reservoirs to satisfy the navigation water levels reduces the energy production for the next period. Therefore, a tradeoff on these two opposite tendencies should be made to get the best out of them.

MIKE11 modelling system has been used to model the hydrodynamics of the Apure River. Also, the hydrological model NAM was used to generate inflows from the subcatchments. However, running MIg11 and NAM computation modules separately from their interfacing modules appeared impossible. This made effective optimization difficult. Because of that, the neural network tool NNN was used to model the behavior of the MIKE 11 -NAM model. The neural network was trained with the results of a hydrological-hydrodynamic model to mimic the behavior of the river at the specific locations in which there is a navigation problem. The use of NNN allowed to generate a standalone program with a quick response so that it could be used in the optimization loop.

For optimization a dynamic programming approach algorithm has been implemented using Pascal language. Such approach allowed to take into consideration the conflicting objectives of the power production and navigability. The objective function was the squared deviation of the difference between the target power and the generated power at every power plant. The navigability requirements have been put into constraints. Traditionally, the use of the dynamic programming approach suffers from two major disadvantages. First, the so called "curse of dimensionality," which means an increase on the amount of high speed memory and running time. Secondly, the dynamic programming optimization relies heavily on the objective function. These two problems are addressed in the report.

A finite difference model for simulation of short wave behaviour

Babak Banijamali -

The undertaken research may be broadly divided into the study of the .physical aspects of Boussinesq-like modelling in a generalised sense, on the one hand, and the investigation of the performance envelope of a proposed three-level scheme, in one and two horizontal dimensions, on the other. The one-dimensional model was shown (in theory and in practice) to be relatively advantageous, both from accuracy and algorithmic points of view. Specific short wave comparisons to reported results were performed in the process. The two-dimensional generalisation was linearly-analyzed, implemented and run to vouch for its favourable stability behaviour. The inherent algorithmic merits of a three-level scheme, particularly for Boussinesq-like modelling, were elucidated. The preliminary tests to investigate the two-dimensional model were performed in agreement with the hypothesis that the generalisation of the model to a final, fine-tuned, industrial model may be a very plausible future option.

Design and implementation of a modelling system for water distribution network: An object oriented approach

Bogale M. Mariam -

An object-oriented programming approach to the development of hydraulic simulation models is attracting increasing attention. The potential of object-oriented technology lies basically on the promise to produce more maintainable and extendable information tools with less effort and expense. The implementation of an object-oriented modelling environment built upon the characteristics of object-oriented programming results in a simulation system that may greatly improve the capabilities of hydroinformatics systems. Current simulation systems deficiencies include lack of modularity, minimal provision for model component reuse, often limited flexibility, and complex model specification format. Object-oriented modelling systems provide improved capabilities in these areas while maintaining a comparable or reduced design effort.

In the framework of the current M.Sc study the prototype architecture for a modelling environment has been designed. It includes graphical features for logical configuration of simulation entities and icon based capabilities for system definition. To quantify the purpose, prototype construction of a water distribution network simulation model was selected as a case study. The system has been implemented in Object-oriented Borland Pascal under Microsoft Windows operating environment. The choice of object-oriented methods for the development of mode]ling tools of this class proven to be justified, resulting in the open flexible architecture of modelling system, and the use of natural object-based representation of a water distribution network

Data Assimilation and Parameter Estimation in a 2-D Advection-Dispersion Model

R. Canizares -

Inspired by previous studies carried out in the fields of Meteorology and oceanography, the interest of Data Assimilation in Hydraulic modelling is growing considerably. The possibility to apply new technologies with the observation of the natural processes has increased the amount of data that can be incorporated into the models.

Numerical models provide a very useful tool for studying hydrodynamic processes. But for numerical models to be useful, they must be proven to be valid for the numerical dynamics and initialisation parameters. This can only be done by referencing with known observations of the system. With the aim of interfacing model and data efficiently, the so called data assimilation methods must be applied.

The objective of this study is the application of a Kalman filter to the 2-D modelling system MIKE 21 in its Advection-Dispersion module, in order to assimilate data and correct some of the initialization parameters, such as the diffusion coefficient or some unknown source concentration.

When the Kalman filter is applied to assimilate data by correcting the model output, the filter can be time invariant under certain circumstances. To avoid the large computational time associated with time dependent Kalman filter applications the filter is applied on the diffusion part of the transport equation only. Under this assumption the Kalman filter is time invariant when the diffusion coefficient is constant in time and once the initial assumption errors have been removed. The time invariant Kalman filter could also be calculated off-line, using special algorithms.

When the Kalman filter is applied for parameter estimation, the filter becomes time dependent. In this case the Kalman filter must be recalculated at every time step, until the parameter is estimated.

The performance of the filter is tested by assimilating known information into two models: i) a channel where advection predominates and ii) the Donagal Bay model with semi diurnal water level fluctuation at the boundary. The results show very good performance of the Kalman filter in the correction of model output and for parameter estimation. An unknown diffusion coefficient and an unknown source concentration have been estimated successullly. Parameter estimation when the measurements are corrupted by noise has been also studied, with different degrees of success.

Calibration of Hydrodynamic Numerical Models Using Global Optimization Techniques

Anghel Constantinescu -

A major decision to be taken when constructing a mathematical model of natural phenomena regards the values of model parameters. The problem is known to be complex, due to the lack of "perfect" scientific knowledge, complicated structure of processes and phenomena involved, measurement and human errors in data collection and processing, etc.

For hydrodynamic models the parameters to be calibrated fall into two categories:

  • physical, such as roughness, parameters of structures, wind, rain, hydraulic transmissivity, tidal amplitude, etc.; in many cases there is not enough recorded data;
  • numerical, such as time-step, space-step and others.

Once the set of parameters governing the behaviour of the model is found, the model can be used to predict a number of events that differ reasonably widely from the known event(s).

This study addresses the general problem of model calibration, discusses the major difficulties one can meet during calibration trials, and model error functions used. Main attention is given to the problem of automatic calibration, using optimization techniques.

The available software (the prototype of a global optimization tool GLOBE including several algorithms) was used to calibrate the two different hydrodynamic numerical models, dealing respectively with

  1. evolution of chloride concentration in the lake Breukeleveense Plas;
  2. tidal driven 2-D depth-averaged flow in the Gulf of Thailand.

Software coupling the optimization tool GLOBE with each hydrodynamic model was developed.

The findings of general validity, along with specific aspects of automatic calibration, are presented. Different global optimization algorithms are compared in its efficiency for solving calibration problems. The results show that the use of global optimization routines allows for efficient and effective calibration of models of different nature. Recommendations for the further research are given.

Elements of object-oriented simulators

P. Ingeduld -

Object-oriented design and object-oriented programming is the direct response to the increasing complexity of modern software applications, complexity that requires designers and programmers to change their habits, programming styles and their way of thinking. Inheritance and encapsulation are extremely effective means for managing complexity. An improvement in the quality of scientific software systems can be achieved by implementing the technology of object- oriented design.

The objective of this study is a trial attempt to use this technology for numerical simulators. The example of object-oriented simulator of one-dimensional unsteady pressurised flow is given to demonstrate the possible advantages and limitations of such a design.

The study is divided into five chapters. Chapter 1 is an introduction to the problem. Chapter 2 describes the common features of object-oriented design and object-oriented programming. Some of the object-based and-object-oriented languages are discussed and analysed. Chapter 3 gives the brief theoretical consideration of the problem of one-dimensional unsteady flow in a pipe network and shows the algorithmic incorporation of hydraulic structures into the respective numerical scheme. Chapter 4 gives a variant of the conceptual design of object-oriented numerical simulator and shows the advantages of this representation - modularity of the system being decomposed into the set of cohesive and loosely coupled modules. Chapter 5 describes practical aspects of such a simulator using Borland Pascal object-oriented language. This implementation follows the steps taken to create an object hierarchy of hydraulic structures (valves), identifying both the class and object hierarchies within a complex software system. In such an implementation, hydraulic structures are represented as software objects sending messages to each other.

The advantages and the main risks of object-oriented design are discussed in the Conclusion of this study. The approach proved to have the good potential for developing flexible, adaptable numerical simulator with open architecture.

Optimal Real-Time Control of Unsteady Combined Sewer Flow.

Hector Martin Garcia -

Two new HP-C/XWindows/Motif applications have been developed to be used in the Mouse Simulator. The Mouse Simulator is the simulation tool for Combined Sewer Systems in one of the pioneers among hydroinformatic systems: The Mouse Online, under development and testing in the Danish Hydraulic Institute in Denmark.

These applications are:

The Distributed Rain Feature (DRF) for spatial distribution of the rain in the catchment and the Sensor Drift Feature (SDF) for simulation of sensor malfunctions in the Combined Sewer System.

An exhaustive analysis was performed for the Combined Sewer Systems in Aalborg (Denmark) and Goteborg (Sweden). The conclusions derived from this snslysis allowed for the first time to reach exact numeric integrated results describing volumes of untreated overflows into the surrounding natural recipients. On these basis, quantitative criteria for the selection of the best available optimization routine, were given. The benefits of introducing a Mouse Online based control (in relation to more traditional solutions) in a specific city catchment are described as well.

The importance of including features such as the Distributed Rain Feature for a full understanding of the hydraulic response of the Combined Sewer System was also proved. This understanding is a decisive element for finding the optimal control sequence for the flow regulators in the system.

Some preliminary work was also performed in order to extend the existing Rule Based optimization routine providing it with non-site specific knowledge which, in principle could be applied to any Combined Sewer System.

A general outline of the future development perspectives of the Mouse Online hydroinformatic system (for a optimal, real-time control of these systems) is also provided.

WADES: Modelling System for Water Distribution Networks.

T. Nada -

The objective of this study was to develop and design a modelling system for simulation and identification of parameters in water distribution systems. The solution to a problem of a steady-flow in a pipe network is obtained when the flows satisfy both the continuity equation at each junction and the energy equation in each pipe. The governing equations are nonlinear. Neither of the nonlinear systems of equations can be solved directly, so iterative methods are employed.

The choice of the method for linearizing the nonlinear terms is essential, since no method can guarantee for all possible network configurations. In the modelling system WADES which has been developed in the framework of this study, the following methods have been implemented:

  • The Linear theory method
  • The Newton-Raphson method
  • The combination of the two previous ones
  • The solution method based on unsteady state analysis

The method combining the Newton-Raphson procedure with the linear theory method as a simple and robust starting procedure, showed to be especially robust and efficient, and guaranteed to converge in an expeditious manner.

Algorithm for identification of design, operating, or calibration parameters of the network, is also implemented in the modelling system. The algorithm is able to calculate directly the parameters required to meet specified operating conditions. Specified operating conditions represent stated minimum pressure and volumetric flow requirements at designated locations throughout the distribution system. Solutions are optimal in the sense that they exactly meet the specified constraints.

The feasibility, accuracy, and flexibility of WADES algorithms is examined using variety of networks, including Hodeidah water distribution system in Yemen with 182 pipes and 142 demand nodes.

The resulting WAter Distribution modElling System (WADES) is implemented in Borland Pascal programming language (around 4000 lines of Pascal code) under the Turbo Vision interface development framework, and allows for the graphical presentation of the analyzed networks.

Flow simulation with cellular automata

J. Savioli -

Water is and in fact has always, been a precious resource in our world, indeed more and more regions in our planet suffer a lack of quantity and quality of water. This awareness led to the concept that there must be a further understanding of the behaviour of water systems.

Nowadays there is a revolutionary technological process being introduced; the use of digital computers and the consequent software decompositions (algorithmic and object-oriented), changed the conceptualisation of the problems that have been posed and they caused new solution methods to be introduced, indeed there is a change in the whole way of thinking about the world. In hydraulics, this process started with an emerging discipline called computational hydraulics. This was the beginning of an era in which problems were formulated specifically to be solved with digital computers, many approaches were introduced and today it is considered as a viable independent branch of hydraulics.

This new opportunity allowed a new examination to be made of various physical systems that had been described previously along pre-computational lines. Sometimes results were as could have been expected from earlier works but often there were situations in which the results could not have been imagined. Such was the discovery of chaotic systems, with it impacts on classical physics. Indeed chaotic systems were detected when computer simulations became available, and it was such simulation as that allowed chaotic systems to be investigated. In hydraulics there are highly chaotic system; turbulent flows are clear examples of such a behaviour.

Fundamental for the development of new methods is the investigation of models which are simple in construction, yet capture the essential mathematics features necessary to reproduce the complexity which is observed in nature. Cellular automata are some of the most simply oredered chaotic systems known, even simple one-dimensional cellular automata can exhibit chaotic behaviour and sometimes produce self-similar patterns known as fractals. Although cellular automata systems are interesting in their own right, the centre of interest has shifted from their general properties to the possibilities of using them to simulate a variety of systems such as fluid flow, diffusion, crystal growth, biological and ecological behaviour, and so on The main idea behind this study is the construction of a completely object-oriented program based on cellular automaton rules to provide a realistic flow and hence of analysing general flow patterns as well as the general behaviour of the models in order to bring out some some general conclusions. Moreover an investigation of simple matters that appear differently such as the introduction of wall roughness, boundary conditions, initial conditions and other aspects.

Model and parameter analysis techniques

Shen Yiyang -

For environmental, hydraulic or hydrological modeling and forecasting, due to various sources of error and uncertainty of both data sets and models, appropriate analysis and estimation techniques should be applied. From this viewpoint, this study deals with techniques of data set transformation, uncertain model parameters, model calibration techniques, artificial neural netvork application, criterions and other statistical components. The aim of the analysis is to obtain useful practical information from these data- sets; the information subtracted from the data-sets must help the user in the interpretation of the model's results, or, from a more general viewpoint, understand the properties/characteristics of the models and the underlying physical processes.

The following techniques/methodologies are studied in this thesis:

  • Discussion of model and data uncertainties;
  • Data series transformation and scaling;
  • Dataset selection for model calibration and verification (here proper criterions must be defined);
  • Artificial neural networks (ANN) application and a method to avoid overfitting of ANN;
  • Global optimization techniques;
  • General statistical/correlation techniques;
  • Collection and discussion of model simulation error functions;
  • Model sensitivity analysis for parameter variations;
  • Model validation techniques;
  • Few result presentation/visualization techniques.

As examples of the methods mentioned above, we have two applications in this work:

A Decision Support System for Environmental Assessment

Suzana Simic -

The environment in which management must operate, is changing very rapidly in the direction of an increasing complexity which makes decision-making harder than ever before. In order to make rational, informed decisions on complex issues, the capacity to aggregate, integrate, reconcile, organize and communicate knowledge across domains is essential. Clearly, there is a need for tools which support decision-making.

This report gives an overview of the development of a Decision Support System for environmental assessment. Although the assessments' characteristic that is most emphasized is that of being highly context-specific, as shaped by the problem itself, it's purpose and it's audience, assessment is more about integrating and communicating knowledge across domains than about knowledge of the domains themselves. This makes it possible for some of the basic environmental assessment activities to be abstracted from the specific problem context. These activities can be used as a core for building generic environmental assessment tools.

One of the main requirements for DSS to be successful is its ability to change as new relevant knowledge becomes available. New knowledge has two main sources: the decision-making process itself (evolution of the understanding of the problem) and new knowledge about relevant domains.

Prototyping is an approach which provides the most benefits in the development of a DSS which has analysis, design, construction and implementation combined into a single step that is then repeated. By disaggregating the initial problem, one can select an important subproblem which should be built first. The subproblem should be small enough that any support needed can be identified and a simple but usable system can be developed. Expanding, refinement and modifying are performed in cycles. When the user is satisfied with the prototype, the same procedure is applied to the other subproblems.

The Environmental Evaluation System, developed by Battelle Columbus Laboratories, has been selected as a core of the present DSS for environmental assessment. Two DSS prototypes have been built, named CASCADE version 0.1 and version 0.2. Flexibility is provided by storing both static domain knowledge and dynamic system information in relational databases. All tables are indexed, providing a stable internal environment regardless of the underlying information. The system has been built for the user-friendly MS-Windows environment. A context-sensitive hypertext help system has also been built so as to guide the user both in the application and in the understanding of the methodology.

Development of a real-time optimal control system for drainage using an object-oriented approach

Mr. TAN, Tien Ser -

The objective of this study has been to develop and design the prototype of a real-time optimal control system for drainage. The system incorporates a number of mathematical models for flood forecasting, optimal control and supervisory functions. The models implemented in the current work include: unit hydrograph rainfall-runoff model to estimate the inflows into the drainage channel; tide forecasting model using harmonic analysis equation to predict the tide levels at the sea; flow routing model using the de Saint Venant equations to forecast the water levels and flows in the drainage system; dynamic programming optimization model for real-time control and decision support; rule-based expert system for alarm activation.

The review of the previous research work and existing real-time control systems has revealed, that in water resources real-time control, dynamic programming optimization techniques have not been widely used. However, dynamic programming paradigm is ideally suited to solve problems of a stagewise nature as in this study.

The alarm model is implemented as an expert system, where the rules can be modified by the rule editor. As one of the possible enhancements, the experience knowledge base component can be introduced in the future; it would accumulate the data on the system operation, and then the rule base will be trained accordingly.

The suitable mathematical models have been selected and implemented in highperformance computer code, coupled into a cohesive system that performs efficiently enough to be used in real-time application.

Object-oriented methodologies have been applied for the whole cycle of software development; from analysis to design and to implementation. The choice of object-orientation in the present study has proven to be the right choice.

One of the important components of a real-time system is the ergonomical man-machine interface. The use of MS Windows environment and object-oriented programming framework ObjectWindows allowed to construct an efficient graphical interface that suits the needs of the system application.

In general, the objective of the study has been achieved - the system, called DRACOS, has been designed and implemented (the software components written by the author include 6000 lines of Pascal code).

A Classifier System for the Control of an Urban Drainage Network

G. Wilson -

This report gives an overview of the development of a new general control strategy selection echnique for real time control. The technique is a learning classifier system. An application on a large urban drainage system is presented. It is, however, also applicable for many other types of engineering planning problems and possibly represents a watershed in the methodology of engineering control through learning. The learning classifier system is an if : then rule-based system and as such responds almost immediately. It is therefore particularly appropriate for real time and model based control.

Site specificity is reflected in the hydraulic behaviour and the operational objectives of the urban drainage system. In this technique, the system's objectives are represented by a general cost function where an optimal control strategy is one which minimises the cost function. The hydraulic behaviour of the urban drainage system is simulated by a fully dynamic numerical model. In addition, the technique needs no a-priori or specific domain knowledge. As such the learning classifier system is very general.

A control strategy is the time sequence of action decisions performed to achieve the operational objectives. The learning classifier system learns a state x action => cost prediction mapping. Each rule carries with it a cost prediction. Action decisions are based on cost predictions which are learned without supervision. Selecting the action predicting the smallest cost will lead to the achievement of the objectives. The classifier system searches the solution space, updates its cost predictions and learns the mapping. It is shown that this technique parallels in many ways with the stochastic shortest route dynamic programming model. The cost each rule predicts is the sum of discounted future costs associated with the rule's then action. Future costs are discounted in time due to the attenuation of cost prediction certainty in time.

The learning classifier system has been tested by learning to control the very large Goteborg urban drainage system. Results obtained to date demonstrate the potential for possible commercial application for real time control, especially with the developments recommended. The learning itself can take weeks and demands large computing power; however, with today's computers this is merely an inconvenience.

Automatic Model Calibration by Simulating Evolution

Zheng-Yi Wu -

Nature evolution and self-adaptive mechanism has been schematized as an optimization process by introducing different working spaces and environments, which adapts the artificial system to its physical environment to obtain the highest payoff. The algorithm, gleamed from the nature mechanism, called genetic algorithm, has been successfully developed and applied to MOUSE model calibration in the research project.

The model calibration is specified as the minimization of the discrepancy between the simulated and the measured, which is furthermore formulated as three different distance functions. Before the genetic algorithm developed for the minimization purpose is applied to the calibration, it has been tested by using the typical numerical optimization function to find out the best combination of the genetic operators namely uniform crossover with scaling technique.

The results of the calibration tests and a real case calibration Aalborg sewer system calibration have shown that the genetic algorithm is very efficient for improvement and can easily circumvents the curse of the high dimensionality and nonlinearity of optimization problem, but the calibration depends on the data quality and quantity.