| one publication added to basket [82333] | A methodology for an adaptive real time control for flood mitigation based on soft computing techniques
Debebe, A. (1999). A methodology for an adaptive real time control for flood mitigation based on soft computing techniques. PhD Thesis. Vrije Universiteit Brussel. Vakgroep Hydrologie en Waterbouwkunde: Brussel. XV, 216 pp.
|
| Available in | Author |
|
Document type: Dissertation
|
| Keywords |
Control > Flood control
Flooding
Methodology
Real time control
Simulation models
Soft computing
|
| Abstract |
In a time where many speak about effects of global warming as potential sources of extreme weather conditions, flooding has become more threatening to human kind than ever before. This is even more so in big cities where the problem takes another dimension: pollution of the aquatic environment due to combined sewer overflows. Many efforts including construction of either detention basins or some protecting measures as dikes and levees have been used in an effort to reduce these damages.
However, since the construction costs of most of these structures are generally enormous, engineers have been researching possibilities of using real time control in order to achieve an optimized use of existing structures or minimizing costs of new ones.
Although a number of real time control methodologies exist for this purpose, most of them rely on static models for setting control actions. In addition, since most of them are based on numerical optimization techniques, they do not guarantee global optimum and the optimization techniques are often time consuming. Moreover, these models are highly data specific and their operational performance depends upon the science used to build and operate them as well as their ability to respond to dynamic and rapidly changing events.
An adaptive controller based on soft computing (intelligent) techniques which, as opposed to the above mentioned type of controllers, offer a more flexible, self-adaptive and less assumption dependent approach, is presented in this research. To this effect, fuzzy logic controllers coupled with artificial neural networks and genetic algorithms are used for prediction of future events and controlling the system in real time. Adaptivity to the changing behavior of the hydrologic and hydraulic systems is derived from the learning capabilities of artificial neural networks and through genetic algorithms and numerical optimization techniques.
A continuous simulation model is developed in order to mimic the hydrologic system to be controlled. The model consists of various nested blocks through which the drainage system is represented. These blocks include subcatchments, reservoirs, weirs and other structures. The overland flow from the subcatchments is simulated through a conceptual mathematical model. the Nash Cascade, after losses through wetting, depression storage and infiltration are abstracted. A time varying runoff coefficient is used for the abstraction of depression storage and continuous losses. Several types of reservoirs are implemented where most of them have outflows that can be controlled in real time. The implementation of the model permits changes to parameters while simulations are underway, allowing the testing of different control strategies on the fly without requiring the model to be terminated.
Future flows for different horizons are predicted from current rainfall information and past and present inflows using models based on fuzzy logic systems and artificial neural networks. Two approaches were followed while developing these models. Firstly, fuzzy mean clustering methods were applied to design the membership functions and generate the necessary rules. This gave fairly accurate predictions of flows at different forecast horizons. However. the membership functions for the inputs at different lag times could not have the same meaning. which made extraction of hydrologic knowledge and logic difficult.
A second approach, where the necessary membership functions and rules are designed in a semi manual way was then implemented. This resulted in a less accurate model, which needed the tuning of some of its parameters. In order to preserve the classification of the data into the different membership functions, and hence give the membership functions similar meanings in all the inputs involved, only the consequent part of the rules is fine tuned using the techniques of artificial neural network, genetic algorithms and numerical optimization techniques.
An adaptive fuzzy logic controller is then developed which takes the states of the controllable reservoirs and predicted inflows into the reservoirs and any uncontrolled side flows and sets optimal releases from each reservoir. Since the desired outputs, the optimal releases from the controllable reservoirs, are not available in advance, automated ways of generating the controller parameters are not directly applicable. Two approaches were followed in order to tackle this problem. In the first, an optimal controller based on genetic algorithms and numerical optimization techniques, was developed and used which resulted in optimal releases at every time step. These releases are then used to fine, tune the fuzzy controller, which is designed manually based on a knowledge attained from the previous step. The second approach was to use these optimization techniques to tune the manually designed fuzzy logic controller online. The extension of this last approach gives the controller the capability to adapt itself to the changing behavior of the hydrologic system.
Application of the controller was performed on a hypothetical catchment and a real rural catchment in Belgium, the Vondelbeek catchment. An application on the Vondelbeek catchment, which is situated Northwest of Brussels showed that overflows, which are inevitable under a no' control scenario could be avoided completely through the controller proposed in this research. |
|
