Neural network based decision support: modelling and simulation of water distribution networks.

Gabrys, B., 1997. Neural network based decision support: modelling and simulation of water distribution networks. PhD, Nottingham Trent University.

10183008.pdf - Published version

Download (26MB) | Preview


This work concerns the applicability of neural networks for the implementation of decision support (DS) systems in operational control of industrial processes. Decision support has two distinct but closely interrelated aspects: mathematical modelling of physical plants and processes, and the modelling of a decision making process. The first one forms the basis for detailed optimization of operations and the second one attempts to mimic an abstract mental reasoning about plant's operation by human operators. This research attempts to integrate both aspects of decision support within a single computational framework of neural networks. The prototype DS system is validated using case-studies taken from the water industry. The optimal control of water systems is a challenging problem because the models are non-linear and large-scale and measurements are noisy and frequently incomplete. The results of this research are general and are directly applicable to other systems, for example, gas and power utilities or road traffic systems.

In the first part of the project the neural network approach to the state estimation problem and confidence limit analysis for water systems is proposed. Since state estimation process is a computationally demanding task new approaches to solving it are constantly being looked for. The neural networks are one of the possible options. The resulting algorithms are the mixture of the well known and tested ways of solving systems of nonlinear equations (the Newton-Raphson method), the optimization criterions (the LS, LAV and their variations) and a relatively new artificial neural network (ANN) technique of finding the solution to the overdetermined systems of linear equations. The problems of bad data rejection, ill-conditioning, arriving at the solution within a predefined period of time are addressed and suitable ANN techniques are proposed and evaluated.

No state estimator can give accurate results from inaccurate data. A way of utilising the neural networks, that have been used to produce the state estimates, for quantifying the measurement uncertainty impact on the state estimates is shown. Two methods of obtaining the confidence limits in form of upper and lower bounds for each state estimate are investigated. The first method presents the usage of neural networks to find the sensitivity matrix which enables calculation of these bounds. In broader terms the way of finding inverse and pseudoinverse matrices, using ANN, is shown. The second method utilizes the superposition principle where each disturbance is analysed separately and the partial results are gradually combined to produce overall confidence limits.

Finally an integrated neural based system for state estimation and confidence limit analysis has been developed and tested for realistic water distribution network.

The second part of this project concerns the development of flexible fuzzy neural recognition system and its application to water systems' state interpretation task.

First a new general fuzzy neural network for clustering and classification is proposed. It can process both deterministic and fuzzy input patterns, combines the supervised and unsupervised learning techniques within a single training algorithm, grows to meet the demands of the problem and learns on-line.

The problems of fault diagnosis and water state interpretation are then addressed. A completely new approach to bad data detection and identification in water systems based on pattern examinations using neural recognition system is demonstrated. The use of state estimates and residuals with corresponding confidence limits is examined. The extensive performance studies for 24 hour of water network operations with particular emphasis on detection and correct location of leakages are carried out.

Item Type: Thesis
Creators: Gabrys, B.
Date: 1997
ISBN: 9781369313062
Divisions: Schools > School of Science and Technology
Record created by: Jeremy Silvester
Date Added: 28 Aug 2020 14:28
Last Modified: 15 Jun 2023 09:48

Actions (login required)

Edit View Edit View


Views per month over past year


Downloads per month over past year