Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.519433
Title: A Neural Network Computer Model of the Hydrodynamical Flow in the River Medway Estuary atits Confluence with the River Thames
Author: Rees, Lyn Hugh
Awarding Body: London Metropolitan University
Current Institution: London Metropolitan University
Date of Award: 2008
Availability of Full Text:
Access through EThOS:
Abstract:
A one dimensional, linearized, shallow water model finite difference scheme was developed to generate data representing both depth averaged velocities and depth fluctuations above or below the still water level in a river. After applying suitable boundary conditions based on the theory of characteristics, the model was then tested against another numerical model. An artificial neural network (ANN) model for both depth and velocity with zero bottom friction was designed to use as a precursor to a full friction model. The model was extensively trained and tested over a 600 Km length, using generated data, to obtain information on the optimum structure of the neural network and various parameters. The model was then finally trained and validated over a 1200 Km length to avoid the danger of overfitting. Using this frictionless model, it was extended to incorporate the effects of bottom friction. However, it was observed that the ANN was incapable of simulating rapid changes in the data close to the downstream boundary because of possible conflict between the nonlinearized bottom friction and linearized boundary conditions. To overcome this difficulty, the standard bipolar activation function was replaced by a modified LeCun activation function. Subsequently, the neural networks were then re-trained and re-validated. Prior to applying the ANNs to the confluence of the rivers Thames and Medway, the networks were tested for their adaptability to a variation of certain parameters. The models demonstrated good universal approximation capabilities when varying the imposed velocities, still water depths and friction coefficients. Apart from minor discrepancies in generated depth and velocity data at the precise juncture of the two rivers, the networks showed more than adequate performance when simulating the flow in the two rivers.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.519433  DOI: Not available
Share: