Practical Application Of Neural Networks ToPredict DO Concentration

This paper presents a different alternative based on neural networks to predict dissolved oxygen concentration (DO) in water masses. This method allows the solution to mathematical models to be obtained more quickly, thus avoiding excessive computing times. With neural networks, non-linear systems can be modelled quite effectively, and time series prediction tasks can be carried out without the need for any excessively complicated calculations. In this paper, the first step has been to look for a suitable neural network model, the solution being found in feedforward backpropagation multilayer perceptron networks. Subsequently, the NevProp network simulator has been used to train and validate a set of networks which allow the time variation of the DO concentration in a specific environment to be predicted and studied.