The Chirostoma estor estor, commonly known as white fish, is an endemic species of great importance in the area of ??Lake Patzcuaro, located in the state of Michoacan, Mexico. This represents a large income in the local economy, unfortunately it is currently in danger of extinction due to the pollution of the lake and its excessive fishing. National institutions, through aquaculture farming, have made efforts in the conservation of this type of fish, however there is still a lack of penetration of the scientific sector in this problem. This paper presents a computational model for the evaluation of non-ionized ammonium which is highly toxic and of vital importance in aquaculture farming systems. Using an Artificial Neural Network (ANN), a relationship is established between non-ionized ammonium (NIA) and parameters such as pH, temperature and total ammonia (TAN). Data bases obtained from measurements in culture ponds and generated by similar models have been proposed to generate efficient RNA training.
The computational model is divided into four stages (figure 1), in the first a database was simulated using the model proposed by Kennet. In the second stage it has a pre-processing so that the efficiency of the ANN is not diminished by a poor selecti on of patterns. The third stage consists of an ANN whose topology is given by the expression , which indicates that it consists of two inputs, a number of hidden layers with neurons in each hidden layer and one output. Finally, in the last stage the value of NIA is obtained.
The lowest MSE was the criterion for the choice of ANN configuration, both in the training set and the test set, and this was obtained with the topology . The results obtained by the training of the RNA show a learning rate of 99.91% of success in the approach. In the visualization of these results it is important to underline that each parameter (temperature and pH) individually affects differently the behavior of the NIA, for which the evaluation of the RNA for each of these was carried out individually (figure 2).
.
.
.
.
.
.
. . .
Fig. 1. Computational model for the evaluation of the NIA based on artificial neural networks.
F ig. 2. Comparison of the NIA simulated with the ANN NIA, for pH and temperature.