Artificial neural network modeling of the water quality in intensive Litopenaeus vannamei shrimp tanks

We used a backpropagation neural network (BP-NN) model to predict the water quality in intensive (300 PLs/m(2)) Litopenaeus vannamei shrimp tanks. The model has a tan-sigmoid transfer function for the hidden layer and a linear transfer function for the output layer. It was developed using measured water quality data that were generated over 120 days (from 1 July to 28 October 2008) with weekly monitoring in four different shrimp tanks. Nine parameters were selected as input variables: water temperature, pH, total ammonia nitrogen, nitrite nitrogen, nitrate nitrogen, dissolved inorganic phosphorus, chlorophyll-a, chemical oxygen demand, and five-day biochemical oxygen demand. The Levenberg-Marquardt algorithm was used to overcome the shortcomings of the traditional BP algorithm; that is, low computational power and getting stuck in local minima. The number of hidden layer nodes was optimized by a trial and error approach, and five optimal neuron nodes were identified. The computed results for water quality show good agreement with the experimental values. The correlation coefficients of the training, testing, and training + testing sets between computed results and experimental values are 0.990, 0.979, and 0.992 respectively. The simulation results reveal that the BP-NN model efficiently predicts the water quality in intensive shrimp tanks. (C) 2014 Published by Elsevier B.V.