I discussed a master’s thesis at the College of Engineering at the University of Basrah (prediction of dissolved oxygen in the Shatt Al-Arab River “Iraq” using artificial neural networks)
The thesis presented by the student Zainab Abbas Khudair dealt with the evaluation, training and testing of the application of artificial neural network models to model the water quality of the Shatt Al-Arab "Basra - Iraq", and it dealt with the use of the ANN model to predict the dissolved oxygen in the river. The ANN model is performed using experimental data taken from the Ministry of Environment during the period 2009- 2014
The forward propagation algorithm was used to train all models, (ANN) was trained using the Levenberg-Marquardt algorithm with different types of algorithms (training equations), to get the best performance of the proposed model. A trial and error method was used to determine the final ANN architecture. The results of the ANN models were selected based on the correlation coefficient (R) and mean square error (MSE). Also, using two hidden layers gave better results than one hidden layer. Three models of artificial neural network studies were studied in predicting dissolved oxygen in the Shatt al-Arab.
In the first pilot study, the proposed model (ANN-M1) evaluated and estimated the water quality of the Shatt Al-Arab River. A total number of (538) samples were collected as a data set and 80% and 20% of the data sets were ANN trained and tested.
While in the second model study, the second proposed model (ANN-M2) was the evaluation and prediction of dissolved oxygen in the Shatt al-Arab River. A total of (570) samples were collected as a data set. In the third pilot study, the proposed model (ANN-M3) was used to predict dissolved oxygen in the Shatt Al-Arab River. (32) samples were collected as a data set for the current year.
In the fourth model study (ANN-M4) it was used based on the data measured in the current study of 2019. In this model, the dissolved oxygen concentration was predicted based on the relationship between dissolved oxygen (which represents the output variable of the ANN model) and variables such as temperature , pH, turbidity, conductivity, chemical oxygen demand, biochemical oxygen demand, phosphate, and total dissolved solids (representing the input variables for the ANN model). A total number of (32) samples were collected as a data set containing 24 data for the training data set. and 8 data for the test data set for the current study
The aim of the thesis is to predict the concentration of dissolved oxygen using numerical or statistical models such as those of artificial neural networks (ANNs).
The thesis concluded in the first case. The Levenberg-Marquardt (trainlm) training function was the best among the other training functions that were used in this study with one hidden layer and two hidden layers.. As for the second case, the results showed that the neurons used [14,9] are in two layers. Two hidden samples were the best result with training: (correlation = 0,99982 and mean square error = 0.00003494) Test: (correlation = 0,9686 and mean square error = 0.00235.) This model was very effective and better than the first model.
The third case also concluded that using [10,8] neurons with two hidden layers was the best result (and also training: (correlation = 1, mean square error = 9,684 * 10-30) test: (correlation = 0,9943, mean error square = 0). ,0233) indicates that the proposed neural network model gave very good prediction and more accurate results.
And in the fourth case, the results showed the minimum mean square error = 6.145 * 10-17 and correlation = 1 for training and 0.91837 for testing.
