*3.1. Artificial Neural Network Interpolation*

In this study, we used MATLAB to construct a multilayer perceptron neural network to fill in the missing data in each dataset. The parameters of MLP were set as follows: the number of implied layers was two, the optimization algorithm was the conjugate scalar gradient algorithm, and the minimum relative change in the training error rate was 0.001. The artificial neural network training images for the DO, CODMn, and NH3-N datasets are shown in the Figure 7.

As shown in Figure 7, the coefficient of determination was 0.99488 for the DO dataset, 0.99317 for the CODMn dataset, and 0.99525 for the NH3-N dataset. It is clear that the fit of each dataset was good. Therefore, the model could be used to estimate the missing values in the DO, CODMn, and NH3-N datasets.
