*3.2. Results of Wavelet Transform Model*

Figure 8 shows the original images of DO, CODMn, TP and NH3-N and the images after db5 wavelet decomposition.

However, after summing the data signals of each frequency band after wavelet reconstruction into the original signal, there is a certain error between this reduced data and

the original data that is determined by the characteristics of the computer and is an error that cannot be eliminated. If this error value is too large, the experimental results will not be credible. Therefore, in this research, in order to verify the experimental accuracy of the model, the difference between the reduced data and the original data after wavelet reconstruction of the four indicators in the experiment was calculated. The errors are shown in Table 4.

**Figure 7.** Residual images of ANN training on DO, CODMn, and NH3-N datasets.

**Figure 8.** Original images and three-layer decomposition images of DO, CODMn, TP, and NH3-N datasets.

From Table 4, it can be found that the error of CODMn was the largest (6.75 × <sup>10</sup><sup>−</sup>16) and the error of DO was the smallest (9.1 × <sup>10</sup><sup>−</sup>16). An error value in this range interval has a negligible effect on the experimental results, which proves the validity of the experiment.
