Modeling amicarbazone degradation in the O3/UV process using artificial neural networks

Recently, an increase in the contamination of water bodies by industrial chemicals, including herbicides, has been observed. In this context, advanced oxidation processes (AOPs), such as ultraviolet radiation-assisted ozonation ( $$\:{\text{O}}_{3}/\text{U}\text{V}$$ ), are being considered a promising solution for removing toxic pollutants. In general, AOPs involve complex chemical reaction mechanisms and may also include photochemical reactions and mass transfer steps. Thus, the phenomenological modeling of these processes is not trivial. Empirical modeling techniques may present a viable alternative, like artificial neural networks (ANNs), which is a very important Artificial Intelligence technology. In this work, experimental data from a previous work of a partner group was applied to develop ANN models that were proposed to model the amicarbazone degradation by $$\:{\text{O}}_{3}/\text{U}\text{V}$$ , to predict TOC and amicarbazone concentration. The obtained models were able to satisfactorily represent the degradation behavior of amicarbazone, presenting coefficients of determination ( $$\:{\text{R}}^{2}$$ ) greater than or equal to 0.986 and mean square errors less than or equal to 6.56 × 10-4. Analyzing the simulations obtained from the ANN models revealed that increasing the pH and ozone concentration led to higher TOC mineralization. However, the electrical power of the UV radiation source only slightly increased TOC removal. Regarding amicarbazone oxidation, it was found that all of the studied variables had a positive impact on the process. Ozone concentration had a substantial impact, while pH and electrical power had more modest influences. This work provided a reliable model for simulating the ultraviolet radiation-assisted ozonation of amicarbazone in aqueous systems. This process is very complex, involving multiple chemical reactions with different kinetics, mass transfer stages and photon transfer, among other phenomena. Machine learning properly captured these phenomena for the entire set of adopted experimental conditions.