Inverse design of three-dimensional electrocatalytic oxidation process for water purification via machine learning and DFT-based descriptors

Three-dimensional electrocatalytic oxidation (3DECO) process has become a research hotspot in the treatment of refractory organic pollutants because of its no need for no chemical reagents and high degradation efficiency. However, its complex process parameters and nonlinear reaction characteristics restrict the rapid acquisition of efficient operating conditions. In this study, a reverse design framework based on machine learning is proposed to optimize the water purification process of 3DECO. By integrating 5704 groups of experimental data, a multidimensional database covering quantum chemical parameters of pollutants (such as Fukui index and HOMO energy level) and reaction conditions (such as current density and electrolyte concentration) was constructed, and the prediction performance of multiple linear regression (MLR), support vector machine (SVM), random forest (RF), and back propagation neural network (BPNN) was systematically compared. The results show that the introduction of reaction condition variables significantly improves the performance of the model. The determination coefficient (R) of the BPNN model in the external test set is 0.91, and the root mean square error (RMSE) is 0.74. Based on the characteristic importance analysis of SHAP (Shapley Additive Explanations), it is revealed that current density, particle electrode load, and pollutant band gap energy (Egap) are the key control factors. Through the particle swarm optimization (PSO) algorithm, the customized operation parameters are designed in reverse. The experimental results show that the removal rate under the optimized conditions is 99.45 %, and the relative error between the experimental value and the predicted value is less than 5 %. This study provides a new data-driven paradigm for the intelligent design and large-scale application of 3DECO technology, and reveals the future improvement direction of model interpretability and cross-scene generalization ability. This study provides an explainable and verifiable machine learning-driven solution for efficient and low-carbon 3DECO process parameter configuration, and also provides a path for data-driven environmental process optimization.