基于MLP-KAN神经网络的真空膜蒸馏废水处理系统性能预测研究OA

Vacuum membrane distillation(VMD)is an innovative technology for resource recovery of high-salinity wastewater.To realize the intelligent operation of VMD system,a hybrid prediction model integrating multi-layer perceptron and Kolmogorov-Arnold network(MLP-KAN)was established.Firstly,a VMD experimental device was built,and multi-condition experiments were carried out with sul-furic acid solution as feeding to acquire 305 groups of sample data.The data were subsequently divided into training,validation and testing sets at the ratio of 7.0∶1.5∶1.5 for prediction analysis of permeate flux.Research results indicated that predicted values matched actual values well.The determination coef-ficient(R2)reached 0.971 for the testing set and 0.995 for the training set,while the corresponding root mean squared error(RMSE)was 0.099 2 and 0.044 7 kg/(m2·h),respectively.The MLP-KAN model possesses outstanding stability,generalization performance and robustness.It can accurately pre-dict the permeate flux,which offers solid technical foundation for parameter optimization and intelligent operation of the VMD system.