PSO-BP-ANN for Prediction of Chlorophenols Removal in An Electro-oxidation System

Author:Mei Zuo

Supervisor:wang jia de


Degree Year:2019





Chlorophenols(CPs)have been widely used in manufacture industries,such as insecticides,herbicides,fungicides and dyes which were toxicity,resistance to biodegradation,easily enriched through the food chain and carcinogenicity.Chlorophenols were listed of priority pollutants in all the countries.Electrochemical oxidation CPs is considered as a promising alternative for its fast reaction rate,high energy,no reagents need to be added and environmental compatibility.The ORP is a measure indicator of the oxidizing or reducing properties of solutions.ORP reflects the solution’s tendency to accept or donate electrons and provides insight into the state of the reaction system that other parameters cannot reveal.The influences of current density,pH and supporting electrolyte(Na2SO4)concentration on ORP were discussed.The relationships betweenΔORP and COD removal efficiency in various current densities,pH and supporting electrolyte concentration were investigated.Meanwhile,a multi-parameter linear relationship among ORP,current density,original pH,sodium sulphate concentration,reaction time and COD removal efficiency and TEC were established with the R2 of 0.8878and 0.93223,which can reflect ORP monitoring can be used in online electrochemical control for 2-CP wastewater treatment.Back propagation(BP)algorithm network was established with five input parameters(ORP,current density,electrolyte concentration,initial pH value and electrolysis time),10 hidden layer nodes and two output parameters(COD removal efficiency and TEC).The experimental results showed that the logsig activation function as transfer function from the input layer to hidden layer,and the pureline function from the hidden layer to output layer under the two prediction score metrics:coefficient of correlation and mean square error.The trainlm was chosen as training function by analyzing the function performance curve,the iteration times and the best calibration errors of different training functions.And the 10 was the best number of nodes in hidden layer.The BP-ANN for prediction of COD removal efficiency and TEC for testing data showed correlation coefficient of 0.9344 and 0.9355,mean square error of 0.0137232 and 0.013127 respectively.But BP-ANN has some problems,such as large error of individual points,uncertainty of initial weights and thresholds,slow learning convergence and easy to fall into local minimum.Prediction accuracy was increased by using particle swarm optimization coupled with BP–ANN to optimize weight and threshold values.The PSO–BP–ANN possessed a swarm size of 50,maximum iteration of 200,C1 of 1.5,and C2 of 1.5 was identified as the best model for predicting 2-chlorophenol degradation through electro-oxidation.The PSO-BP-ANN for the efficient prediction of COD removal efficiency and TEC for testing data showed correlation coefficient of 0.99 and0.9944,mean square error of 0.0015526 and 0.0023456 respectively.Compared with BP-ANN,PSO-BP-ANN had higher accuracy and could effectively predict the electro-oxidation process.The weight matrix analysis indicated the correlation of the five input parameters was current density 18.85%,initial pH 21.11%,electrolyte concentration 19.69%and electro-oxidation time 21.3%,ORP 19.05%.