A novel study on activated carbon production based on artificial neural network model: An experimental and artificial intelligence method approach

There are many factors affecting the activated carbon (AC) production process. The process simulation and product prediction methods of activated carbon are still lacking. In this work, the production of activated carbon was conducted on a tubular reactor. The influence of activation temperature, time, steam/char ratio and CO2/char ratio were investigated. Further, a novel artificial neural network (ANN) model was built to predict the production of activated carbon. Different raw material samples were used and in a total of nine influencing variables are used in the input layer. A total of 113 groups of datasets were collected and used for ANN model training, testing and model validation. Results showed that logsig was more suitable to be used as the transfer function in the hidden layer and the optimized neuron number was fifteen, where the MSE value and R-Squared value were 0.010967 and 0.967 respectively. Back propagation(BP)-ANN model could be used for the prediction of the AC yield and the specific surface area. This method is effective in predicting both physical activation and chemical activation. The prediction of AC yield in the process of steam activation is highly accurate. Although there is a certain deviation in the absolute value of the predicted data, the predicted results show a good consistency with the experiment in the variation trend for CO2 activation and KOH activation process. Impact analysis of input variables on outputs indicated that for both AC yield and specific surface area, the oxygen content, activation temperature and CO2/char ratio were the three major influencing variables.