Performance prediction of vacuum membrane distillation system based on multi-layer perceptron neural network

Evaporation capacity is an important evaluation indicator for a vacuum membrane distillation (VMD) system. Accurate and efficient prediction of evaporation capacity has always been a difficult problem for the VMD system. This paper developed a multi-layer perceptron (MLP) neural network model, to effectively predict the evaporation capacity of the VMD system. With feed concentration, feed temperature, feed flow rate, vacuum side pressure and membrane area as input variables and evaporation capacity as output variable, 370 experimental data were divided into training set and testing set in a 4:1 ratio. The prediction performance was analyzed by comparing with convolutional neural network (CNN), long short-term memory network (LSTM) and gated recurrent unit (GRU). The results showed that the goodness of fit (R-2) in the training set and testing set for MLP, CNN, LSTM and GRU were 0.99, 0.95, 0.94, 0.95 and 0.98, 0.94, 0.93, 0.94 respectively, four models could effectively predict the evaporation capacity due to the good coincidence between prediction and real values under an acceptable error level. Moreover, the sample numbers with relative error (<5 %) between prediction and real values accounted for 56.5 %, 43.8 %, 23.5 % and 25.4 % of the total samples in MLP, CNN, LSTM and GRU. Obviously, MLP exhibited highest accuracy and stability than that of CNN, LSTM and GRU for the current VMD system.