Soft-Sensing Modeling Method of Vinyl Acetate Polymerization Rate Based on BP Neural Network

Grey neural network prediction model based on fruit fly optimisation algorithm and its application

“…Comparing the predictive results of this FOA_GNN model and relevant prediction models mentioned in the references (Huang et al, 2009(Huang et al, , 2010Tang, 2008), the FOA_GNN model gave satisfactory forecasting performances. Obtained by FOA_GNN model, the values of MPE, MNE, SSE, MSE, RMSE, MAPE and MAE are 0.0041, -0.0175, 0.0023, 6.33e-005, 0.0079, 1.99% and 0.0065, respectively, far blow those performance values obtained by the other five models.…”

“…Using the fruit fly algorithm to optimise the 'whitening' parameters in the G(1, N) prediction model and define the initial weights and biases of GNN, we can establish the GNN prediction model based on FOA. Utilising this model to predict the VAC polymerisation rate in the production of PVA and verifying it with actual cases, we have found that this model is more accurate than standard GNN prediction model and other relevant prediction models mentioned in Huang et al (2009Huang et al ( , 2010 and Tang (2008).…”

“…Apparently, it is evident that the FOA_GNN prediction model is superior to the GNN prediction model in terms of predictive accuracy and correlation of the VAC polymerisation rate (i.e., actual vs. predicted) by above conclusions Liu, 2011a, 2011b;Mohamed et al, 1999). In order to directly reflect the advantages and disadvantages of prediction model, this paper also compare four VAC polymerisation rate prediction models -BP [3] , GA-BP [4] , ACGA-BP [4] and RBF [5] mentioned in the references (Huang et al, 2009(Huang et al, , 2010Tang, 2008) with GNN prediction model and FOA_GNN prediction model in the different performance evaluation indexes, and the simulation results are reported in Table 1. The maximum positive error (MPE), maximum negative error (MNE), sum of squared error (SSE), mean squared error (MSE), root-mean-square error (RMSE), mean absolute percentage error (MAPE) and MAE were used to assess the performances of different prediction models.…”

“…In recent years, the neural network prediction method has been widely used to model and control dynamic processes as a data mining because of its extremely powerful adaptive capabilities in response to nonlinear behaviours (Zhang et al, 2014). Huang et al (2009Huang et al ( , 2010 proposed to utilise two prediction methods, based on the adaptive compete genetic neural network and BP network, while Tang (2008) presented a prediction method of the VAC polymerisation rate, based on radial basic function (RBF) network. These works greatly improved the prediction precision of the VAC polymerisation rate.…”