Modeling And Predictive Control For Purification Process Of VOC Waste Gas With Trickling Biofilter Based On Neural Network

Multi-layer Artificial Neural Network has the ability of approach non-linear function at any precision. It can model non-linear process and therefore is an important way to solve non-linear problem. Nowadays it more and more attracts the world's attention.Based on the non-linear parameters of the treatment of VOC waste gas by trickling biofilter, the treatment process can be modeled by using artificial neural network. Traditional back propagation neural network uses Steepest Descent Algorithm to train network. Steepest Descent Algorithm has some inherent defaults such as low convergence, easily getting into local minimum and so on. Levenberg-Marquardt Algorithm can solve the low convergence, but it also can not avoid falling into local minimum. Genetic Algorithm, a global optimization algorithm, has the advantage of greater probability of global convergence. The paper applies the hybrid algorithm composed of a Modified GA and Levenberg-Marquardt to train the linkage weights of ANN, which makes use of the experiment data samples. The MGA can improve the convergence ability. The result shows that MGA-LMBP Algorithm is good in training ANN provided by this paper.Compared with general predicting methods, predictive control method based on ANN more easily fits into non-linear fields. In this paper, author established multi-step predictive controller with multi-variable based on ANN. The control variables are inlet toluene flux and circular liquid flux. The output is toluene removal efficiency. To simulate the steady removal VOC waste gas degradation parallel plat plates theoretical model, the result shows that the toluene removal efficiency curse can track the dentiform and sine wave reference curves and five-step predictive control can track reference curve closely compared with one step predictive control. In the end of the paper, the ANN model multi-step predictive controller, compiled with Visual C++ 6.0, can train the treatment of VOC waste gas by trickling biofilter ANN and be used to watch and control the treatment process. It also has the ability to modify the parameters of training algorithm and predictive control algorithm.