Research On Soft-sensing Of Polyvinyl Chloride Production Process Based On Neural Network

As a result of the conditions of polymerizing process field and the lack of mature detecting device, the conversion rate of vinyl chloride monomer(VCM)is difficult to obtain real-time and the direct control of quality. Therefore a hybrid method utilizing artificial neural network and soft sensor is proposed to forecast the conversion rate of VCM, and this method has important significance. The specific works are as follows:First of all, this paper introduces polyvinyl chloride(PVC) polymerizing process and the basic principle of soft sensor and the application of soft sensor based on neural network. It introduces the learning algorithms of radial basis function(RBF) neural network and the common cultural difference evolution algorithm in detail, optimizes the RBF neural networks weights though combining of cultural algorithm (CA)and differential evolution algorithm(DEA); Kernel principal component analysis(KPCA) method is adopted to select the auxiliary variables of the soft-sensing model in order to reduce the model dimensionality; The soft-sensing model of VCM conversion rate is established by using four methods including gradient method, clustering method, orthogonal least squares method and culture difference evolution algorithm(CDE), then forecasting and simulating.Secondly, this paper introduces the dynamic fuzzy neural network(DFNN), research adjustment methods of the network of different parameter and compares the performance. Then soft-sensing model of VCM conversion rate is established by using three methods including the Kalman filter(KF) algorithm, linear least squares(LLS) algorithm and extended Kalman filter(EKF) algorithm; gets a new polymerizing process soft-sensing model by correcting the model with model migration method ,last use these methods to forecast and simulate.Finally, this paper introduces the echo state network(ESN) and artificial fish swarm algorithm(AFSA), using AFSA to optimize the output weights of echo state network, then establishing soft-sensing model of VCM conversion rate with this method and simulating. In conclusion, simulation results show that three kinds of neural network soft-sensing model can obtain good prediction effect, and significantly enhance the predictive accuracy and robustness of the technical-and-economic indexes and satisfy the real-time control requirements of PVC polymerizing production process.