Elman Neural Network Soft-Sensor Model of Conversion Velocity in Polymerization Process Optimized by Chaos Whale Optimization Algorithm

Abstract

According to the requirement of real-time monitoring of the conversion rate of vinyl chloride in the production process of polyvinyl chloride polymerization and the nonlinearity of the industrial data, the Elman neural network with strong nonlinear performance is chosen to build the soft-sensor model. However, because of the early stage of the Elman neural network to train the connection weights between the layers, the training effect is difficult to guarante with the connection weights. So the whale optimization algorithm (WOA) is adopted to optimize the Elman neural network, to avoid it falling into the local optimum. At the same time, to solve the problem that the position of the search agent is randomly distributed in the initialization process of the WOA algorithm, and to introduce the idea of chaos, a chaos WOA (CWOA) based on the idea of chaos is proposed to improve the diversity of all search agents and egocentricity of agent search by utilizing the chaotic features. At the end of this paper, considering that the input vector dimension is too large, the neural network topology is very large, which will lead to the complexity of the training process. Therefore, the locally linear embedding method is introduced to reduce the dimension of high-dimensional input vectors. The simulation results show that the chaotic whale algorithm can significantly improve the prediction accuracy of economic and technical indexes of PVC polymerization process, which especially has a significant improvement in the prediction effect in the early stage and meets the requirements of real-time control of the production process of the polymerization reactor.