Modeling And Hdp Control For Clarifying Process Of Sugar Mills Based On Echo State Networks

The sugar industry is one of the pillar industries in Guangxi, and the quality of sugar relates to our daily life and production. Because the clarifying process is the important part of the sugar production process, so the optimal control of sugar mills clarifying process is critical. An accurate and effective control object model is the basis of the optimal control. As the clarifying process of sugar mills is a large time delay, strongly coupling, strongly nonlinear and complex multi-input multi-output process, it is difficult to model this process based on mechanism. Therefore the neural network modeling method based on data is used to model this process. Echo State Networks (ESN) is a novel recurrent neural network with the characteristics of fast training and never fall into local minimum point. In this paper, this network is used to model the clarifying process of sugar mills. After analyzed the characteristics of the sugar mills clarifying process, select the model input and output parameters, and then build the clarifying process model based on ESN with the sugar mills history data. In order to verify the effectiveness of the model, this model is used to generalization simulation and compared with the traditional BP network model and RBF network model. The results show that the accuracy of the model based on ESN is better than the model based on BP or RBF and the ESN model reflects the properties of sugar mill clarifying process well.After get the model of sugar mills clarifying process, an intelligent control algorithm called adaptive dynamic programming (ADP) is used to control this process. In order to solve the problem of slow convergence and easy to fall into local minimum point of the adaptive dynamic programming based on BP neural network, the adaptive dynamic programming based on echo state networks is proposed, and then using this algorithm to control sugar mill clarifying process.According to Bellman principle of optimality, ADP based on ESN algorithm is deduced. To verify the effectiveness of this algorithm, this algorithm is used to simulation control the sugar mills clarifying process model that built based on ESN. And then compared with traditional ADP control based on BP neural network. The results show that the HDP based on ESN can be a good improvement of HDP based on BP network in the implementation easy to fall into local min-point. And HDP based on ESN has rapid convergence.