Application Of Echo State Network Based On Cuckoo Search In Process Industry

With the desire for quality improvement of process industry,the requirements of control technology for process industry are more and more rigorous.Process industry is characterized by high nonlinearity,severe coupling between variables and complexity of dynamic changes.Although the researches on the characteristics have been experienced a long development,along with many considerable research results,a mature theoretical system like linear control field has not formed.Therefore,it is of great significance to study the identification and control of industrial process with complex nonlinear dynamic characteristics.By mapping complex nonlinear time series,Echo State Network(ESN)is applied to modeling and control for process industry in this paper.In this paper,a new scheme of reservoir generation is proposed to improve the poor generalization performance of ESN,which is caused by the completely random generated reservoir.The scheme simulates the pattern of information exchange between biological neurons,which adjacent nodes information exchange is more obvious than the others,and introduces an adjacent strength matrix to represent the potential ability of the interaction between the neuron nodes in the reservoir,which can optimize topology of reservoir.The optimized reservoir has a more stable structure,thus avoiding determining sparseness of the reservoir weights matrix,which is difficult to choose.The stable structure makes it easier to employ swarm intelligence optimization algorithm to optimize the parameters of reservoir.Based on the characteristics of Cuckoo Search(CS)algorithm with little parameters,fast convergence speed and strong global search ability and simple realization process,this paper presents a Cuckoo Search based Echo State Network(CSESN)algorithm,using CS for the optimization of reservoir parameters.The CSESN is tested by benchmark Lorenz chaotic system,of which only the x component is used,under different scale.The experimental results show that the accuracy is enhanced by CSESN compare to traditional ESN.In this paper,Continuous Stirred Tank Reactor(CSTR)is used as an example to study it in the modeling of process industrial systems.Based on significant improvement of modeling accuracy in CSESN,this paper presents a neural network based direct inverse model control method based on CSESN.Due to the poor stability and robustness of open-loop control of direct inverse model,a feedback channel is introduced to modify the controller online.Through Recursive Least Squares(RLS)algorithm,the linear output layer in the CSESN can be identified online,and the inverse model is able to track the target system,thus realize the closed-loop control.A simulation control experiment of steam water heat exchanger is made to show the effectiveness of the proposed control method.The experimental results show that the direct inverse model control based on CSESN is feasible in the process industry control.