Research And Application Of Ensemble-based Input Attribute Partitioning ESN

The scale development of the process industry is growing,and the effective establishment of a process model is of great significance for the optimal control of the production process.Currently,data-driven intelligent modeling has become a research hotspot in the field of control science and engineering.Neural network technology is a typical data-driven technology,which has been widely used in process modeling.Among them,Echo State Network(ESN),as a typical recurrent neural network,has fast training speed,strong generalization ability,and can effectively process dynamic time series data.Therefore,it is important to study new and reliable ESN for complex industrial process modeling.theoretical significance and practical value.By referring to relevant literature,the main work of this paper is as follows: Aiming at the problem that neuron redundancy in the traditional ESN reserve pool affects the modeling accuracy,this paper first proposes a multi-reserve pool ESN model(Multi-Reserve Pool ESN,MRP)based on input attribute space division-ESN).In the MRP-ESN model,the input space is divided by the k-shape algorithm,and a multi-reserve pool ESN model is established according to the division results.In order to further improve the model accuracy,based on the integration idea,an MRP-ESN model(Multiple Activation Functions-MRP-ESN,MAF-MRP-ESN)that integrates different activation functions is proposed.Firstly,the MRP-ESN based on sine function,cosine function and hyperbolic tangent activation function is established,and finally the three models are integrated to further improve the modeling accuracy.The model proposed in this paper is verified by UCI data and process industry data.The simulation results show that the model proposed in this paper is feasible.Among them,MAF-MRP-ESN has the best model performance and provides reliable and reliable performance for complex petrochemical production processes.smart model.