Prediction Of Strip Shape And Rolling Force Based On Improved ELM Method

Iron and steel industry is an important pillar industry of national economic development,which represents the level of national economic development and comprehensive national strength.With the continuous advancement of the modernization process of various countries,all walks of life have put forward higher requirements for the quality of strip.How to improve the quality of strip products has become a hot research direction.Improving the quality of finished strip can be refined to improve its dimensional accuracy.The important evaluation index of dimensional accuracy is shape quality,and the main influence factor is rolling force.Therefore,it has become the most important to establish the high precision shape prediction model and rolling force prediction model.At present,the prediction models of plate shape and rolling force based on neural network mostly adopt BP neural network.Due to the slow modeling speed,it is easy to fall into the local optimal solution and the low prediction accuracy,which greatly limits its application in industry.Therefore.Extreme Learning Machine(ELM)is adopted for modeling.Based on the original algorithm,reliable and accurate prediction models of the shape and rolling force are established.In the aspect of shape prediction,a Two-hidden-layer ELM(TELM)shape prediction model based on Error Minimization(EM)principle is established.According to the principle of error minimization,the model finds the optimal network structure of TELM by adding nodes to the existing network batch by batch(the number of nodes in each batch is variable,which can be one or more).As the number of nodes increases,the parameters of the second hidden layer are updated incrementally by block matrix;Secondly,the idea of LU decomposition is introduced when calculating the output weight of the network,and the traditional generalized inverse solution method is transformed into a simple four-order operation,which simplifies the calculation process of the model.The simulation results show that the proposed EM-TELM method based on LU decomposition improves the modeling speed to some extent.In terms of rolling force prediction,a Multiple hidden layer ELM(MELM)rolling force prediction model is established.In this model,a MP generalized inverse solution method based on LU decomposition is proposed to replace the original generalized inverse calculation method in MELM algorithm.This method can effectively improve the modeling efficiency of MELM.At the same time,the Genetic Algorithm(GA)is introduced to automatically determine the network structure of MELM,and the improved Particle Swarm Optimization(IPSO)is used to optimize the first hidden layer parameters of MELM.The simulation results show that the proposed IPSO-GA-MELM rolling force prediction model based on LU decomposition can accurately determine the network structure of MELM based on actual data.At the same time,the accuracy and speed of the model are improved,and the prediction error is controlled within 4%.