Based on analyzing the influential factors and the characters of flatness forecasting, a hybrid optimized algorithm for RBF neural network based on modified particle swarm optimization (MPSO) is introduced in the paper to forecast the flatness. The chaotic optimization algorithm is introduced to decide the parameters of PSO. The number of units in RBF hidden layer is determined by using the rival penalized competitive learning (RPCL) algorithm. Centers, widths of basis functions and weights of neural network are estimated dynamically in global space with MPSO. The proposed model is trained and tested based on the field data collected from 1220 cold rolling mill. The simulation results show that the RBF flatness forecasting model based on MPSO algorithm has the highest forecasting accuracy among BP neural network model and LS method, which is usually used in the real production. So the proposed model has a good prospect for the flatness forecasting.
[1]
Michael Gribskov,et al.
Rival penalized competitive learning (RPCL): a topology-determining algorithm for analyzing gene expression data
,
2003,
Comput. Biol. Chem..
[2]
Zhao Xiao-Mei.
IMPROVED CHAOS OPTIMIZATION ALGORTHM AND ITS APPLICATION
,
2006
.
[3]
Ust Beijing,et al.
CVC Tandem Mill Flatness Predictive Control Model Based on BP Neural Networks
,
2000
.
[4]
Haralambos Sarimveis,et al.
A new algorithm for developing dynamic radial basis function neural network models based on genetic algorithms
,
2004,
Comput. Chem. Eng..
[5]
Cheng Lu,et al.
Modeling and optimization of threading process for shape control in tandem cold rolling
,
2003
.