Research On Symmetric Extreme Learning Machine With Shuffled Frog Leaping Algorithm

Extreme learning machine (Learning Machine Extreme, ELM), as a recently emerging feedforward neural network algorithm, and the traditional sense of the feedforward neural network algorithm has a great difference. Elm algorithm weights of input layer and hidden layer offset value selection is random, output layer weights is calculated without the continuous iteration to adjust the network parameters, which also to a great extent solves the disadvantage of feedforward neural networks learning efficiency is low. However, ELM algorithm parameters of random selection, it is inevitable to bring the ELM generalization performance is greatly reduced, if the selection of the initial parameter is not appropriate, and even greatly affect the performance of ELM prediction.Based on the above problems, this paper the elm algorithm were a series of improvements, learning machine algorithm (SFLASym-ELM) a shuffled frog leaping optimization of symmetric limit is proposed. Based on the analysis of the characteristics of strip steel data, the prior information is put forward to improve the network structure of limit learning machine, which can reduce the number of hidden layer nodes of ELM. At the same time, also affect strip exit thickness of various relevant factors analysis, the principal component analysis method and the ultimate learning machine combination, will extract the strip thickness main factors as the elm network input, so as to optimize the network structure of elm. Based on extreme learning machine in the input layer of weights and value in the initial state randomly selected caused output weighting matrices larger problem of hidden layer of bias, on the third chapter in the improved algorithm Sym-ELM using shuffled frog leaping algorithm (SFLA) were further optimized, and puts forward the SFLASym-ELM algorithm. SFLASym-ELM model by applying shuffled frog leaping algorithm to select input weights Sym-ELM algorithm and hidden layer offset value can effectively less predictive error.The research object of this paper is to predict the thickness of strip in rolling process. By using the software ibaAnalyzer of strip thickness at the exit of each factor analysis, and using principal component analysis method is used to calculate the correlation coefficient of each factor and the strip exit thickness, ultimately chose and strip exit thickness larger correlation factors as input SFLASym-ELM into, and the prediction of strip exit thickness based in MATLAB for offline strip exit thickness prediction experiments, and the experimental results with the traditional elm algorithm, BP algorithm for the prediction of the results were compared and found the SFLASym-ELM algorithm has less prediction errors and to verify the SFLASym-ELM count method has higher prediction accuracy, strip production of online has better guiding significance.