Font Size: a A A

Dynamic Extreme Learning Machine Based On Particle Swarm And Its Application

Posted on:2020-06-25Degree:MasterType:Thesis
Country:ChinaCandidate:H M WangFull Text:PDF
GTID:2438330602457848Subject:Mathematics
Abstract/Summary:PDF Full Text Request
The Extreme Learning Machine(ELM)is a new type of single-layer feedforward neural network,which is essentially an embodiment of applying the idea of least squares to neural networks.This paper mainly combines particle swarm optimization algorithm with dynamic extreme learning machine,and then applies it to regression and classification problems.The current research direction of extreme learning machines mainly focuses on the optimization of random parameters,the selection of the number of hidden layer nodes and the improvement of stability.Researchers have gradually proposed a variety of extreme learning machines such as Re-ELM,Dynamic Extreme Learning Machine(D-ELM),and Particle Swarm Based Extreme Learning Machine(PSO-ELM).Among them,the dynamic extreme learning machine(D-ELM)is currently a hot research direction.In this thesis,the particle swarm optimization algorithm is integrated into the dynamic extreme learning machine,and dynamic extreme learning machine(PSO-D-ELM)based on particle swarm optimization algorithm is proposed.The algorithm uses particle swarm optimization to optimize random parameters and dynamic limit learning machine to optimize the number of hidden nodes.In order to verify the superiority of the proposed PSO-D-ELM algorithm,it is tested by using the data in the UIC database.It is also a test data set commonly used in extreme learning machines,including octane content dataset,Abloe dataset and iris dataset.At the same time,ELM,D-ELM,PSO-ELM,Re-ELM and Re-D-ELM algorithms were compared with PSO-D-ELM.The results show that the PSO-D-ELM prediction error is smaller than that of other extreme learning machine algorithms.The prediction error of the near-infrared spectral data set is only 0.03,and the prediction error of the Abloe data set does not exceed 0.12.In the classification identification of irises,PSO-D-ELM is the highest average accuracy rate of more than 99%.Through multiple comparison experiments,the experimental results show that the stability of PSO-D-ELM is also the best.After the algorithm is proved to be excellent,in order to explain its wide application in classification and prediction,it is applied to the classification identification of electrical equipment and the prediction of oil gas production,and also compared with the effects of the other five extreme learning machines.The results show that it can be effectively identified in the classification and identification of electrical equipment.The error in the prediction of gas production in oil fields is only 0.16,which is more stable than other five types of extreme learning machines.That is to say,the PSO-D-ELM algorithm has high accuracy in classification and prediction.Compared with some extreme learning machines,it can be expected to have wider application in other fields.
Keywords/Search Tags:Least squares method, extreme learning machine, dynamic extreme learning machine, particle swarm optimization
PDF Full Text Request
Related items