An Online Sequential Multiple Hidden Layers Extreme Learning Machine Algorithm

Because of the broad application prospects of the artificial neural network,many scholars favor it very much and have achieved many important achievements at the same time.However,the slowly operation speed of the neural network(e.g.BP neural network)limited its application in the engineering process.In 2006,the extreme learning machine(ELM)neural network greatly enriched the development of feedforward neural network and reduced the computation time of feedforward neural network.But,extreme learning machine also shows its own weakness when dealing with the input dataset with complex noises or high-dimensional information,which leads the great decline of classification regression accuracy of the established network model.In recent years,because the rapid development of the ELM,many predecessors also put forward many productive ideas to improve the plight of ELM.Based on the previous studies,this paper proposes its own improved algorithm.Firstly,this paper improve the ELM algorithm by adding multiple hidden layers(the number of hidden layers is greater than or equal to 3)based on the three-layer structure of the network,and form a multiple hidden layers ELM network.At the same time,the MELM theory inherits the randomness of the ELM technique for the first hidden layer(connection weights between the input weights and the first hidden layer and the bias of the first hidden layer),and introduce a method to calculate the weights and bias parameters of the residual hidden layers.In practice,the original datasets are not immutable,and change its data characteristics over time.But the networks model that established by the above MELM and ELM are fixed,and the parameters can't be changed arbitrarily,which would cause deviation.Based on the above MELM and the mature research results,this paper adds multiple hidden layers to the extreme learning machine,and constitute the online sequence multiple hidden layers extreme learning machine(FOS-MELM).The FOS-MELM algorithm can handle the timeliness datasets batch by batch(e.g.stock datasets,climate datasets),improve the training accuracy and reduce the impact on the behind training process by discarding the outdated data.Also,the FOS-MELM can obtain a high accurate model by removing a series of unavoidable noise signals based on the MELM model.At the same time,the algorithm can update the parameters of the model over time,fine tuning the model,make it more consistent with the direction of datasets changing,and ensure the accuracy of the networks model.