Towards An Optimal Kernel Extreme Learning Machine Based On Swarm Intelligence Optimization With Its Applications

Single hidden layer feedforward neural network(SLFN)with its good learning ability has been widely applied in many fields,but the weights and bias of SLFN need to be corrected iteratively,so its development and application are restricted due to the inherent slow training speed and worse generalization performance.Extreme learning machine(ELM)algorithm is a kind of new SLFN,which has the advantages of fast learning speed,good generalization performance and the weights of the input layer and the hidden layer and the bias of the hidden layer are generated randomly.Kernel extreme learning machine(KELM)is combinated kernel function and ELM,and then KELM can reached the more robust and better generalization performance than the original ELM.However,the performance of this method is decided by the key parameters in the practical cases.In this paper,the key parameters and feature selection of KELM are explored;in addition the selection methods of the KELM-based swarm intelligence optimization are proposed and successfully applied in disease diagnosis and financial bankruptcy prediction.The swarm intelligence is a new kind intelligenc computing method which abstracts the math model from the hunting behavior of the creatures in the natural world.Grey wolf optimization algorithm(GWO),Moth optimization algorithm(MFO)and Whale optimization algorithm(WOA)as the latest intelligence methods have been widely paied attention by scientific researchers and engineering technicians due to their strong global search ability.In order to further explore the ability of the swarm intelligence optimization for KELM model selection,this paper hereby proposes improved GWO-based opposition-based learning,improved MFO-based chaos and improved WOA-based multi population,and then construct the KELM models for solving the practical problems.The main work of this paper includes the following asepects:(1)For the traditional GWO,an opposition-based learning strategy is introduced in the process of population update(BGWO),which can increase the diversity of the population for each agent.Then the gray wolf particle,which has the best fitness function value in the search space,is iteratively fulfilled to train the KELM(BGWO-KELM).The strategy for parameter optimization and feature selection simultaneously is also adopted and then the model is successfully applied to the prediction of corporate bankruptcy.(2)In this paper,the chaos strategy is introduced into the original MFO(CMFO),with the chaos sequence value completing the initialization of population and chaos disturbance mechanisms in the search process,which can increase the diversity of the initial population and makes the moth particles jump out of the local optimum at the same time.Subsequently,the CMFO algorithm is applied to the KELM model selection problem and feature selection simultaneously and applied to intelligent diagnosis of disease(CMFOFS-KELM).(3)This paper also proposed the multi population WOA(MWOA)using the K-means to divide the population into k independent sub group.And then applied the MWOA to complete the parameters tuning and feature selection,which is successfully applied to the disease diagnosis.