Research On Feedforward Neural Network Based On Electromagnetism-like Mechanism Algorithm Optimization

Feedforward Neural Network is one of the most basic neural networks.Each layer of neurons is only connected with the previous layer of neurons,and there is no feedback between the other layers.It is one of the most widely used and rapidly developed artificial neural networks.Extreme Learning Machine(ELM)is a novel Feedforward Neural Network Algorithm.The input weights and hidden layer bias of the network are generated randomly,and the output weights of the network are obtained by solving M-P generalized inverse.In the process of training,only the number of hidden layer neurons needs to be optimized and the appropriate activation function needs to be selected.Different from the traditional BP algorithm which uses a gradient descent algorithm to adjust the network parameters,ELM does not need repeated iteration and a lot of parameter adjustment.It has the advantages of simple structure,fast learning speed,less manual intervention,and has good generalization performance.It is widely used in pattern recognition,disease diagnosis,image quality evaluation,and other fields.However,a large number of studies show that although ELM network can greatly improve the training speed of the network by randomly generating input weights and hidden layer bias,it is also easy to generate bad network parameters,thus affecting the generalization ability and stability of the network.In this paper,by optimizing the network parameters of Extreme Learning Machine as an example,a research method based on Electromagnetism-like Mechanism Algorithm to optimize the Feedforward Neural Network is introduced.In order to better optimize the network parameters,two improved Electromagnetism-like Mechanism Algorithms are proposed.The effectiveness and reliability of the proposed algorithm are verified by simulation experiments and tests on multiple data sets.The main research work of this paper is as follows:(1)Improvement of Electromagnetism-like Mechanism AlgorithmAfter analyzing the problems of population diversity loss,search information loss,and resultant force calculation redundancy of the original Electromagnetism-like Mechanism Algorithm,a new Electromagnetism-like Mechanism Algorithm called NEM is proposed by modifying the charge calculation formula;By using the idea of population division to optimize the iterative mechanism of the Electromagnetism-like Mechanism Algorithm,a Subpopulation Electromagnetism-like Mechanism Algorithm(SEM)is proposed.The experimental results show that the new charge formula improves the utilization of population information and population diversity,and the improved NEM algorithm achieves better convergence accuracy;The idea of Subpopulation evolution is introduced to reduce the computational complexity of the algorithm.(2)Optimization of Feedforward Neural Network based on the improved Subpopulation Electromagnetism-like Mechanism AlgorithmIn order to get better network parameters,the improved Subpopulation Electromagnetism-like Mechanism Algorithm is used to optimize the input weights and hidden layer bias of the forward neural network.Taking the Extreme Learning Machine as an example,the Extreme Learning Machine optimized by the Subpopulation Electromagnetism-like Mechanism Algorithm is proposed,The optimized network parameters improve the prediction accuracy and generalization performance of the network,and the introduction of the idea of subpopulation greatly reduces the training time of swarm intelligence algorithm in optimizing the neural network;Finally,the SEM-ELM algorithm is applied to handwritten numeral recognition,the reliability and practicability of the algorithm are further verified.