Partial Learning Machine:Concept,Algorithms And Applications

In recent years,the research of artificial neural network has become one of the hottest topics in the field of artificial intelligence.The research results in this field have been widely applied to various problems,such as image recognition,automatic translation,speech signal recognition,etc.In this paper,a new optimization algorithm for feed-forward neural network—partial learning machine(PLM)—is proposed.Its name stems from the fact that the algorithm only optimizes partial parameters of the network.The parameters of the network needed to be optimized by the PLM algorithm are the hidden layer neuron biases and the output layer weights.In this paper,we proposed two different types of PLM algorithms,which are PLM_PSO based on particle swarm optimization and least square method,and PLM_G based on gradient descent method and least square method.The numerical results of the algorithm in this paper show that the training time of the PLM_G algorithm proposed in this paper is much less than that of the BP algorithm and the PLM_PSO algorithm on the public data set.Meanwhile,compared with BP algorithm and ELM algorithm,the test error of PLM_G algorithm is reduced by 12.25%and 8.08%,respectively.The main research work of this paper is as follows:(1)PLM_PSO algorithm is proposed based on particle swarm optimization and least square method.The PLM algorithm needs to adjust the hidden layer neuron biases and the output layer weights.Where the hidden layer neuron biases is optimized by the PSO algorithm and the output layer weights is calculated by the least square method.(2)PLM_G algorithm is proposed combining with gradient descent method and least square method.The hidden layer node biases of this algorithm is optimized by the gradient descent method,and the output layer weights are still calculated by the least square method.The PLM_G algorithm solves the problem that the PLM_PSO algorithm suffers long training time due to the increase of hidden nodes.(3)Pretreatment for axle box vibration data of high speed train bogie.By analyzing the vibration data of the axle box,we find that there exists noise in the collected data.Thus,it is necessary to de-noise the original data.The wavelet transform is an appropriate algorithm forp this.(4)The PLM algorithm proposed in this aper is applied to practical problems.By analyzing the vibration data of the axle box of high-speed train bogie,we find that the vibration value of the axle box is affected by the operating mileage,running speed and acceleration of the high-speed train.Therefore,the proposed PLM algorithm is applied to this dataset.The obtained model can be used as a reasonable basis for guiding the maintenance cycle of the high-speed train bogie axle box.At the same time,the model can also be used to guide the maximum operating speed and acceleration of high-speed trains under a certain operating mileage when the axle box vibration safety threshold is set.