Data classification method and system, equipment and information data processing terminal

Abstract

The invention belongs to the technical field of machine learning and intelligent computing, and discloses a data classification method and system, equipment and an information data processing terminal. The data classification method comprises: dividing each data set into a training set and a test set; initializing the structure and partial network parameters of a single hidden layer feedforward neural network (SLFN), including the state of hidden layer neurons, an input weight and hidden layer bias, and calculating an optimal output weight; evaluating fitness function values of the generated chromosomes; optimizing a network structure and connection parameters at the same time through a collaborative genetic algorithm (CGA-ELM) based on an extreme learning machine, and obtaining an optimal solution to serve as a final single-hidden-layer feedforward neural network; and testing network performance by using the test set, and outputting average classification accuracy. The generalization ability of the CGA-ELM provided by the invention is obviously superior to that of CGA and ELM. Compared with other advanced algorithms, the CGA-ELM can achieve higher recognition capability on the premise of keeping convergence speed, and has higher competitive capacity and better generalization performance.

Description

technical field [0001] The invention belongs to the technical field of machine learning and intelligent computing, and in particular relates to a data classification method, system, equipment and information data processing terminal. Background technique [0002] At present, the artificial neural network (ANN) is a nonlinear adaptive information processing system composed of a large number of processing units. It is proposed on the basis of modern neuroscience research results, trying to process information by simulating the way the brain's neural network processes memory information. Feedforward neural network (FNN) is the earliest simple neural network invented in the field of artificial intelligence. In FNN, information is unidirectionally propagated from the input layer to the output layer through the hidden layer, which has attracted extensive attention due to its simple structure and good performance. The approximation theorem of ANN shows that a trained multi-layer ...

AI Technical Summary

Problems solved by technology

[0033] (1) The existing feedforward neural network training algorithm has limitations, such as slow convergence speed, easy to fall into local optimum, and strong dependence on initial weights, etc.

[0034] (2) There is a large training error in the existing neural network, and the neural network structure is relatively complex

[0035] (3) The randomness of ELM easily leads to fluctuations in its learning performance, which further leads to unstable classification performance of SLFN, and only calculates network parameters with a fixed structure

[0036] (4) Due to the inappropriate learning step size, the existing learning method based on gradient descent tends to converge to a local optimal solution, and needs to solve the gradient information and iterative calculation, resulting in a large amount of calculation

[0037] The difficulty of solving the above problems and defects is: it is necessary to design an algorithm to optimize the network, so that the performance of the network is good enough, and at the same time the obtained network is compact, and it is necessary to overcome the shortcomings of easy to fall into local optimum, reduce the amount of calculation, and improve the calculation efficiency. speed

Images