Simplified width learning system based on L1 and L2 norms

Abstract

The invention discloses a simplified width learning system based on L1 and L2 norms, and the method specifically comprises the following steps: 1, obtaining training input data and training output data, obtaining test input data and test output data, N1 and N2 being the number of training test data samples respectively, and T1 and T2 being the dimensions of the input data and the output data respectively; 2, constructing a width model; 3, replacing a regular term in an objective function of the standard width learning system to serve as a new objective function, and 4, for the new objective function, iteratively solving an output weight W by adopting an augmented Lagrange multiplier method; 5, obtaining the output Y of the simplified width learning system according to a formula; wherein Wmis the whole weight connecting the feature node layer and the enhanced node layer to the output layer, and is obtained by W connection. According to the system, the network structure of a width learning system can be effectively simplified, nodes with low correlation with results can be effectively removed, and prediction requirements of related indexes in a complex industrial process can be met.

Description

technical field [0001] The invention belongs to the technical field of industrial process modeling, in particular to a simplified width learning system based on L1 and L2 norms. Background technique [0002] The continuous high-speed development of artificial neural networks has brought data analysis to an unprecedented level. Through machine learning methods, a series of complex industrial process control and optimization problems have been solved. Among them, the most popular method is deep learning. However, the deep learning network has many input nodes and deep network layers. Even in order to improve the accuracy, more network layers and a large number of hyperparameters are needed to continuously iterate the operation. As far as the structure of the algorithm is concerned, this method will make the structure of the algorithm very complicated, and the work of analyzing this structure will become cumbersome. For training, a large number of calculations leads to increa...

AI Technical Summary

Problems solved by technology

The existence of such nodes in the network will not only make the model too complex, resulting in time-consuming training process, but also risk reducing the generalization of the model

This problem hinders its development and application in some fields, so it is necessary to find a network structure simplification method to improve training efficiency and make BLS structure simplification easy to analyze

Images