Virtual sample generation method based on improved ant lion optimization algorithm

Abstract

The invention provides a virtual sample generation method based on an improved ant lion optimization algorithm. The virtual sample generation method comprises the steps of determining a lower boundary and an upper boundary of a generated virtual sample according to an overall trend diffusion technology; determining the number of neurons of a hidden layer of the ELM model through a test parameter method, and enabling the mean absolute percentage error predicted by the ELM model to be smaller than 10%; improving an ant lion optimization algorithm by introducing ant lion active Gaussian variation and a self-adaptive walk boundary mechanism, randomly selecting a real sample through the improved ant lion optimization algorithm, and generating a plurality of virtual samples around the real sample. According to the virtual sample generation method, the defect that the prediction precision is not high when a traditional machine learning model processes a small sample problem is overcome, the advantages of an improved ant lion optimization algorithm can be fully utilized, the virtual sample with higher reliability is generated, an original data training set is expanded, and the method has the advantages of being small in error, high in applicability and high in precision.

Description

technical field [0001] The invention relates to a virtual sample generation method based on an improved antlion optimization algorithm, and belongs to the research field of machine learning small sample problems. Background technique [0002] Predicting the future state of complex systems and designing such systems is very expensive, time-consuming, and computationally intensive. To overcome these complexities and save significant cost, time and energy, modeling can be utilized. As data-driven modeling methods are widely used in many fields to build predictive models, many algorithms have been proposed to learn data trends by utilizing structured datasets collected from specific domains. Such algorithms rely on data, that is, sufficient data is a necessary condition to ensure that the model is more accurate in classification and regression applications. The more valid data, the higher the prediction accuracy of the model. [0003] But in the real world, the data collectio...

AI Technical Summary

Problems solved by technology

Solving the small sample set problem remains a long-term challenge

Images