Research and Application of Virtual Sample Generation Method Based on Conditional Generative Adversarial Network

Decay‐like fracture diagnosis of composite insulator based on small sample virtual expansion and radar graph mapping

“…The specific implementation process of MTD‐VSG and TTD‐VSG methods is consistent in Ref. [9, 17]. The latest feature representation‐based VSG applied in the fault diagnosis are PSO‐VSG and CGAN‐VSG [25].…”

“…According to the established network model, the virtual sample in the original high‐dimension can be predicted. Select the feasible virtual samples based on the asymmetric acceptable extension domain. The asymmetric acceptable extension domain is calculated according to the following [17, 19, 27]: $$$\left\{\begin{array}{c}{B}_{\mathrm{L}}=M-\frac{1}{1-{e}_{\mathrm{L}}}\times \left(M-{X}_{\mathrm{min}}\right)\\ {B}_{\mathrm{U}}=M+\frac{1}{1-{e}_{\mathrm{U}}}\times \left({X}_{\mathrm{max}}-M\right)\end{array}\right.$$$ $$$\left\{\begin{array}{c}{e}_{\mathrm{L}}=\frac{{N}_{\mathrm{L}}}{{N}_{\mathrm{L}}+{N}_{\mathrm{U}}+1}\\ {e}_{\mathrm{U}}=\frac{{N}_{\mathrm{U}}}{{N}_{\mathrm{L}}+{N}_{\mathrm{U}}+1}\end{array}\right.$$$ $$$M=\left\{\begin{array}{ll}{x}_{[(n+1)/2]}& \,\text{if}\,n\,\text{is}\,\text{odd}\,\\ \frac{1}{2}({x}_{[n/2]}+{x}_{[1+n/2]})& \,\text{i...$$…”

“…Select the feasible virtual samples based on the asymmetric acceptable extension domain. The asymmetric acceptable extension domain is calculated according to the following [17, 19, 27]:…”

“…To solve the hard concepts problems and visualize the data [16, 17], this paper constructs the new features by the radar graph mapping model.…”

A novel virtual sample generation method to improve the quality of data and the accuracy of data-driven models

“…Wang [15] added disturbance based on training samples to obtain new virtual samples and used these virtual samples to make the model have a better recognition rate. The idea based on distribution is mainly to determine the range and probability of virtual sample generation according to the domain distribution of small sample data [16][17][18][19][20][21][22][23][24][25][26]. Der Chiang Li [16] proposed mega-trend-diffusion (MTD), which uses a common diffusion function to spread a group of data and determine the possible coverage of the data set based on group consideration to generate reasonable virtual data.…”