T-Gen: Time series data generator for inverse analysis of machine learning models

Abstract

A method was developed to generate diverse time series data automatically. Following the discrete Fourier transform (DFT) for time series data, the real and imaginary parts of the complex numbers in the frequency domain representation were modified using random numbers to generate new ones. Then, an inverse DFT was performed to generate virtual time series data. This innovative approach can propose diverse time series data in case studies involving actual processes and design time variation in each process variable to meet the target values of objective variables through the inverse analysis of machine learning models constructed with process data.

Keywords

• Inverse analysis
• Time series data
• Process design
• Soft sensor
• Process dynamics
• Fourier transform