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

In the chemical production process, due to certain physical limitations and high measurement costs, it is sometimes difficult to obtain plenty of data points to improve the accuracy of the prediction model. To solve this problem, this paper proposes a novel virtual sample generation method to reasonably expand training sets, aiming to promote accuracy of prediction model. First, the outliers of the original samples are found by local outlier factor. Then, iterative midpoint interpolation is performed on each outlier to obtain a new sample x with a more uniform distribution, so as to fill the scarce area as much as possible. Second, the corresponding output of these new virtual samples are generated by a generative model based on conditional generative adversarial network. To observe the effectiveness of the proposed method, we used standard function dataset and purified terephthalic acid (PTA) production dataset for verification. Experimental results show that our method effectively improves the accuracy of the prediction model.