Complex product network change prediction method based on GANs with small sample data

Complex product network change prediction can significantly reduce product redesign time. The accuracy of change predictions often depends on the richness of the historical sample data of this product. Aiming at the problems of less change data and low prediction accuracy in product design change prediction, this paper proposes an improved generative adversarial network using multiple discriminators based on real historical data and generated data called sdrgGAN to improve the data generation. Generated data and real historical data both are applied in the product intensity prediction. The proposed sdrgGAN has improved the multi-feature fusion ability based on feature similarity. The discriminator part is composed of several discriminators. This design improves the discriminant ability of the discriminator and is helpful in generating more realistic data. In the experiment, real historical change data of a TV set is applied. Convolutional Neural Network (CNN) and Long Short Memory Model (LSTM) are used for the prediction of a product change intensity. The experimental results demonstrate the effectiveness of the designed GANs in dealing with small sample problems.