Data-Driven Method for Voltage Sag Consequence State Recognition for Industrial Users

Abstract

With the advancement of automation in industrial production, the sensitivity of user equipment and processes to voltage sags has progressively increased. Voltage sags are highly likely to cause adverse consequences for production. Voltage sag events exhibit different characteristics, resulting in varying consequence states for users. Accurately identifying these consequence states is essential for determining the user's voltage sag mitigation needs and planning the optimal mitigation strategy. However, due to the low-frequency, high-damage nature of voltage sags, events with more severe consequences are less likely to occur, resulting in insufficient sample data. Furthermore, users are unable to provide detailed sample data due to the protection of production information, making data-driven assessments of industrial process voltage sag consequences even more challenging. To address these challenges, this paper proposes a method for identifying the consequence state of voltage sags in industrial users based on a data-driven method. First, voltage sag event features and consequence category labels for industrial processes are established. An improved semi-supervised fuzzy C-means (SSFC) algorithm is introduced to classify the consequence states of industrial processes. Second, a data augmentation technique based on the least squares generative adversarial network (LSGAN) is applied to expand the dataset of voltage sag samples with the consequence category labels. Next, based on the augmented dataset, a recognition model with VTC-Attention-bidirectional long short-term memory (VA-BiLSTM) is developed to explore the latent features of voltage sag consequences in industrial processes. A recognition library for voltage sag consequence states is created, allowing industrial users to input easily obtainable voltage sag data and obtain corresponding consequence states. Finally, a case study involving a manufacturer in South China is conducted to validate the effectiveness of the proposed method.