A Health Assessment Model for Communication Power System Based on Auto-Encoder and Generative Adversarial Network

Accurate health assessment can improve the service level of power grids, reduce operation and maintenance costs, and ensure the stable operation of communication power system. However, current assessment methods do not fully exploit the complex temporal information among power data. For this reason, this paper proposes a health assessment model based on Auto-Encoder (AE) and Generative Adversarial Network (GAN) to realize the health assessment of communication power system. First, an auto-encoder is constructed by using a long short-term memory neural network (LSTM) and global Luong attention to fully exploit the complex temporal information among the electric power data. Then, a generative adversarial network model is constructed with the constructed self-encoder as a generator and a multilayer perceptron (MLP) as a discriminator, which enhances the characterization of electric power data through their mutual confrontation. Finally, the health factors are extracted by comparing the errors between the reconstructed data of the generator and the original time-series data, and are inputted into the scoring model to complete the evaluation of system health. The experimental results show that the model proposed in this paper can effectively improve the health assessment of the communication power system and provide reference for the stable operation of the system.