Uncertainty-Aware Voltage Control for Distribution Substations Based on Bayesian Deep Learning and Probabilistic Graph Self-Encoder Fusion

Voltage control in low-observable distribution networks faces significant uncertainty challenges. In this paper, an uncertainty-aware voltage control method based on the fusion of Bayesian deep learning and probabilistic graph self-encoder is proposed. The method first utilizes Bayesian deep learning to establish a probabilistic model of the network topology and inter-node dependencies, and then realizes efficient encoding and decoding of the system state through a probabilistic graph self-encoder. This fusion approach combines the uncertainty quantification capability of Bayesian inference and the structure learning advantage of graph self-encoders to achieve multi-scale uncertainty modeling and dynamic decision optimization. Specifically, this paper develops a Bayesian deep learning model augmented by graph attention network to effectively capture the complex topology of the network; meanwhile, an adaptive conditional probabilistic generative adversarial network is proposed to generate realistic and diverse probabilistic scenarios. By fusing these two techniques, this method achieves adaptive probabilistic scenario generation and screening, as well as uncertainty-based dynamic control strategy optimization. Experimental results show that the fusion method significantly improves the robustness and adaptability of voltage control in low-observable distribution networks, and provides a new idea for solving the uncertainty problem of complex distribution systems.