Online Voltage Regulation of Active Distribution Networks: A Deep Neural Encoding-Decoding Approach

Voltage regulation plays a vital role in active distribution networks, which can be posed as mixed-integer quadratic programming (MIQP). To alleviate the real-time MIQP solution burden, this article proposes a deep neural encoding-decoding approach for online voltage regulation, including offline encoding and online decoding processes. Offline encoding extracts the integer variables and active constraints of the MIQP regulation problem and reduces the MIQP into quadratic programming (QP). The KKT-based solution matrices derived from the reduced QP are stored via the compact triangular matrix factorization way. Deep neural network (DNN) is trained to
encode
the optimal voltage regulation solution into the regulation strategy. Online decoding utilizes the trained DNN to identify the optimal regulation strategies, which selects the top-
k
most likely strategies under various scenarios. Then it
decodes
the
k
strategies into precomputed KKT-based linear solution matrices for real-time regulation. The above online voltage regulation can be calculated efficiently without any solver. Case studies verify the proposed approach's voltage regulation effectiveness and calculation efficiency.