Varying Condition SCOPF Optimization Based on Deep Learning and Knowledge Graph

Abstract

Security-constrained optimal power flow (SCOPF) is a vital task for independent system operators (ISO) in daily scheduling. However, the large number of inequality constraints bring us big challenges to solve large-scale SCOPF in real time. This paper proposes a fast solution method for SCOPF by predicting active constraints based on machine learning approaches. Namely, deep neural networks (DNNs) are employed to predict active security constraints based on historical data, thus accelerating the SCOPF calculation. Active margin functions are proposed to quantify how likely these security constraints will be active, thus improving our prediction accuracy. Knowledge graph is adopted to record system working conditions, pertinent learning results and their relationship, thus improving the transferability of the learning model under varying operation conditions. Simulations have been done on IEEE 30-bus, 118-bus and 300-bus systems to demonstrate the effectiveness of the proposed DNN approach. The influence of artificial parameters and the effectiveness of the knowledge graph are also illustrated.