Knowledge Transfer for Long-term Voltage Stability Assessment Between Power Grids Based on Deep Domain Adaptation Networks

For machine learning techniques, it is difficult to transfer knowledge between different domains due to the system discrepancy and limited training data. In this paper, we implement a knowledge transfer system (KTS) for long-term voltage stability assessment between power grids. First, system behaviours are converted to heatmaps with a more general method to avoid negative transfer. Then, a deep domain adaptation network (DDAN) is introduced to learn domain-invariant representations with strong semantic separations by adding a maximum-mean-discrepancy calculator. The KTS based on DDAN is performed to transfer knowledge from IEEE 39-bus system to IEEE 14-bus system. Case studies are also given to show its potential under those scenarios of data shortage.