Operational Behavior Diagnosis of Protective Relays Based on Temporal Convolutional Network

Protective relays are crucial for ensuring the safe and reliable operation of power systems. Traditional diagnosis for relay operation often heavily depend on the expertise of substation personnels. However, as the power grid expand, it faces the issue of insufficient operation and maintenance staff, which can cause delays and inaccuracies in fault analysis. Recent advancements in artificial intelligence and substation data acquisition techniques have brought new solutions for operational behavior diagnosis. To address the issue, this paper proposes a novel behavior diagnosis approach for protective relays utilizing temporal convolutional networks (TCN). First, the behavior diagnosis of relays is converted to a binary classification task and formulated. Second, the fault recordings are treated as the time series input, and one-dimensional convolution is calculated along the time dimension. Then, global average pooling is applied to adapt to varying sampling points. Finally, case studies are conducted in the PSCAD simulation environment. The result shows the classification accuracy reaches 99.95% in 0.209 seconds, which demonstrate the effectiveness and efficiency of the proposed method.