Data-Driven Hidden Fault Mechanism and Early Warning Method for Ultra-High Voltage Direct Current Control and Protection System
Weitao Li, Zihao Wan, Ao Shen, Kaige Zhao · 2024
Aiming at the problem that the alarm information of Ultra-High Voltage Direct Current (UHVDC) control and protection system is dense and leads to hidden faults that are difficult to be uncovered, a short text classification model incorporating adversarial training of self-attentive multilayer bi-directional long-and-short-term memory networks (Con-Att-BiLSTMs) and an associative inference method combining the topology of UHVDC control and protection system are proposed. The text training set is classified in different proportions for adversarial training, which improves the robustness of the model. The semantics are extracted using multilayer bi-directional short-term memory networks, and the semantic information is weighted and strengthened using the self-attention mechanism layer. The loss rate is minimized by the SoftMax function. Then, combined with the correlation reasoning method of the topology of the UHVDC control and protection system, the hidden faults of the UHVDC control and protection system are reasoned out. The experimental results show that the method can effectively extract the alarm information of the UHVDC control and protection system and get the hidden faults.