Low-latency Intelligent Identification Method for Abnormal Communication State of Off-grid Microgrid

There is a delay in the communication state of offgrid microgrids, but the abnormal state recognition process relies too much on historical data and experience to set the normal range threshold of the state parameters. However, the fixed threshold cannot reflect the communication changes in time, which easily leads to low recognition efficiency. Therefore, a low-latency intelligent recognition method for abnormal communication state of off-grid microgrids is designed. Sensors are used to monitor signal changes, and signal data is mapped to a specific interval. The signal is decomposed into low-frequency and high-frequency information, and its wavelet packet energy entropy is regarded as a fault feature. Based on the extracted features, a low-latency intelligent recognition model for abnormal communication state of off-grid microgrids is established by combining BP neural network, and the abnormal state category is determined by comparing probabilities. Experimental results show that the recognition delay of the intelligent recognition method is stable at $\mathbf{3 0} \sim \mathbf{5 0}$ s and has no cumulative effect. The AUC value of the ROC curve reaches $\mathbf{0. 9 7}$. It has high sensitivity to both sudden and gradual abnormalities. The confusion matrix shows that the overall accuracy rate is as high as 98.1%. It can meet the stringent requirements of millisecond response and low false alarms of off-grid microgrids, and provide highly reliable and stable real-time detection technology support for microgrid communication security.