SOABiLSTM-based nondestructive evaluation of electromagnetic interference in Coal-Rock fractures

Accurate rockburst risk assessment is vital for mine safety. Electromagnetic radiation (EMR) monitoring is widely used, yet interference from underground operations and equipment degrades data quality. We propose an intelligent method to recognize EM interference using a Sparrow Search Algorithm–optimized bidirectional long short-term memory (SSA-BiLSTM). We first extract dimensional and dimensionless time-domain features (e.g., maxima, RMS, variance, standard deviation, kurtosis, waveform, peak and pulse factors) from large sets of effective EM signals and diverse interference types. The SSA tunes BiLSTM hyperparameters to improve sequence learning without manual parameter adjustment. We evaluate the model on labeled underground EM datasets collected under varying periods and operating conditions. The approach achieves 94.6% accuracy on a held-out test set and 90.6% accuracy on an independent, temporally separated dataset, indicating strong generalization. Compared with conventional manual identification and non-optimized deep networks, the SSA-BiLSTM delivers higher accuracy and efficiency, supporting reliable EMR monitoring and practical deployment for rockburst early warning.