相比现有的干扰抑制技术（自适应滤技术、自适应干扰对消技术），盲源分离技术能够分离混合在一起的多个信号，计算复杂度低，鲁棒性强。但盲源分离技术难以全面覆盖井下复杂多变的干扰源，同时缺乏对处理后信号成分的自动分析与评估机制，不仅限制了通信效率的提升，还可能因干扰残留而引发安全隐患。针对上述问题，提出了一种基于神经网络的矿用5G通信信号传输干扰监测抑制方法。通过分析井下主运输大巷、综采工作面和变电所等区域的干扰源特点，指出毛刺干扰及串扰信号的抑制和处理是5G抗干扰问题的关键。采用盲源分离技术初步分离矿用5G通信信号中的干扰成分，利用神经网络对分离后的信号进行特征提取及深度分析，精准识别并量化其中残留的干扰信号，一旦监测到干扰信号超出预设阈值，将自动触发新一轮的干扰抑制流程，形成迭代优化的闭环控制。实验结果表明：① 在100 MHz全带宽发送的环境中，使用矿用5G通信信号干扰监测抑制方法能够对毛刺干扰与串扰信号实现13 dB的干扰抑制增益，比使用盲源分离干扰抑制方法效果提升约117%及86%。② 矿用5G通信信号干扰监测抑制方法较盲源分离等传统干扰抑制技术，信噪比平均提升了15.56%，误码率平均降低了21.88%，能够显著提升信号质量。

Compared to conventional interference suppression techniques, such as adaptive filtering and adaptive interference cancellation, blind source separation (BSS) offers the advantage of separating multiple mixed signals with lower computational complexity and higher robustness. However, BSS has limitations in addressing the complex and dynamic interference sources found in underground mining environments. It also lacks automated mechanisms for analyzing and evaluating processed signal components, which not only hampers communication efficiency but may also lead to safety risks due to residual interference. To overcome these challenges, this study proposed a neural network-based interference monitoring and suppression method tailored for mine-used 5G communication signal transmission. By analyzing the characteristics of interference sources in key areas such as main haulage roadways, fully mechanized mining faces, and substations, the study identified the suppression and mitigation of spike interference and crosstalk signals as critical for 5G anti-interference performance. The proposed method initially employed BSS for the preliminary separation of interference components in mine-used 5G communication signals. It then leveraged a neural network for feature extraction and deep analysis of the separated signals, enabling precise identification and quantification of residual interference. If the monitored interference signal exceeded a preset threshold, the system automatically triggered a new round of suppression, forming an iterative and optimized closed-loop control process. Experimental results revealed that: In a 100 MHz full-bandwidth transmission environment, the proposed method achieved a suppression gain of 13 dB for both spike interference and crosstalk signals, showing improvements of approximately 117% and 86%, respectively, over BSS-based interference suppression alone. Compared to traditional techniques, such as BSS, the proposed method enhanced the signal-to-noise ratio (SNR) by an average of 15.56% and reduced the bit error rate (BER) by an average of 21.88%, which could significantly improve signal quality.