Abstract: The existing gas disaster warning systems suffer from issues such as poor data transmission stability, high latency, and delayed warning responses. To address these problems, a gas emission anomaly warning device with multi-parameter sensing, multi-mode communication, and independent operation capabilities has been developed to enable real-time on-site monitoring and accurate warning of gas disasters. The device integrates an environmental parameter sensing module, a power module, a multi-mode communication module, an analysis and warning module, and an accident alarm module. It can simultaneously collect seven environmental parameters, including methane and wind speed, and supports five modes of data transmission. It is adaptable to a wide voltage range of 127 V to 660 V underground and can operate independently without relying on safety monitoring systems, achieving real-time on-site monitoring and warning of gas disasters. The research also established gas over-limit warning technology based on Long Short-Term Memory(LSTM)artificial neural networks, gas emission anomaly warning technology based on gas monitoring statistical data, and a warning indicator system for coal and gas outbursts, enabling accurate monitoring and warning of hazards such as abnormal gas emissions, gas over-limit, and coal and gas outbursts. The warning device has completed its industrial trials at the Xiangshan Mine of Shaanxi Coal and Chemical Industry Group Hancheng Mining Co., Ltd. The test results show that the warning device can predict the methane concentration 5 minutes ahead with an average absolute percentage error of 0.98%. The coal and gas outburst indicator A can effectively reflect the change trend of the gas desorption indicator K₁ in drill cuttings, and it has demonstrated good warning effects for both gas exceedance warnings and coal and gas outburst warnings.