Abstract: Accurately identifying the type of microseismic signals is of great significance for monitoring, warning, and preventing dynamic disasters in rockburst mines. Leveraging the engineering context of Nantun coal mine, we classify five types of microseismic signals, namely mine earthquakes, blasting, coal-rock fractures, knocking, and noise. We establish a sample database comprising five types of microseismic signals. We analyze the parameters of microseismic events, including maximum amplitude, duration, and main frequency, and develop a BP neural network prediction model for microseismic event types. The results indicate that mine earthquake events are associated with the fracturing of high-hardness rock above the coal seam, and their energy generally exceeds 10⁵ J, which can trigger more than half of the sensors. These events are characterized by high maximum amplitude and high average amplitude (with a threshold of 3×10⁻⁴ m/s). Their duration typically does not exceed 6 seconds, and the main frequency is generally below 10 Hz. Blasting events typically have energy levels of not less than 10⁴ J, capable of triggering more than half of the sensors in the mine. These events exhibit high maximum amplitude and high average amplitude, with durations generally less than 2 seconds. Coal-rock fracture events are associated with the periodic rupture of strata, and their energy typically does not exceed 10⁴ J, triggering between 40% and 65% of mine sensors. These events are characterized by low maximum amplitude and low average amplitude, with durations ranging from 0 to 1 seconds, and a relatively uniform distribution of main frequencies. Knocking events are associated with human testing and typically trigger fewer than 4 sensors. These events exhibit high maximum amplitude and low average amplitude, long duration, and a uniform main frequency distribution within the 0-80 Hz range. Noise events are related to electromagnetic interference and other factors. Most microseismic sensors experience erratic vibrations, with generally fewer than 6 sensors being triggered. These events are characterized by low maximum amplitude, low average amplitude, extended duration, and irregular main frequency. The established BP neural network prediction model achieves an accuracy rate of 88.3%. The method boasts a low misjudgment rate and ease of obtaining characteristic parameters, offering a novel approach for rapidly and accurately identifying microseismic events in rock burst mines.