The landing phase is a critical part of the flight process, and heavy landings pose a significant threat to flight safety. When an aircraft experiences a vertical acceleration (VRTG) of 2g or higher during landing, the fuselage structure is subjected to immense stress, which can easily cause damage to the aircraft and, in severe cases, lead to casualties. Therefore, accurately predicting the VRTG during landing and issuing early warnings for heavy landing risks is crucial for ensuring flight safety and reducing accident rates. This project focuses on early warning systems for heavy landings by analyzing real-time aircraft sensor parameters to predict future VRTG values. The key technology employed is time-delay neural networks (TDNN), a unique network structure that effectively captures temporal correlations in time-series data. Compared to traditional prediction methods, this approach demonstrates stronger adaptability and processing capabilities for the complex and dynamic parameters involved in aircraft landings. The experimental validation and tests using actual flight data show that the mean squared error (MSE) of this method’s predictions can reach the order of 10^-3. This metric indicates that the method achieves extremely high precision in predicting VRTG during aircraft landings.