A Landing-Point Prediction and Control Algorithm Based on a Bidirectional Long Short-Term Memory Neural Network Model

To address the challenges of discrete control and real-time decision-making for impulse-controlled vehicles, this study proposes a landing-point prediction and control algorithm based on a Bidirectional Long Short-Term Memory (Bi-LSTM) neural network. The algorithm employs the Bi-LSTM network to accurately predict the vehicle landing point in real time, and subsequently computes the vehicle–target deviation to determine the optimal impulse-ignition parameters. Simulation results demonstrate that the Bi-LSTM prediction model significantly outperforms traditional BP, LSTM, and GRU networks, achieving a final landing-point prediction error with a mean per-axis RMSE of 1.96 m (three-dimensional error norm of 3.87 m). Based on these prediction outcomes, two control strategies—“nearest-point ignition” and “ignition delay”—are compared. Monte Carlo experiments indicate that both strategies effectively enhance landing accuracy, with the nearest-point ignition strategy showing superior performance, reducing the circular error probable (CEP) from the uncontrolled 72.60 m and 124.82 m (at launch angles of 23° and 51°, respectively) to 7.20 m and 10.30 m. This study provides a high-precision, terminal-phase control solution for impulse-controlled vehicles, with implications for the low-cost onboard implementation of discrete impulse control.