High Computationally Efficient Predictive Entry Guidance with Multiple No-Fly Zones

Abstract

This study proposes a high computationally efficient data-driven predictive entry guidance method for hypersonic vehicles under multiple no-fly zones. The method uses a reduced-order motion-model-based semianalytic guidance framework to obtain a trained neural network that only requires two-dimensional input. First, the sixth-order entry dynamic motion model is simplified to a third-order model by considering height as the independent variable. Second, based on the reduced-order motion model, a novel exponential function is introduced to yield a semianalytic range-to-go expression in longitudinal guidance. Third, to generate sample trajectory data for training the neural network, the semianalytic guidance framework is supported by the reduced-order motion model with the semianalytic range-to-go expression. Then, a new dynamic lateral guidance reversal logic based on a chain mode strategy is employed to avoid no-fly zones with different configurations and numbers. Finally, to obtain real-time trajectory online, a data-driven online predictive guidance method is proposed based on a back propagation neural network trained by sample trajectory data generated by the semianalytic guidance framework. The proposed method overcomes the drawbacks of most predictor–corrector guidance methods; i.e., the corrected guidance parameters are heavily dependent on the initial values of each iteration in each guidance cycle. Advantageously, the proposed method greatly reduces the online command calculation time in one guidance cycle and only requires two input data to train the neural network, i.e., height and range-to-go, thus yielding results that are close to the engineering reality. The effectiveness of the proposed method is verified through simulations.