Research on Ballistic Planning Method Based on Improved DDPG Algorithm

Aiming at the ballistic planning problem given the target position indication under the condition of satisfying the missile capability boundary and process constraints, an improved deep deterministic policy gradient (DDPG) algorithm is proposed. This method solves the problems of the traditional trajectory planning method, such as strong dependence on model, low adaptability to environment and lack of real-time performance. The agent is firstly trained by this algorithm at the initial stage in a simple environment through a guided training based on the slow training speed of deep neural network, and then the generated strategy is adopted as an initial strategy for retraining. In the process, the sampling strategy of experience-first playback is applied to the gradient algorithm of deep deterministic strategy, and the Ornstein-Uhlenbeck (OU) noise model is employed for increasing the exploration ability of the agent. Then, this work develops a comprehensive simulation model according to an actual environment, where an improved DDPG method is thus introduced into the ballistic planning field by combining with performance indicators for planning and the trajectory can also meet the constraints. In order to verify the effectiveness of this method, the Gaussian pseudo-spectrum method is also used to compare with the proposed method in this paper. The simulation results show that the ballistic planning time of this improved algorithm is only within 10ms, which is about 8 times shorter than that of Gaussian pseudo-spectrum method. The corresponding ballistics that satisfy the constraints can be quickly generated, which has a good convergence and robustness, and is also suitable for solving online ballistic planning problems.