Neural network-based simulation and prediction of precise airdrop trajectory planning

Abstract

The ram-air parafoil, which can be controlled to achieve stable and precise landing, plays an important role in the field of precision airdrop. The overall goal of the paper is to discuss the effectiveness of three methods in trajectory planning and apply them to simulations and predictions under certain random conditions, thereby improving the landing precision and avoid complicated derivations in traditional dynamics. A trajectory planning model based on a back propagation neural network (BPNN) is proposed. Considering the influence of the apparent mass and random wind field, genetic algorithm (GA) is the landing points accuracy optimization algorithm supplying a database verified by Kane equation (KE) model on which the BPNN is trained, verified, and tested. BPNN is the model to predict a large number of airdrop landing points data after training. Simultaneously, the effects of the different control methods and random wind speeds on the dynamic characteristics of the parafoil are analyzed. In KE, GA, and BPNN, the BPNN model features the highest landing point precision among the three methods. The radius that 95% landing points are lied in of BPNN can reach 6.30 m, which is only 77.0% of the GA and 54.6% of the traditional KE model. In addition, the influence of cutting-in angle and transition radius on the landing point error is analyzed. Our results indicate significant potential application of GA and BPNN in the field of precision airdrop.