Research on Flight Attitude Prediction Method for Multi-rotor UAV Based on CNN-LSTM-attention Model

This paper proposes a unmanned aerial vehicle (UAV) flight attitude prediction method utilizing a convolutional neural network (CNN) and long short-term memory (LSTM) network and Attention mechanism. It has become particularly important to predict flight attitude accurately with the wide application of UAVs in many fields such as, aerial photography, logistics, and surveillance. The proposed method uses CNN to extract spatial patterns of UAV flight data, and LSTM to learn the temporal dependencies of these features to capture dynamic changes in flight attitude. The model introduces attention mechanism, empowering it to prioritize the parts of the data that are more critical to the prediction results. Experimental results with a certain type of multi-rotor UAV flight parameters show that the proposed model predicts the pitch angle with high prediction accuracy and keeps the error low. Through comparative experiments, the CNN-LSTM-Attention Model, compared to simple CNN and simple LSTM models, has improved accuracy, slightly decreased error, stronger generalization ability, and can effectively predict the UAV flight attitude.