关键词: 路径规划, 神经网络, 训练集构建, 卷积神经网络-长短期记忆-注意力模型, 霜冰优化算法

Abstract:

To address the high complexity of multi-unmanned aerial vehicle （UAV） cooperative path planning and the poor global planning capabilities of traditional neural network models in task scenarios, an improved convolutional neural network （CNN）-long short-term memory-Attention （LSTM-Attention） model is proposed. Firstly, an enhanced rime optimization algorithm is integrated with traditional CNN models, LSTM model and Attention mechanisms. This dynamically allocates weights and hyperparameters for CNN and LSTM, enhancing the balance between local and global to prevent local optimum while improving the model’s exploration and accuracy. Subsequently, experiments are conducted on UAV path planning, with simulations exercise for collaborative UAV search. Finally, a metric evaluation system of systems for UAV path planning is established. Experimental results demonstrate that compared to five neural network models such as back propagation neural network, radial basis function neural network, CNN, LSTM, and CNN-LSTM. The proposed model achieves a 12.7% to 23.5% improvement in prediction accuracy, a 26.7% acceleration in convergence speed, and an 18.9% increase in obstacle avoidance success rate. The proposed model effectively balances global planning capabilities with local search capabilities.

Key words: path planning, neural network, training set construction, convolutional neural network-long short-term memory-attention model, rime optimization algorithm