基于Bi-LSTM和多头自注意力的空战目标意图识别模型

Air Combat Target Intention Recognition Model Based on Bi-LSTM and Multi-Head Self-Attention

在实时空战中,准确预测敌方的目标意图对及时调整我方空战战术起到关键作用。然而,现代化的空战数据通常具有时序性、多样性和复杂性等特点。此外,传统算法仅依赖当前时刻的数据做出决策从而导致目标意图识别准确率下降。针对这些问题,本文提出了一种基于双向长短期记忆网络(Bi-LSTM)和多头自注意力(MHSA)的空战目标意图识别模型。首先,对目标状态数据进行预处理生成时序序列特征,然后使用Bi-LSTM神经网络捕捉特征序列中的双向时序关系,有助于模型更好地学习特征序列之间的长期依赖关系;其次,MHSA通过多个独立的自注意力机制将Bi-LSTM提取的特征映射到不同的子空间,从而引导模型学习时序特征不同角度的关联关系。最终通过SoftMax层输出敌方目标意图。试验结果表明,该方法有效提高了目标意图的准确率,对及时灵活调整空战策略具有科学合理的指导意义。

In real-time air combat,accurately predicting the enemy's target intention plays a key role in timely adjusting our air combat tactics.However,modern air combat data is usually characterized by temporal sequence,diversity and complexity.In addition,traditional algorithms only rely on the current moment data to make decisions,which leads to a decrease in the accuracy of target intention recognition.To address these issues,an air combat target intention recognition model is proposed based on bi-directional long short-term memory(Bi-LSTM)and multi-head self-attention mechanism(MHSA).First,the target state data is preprocessed to generate temporal sequence features,and then the bidirectional temporal relationships in the feature sequences are captured using the Bi-LSTM neural network,which helps the model to better learn the long term dependencies between the feature sequences;second,the MHSA maps the features extracted by the Bi-LSTM to different subspaces through multiple independent self-attention mechanisms,which guides the model to learn the correlation relationship between different angles of the temporal features.Finally,the enemy target intent is output through the SoftMax layer.The experimental results show that the method effectively improves the accuracy of target intent,which is of scientific and reasonable significance for the timely and flexible adjustment of air combat strategy.