An online navigation ratio identification model based on the gated recurrent unit (GRU) and a state estimation extended Kalman filter (EKF) are proposed under the scenario in which multiple enemy missiles attack a stationary target using a time-cooperative guidance law. The navigation ratio identification is solved as a dynamic problem, and the time-varying navigation ratios of each missile, instead of the effective navigation constants and cooperative gains, are identified in this paper. In other words, the simplified assumption that the true value is within a known finite set, which is generally adopted in a conventional identification-estimation scheme such as multiple-model adaptive estimators (MMAEs) or interacting multiple-models (IMMs), is discarded. To increase the training speed and identification accuracy, the improved multiple-model mechanism (IMMM) is adopted, and a multiple-model layer, in which regimes representing different values are set, is connected behind a conventional neural network. Since the navigation ratios are identified online, the connections between missiles are decoupled, and only one filter is required for each missile. This could greatly reduce the computational burden of onboard computers. The effectiveness of the proposed online identification model and the performance of the state estimation filter are demonstrated through numerical simulations.

提出了一种基于门控循环单元（GRU）和状态估计扩展卡尔曼滤波器（EKF）的在线导航比识别模型，适用于多发敌方导弹使用时间合作制导律攻击静止目标的场景。导航比识别作为一个动态问题来解决，本文识别的是每枚导弹的时变导航比，而不是有效导航常数和协同增益。换句话说，在多模型自适应估计器 (MMAE) 或交互多模型 (IMM) 等传统识别估计方案中通常采用的真实值在已知有限集中的简化假设被丢弃. 为了提高训练速度和识别准确率，采用改进的多模型机制（IMMM），在传统神经网络后面连接了一个多模型层，其中设置了代表不同值的机制。由于导航比是在线识别的，导弹之间的连接是解耦的，每枚导弹只需要一个滤波器。这可以大大减轻机载计算机的计算负担。通过数值模拟证明了所提出的在线识别模型的有效性和状态估计滤波器的性能。每枚导弹只需要一个过滤器。这可以大大减轻机载计算机的计算负担。通过数值模拟证明了所提出的在线识别模型的有效性和状态估计滤波器的性能。每枚导弹只需要一个过滤器。这可以大大减轻机载计算机的计算负担。通过数值模拟证明了所提出的在线识别模型的有效性和状态估计滤波器的性能。