基于LSRBF神经网络的空战目标威胁评估

Target threat assessment based on LSRBF neural network for air combat

提出了一种基于Log-Sigmoid型径向基(简称LSRBF)神经网络的空战目标威胁评估方法。采用威胁指数法量化各因素的威胁度,运用工程模糊集方法确定因素的权重系数,在此基础上合成目标总的威胁指数,作为网络的初始训练样本。根据专家经验对不合理样本进行调整校正,得到最终的训练样本,供LSRBF神经网络训练使用。采用标准梯度下降法与指数梯度下降法相结合的学习算法,保证网络具有较强的鲁棒特性。仿真实验结果表明,LSRBF神经网络具有很好的函数逼近性能,可以成功地完成空战目标的威胁评估。

A method of target threat assessment based on Log-Sigmoid Radial Basis Function （LSRBF） neural network for air combat is proposed. The threat level of each factor is quantified by threat index method, and the weights of all factors are determined by engineering fuzzy set method. Then the initial training patterns are formed. According to expert experiences, an adjustment is made for adjusting the unreasonable training patterns. The resulted training patterns are supplied to LSRBF neural network. The learning algorithm is composed of traditional gradient descent method and index gradient descent method, thus can ensure high robustness of LSRBF neural network. The simulation results show that due to the perfect function approximation capability, LSRBF neural network can be used successfully to accomplish target threat assessment for air combat.