基于免疫算法的自组织神经网络在效能评估中的应用

针对自组织神经网络自身的局限,将免疫克隆选择算法的克隆和变异机制引入SOM的学习算法中,提出一种免疫自组织神经网络模型,并建立了模型的学习算法。该学习算法用免疫克隆选择算法的克隆算子和变异算子改进自组织神经网络中的邻域大小和权值调整规则,使每个神经元的权值学习率和邻域大小随神经元的亲和力发生变化,从而克服了自组织神经网络分类效果受样本输入次序影响的弱点,且在很大概率上保证网络收敛到全局最优解。性能仿真结果说明该学习算法比自组织神经网络学习算法具有更好的分类准确性和泛化性能。将该模型应用雷达电子战装备的作战效能评估中,结果表明免疫自组织神经网络模型比自组织神经网络模型分类更合理。

Self-organizing Map Neural Network Based on Immune Algorithm and Its Application in Effectiveness Evaluation

Aiming at the deficiency of self-organizing map neural network (SOM),an immune self-organizing map neural network (ISOM) model was proposed by combining the immune clonal select algorithm with self-organizing map neural network,and a learning algorithm was given out.The ISOM made use of immune clonal operators and mutation operators to improve neighborhood size and weight update rule,which made the neurons learning rate and neighborhood sizes changing with its affinity and overcame the disadvantage of SOM cluster result disturbed by the order of learning samples and ensured network convergence to global optimization by and large.The simulation results show that ISOM algorithm is superior to SOM in classification accuracy and generalization performance.The application of the model to system effectiveness evaluation of electronic warfare equipment shows that ISOM is more reasonable than SOM.