AP聚类改进免疫算法用于航空发动机故障诊断

在免疫算法训练过程中引入近邻传播(AP)聚类与熵权法,对训练样本进行聚类与权值计算,将权值引入免疫算法中样本选择阈值的计算,以解决训练过程采用固定选择阈值所造成的检测器在部分区域过拟合,部分区域欠拟合的问题。结果表明:改进的免疫算法用于典型非线性函数的寻优时,迭代性能均优于传统免疫算法,并在大部分情况下优于粒子群算法与量子遗传算法,在进行某型发动机故障诊断的实例实验时,改进后的算法的诊断准确率达到98.06%,高于传统免疫算法的92.60%。 

AP clustering improved immune algorithm for aeroengine fault diagnosis

In the process of immune algorithm training, the affinity propagation(AP) clustering and entropy weight method were introduced, the training samples were clustered and weighted, and the weights were introduced into the calculation of the sample selection threshold in the immune algorithm to solve the problem of a fixed selection threshold in the training process, which led to over fitting of the detector in a partial area, and under-fitting of the partial area. Result showed that, when the improved immune algorithm was used for the optimization of typical nonlinear functions, the iterative performance was better than the traditional immune algorithm. In most cases it was better than the particle swarm optimization algorithm and the quantum genetic algorithm, in the case of a certain type of engine fault diagnosis. The improved algorithm had a diagnostic accuracy of 98.06%, which was higher than 92.60% of the traditional immune algorithm.