论文部分内容阅读
为了提升储备池的动态适应性能,克服回声状态网络(Echo State Network,ESN)输出权值求解的病态不适定问题,平衡其拟合与泛化能力,提出一种基于L_(1/2)范数正则化的塑性回声状态网络故障诊断模型。在储备池构建中引入BCM规则(Bienenstock-Cooper-Munro rule)对连接权矩阵进行预训练,并在目标函数中添加L_(1/2)范数惩罚项以提高稀疏化效率,利用一个光滑化的L_(1/2)正则子克服迭代数值振荡问题,最后采用半阈值迭代法对模型进行求解。将模型应用于机载电台的故障诊断问题中,仿真结果证明了模型的有效性和优越性。
In order to improve the dynamic adaptability of reserve pool and overcome ill-posed ill-posed problems of Echo State Network (ESN) output weights and to balance its fitting and generalization ability, an L_ (1/2) Number Regularized Plastic Echo State Network Fault Diagnosis Model. BCM rule (Bienenstock-Cooper-Munro rule) is introduced into the reserve pool to pre-train the connection weight matrix, and L_ (1/2) norm penalty term is added to the objective function to improve the sparsity efficiency. By using a smoothing The L_ (1/2) regular subproblem overcomes the problem of iterative numerical oscillation. Finally, the model is solved by the semi-thresholding method. The model is applied to the fault diagnosis of airborne stations. The simulation results show the effectiveness and superiority of the model.