论文部分内容阅读
提出一种可用于非线性系统建模的鲁棒自联想神经网络,将该网络映射层和解映射层分别作为2个子网络进行训练,提高了收敛速率。在网络训练目标函数中加入网络权值限制项,抑制了训练过程中网络权值的异常调整,提高了模型的准确性和鲁棒性。在解映射子网络训练结果集和原输入数据集中增加了扰动数据,构成映射网络的扩展训练样本集,提高了整个网络的鲁棒性。该文给出基于所提网络模型的传感器故障诊断方法及诊断流程,并以某300 MW机组热力系统为对象进行算例分析,结果表明该文方法应用于传感器故障诊断时能够实现对故障测点的快速准确定位,并对各变量值进行精确重构。
A robust self-associative neural network that can be used in nonlinear system modeling is proposed. The network mapping layer and de-mapping layer are respectively trained as two sub-networks to improve the convergence rate. By adding the network weight limit term to the network training objective function, the abnormal adjustment of the network weight in the training process is suppressed, which improves the accuracy and robustness of the model. The disturbance data are added to the de-mapping sub-network training result set and the original input data set to form an extended training sample set of the mapping network, which improves the robustness of the whole network. In this paper, a fault diagnosis method and diagnosis process based on the proposed network model are given. An example is given to analyze the thermal system of a 300 MW unit. The results show that this method can be used to detect fault points Fast and accurate positioning, and accurate reconstruction of the value of each variable.