基于局部异常因子结合神经网络的制冷剂充注量故障诊断

Refrigerant Charge Fault Diagnosis Based on Local Outlier Factor Combined with Neural Network

为提升多联机系统故障检测率,本文提出了一种基于局部异常因子结合神经网络的多联机故障诊断方案,并进行制冷剂充注量实验验证该方案的可行性。研究通过局部异常因子(Local Outlier Factor,LOF)法剔除实验原始数据中的异常值,再构建反向传播(Back-Propagation,BP)神经网络进行制冷剂充注量故障诊断,同时优化模型隐含层节点数,进一步提升故障检测率。结果表明:LOF法能有效剔除多联机异常值;较BP神经网络,最优隐含层节点数的LOF-BP神经网络诊断性能增强,整体检测率提高至98.97%。

In order to improve the correct diagnosis rate (CDR) of variable refrigerant flow (VRF) system, this paper proposes a fault diagnosis based on local outlier factor (LOF) combined with neural network in variable refrigerant flow system. In addition, the feasibility of the method is verified by the refrigerant charge experiment. The strategy uses the local outlier factor method to eliminate outliers in the original data, and constructs back-propagation (BP) neural network for refrigerant charge fault diagnosis. Simultaneously, the experiment optimizes the number of hidden layer nodes of the model and the correct diagnosis rate is further improved. The results show that the outliers in variable refrigerant flow system are effectively eliminated by the LOF method. Compared with the BP neural network, the overall correct diagnosis rate of the LOF-BP neural network with the optimal hidden nodes is increased to 98.97%, and the diagnostic performance is enhanced.