基于集成神经网络的多故障诊断方法

铝电解过程是一个非线性、多耦合、时变和大时滞过程,受强电场、强磁场、强热场交互干扰,形成了复杂多变的槽况特征,故障种类繁多,发生频繁,有效地故障预报和诊断,对电解系列平稳供电,节约电能、提高铝的产量和质量有重要意义。根据铝电解过程故障特点,提出了基于主成分分析的集成神经网络铝电解多故障诊断方法,建立分层故障诊断模型结构,包括子神经网络层和决策融合神经网络层,子神经网络模块采用了改进型的Elman神经网络,强化信息的记忆功能,并通过主成分分析优化了神经网络结构;决策融合神经网络通过各子网络传递的相关信息,进一步验证对子神经网络诊断结果和复合故障进行综合决策。仿真结果表明,具有良好的诊断效果,验证了该故障诊断方法的可行性和有效性。

Research on Multi -fault Diagnosis Method Based on Integrated Neural Network

Aluminum electrolysis is a nonlinear,more coupling,time-varying,and a large delay process,which is interfered by the interaction of strong electric field,strong magnetic field,and strong heat field.Because of that it forms the complicated and changeable characteristics of slot status and concerns a wide variety of faults which happen frequently.So it is very important for power supply smoothing of electrolytic series,energy saving,and the output quality improvement of the aluminum.According to the diagnosis characteristics of the aluminum electrolysis process,the study shows the integrated aluminum electrolytic multi-fault diagnosis method based on the principal component analysis and builds the layered fault diagnosis model structure including the sub-neural network layer.Improved Elman neural network is adopted in the sub-neural network module which strengthens the information of memory,and the neural network structure is optimized through the principal component analysis.A further validation of the comprehensive decision of the diagnosis results of the sub-neural network and complex fault is made by the decision fusion neural network model through the relevant information passing by each sub-network.Simulation results not only show the advance of the diagnosis effect but also verify the feasibility and effectiveness of this fault diagnosis method.