粗糙集-神经网络在铝电解故障诊断中的应用

目的通过对铝电解生产过程中的故障进行有效地诊断来提高铝的生产效率和节约能源．方法把粗糙集和神经网络结合起来应用在铝电解的故障诊断中．先用自组织特征映射网络（SOM）对初始数据进行离散化后得到决策表，然后用粗糙集理论对决策表进行约简得到最简决策表，根据最简决策表设计BP神经网络对铝电解中的故障进行诊断．结果用粗糙集对神经网络的输入数据进行预处理可以简化神经网络的结构，减少计算量和训练时间，从而提高整个诊断系统的诊断效率、故障诊断准确率在90％以上．结论该方法能够对铝电解中的故障做出正确的诊断．

The Application of Rough Set and Neural Network to the Fault Diagnosis of Aluminum Electrolysis

Based on effective diagnosis of the fault in the production process of aluminium electrolysis, this paper tries to advance the productivity of aluminum and conservate energy. The method of it is to apply rough set and neural network to fault diagnosis of aluminium electrolysis. The primary data was discreted by SOM and it made the decision table got. Then the paper simplified the decision table by rough set and got the easiest decision table, and designed BP Neural Network based on the easiest decision table to diagnose the fault of aluminium electrolysis. The pretreatment on incoming data of neural network with rough set can simplify the structure of neural network and reduce the amount of computation and training time. As a result, the diagnostic efficiency and accuracy rate of fault diagnosis can be increased, which is over 90%. This method can diagnose the fault of aluminium electrolysis correctly.