基于深度学习LSTM的空调故障诊断

Air conditioner fault diagnosis based on deep learning LSTM

目的:对空调的故障进行及时的检测排查并减少能源消耗,以减少人力维护成本,保证空调系统高效运行。方法:利用深度学习对空调系统进行准确诊断,使用长期短期记忆(LSTM)神经网络,针对空调的冷水机组故障数据,利用时间序列特性,搭建LSTM分类模型,同时对参数进行优化调整并进行交叉验证,以确定最优的LSTM模型参数及准确地对空调故障标签进行分类。结果:在五种不同空调故障严重程度下,该模型能够较准确的对空调故障进行诊断。结论:通过对比传统循环神经网络和它的另一个变体门控循环网络,该模型故障诊断的准确率较好,并且对新样本的适应性较好。

Aims:This paper aims to timely detect the air conditioner faults,reduce energy consumption and maintenance cost,and guarantee air conditioner operation effect.Methods:The LSTM classification model was built by using the deep learning method to diagnose air conditioner faults.The LSTM model parameters were optimized.Results:The built model can diagnose five severse air conditioner faults accurately.Conclusions:By comparing the traditional cyclic neural network with another variant gate-controlled cyclic network,the model has better fault diagnosis accuracy and better generalization performance.