摘要
神经网络应用于设备状态参数预测时,如果训练样本繁多,神经网络要学习多种规律,常常导致其训练过程的无所适从,使神经网络收敛缓慢,甚至不收敛。为了解决这一问题,提出利用时序相似性搜索来选择训练样本。通过相似性搜索,保证训练样本集中反映某一类规律,神经网络只需要学习这一类规律,因此,学习速度和精度得到了很大提高。以发电机主蒸汽流量为例,对提出的方法进行了验证,并且与时序模型法及一般的神经网络方法进行了比较,结果证明所提出的方法大大减小了预测误差。
When the neural network(NN)is used in the machine condition prediction,the training samples wiTHlots of irregular rules make the NN converge very slowly or even not convergent.To solve this problem,we propose a method to select the training samples for the NN based on the similarity search of time series.Through the similarity search,the training samples represent one kind of rules.The NN trained by these samples just need to learn this kind of rules,therefore the learning speed and accuracy are boTHimproved.We take the prediction of the main gas flow in the dynamotor as an example.The performance of the proposed method is compared wiTHthat of the time series model and general NN model.The results prove that the predicted error of the proposed method is significantly reduced.
作者
张蕾
ZHANG Lei(School of Mechanical Engineering,Shanghai Dianji University,Shanghai 201306,China)
出处
《上海电机学院学报》
2020年第6期328-332,共5页
Journal of Shanghai Dianji University
关键词
时序相似性
状态预测
神经网络
machine condition prediction
similarity search
time series
neural networks