关键词: 大数据, 订单完工期, 回归预测, 深度置信网络&mdash, 深度神经网络模型, 离散制造车间

Abstract: Accurate order completion time prediction is an important guarantee for production planning,scheduling,and on-time delivery in discrete manufacturing workshop.On the basis of multi-resource and massive manufacturing data,a prediction model based on Deep Belief Network(DBN)-Deep Neural Network(DNN)was designed to rapidly predict the order completion time of manufacturing system with big data characteristics.In pre-training process,DBN with ReLU activation function was trained to extract features.In predicting process,the weight and bias parameters of DBN were transmitted to DNN as initialization parameters.The dropout layer and L2 regularization were applied to avoid overfitting problems.Taking 10 000 samples with 1059 features from a spacecraft manufacturing workshop as data set,several numerical experiments were carried out to verify the feasibility of the proposed method.The comparisons with multi-hidden-layers Back Propagation(BP)neural network,Principal Component Analysis(PCA)-BP neural network and PCA-Support Vector Regression(SVR)showed that the established prediction model had better performance in terms of accuracy and applicability.

Key words: big data, order completion time, regression and prediction, deep belief network-deep neural network model, discrete manufacturing workshop