A DeepAR based hybrid probabilistic prediction model for production bottleneck of flexible shop-floor in Industry 4.0

Abstract

Predicting production bottleneck plays an important role in dynamic production process optimization and throughput improvement of flexible shop-floor. However, great challenges aroused by the high uncertainty of shop-floor environment and the interactive effects from many influencing factors still remains on production bottleneck prediction. Therefore, a Deep Auto-Regressive Neural Network with hybrid model (H-DeepAR) is developed to provide production bottleneck probabilistic prediction. In this proposed model, convolution neural network (CNN) and multi-head attention mechanism (MHA) are constructed to capture high-level representations from the multivariate input data. Then, a gated recurrent unit (GRU) is leveraged to capture temporal correlations of production bottleneck. After that, fully connection layers (FCN) are designed to make production bottleneck probability prediction. Finally, a case study based on aircraft overhaul shop-floor is illustrated to validate the effectiveness and superiority of the proposed method. The results indicate that the proposed model yields better performance than the benchmark models.