Product quality time series prediction with attention-based convolutional recurrent neural network

The product quality is the key index to measure the process of the industrial manufacture. Thanks to the ever-expanding scale of time-series data, the deep learning technology can be regarded as the effective approach to predict the future product quality accurately. In this article, the product quality with time series data is considered and an attention-based convolutional recurrent neural network (ACRNN) is proposed for the prediction of the product quality. Firstly, by reconstructing the time series data into the two dimensions the convolutional layers are built to compress the information of the product quality data, and the more comprehensive features can be extracted. Furthermore, to ensure the prediction accuracy of the time series data process and extract the 2-dimmension feature of the data, the long short-term memory (LSTM) based on recurrent neural network (RNN) layers are constructed. After that, the fully connection layers with attention mechanism is applied to improve the calculation efficiency and accuracy of the models. Finally, the test experiments on the time series data of the industrial product are given and the comparisons show that the effectiveness of the proposed algorithm framework.