Abstract
Soft sensors based on deep learning have been growing in industrial process applications, inferring hard-to-measure but crucial quality-related variables. However, applications may present strong non-linearity, dynamicity, and a lack of labeled data. To deal with the above-cited problems, the extraction of relevant features is becoming a field of interest in soft-sensing. A novel deep representative learning soft-sensor modeling approach is proposed based on stacked autoencoder (SAE), mutual information (MI), and long-short term memory (LSTM). SAE is trained layer by layer with MI evaluation performed between extracted features and targeted output to evaluate the relevance of learned representation in each layer. This approach highlights relevant information and eliminates irrelevant information from the current layer. Thus, deep output-related representative features are retrieved. In the supervised fine-tuning stage, an LSTM is coupled to the tail of the SAE to address system inherent dynamic behavior. Also, a k-fold cross-validation ensemble strategy is applied to enhance the soft-sensor reliability. Two real-world industrial non-linear processes are employed to evaluate the proposed method performance. The obtained results show improved prediction performance in comparison to other traditional and state-of-art methods. Compared to the other methods, the proposed model can generate more than 38.6% and 39.4% improvement of RMSE for the two analyzed industrial cases.
Highlights
Several hardware sensors supply data for monitoring and control process optimization in industrial production processes [1]
The narrowest box ranges of eMISAEL and MISAEL indicate the best prediction performance between the six compared methods mainly because MISAEL extracts non-linear features, selects the most relevant representations, and copes with the dynamicity of the process
EMISAEL and MISAEL are the models with the narrowest box ranges mainly because MISAEL extracts non-linear features from massive unlabeled data, retains only the relevant representations, and addresses process dynamicity
Summary
Several hardware sensors supply data for monitoring and control process optimization in industrial production processes [1]. The unsupervised layer-wise pre-training and supervised fine-tuning procedures allow deep structures to outperform the prediction performance of traditional techniques for soft-sensing. Proposed LSTM soft-sensor models for predicting boiling points of heavy naphtha and aviation kerosene in [43] Those works do not use unlabeled samples for unsupervised pre-training, which may cause poor feature representation. In [45], an Xgboost is used to select features, acting as an encoder to feed a soft-sensor based on LSTM that extracts dynamic information hidden in-process data. A novel semi-supervised soft-sensor modeling based on deep representative learning is proposed to enhance soft-sensing prediction performance. A deep representative learning method extracts high-level features from unlabeled data and eliminates non-relevant representations and highlights relevant information for efficient soft-sensing development.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
