LSTM Soft Sensor Development of Batch Processes With Multivariate Trajectory-Based Ensemble Just-in-Time Learning

Feifan Shen,Jiaqi Zheng,Xiushui Ma,Lingjian Ye

doi:10.1109/access.2020.2988668

Abstract

To implement the quality prediction scheme for batch processes, long short-term memory (LSTM) neural network is a feasible tool to handle with the process dynamics and nonlinearity. However, a global LSTM soft sensor suffers a decline in performance facing batch-to-batch variations. To overcome the batch diversity problem and take advantage of LSTM model, a multivariate trajectory based ensemble just-in-time learning strategy is proposed in this paper. Different trajectory based similarity measurements are designed to extract historical batch trajectories which share similar spatial positions and trends. For each selected trajectory, an online local LSTM soft sensing model is constructed and the real-time quality prediction result for each local model can be obtained. Then, a weighting parameter is determined for each model by cross validation. Bringing together quality prediction results from different local models, the ensemble prediction result can be finally figured out. Two case studies are carried out to prove the effectiveness of the proposed methodology including a fed-batch reactor and the fed-batch penicillin fermentation process.

Highlights

Nowadays, the proportion of batch processes in modern industry is increasing rapidly due to the growing demand of high-value-added products [1]–[5]
Three steps are designed for a typical just-intime learning (JITL) based soft sensor, which consist of similar sample extraction, online local modeling and quality prediction
It can be inferred from the prediction results that the proposed trajectory based ensemble JITL (TEJITL)-long short-term memory (LSTM) soft sensor provides a smaller root mean squared error (RMSE) value and a larger R2 value that any individual JITL-LSTM strategy, which means the quality prediction performance can be significantly improved by the use of the multivariate trajectory based ensemble strategy

Summary

INTRODUCTION

The proportion of batch processes in modern industry is increasing rapidly due to the growing demand of high-value-added products (e.g., food, pharmaceuticals, semiconductors, polymers, etc.) [1]–[5]. F. Shen et al.: LSTM Soft Sensor Development of Batch Processes With Multivariate Trajectory-Based Ensemble Just-in-Time Learning. Three steps are designed for a typical JITL based soft sensor, which consist of similar sample extraction, online local modeling and quality prediction. The performance of the JITL/EJITL strategy used in continuous processes is limited since it only takes one single query sample for similarity measurements, where the feature of the current batch trajectory is ignored. Several LSTM soft sensor models can be constructed with the extracted batch trajectories and used for the quality prediction of online query samples. Combined with the nonlinear dynamic LSTM soft sensing model, ensembled quality prediction results can be obtained and the average prediction error is expected to be smaller than a single JITL technique based LSTM for batch processes.

PRELIMINNARIES

MULTIVARIATE TRAJECTORY BASED EJITL

Findings

CONCLUSION