An Advanced Robust Estimator-Based Enhanced LSTM Neural Network for Data Reconciliation Considering Measurement Uncertainty

During evaporation process for alumina production, various measuring instruments are inevitably affected by strongly alkaline and high-temperature conditions, resulting in measurements with random error. In addition, due to the improper operation and interference of instrument leakage, the measurements have gross errors with varying degrees of uncertainty. The accuracy of the process modeling, optimization, and control are affected by the inaccurate measuring information for industrial production. Therefore, an advanced robust estimator-based enhanced long short-term memory (LSTM) neural network for data reconciliation considering measurement uncertainty is developed in this article. First, a new robust estimator (NRE) is present to address the measurement errors with uncertainties, effectively eliminating gross errors and reducing random errors. Moreover, considering that the complex industrial process constraints are often not readily accessible, the loss function is replaced by the present robust estimator which is introduced into the LSTM network, and an advanced robust estimator-based enhanced LSTM network is utilized to extract process correlations for solving data reconciliation. Furthermore, the developed data reconciliation strategy is verified through numerical experiments and two actual industrial cases. The results demonstrate a marked enhancement in the accuracy and reliability of the measured data, which is crucial for stable operation, efficient production, and green energy saving of industrial processes.