A Novel Spatio-Temporal Adaptive Prediction Modeling Strategy for Industrial Production Process

Abstract

Evaporation is a key process for recycling resources and reducing environmental pollution in alumina production. Its outlet liquid material concentration is a significant production indicator for evaluating evaporation quality, and also an important basis for adjusting evaporation operation parameters. However, the quality detection of sodium aluminate solution lags behind, and the delayed production information affects the accuracy of optimization and control. Therefore, to achieve efficient and green production, a novel spatio-temporal prediction model based on mutual information is presented in this paper. First, data reconciliation is applied for preprocessing to obtain the high-quality process production information. Besides, the process mechanism model is constructed through utilizing process knowledge and balance principle. Taking into account the nonlinearity and time-varying characteristics, a spatio-temporal data-driven model with mutual information and moving window is established for mechanism error compensation. Finally, an evaporation industrial process is applied to illustrate the feasibility of the proposed prediction model, and more than 90% prediction error within the ±2% error range, which demonstrates that the proposed prediction model improves the prediction condition and performance effectively.