Layered online data reconciliation strategy with multiple modes for industrial processes

Data reconciliation is a crucial technique to improve the accuracy of the measured data in industrial processes. However, most traditional data reconciliation researches mainly focused on global modeling for single mode processes, but little attention was paid to multimode processes. In this paper, a layered online data reconciliation strategy based on Gaussian mixture model is proposed for complex industrial processes with multiple modes. In the proposed data reconciliation framework, Gaussian mixture model is first used to identify and partition different operating modes from process data. Then, layered data reconciliation models are established for each operating mode. In the online data reconciliation step for new data, it is reconciled with the trained reconciliation models from different modes and its posteriors corresponding to different modes are calculated for new data. Finally, the reconciled result is obtained by the weighted sum of individual reconciled data in each operating mode. The effectiveness and feasibility of the proposed data reconciliation strategy are validated through a real industrial application on the sodium aluminate solution evaporation process.