Evaluating and predicting energy efficiency using slow feature partial least squares method for large-scale chemical plants

As demands for energy conversion and management are increasing, accurate evaluation and prediction of energy efficiency are the prerequisites of making production strategies to improve economic benefits in the chemical industry. With the characteristics of slow variations in the chemical process and temporal correlations in the sampling data, it is more difficult for conventional estimation methods to precisely evaluate and infer the energy efficiency. To overcome these difficulties, this paper proposes a novel energy efficiency evaluation and prediction method, named slow feature partial least squares. The proposed method is integrated with slow feature analysis and partial least squares. Slow feature partial least squares can extract dynamic features from temporal behaviors of chemical products and energy media in a supervised manner and construct the model relationship. With the established model, not only are the energy efficiency levels evaluated accurately, but the tendency of energy efficiency is also inferred in advance. The effectiveness and practicality of the slow feature partial least squares method are validated by a mathematical example and an actual ethylene process. The energy efficiency results provide great support for the enhancement of energy efficiency and the development of production strategies in chemical plants with high energy consumption.