Synchronous Prediction of Multiple Energy Consumption for Cement Clinker Calcination Based on DFF-KAN Model

The largest energy consumption process in cement production is the clinker calcination process. It is of paramount importance to accurately predict the consumption of coal and electricity during this process to optimize production processes, enhance output, and reduce energy consumption. In order to address the issue of low energy consumption prediction accuracy resulting from time-varying delay characteristics and the strong coupling of production data, the dynamic feature fusion (DFF)-Kolmogorov-Arnold Network (KAN) algorithm combining DFF and Kolmogorov-Arnold is proposed. The DFF layer combines Taylor expansion and Hilbert transform to amplify the local features of the data and analyze the high and low frequency characteristics, respectively, so as to realize the feature extraction of long sequence time dependence. The adaptive weight allocation strategy is combined with KAN for sequence modeling of the extracted feature data. Compared to five mainstream time series prediction algorithms, the proposed DFF-KAN algorithm achieves higher prediction accuracy with fewer model parameters and shorter training time.