Search

• A dynamic time-delay calculation algorithm based on MI and adaptive sliding window (MIDTA) is proposed. • MIDTA algorithm is used to solve the problem of long time-delay of cement kiln. • An algorithm of residual network and bidirectional novel gated recurrent unit (ResNet-BiNGRU) is proposed. • A Gaussian-weighted multi-model fusion method is proposed. • MIDTA-ResNet-BiNGRU multi-model fusion algorithm can be used to measure the temperature of cement rotary kiln with high precision. China's cement production energy consumption accounts for more than 10% of the country's total energy consumption, and its emissions account for about 60% of the country's industrial emissions. The temperature of a cement rotary kiln directly affects the quality of the final product, the energy consumption of a plant, and emissions. Therefore, it is necessary to measure, to a high degree of accuracy, the temperature of a cement rotary kiln to reduce energy consumption, emissions, and improve product quality. However, the conditions in cement kilns are complex and variable; and there are large time-delays, while temperatures are affected by historical working conditions, making temperature measurement extremely difficult. In this paper, a modelling algorithm with a residual network and a bidirectional novel gated cycle unit is used to predict the clinker exit temperature of the rotary kiln. In addition, to solve the problem of large time-delays in cement kilns, a dynamic time-delay calculation algorithm based on Mutual information and adaptive sliding windows is proposed for time series data reconstruction, and finally a Gaussian-weighted multi-model fusion is applied to the prediction results for cement rotary kilns that produce a wide range of working conditions. To verify the advantages and feasibility of the proposed method, experimental validation is carried out using original production data from a large cement manufacturing enterprise in South China. The results demonstrate that the single step prediction time of the presented approach is 0.66 s, which can predict cement kiln temperatures quickly and accurately, compared to a single conventional forecasting algorithm, the average temperature difference dropped from 11.76 °C to 3.51 °C, R2 increased from 0.43 to 0.89, 94% of the depicted temperature errors are at ± 10 °C This will provide an effective solution for online decision-making during actual cement production. [ABSTRACT FROM AUTHOR]