Abstract: In order to solve the problem that the influence of output value on the delay of process parameters is not considered in the prediction model of cement kiln temperature, a method of fuzzy curve dynamic time delay estimation and time series weighting is proposed. The total delay of each parameter was determined according to the fuzzy curve, and the dynamic delay was calculated by the adaptive sliding window. A Gaussian function was introduced to carry out timing weighting to obtain the most causal input-output samples. The stratified extreme learning machine was used to train and establish the temperature prediction model of cement kiln head. The experimental results show that considering the dynamic time delay information and time series weighting can improve the prediction accuracy of the cement kiln temperature prediction model.