In cement production, clinker is the key component, and its quality directly affects the overall performance of cement. The content of free calcium oxide (f-CaO) in cement clinker is one of the important parameters to evaluate the quality of cement clinker. In order to make up for the lack of timeliness of traditional laboratory methods, this paper constructs an efficient and accurate soft sensor model for cement clink f-CaO content, which combines channel attention mechanism and long short-term memory network. The feature extraction method combined with attention mechanism was used to extract the features of the data set. Then, the features were input into the LSTM network for learning, so that the model could focus on the core feature channels in an adaptive manner. Due to the poor prediction effect of LSTM on data with large volatility, the above soft sensor model is improved. Based on the original model, the classification module and weighting module are introduced to modify the prediction results of the LSTM network, so that the model can be more flexible to adapt to the differences between different categories. The accuracy of the model prediction is improved. The experimental results show that the coefficient of determination (R2), Root Mean Square Error (RMSE) and Mean Absolute error (MAE) of the improved SE-LSTM prediction model for cement clink f-CaO are significantly improved compared with the ordinary LSTM and SE-LSTM prediction models. Therefore, the improved model proposed in the prediction of cement clinker f-CaO content improves the prediction accuracy and achieves the expected prediction effect.