When sampling the internal cooling water flow, inlet and outlet valve temperature, main pump and return water pressure of the valve cooling system in the converter station, data loss or abnormality often occurs. Meanwhile, due to the timing of the above operating parameters of the valve cooling system, the accurate evaluation of the cooling capacity of the valve cooling system can be achieved by completing and predicting effectively key parameters such as the inlet water temperature of the internal cooling system. In this paper, a sampling data processing method based on mean filling is proposed to make the sample set more perfect and fit the actual operation of valve cooling system in converter station. By analyzing the time-series properties of valve cooling system operation parameters, a time series prediction method based on Principal Component Analysis and Long-Short Term Memory (PCA-LSTM) model is proposed. By predicting the key operating parameters of converter valve, the cooling capacity of the system can be effectively evaluated. The simulation shows that the average absolute percentage error of the prediction model is about 4.96%, which proves the effectiveness of the model.