Abstract: The processes in converter steelmaking blowing stage mainly include oxygen supply, slag discharge and bottom blowing. And the quality of molten steel at the end is directly affected by the stability of blowing process. Traditional static control method obtains the blowing process model based on material balance and heat balance, without considering the strong coupling relationship between the raw materials and the process parameters, resulting in a low reliability. What’s more, the data types of raw materials and process parameters are scalar and time series respectively. In order to extract the features of the complex mixed data mentioned above, a process model extraction method of converter steelmaking based on improved autoencoder (IAE) is proposed. The IAE method is based on the autoencoder, including fully connected modules, long short-term memory network modules, one-dimensional convolution modules and a batch K-Means module. In the encoder, the fully connected modules are used to extract nonlinear features of scalar data, long short-term memory networks are used to extract long-term dependent features of time series, and one-dimensional convolutional modules are used to extract local features of time series. As a result, the hidden vector is obtained by mapping the original high-dimensional data to a low-dimensional feature space through the encoder. Then the hidden vector is input to the batch K-Means module to update the cluster center and calculate the cluster loss. On this basis, the decoder reconstructs the hidden vector back to the original space to obtain reconstructed data, which is used to calculate the reconstruction loss. The IAE model is trained jointly with the clustering loss and the reconstruction loss. Finally, the cluster center of the original data and the cluster category of each sample are obtained. The closer the sample is to the cluster center, the better the process parameters are controlled. At the same time, samples within the same cluster category are closer in process operation. Therefore, the oxygen supply, slagging and bottom blowing process of the closest samples are taken as the process model of this type of samples. The effectiveness of the IAE model is evaluated using the end-point quality index of real data from converter steelmaking. The average hit rate of the endpoint carbon content within the error range of ±0.02% is 93.05%, the average hit rate of the endpoint temperature within the error range of ±20°C is 92.10%, and the average double hit rate within the error range of ±0.02% carbon content and ±20°C temperature is 90.38%. The results show that the process model extracted is helpful to improve the endpoint hit rate.