The Study Of BOF Steelmaking Assistance System

The subject of the dissertation is data-driven simulation of the steel production process and system development, from public technology research projects in Zhejiang Province. The main purpose of this dissertation is to research some questions existed in five 50-tons steel converter in the Jiangsu yonglian steel and iron company such as low steelmaking automation, full personal experience in steelmaking control, low blowing end point hitting ratio. Based on analyzing blowing process on the scene and a large amount of production, combined with neural network technology, a BOF decision support system is built, to assist the operator to make control decisions in the end of steel making. The functions of the end point prediction and the optimization of end point control parameters are decision support system's two core functions.In the realization of system function to forecast the end, based on study of the theory that use RBF neural network to model, according to network architecture design, data preprocessing, learning algorithms selecting, and several other steps, a static endpoint prediction model is built in this paper. The quantum particle swarm optimization is selected as the learning algorithm for RBF networks. Compared with traditional methods, the quantum particle swarm optimization has more global search capability, difficult to be trapped in local optimum, so it can improve the network generalization. And it has few control parameters, so the algorithm process is simplified. Forecast model uses parameter-adaptive method, using the latest data to update the model parameters to improve the prediction hit rate.In the realization of system function to optimize the end control parameters, the paper analyzes the distribution of BOF control parameters. An algorithm is proposed to find control parameters with optimal cost, to help the operator to find a control parameters solution that has least cost. The solution space of control parameters is divided into several areas in this method. And the quantum particle swarm algorithm is selected to search a specific area. System's software is implemented with a combination of C #, Oracle database and matlab, and its user interface easy to use. For the system functions, prediction and control parameters optimization, offline matlab simulation and online performance test were carried out. The results show that the system has good performance, can assist steel operator to making decisions and can improve prediction accuracy of the converter end C content and temperature.