Study On Charging Optimization For Bell-type Annealing Furnace And Gas Holder Level Prediction In Steel Industry And Their Applications

With the popularization of integrated automation and energy technology in steel industry, the manufacturing executive system and energy management system have been successfully applied in steel industry. As the core function of MES and EMS, the ability of production planning and energy balance scheduling make a direct influence on the utilization effect of enterprise's resource and energy. Based on the projects of National High-Tech Research and Development Programme, the main problems and characteristics of resource and energy optimization in steel plant are analyzed in this dissertation, in which the optimal charging for bell-type annealing furnace and the on-line prediction for gas holder level in steel production process are studied in detail. The dissertation has mainly carried on the following researches.The optimal charging for batch-type annealing furnace in cold rolling process is studied. Firstly, the coils are merged, and then the charging coils are effectively combinated. At the stage of coils merging, a multi-objective and multi-knapsack model is established, based on which, a clusters method acquires the classification of coils and the knapsack model's center. And an adaptive quantum genetic algorithm is developed to obtain the optimal coils merging results rapidly. At the stage of optimal charging combination, a multi-knapsack model with multi-furnace type and uncertain furnace numbers is suggested, where the furnace numbers upper limit is obtained by lagrangian relaxation heuristic algorithm and a new partheno genetic algorithm is proposed for solving the model. The charging results with the least energy consumption are obtained.The prediction problem for coke oven gas holder level in coke-making process is studied. A prediction model based on the least square support vector machine is established through analyzing the change characteristics of gas production-consumption and holder level. Considering the influence of model's parameters and samples on the prediction precision, a gradient grid search algorithm is proposed to opitimize the model's parameters, and an effective big samples selection method is suggested to build the training samples.The on-line prediction problem for blast furnace gas holder level in steel-making process is studied. Considering the practical level with frequent and great fluctuation, this problem is regarded as a general nonlinear regression problem with non-flat variation. A new multiple kernel learning based least square support vector regression is constructed based on a reduced gradient algorithm, which can rapidly give the resulting regressor under the optimal linear combination of kernels for predicting the various change trend of holder level.The multi-holder level prediction problem for linz donaniz gas in iron-making process is studied. A consistent T's grey relation analysis is developed to determine the main influencing users of holder level. According to the parallel running mode of several holders in practice, a multi-output least square support vector machine regression algorithm is proposed to establish the prediction model, in which the model's weighted values and bias are fast given by solving a series of linear equation system. The developed model can accurately predict the level trend of each LDG holder.Based on the studies mentioned above, two software systems for bell-type annealing charging optimization and gas holder level prediction are developed in combination with software engineering technology. The application in cold rolling plant and energy cernter of Shanghai Baosteel Co. Ltd. show that the system can improve the level of the utilization of resource and energy in steel enterprise, decrease the production cost and reduce the enviorenment pollution.