Application Research of Electric Arc Furnace in Metallurgy

In the mid-1980s, countries around the world set off a wave of research and application of artificial neural networks. Early studies on the application of neural networks in metallurgy in the United States, Japan and other countries have solved some important technical problems in metallurgical production and achieved good results. For example, Nippon Steel Oita Plant uses BP network to predict the distribution pattern of gas flow in the furnace, and combines with the expert system to predict the furnace condition and guide the operation. The correct rate is over 90%. The intelligent Electric Arc Furnace IAF developed by the US company is implemented by neural network. Online adjustment of electrode position, applied to NorthStar Steel's power saving effect of 30.89%, saving electrodes 30.89%, improve productivity), this system has been applied to several companies including China's Guangzhou Iron and Steel Company. The development and research of artificial intelligence and neural network in the production of electric Arc Furnace steelmaking is of great significance and practical value for ensuring high quality, high production, safe operation, energy saving and consumption reduction of electric arc furnace.

1 Neural network problems in the application of electric arc furnace steelmaking Neural network-based computers are introduced into the intelligent electric arc furnace (IAF), which uses three neural network models, which are electric furnace simulators and regulators. And the neural furnace controller, consisting of more than 200 neurons, all using the extended -Bar-adaptive algorithm.

The electric furnace and regulator simulator network is combined into a neural furnace controller that learns how to adjust the electrodes to achieve a set of setpoints controlled by the furnace. When the disturbance of the grid flicker is minimized, this conclusion is as follows: (1) The outer wall of the composite elbow is made of steel pipe, and the inner wall is made of high-chromium cast iron, so that the elbow has the characteristics of wear resistance and corrosion resistance of the high alloy cast iron, and The high impact resistance solves the long-standing contradiction between the wear resistance and weldability of the conveyor industry. (2) In the development process of composite elbow, numerical simulation technology can quickly find design defects, and provide theoretical basis for product design and definition, shorten product development cycle and reduce development cost.

The foundry technology network achieves the set values for current and power factors. The weighting factor of the network is adjusted once every 15s. This technique allows continuous improvement of the furnace's prediction and adapts it to changes in operating conditions, thus automatically compensating for changes in scrap loading, voltage, electrode length and system impedance. This adjustment makes sense because the ideal operating conditions are not readily available. The modern electric arc furnace steelmaking process includes a large system of power supply equipment, electric arc furnace and secondary refining equipment. How to establish the whole process of electric arc furnace steel production and the model of large system is an important subject of applied basic research.

According to the actual situation of China's metallurgical enterprises, using the successful operation of the foreign successful arc furnace control system, the expert control system has a good effect in terms of power balance and power factor in practical applications, but its biggest disadvantage is that only It can react to the state of the electric arc furnace and then adjust it through the actuator, so that the adjustment often lags behind the actual state of the electric arc furnace, causing large fluctuations in the current, causing the power factor to deteriorate, and the generated higher harmonics become the grid. The public hazard reduces the overall benefits of EAF operation. In response to this situation, an electric arc furnace neural network prediction system is added to the original system, and the state of the electric arc furnace at the next moment is estimated, and the output of the expert system is optimized and compensated by a specific optimization program to the expert. The system increases the estimation ability, so that the arc current can reach a relatively stable state, reducing the reactive power and reducing the damage to the power grid becomes inevitable.

In the arc furnace electrode control, due to the difficulty of three-phase decoupling, the current control is mostly based on single-phase consciousness, which leads to electrode misalignment in the electrode adjustment, which affects the operating efficiency of the electric arc furnace. On the basis of neural network estimation and optimization, the neural network is used to decouple the three phases of the electric arc furnace, and a new method of electric arc furnace control based on three-phase consciousness is developed, so as to further improve the comprehensive operation efficiency of the electric arc furnace. Row.

2 neural network technology in the electric arc furnace steelmaking process application neural network can learn, identify and adapt to the continuous changes in furnace load characteristics during the smelting process, adjust the furnace's working point setting, so that the furnace always runs within a reasonable range, can Achieve more sophisticated control than traditional technology, and get more advanced production technology indicators. Most of the neural networks adopt the BP network model, which has the advantages of simple structure, good adaptability, strong generalization ability, stable working state, and ability to approximate arbitrary functions with arbitrary precision, thus obtaining a wide range of applications. The shortcomings of BP network are characterized by long learning time and slow convergence. In order to eliminate these defects, a common method is to use a variable step size algorithm, a momentum item, etc., in addition to other models such as a self-organizing network. What models, topologies, and algorithms are used? There are no mature theories or rules at present, which are mainly determined by experience and many experiments.

The ability of neural networks combined with mathematical models, expert systems, and fuzzy techniques to improve problem solving will be an important direction for neural networks in metallurgical applications. In the electric arc furnace steelmaking production process, the neural network can be studied in the following aspects: (1) Optimized the design of the electric arc furnace expert system through network estimation. The performance of this system mainly depends on the fidelity of the arc furnace neural network model. . Therefore, the modeling of the object of the electric arc furnace is one of the main research issues, and a lot of work needs to be done in the selection of the network type and the corresponding learning algorithm.

(2) Research and development of electric arc furnace electrode control method based on three-phase consciousness. Based on artificial intelligence heuristic search method and artificial neural network optimization technology, the electrode lift self-optimizing intelligent control system for DC electric furnace is designed to automatically find electric furnace. The optimal position of the electrode and automatically maintain the optimal working state of the electrode of the electric furnace when the environmental conditions change.

(3) On-line adaptive adjustment of neural network's online adaptive adjustment neural network prediction model, and its adaptive adjustment based on three-phase consciousness neural network controller. It mainly focuses on the research of neural network related algorithms. The ultimate goal is to make the system adaptively adjust its own parameters online to meet the characteristics of the control object due to the ablation of the electrode and the loss of the lining.

(4) Optimized program design of the optimal power supply curve of the electric arc furnace The key to this problem lies in: maximizing the optimization of various important parameters for the ideal state of the operation process of the electric arc furnace, and researching the optimization strategy to make the system have an automatic search. The ability of the optimal power supply curve achieves the purpose of intelligence and improves the operating efficiency of the electric arc furnace.

(5) Research on intelligent quality prediction and control technology for electric arc furnace steelmaking end point and research and develop the end temperature and carbon content intelligentity of electric furnace steelmaking production process with self-learning and self-adaptive characteristics based on artificial intelligence expert system and artificial neural network Forecasting methods, computer application software and endpoint quality intelligent control technology.

(6) Research and development of intelligent management system for electric arc furnace steelmaking production process The intelligent management system for operation planning, production process scheduling, process conditions and quality assurance, raw material ratio and adjustment for electric steelmaking production process.

(7) Research and development of intelligent monitoring system for electric arc furnace steelmaking production process Intelligent monitoring of normal working conditions in electric furnace production process, intelligent monitoring and fault diagnosis of abnormal working conditions, method and technology for intelligent monitoring of environmental pollution and power grid interference.

(8) Research and development of intelligent integrated automation system for electric arc furnace steelmaking production process Based on multi-agent, distributed artificial intelligence, large-scale expert system technology, intelligent control of electric furnace steelmaking production process, intelligent management, intelligent monitoring integrated intelligent integrated automation system.