Study Of Soft-sensing For Crucial Parameters Of Aluminum Electrolysis Process

Maintaining the dynamic balance of energy and materials in the pre-baked aluminum reduction cell is the key to the stable operation of the aluminum electrolysis production.It is also the premise of emission reduction,energy saving and quality improvement of electrolytic aluminum production enterprises.However,due to the limitation of the current development level of process detection technology,many key parameters affecting material balance and heat balance are still unable to be detected online or quickly,considering that the electrolyte has high temperature and strong corrosive properties during aluminum electrolysis production.As a result,the implementation of advanced process control strategies is difficult and the power consumption is high.Based on the extreme learning machine algorithm,a soft sensor modeling method for parameters of aluminum electrolysis was proposed in this paper.The main work is as follows:Firstly,the online sequential extreme learning machine algorithm was used to enhance the dynamic tracking capability of the system.Moreover,the meta-heuristic algorithm was employed to find the optimal values of input weight and hidden layer bias in the network of extreme learning machine,so as to reduce the stochastic error.Secondly,the particle swarm optimization algorithm that commonly used in meta heuristic algorithm is analyzed,and the reason that which is easy to fall into premature convergence phenomenon is summarized.From this,dynamic decay weights and learning factors were added to the speed update formula to balance global search and local search capabilities.In addition,the initial population was filtered by using the reverse learning mechanism to improve the convergence rate of the particle swarm optimization algorithm.To testify the algorithm,the secondary variables were selected by analyzing the influencing factors of each parameter;furthermore,the data of 10 sets of 500 kA aluminum reduction cell in the field of an electrolytic aluminum factory were collected.The optimized algorithm was applied in the soft-sensor modeling of electrolyte temperature,molecular ratio and alumina density in aluminum electrolysis process.Experimental results show that the proposed soft-sensor model has higher prediction precision and generalization performance.The predictive control of alumina concentration based on extreme learning machine was proposed.The online sequential extreme learning machine optimized by particle swarm optimization was used as the predictive model.And the receding horizon optimization strategy was used to control the time interval of alumina material addition.Furthermore,the predictive control problem with restricted constraints was solved by particle swarm optimization algorithm.The experimental results show that the model has effective predict and control performance,which is of great significance to ensure the material balance and increase the aluminum yield in the aluminum electrolysis process.