Research On Control Of The Alumina Concentration Based On Neural Network

The main object of aluminum electrolysis process control is the control of aluminaconcentration, so accurately judging of alumina concentration is the basic step of thealumina concentration control. Aluminum electrolysis process is a nonlinear,multivariable coupling, time-varying and large time delay industrial production process,accompanied by a variety of complex physical and chemical changes. Aluminumelectrolyte is highly corrosive, and its production is carried out under high temperatures,strong magnetic field and a dynamic state of motion. The complex production processdetermines that the process parameters are uncertain, not continuously measurable.Therefore, it is difficult to control the alumina concentration.After analyzing the aluminum electrolysis process and real-time data, this paperchooses to control the alumina concentration combined with the knowledge of neuralnetwork, genetic algorithm and expert knowledge. For specific cell state of electrolyticcell, after effectively measuring the alumina concentration, correct feeding action istaken to make the alumina concentration controlled within the target range and keep thealuminum electrolysis process steady.The main contents of this paper include:①Cell resistance generated from the online measured cell current and voltage is animportant parameter of aluminum electrolysis process. Considering the characteristic ofcell resistance time series, BP neural network, which has strong ability of nonlinearmapping and self-study, is used to predict the next time cell resistance. Geneticalgorithm is used to optimize the weight and threshold of BP neural network to preventthe network training from slow convergence, easily running into the local minimum.②The R~C characteristic curve giving the relationship between cell resistanceand concentration is uncertain, and it will change according to the cell state ofelectrolytic cell. Three characteristic curves under the different cell states based onexperience and plant data of Guizhou aluminum electrolysis factory are got. As thestructure of LVQ neural network is simple, and classification processing can becompleted by internal interactions, then this paper chooses to use LVQ neural networkto recognize the real time cell state. From the R~C characteristic curvescorresponding to each cell state, the alumina concentration is got.③Reasonable feeding action is chosen according to the cell resistance information, and the control parameters are adjusted depend on real time cell state and aluminaconcentration. The feed action is taken to control the alumina concentration in the targetrange, and realize the optimal control of aluminum electrolysis process.