Predictive Fuzzy Control For The Temperature Of Aluminum Electrolysis Process

In the aluminum electrolysis process, the main reason for effecting current efficiency is the temperature of aluminum production cells. As the temperature is a kind of time-delay variable, forecasting is supposed to be used in the field. Fuzzy-expert system has widely used in nowadays aluminum production industry, and has already achieved a great success. The main work of the paper is based on the achievement which the predecessors have made. Add a module with predicting function into system to improve the performance of the original fuzzy-expert system.The kernel of the paper is to predict the future temperature of aluminum production cells with forecast model built based on transformation of orthogonal matrices by use of historical data of temperature, working voltage, aluminum tapping amount and fluoride addition times. The method simplified a multiple regression problem into several simple regression problems and converted the variety of production parameters into the transformation of orthogonal matrices. The process of analysis and forecasting were completely by mathematical method. Reverted to actual production parameters in the end. The predicted value and the actual value were very close. Choosing two aluminum production cells for experiment for over 50 days, calculating the predicted values by Matlab. It was showed that the standard deviation of the predicted value and the actual value were 1.2634 and 0.9358 respectively, which were close enough to be accepted. Compare to the classical multiple regression prediction, whose standard deviation of the predicted value and the actual values were 2.8062 and 1.6758 respectively, the prediction accuracy has been enhanced. It was found in experiments that some parameters of the model affected the accuracy of the prediction. Although this paper has given a method for choosing proper parameters, which can guarantee the acceptable deviation of the predicted value and the actual value, through lots of attempts, some better parameters of the model for enhancing the prediction accuracy were found. There is no good solution for acquiring those parameters scientifically.Add the module into Fuzzy-expert system for constructing a predictive-fuzzy system. Evaluating the predictive value, making sure whether the result is satisfied. If the result isn't good enough, the module would give the original system some feedback in order to adjust the result, till the result is satisfied; send the decision-making to be executed. The stability of this method was dependent on the stability of the original fuzzy system, as well as the accuracy of the prediction model.