The Control Of The Temperature Of Aluminum Electrolysis Process Based On The General Dynamic Fuzzy Neural Network

In the aluminum electrolysis industry, the current efficiency is one of the most important technical-economic indicators, while electrolys is temperature is also the most important factor to influence the current efficiency, hence the determination of Aluminum tapping volume and A1F3addition is very important to control the temperature of aluminum production cells and the overheated temperature precisely. As aluminum electrolyzing is a large-lag industrial process, the effect of controlling temperature cannot be appeared instantly, which proposes a huge challenge to the traditional control algorithm. In this thesis, the general dynamic fuzzy neural algorithm predicts the control effect of some specific parameters in advance so as to modify the control parameter dynamically and achieve the control performance the system acquired.The article is based on general dynamic fuzzy neural network, constructing an aluminum electrolytic prediction system to forecast the aluminum tapping volume and A1F3addition at a given temperature of aluminum production cells that we need. Through the adaptive analysis of operation law of high current efficiency and low-energy electrolytic cell, the system trained the corresponding rales and applied to the inefficient electrolytic cell algorithmically, so as to control the temperature of aluminum production cells and the overheated temperature and improve the efficiency of aluminum electrolysis. The system has been applied in practical industrial data shows that the algorithm validity in the industrial control systems by simulation experiment. The advantage of this algorithm is to build the model automatically and obtain expertise rules by practical sample without expertise knowledge in the aluminum electrolysis field. The system model we got has the characteristic of compact structure, avoid over fitting and bring great convenience to decision on aluminum electrolysis process.Besides, the thesis constructs a predicting control system by adding network verification module into general dynamic fuzzy neural network. The network can get the predicting error of the model through comparing the value between the output and the predicting output, so as to get a more precise decision value by re-training the model.At last, the author develops software by mixed programming among access, MATLAB and visual C++. Its main function is to achieve off-line train of predicting model and predict the Aluminum tapping volume and A1F3addition, and display of historical data and curve display.