Cell Status Diagnosis for the Aluminum Production on BP Neural Network with Genetic Algorithm - Advanced Research on Computer Education, Simulation and Modeling - page 146

Cell Status Diagnosis for the Aluminum Production on
BP Neural Network with Genetic Algorithm
Shuiping Zeng, Jinhong Li, and Lin Cui
North China University of Technology, Beijing, 100144
[e-mail]
Abstract. To diagnose the status of aluminum production cell, we set up a
diagnosis system. This system was based on BP neural network with 10 inputs
and 3 outputs. It used frequency energy of cell resistance as characteristic
vectors. Also genetic algorithm was used to optimize the initial weights and
threshold value of BP network. After designed and tested this software system,
three kinds of status were successfully diagnosed, which are anode lesion, liquid
aluminum fluctuation and normal condition. As a result, by sampling data on site,
the status diagnosis accuracy is larger than 83%.
Keywords: Aluminum production cell, Genetic algorithm, BP neural network.
1 Introduction
Aluminum is produced by electrolysis, with alumina as raw material, cryolite-alumina
melts as solvent carbon block as anode and liquid aluminum covered on the carbon as
cathode. Direct current is injected. After the electrolytic reaction, carbon dioxide mixed
with carbon monoxide is produced at anode and aluminum at cathode. The analysis and
diagnosis of aluminum production cells status is important to the aluminum production
industry. It is also a bottleneck for this industry to improve the automation [1].
Aluminum production cell is the major equipment for aluminum production, which
status of is not only related to the economic technical indexes, but also affecting the
cells life and daily production. Aluminum production cell is a nonlinear, more
coupling, time-varying and large delay process. In this process, the material balance
and energy balance are constantly changing, influencing and restraining each other, so
that some status maybe happen, such as anode effect, anode lesion, cold cell, hot cell,
etc. Once some of them happen, large economic losses will be caused. It has been an
urgent problem to reduce the energy consumption of aluminum electrolysis, to increase
cell age, and to reduce the economic loss [2].
Cell status or Fault diagnosis technology in aluminum production mainly deal with
the fuzzy expert system, neural network, wavelet transform and the combination of
other technology methods. Good results have been achieved in some extent from the
papers. But systems with good performance in industry process are not reported. The
intelligent diagnosis research of the aluminum production cells has great progress in
theory but most of the research remains in the laboratory [3-5].