Study On Intelligent Control For Aluminum Reduction Cell Based On Data-driven

Aluminum electrolysis is an energy-intensive process industry, as China hasproposed to implement the scientific concept of development, and build the goal ofsaving the community, taking full advantage of today’s advanced informationtechnology to the aluminum electrolysis industry, and achieving the low pollution, highefficiency and highly intelligent of the low pollution, high efficiency and highlyintelligent is an important subject that we must face today.The pre-baked aluminum cell is a complex nonlinear time-varying andmultivariable system model, it is uncertainty and has complex and volatile cell slotstatus, a lot of process production parameters can not be measured on line, especiallythe alumina concentration, which is the basic parameters of material control. Thecontrol system database has accumulated a large number of the process production databecause of the long-term operation of aluminum reduction cell. The purpose of thispaper is to use data-driven approach to extraction and mining of the cell slot status ofthe process of aluminum reduction cell and alumina concentration from the parameterdata which is stored in the control system, in order to establish an intelligent control foraluminum reduction cell based on data-driven. This main implementation of this paperis to use wavelet packet analysis technology to extract the instantaneous cell voltagesignal characteristics in the current period of time, combined with the power input in thecurrent period of time, a series of current setting voltage of state parameter cell slotstatus to form the predict feature vector, and to use the wavelet neural networkself-learning function to judge the current cell slot status. Focusing on the difficult ofreal-time alumina concentration, this paper proposes an alumina concentration softsensor model based on least squares support vector machine (LSSVM). This paper usesa large number of measurable process parameters on the soft sensor model for trainingto learn to take advantage of training in good soft measurement model to achieve theimplementation of the soft measurement of the alumina concentration. In the controlpart, the alumina NB is divided into the main NB and the aid NB, and use fuzzycontroller to reason the main NB and the aid NB in the different alumina concentrationand to meet the requirements of adaptive control for each aluminum reduction cellproduction, to improve the efficiency, and reduce energy consumption.In this paper, a large number of data-driven approaches are used, this paper promotes the application of a wide range of data-driven pattern recognition and controltechnology in the aluminum industry. After a large number of practical engineeringapplications and a large number of simulation experiments, it greatly improves the levelof information processing in the existing control system. From the experts statistics, itachieves more than90percent accuracy of the cell slot status diagnosis, and the controlprocess is fast and accurate, to achieve energy-saving more than150degrees of the tonsof aluminum.