Soft Sensor Model Of The Current Efficiency For Aluminum Electrolysis Process Based On Status Classification

Abstract:The current efficiency is an important economic indicator in the production of aluminum industry. To improve current efficiency not only can reduce energy consumption, but also can improve the production capacity of aluminum electrolysis. The decision can be provided for management staff on site to measure the real-time current efficiency. Since the current efficiency was hard to be measured accurately by instruments,so it is a inevitable choice to solve this problem used by soft measuring technology. Because of different cell states,the current efficiency predicted by a single model will increase the complexity and lose information. So the model of multiple support vector machines were proposed to solve the soft sensor modeling problems of current efficiency for aluminum electrolysis.The main content was included as the followings:First, the measured status for the process parameters of aluminum electrolytic was introduced,and through the problems of measurement of current efficiency,the soft measurement model was established by SVM; Process parameters which has a important impact on the current efficiency, such as cell voltage, electrolyte temperature, the molecular ratio, were analyzed and collected to become the auxiliary input variables of the model after the pretreatment;Meanwhile, the impact on the current efficiency of the cell states were also analyzed.Then, the sample data was classified by using data mining techniques, and according to the characteristics of data of aluminum cell, the fuzzy C-means clustering algorithm(FCM) was selected to cluster. Because FCM can be lead to local optima easily, so an improved FCM which redefined the Euclidean distance formula was proposed to solve this problem.The effect of clustering was adjusted by changing the size of the smoothing parameter.The result show that FCM improved algorithm can get a more rational division for data samples.Finally, the child support vector machine model was established according to each sub-class clusters and the information of different cell states was contained in each model.When the data was inputed,the current efficiency would be predicited by the various types of sub-models and intergrated by fuzzy memberships.Which cell states belongd by sample points, which weightiness assigned to sub-model was high. Simulation results show that the multi-vector machine model proposed can more effectively and quickly close to the true value, improve the prediction accuracy greatly.