Research And Application Of Frequent Time Sequence Patterns In The Production Data Processing Of The Aluminum Slot

Aluminum cell every day generates a large amount of real-time production data, and these data are normally recorded into sequence in accordance with time. However, these data have not been fully utilized. Generally, the various technological parameters of electrolyzer are adjusted by the manager in accordance with his personal experience, and the user cannot make full use of the large number of historical data of production to make a control decision from a quantitative point of view. To solve this problem, this thesis puts research priority on traditional sequential pattern mining algorithm. In order to arrive at frequent time sequential pattern, the time sequential data of aluminum cell are mined and applied.1 Through the exhibition of various diagrams of original data, the data of initial time sequence is analyzed.2 In the process of data pre-processing, the vacant data are processed by filling average value, removing vacant value, filling vacant value artificially and some other methods. As for noisy data, it can be dealt with sub-box method.3 In connection with the data of single technological parameters, the solution lies in discretization which will separate the data into multiple segments so that the data sequence can be transformed into that of continuous symbols.4 For multi-dimensional data sequence of multiple technological parameters, fuzzy clustering algorithm should be adopted to aggregate the values of multiple technological parameters in the same day into different category values (or the slot status). Namely, they are transferred into a row of continuous symbols so that the time sequential pattern of slot status can be mined.5 Do some research on the classical sequential pattern algorithm. Directed to the characteristics of the large quantity of production data and the need of improving mining efficiency of aluminum cell, a suitable solution in aluminum cell field called PrefixSpan is presented. This algorithm is applied into the analysis mining of historical data in aluminum electrolysis technological parameters.Based on above, a mining system of frequent time sequential pattern in aluminum electrolysis is designed as well as developed. By means of adapting this system into the production data processing of aluminum cell, the single technological parameters with a span in the same slot are mined, so are the multiple technological parameters and the original data of slot status. Rest on the frequent time sequential pattern mined, the development tendency of future production can be foretold, thus a scientific basis could be provided for decision-making.