Research And Development Of The Data Mining Of Bake And Start-up Of Aluminum Reduction Cells

Along with the generalizing application of computer in the aluminum electrolysis industry, the supervisory control system which controls the aluminum reduction cell to automated work was used in the production of all aluminum electrolysis factories. Various kinds of data which reflect the states of cells were automatically collected by supervisory control system. Tremendous amounts of historical data was accumulated in the aluminum electrolysis industry, however such problems as data un-sharable, low integration, difficulty to extract characteristic in mass data was exit in the present system which only can carry on simple data input, inquiry, statistics and so on. More important guidable rules for the management and production of the enterprises in the mass data cannot be discovered. The decision-maker urgently needs the valuable information and the knowledge which should be extracted from the massive data and can be used to instruct the management for aluminum reduction cells to enhance the production benefit. The bake and start-up is one of the key factors affect the aluminum reduction cell's life and consequently influence enterprise economic efficiency, therefore to analysis the data reflecting the conditions of cells at the stage of bake and start-up will be quite important.This article mainly introduces the research and the development about the application of the data warehouse and data mining technology in aluminum electrolysis field. It introduced the elementary theory and the concept of the data warehouse and the data mining. The data warehouse of bake and start-up of aluminum reduction cells has been designed and built. A ETL(Extract Transform Load) tool with many kinds of preprocessing methods has been designed and developed to implement the data extracting, transformation and loading to provide a finished data warehouse.It emphatically research on cluster analysis, classification, association rule algorithm and improve them in view of the characters of aluminum electrolysis data to get more effective and more practical data mining result. Which comprised: 1 the k-means clustering algorithm, improving and simplifying the k-means based on Partheno-Genetic algorithm to eliminate the dependence of classic k-means to the initial central points;2 ID3 classifying algorithm, we proposed an improved ID3 algorithm which can dynamic divide continuous data to subsections in view of the aluminum electrolysis data all is the continual value and the classic ID3 only can process the discrete data, so we can establish an optimized decision tree and get the dataprocessing result more conforming to the demand of production management;3 the association rules algorithm based on the FP-Growth, this algorithm improved according to the distributed characteristic of aluminum electrolysis data can be used to analysis continual aluminum electrolysis data directly and get the rules about the data in different sections which was divided by the algorithm automatically. These rules have more instructive value to the production.An aluminum electrolysis data mining system has been designed and developed, through cluster analysis, classification analysis and association analysis of the data collected at the stage of aluminum reduction cells baked and started by realize and implement the improved algorithms in the data mining system, the abnormal cells was discovered and the classified rule about cells conditions was obtained. The mutual influence relations between various crafts parameter was analyzed and a best craft parameter combination with high current efficiency was found, at the same time the validity and the usability of these improved algorithms had been proved.The system is applied to process the data in the aluminum electrolysis production field to instruct the production effectively and enhance the production efficiency and prolong the life of the cells and provide the basis for the scientific management.