Algorithm Analysis And System Realize Of Sand-dust Storm Data Mining

The paper introduces the domestic and international current situation of the development of sand-dust storm study, data mining and the fundamental knowledge of data mining such as the models of data mining process, criterions, techniques, and mining steps. The Microsoft OLE DB for DM and DMX(Data Mining extensions) also are introduced briefly.The domestic data mining on meteorological data is in beginning. It is a challenge to mine meteorological data due to two reasons. Firstly, meteorological data are a kind of complex data. They are both spatial-data and time-series data. Their storage formats are verity. Secondly, data mining tasks are very complex and the professional data mining algorithms for meteorological data are a few.First, We obtain the data mining tasks by analysing the user's requirements, and transform them to data mining algorithms. Second, we select the relevant data from sand-dust storm data in recent 46 years of 241 observations in Northwest China. A data warehouse was built after the data being cleaned and transformed. Last, We developed an interactive and visual sand-dust storm data mining system. It can be dealt with the data cleaning, data transforming, data mining, model assessing and result displaying on an interactive and visual platform.We try to apply general data mining algorithms such as Association analysis, Regression analysis, Clustering and Spatial analysis to data mining on meteorological data, and obtain useful knowledge.For resolving the problems in meteorological field, we design the circle-region continue algorithms and simply time continue algorithms by simplifying DBSCAN. Aε-Circle scan algorithm and a rectangle scan algorithms were designed based on DBSCAN to resolve its missing cluster problem. They can find whether high dense continuous area exists or not exactly in samples region. We bring forward a constraint-based Continuous Rainfall & Overcast (CRFO) algorithm with cluster assessing . the CRFO uses hierarchical clustering idea . it can effetely resolve time continuous problem in meteorology. We also bring forward period data fetching algorithm in folded data. The algorithm can aggregate folded data of the same period of time every year, which OLAP' drilling and simple SQL aggregating cannot realize.Some significant characteristics (knowledge) of sand-dust storm are discovered by applying our data mining system. The knowledge is helpful for meteorological researchers and forecasters, it can be used to study and forecast the source, moving, start time and occur times of sand-dust storm and so on. The result shows the data mining is effective on meteorological data.It is feasible to implement data mining technology on meteorological data. But many works must be resolved firstly to apply widely, for example to research the suitable data mining algorithms for meteorological tasks, to study model assessing technique to use domain knowledge, and to boost mining efficient.