Research On The Structure Patterns Of The Multiple Time Series

We oriented the complexity multiple streams time series data in military affairs, industry process monitor system, medical diagnose, Robert control, and logistical management, financial management system. It is very important to analysis , estimate , optimize and integrate these data by computer technology, For example , In agriculture, If we can find the relationship between temperature variety, air pressure variety ,insect pest variety, and emblement output variety, So ,We can find the way to enlarge the output .The current research of time series are focus on the finding of the knowledge between the events in the single time series, for example trend, sequential patterns , finding similitude patterns, association rules,finding period,etc. We can't divide the multiple streams time series into singleness times series simply in the research of multiple streams time series , we'll dissever the relation between the events of the multiple streams .although the MSDD can find the dependency relationship of multiple streams, but it haven't the initialization of the events, The express of the time relationship between events is not frank, The cost of the algorithm is expensive (O(n5 )), i can't find much more knowledge in multiple time series, it find the dependency patterns only of the multiple time series, So there need a new more effective, frank, complete algorithm to find the knowledge.This paper analysis the data mining of the single nd multiple streams time series , And draw a conclusion that the relationship between the events of the multiple streams time series are theassociation patterns dependency patterns, sudden patterns,This paper call them are structure patterns, the existing algorithm haven't discuss these patterns, Although MSDD discussed the dependency patterns, However, It ignored the association patterns, sudden patterns, This paper have a definition of the association patterns, sudden patterns and dependency patterns, And have a complete, frank algorithm called TWMA(Time Window Moving And Filtering Algorithm) ,the peculiarity of this algorithm is that events is listed by the time window, by this way, the relationship of the events is clear. This algorithm can discovery the relationship of the events and can discover more knowledge and the cost is more cheap than other algorithm, the algorithm is concision, frank, it take only (O(n3)) time. , in addition ,1 design a data miner by use of VC++, And it is successful to mine the multiple time series of medical data streams, temperature data streams and air pressure data streams .The traditional serial algorithm can't do work well for the data ocean quickly and correctly, It also important to research the parallel algorithm .This paper analysis the main parallel algorithm and models ,And find there are two problema :how to use the capability of the processors and the information number of the transmission between the processors. This paper extend the TWMA to PTWM ,and PTWMA solve the problem by the method of table corresponding ,And the idea that the algorithm need not to transmit the large item sets. PTWMA is an effective successful Algorithm and model to the knowledge discovery of the multiple streams time series.