Research And Application Of Clustering Method Based On Grey Incidence Analysis

Grey incidence clustering method is an important research area of grey system.Most existing grey incidence clustering methods can only be applied to 1-dimension sequences. Besides, all of these models require the length of comparing sequences be equal. If there are uncertain sequences having different length or existing data loss, they need to be padded up or deleted some redundant data before use these methods. Such preprocessing will increase the uncertainty of the system. To solve these problems, the main research contents of this thesis is shown as follows.1. A Novel 1-dimension grey incidence analysis model and corresponding grey clustering methods are proposed. Focused on the problem that 1-dimension grey incidence analysis model can't compare sequences with inconsistent length, on the basis of dynamic time warping distance, a new grey incidence analysis model is proposed. By computing the shortest path of distance matrix between two 1-dimension sequences, the similarity between their geometric curves can be measured. Based on this method,corresponding grey incidence clustering method is constructed which can cluster the sequences with different length directly. The experiment result shows that this method has stronger robustness. Especially when the sequences' length is unequal, its result will be more reliable and accurate.2. Multidimensional grey incidence analysis model and corresponding grey clustering method are constructed. To solve the problem of multidimensional grey incidence analysis model, on the basis of 3-dimension grey incidence analysis model, a new multidimensional grey incidence analysis model is constructed. Refer to3-dimension grey incidence analysis model, observing data of each sequence is regarded as points of m-dimension space, and reference sequence can be obtained from these sequences. As multidimensional dynamic time warping distance can compute the shortest path to measure the similarity between sequences, even if the length of observing sequences are inconsistent, this model can be applied to them directly without preprocessing sequences. On the basis of this model, a corresponding multidimensional grey incidence clustering method is proposed. This method doesn't need to compute the grey incidence degree between each two sequences. It can get a reference sequence from observing data, and each sequence only needs to be computed once. So thecomputation process of this method is easier. The experiment shows that the analysis result of this grey incidence model is more accurate, and the clustering result is far better than traditional one. Finally, it is used to analyze the electricity consumption of an individual household, preferably effect is acquired.