Research On Rough K-means Clustering Algorithm With Space Distribution Based Self-adaptive Weights Measurement And Its Application

Clustering analysis is an important technology in the field of data mining.Rough k-means algorithm has been developed rapidly in recent years.In rough k-means algorithm,data objects are objectively assigned to lower approximation with clear data objects ownership and boundary area with unclear ownership.Rough k-means is a kind of new effective soft clustering analysis method.In rough k-means algorithm,the weight coefficient setting of boundary objects during means iteration has a very important effect on the clustering results.Existing rough k-means algorithms do not fully consider the effect of space distribution of data within clusters in measuring the relative importance weights of lower approximation and boundary area as well as individual weight coefficient of boundary object.This paper synthetically considers the spatial distribution of the lower approximation and boundary region objects,explores self-adaptive weight measurement method of the relative weight of the lower approximation and border areas as well as individual weight of boundary objects in means iteration,and combines with the practical application requirements of photovoltaic power generation forecasting system conducting data preprocessing analysis on photovoltaic forecast data by the designed rough k-means clustering algorithm.The main research contents include:(1)A new rough k-means algorithm with self-adaptive relative weights measurement based on space distance.Traditional rough k-means algorithm mostly choose fixed weights by experience,ignoring the effect of space distribution of objects within the cluster and the differences between clusters.According to the the space distribution of lower approximation and boundary region objects relative to cluster center,an adaptive measurement method of relative weights based on the space distance is designed to dynamically compute the relative weight coefficient of lower approximation and border areas.And a rough k-means algorithm with self-adaptive relative weights measurement based on space distance is proposed.(2)A new rough k-means algorithm with self-adaptive weights measurement based on space distribution of neighbor points.Most of the traditional rough k-means algorithms measure similarity of boundary object to cluster based on the distance of boundary object to the cluster center,ignoring the space distribution of boundary object neighbor points.The weight coefficient of boundary objects in cluster iteration calculation is measured according to the number of boundary objects neighbor points and the distance to boundary objects neighbor points in lower approximation ofoverlapping clusters that boundary objects may belong to.And a new rough k-means algorithm with self-adaptive weights measurement based on space distribution of neighbor points is put forwarded.(3)Data preprocessing of photovoltaic power prediction based on rough k-means clustering algorithm.Neural network is the most commonly used photovoltaic power prediction model.But with the improved requirement of accuracy and increased amount of data,training neural network forecasting model with initial data sample often causes huge network structure and hard training.It is difficult to meet the requirements of accuracy.It contributes to the structure of the neural network model and speeding up the training process of network model model by making an clustering preprocessing on original data.Combined with the practical application requirements of photovoltaic power generation forecasting system,conducting preprocessing analysis on photovoltaic training data by the designed rough k-means clustering algorithm.The result of photovoltaic power prediction with designed rough k-means is compared with the traditional clustering preprocessing method.