Research Of Association Rules Mining And Clustering Analysis On Energy Consumption Monitoring Data

With the rapid development of the economy, the consumption of building energy has gained an increasing percentage, in which a great deal of data are collected and stored. And also, numerous valuable information is contained in these data, which cannot be found by using traditional statistical methods. To cope with this issue, some data mining technologies are utilized in this paper to analyze the associations among the attributes of energy consumption data and the aggregation property of the energy consumption nodes. Hence, a feedback strategy can be established to make guidance for the energy-saving and emission-reduction and make the best use of resources, the main contents are shown as follows:Firstly, for the traditional association rule methods, a huge amount of candidate sets are generated during the mining process, hence the results are highly dependent on the settings of the degrees of support, confidence as well as the threshold values. In this paper an adaptive association rule method incorporated with particle swarm optimization(PSO) and Gravitational Search Algorithm(GSA) is proposed. The proposed algorithm combine the exploration ability of PSO and the exploitation ability of gravitational search, and then the adaptive ability is dependent on the distribution of the population and the evolution condition. Experimental results on the data sets from the well-known UCI library show that, comparing with other methods considered in this paper, the proposed algorithm can be more effectively applied to analyze the rules of association. Moreover, the proposed method is applied for the analysis of the energy consumption database of the Jiangnan university, and the results of the association rules are get to provide guidance for the management of resources.Secondly, as it is known that the redundant attributes cannot be effectively removed by traditional spectral clustering algorithms during the process of nonlinear low rank approximation, resulting in vague distinction of local regions and high computation complexity. To resolve this problem, a novel spectral clustering algorithms based on the feature extraction approach using nonlinear low dimensional embedding is proposed in this paper. The nonlinear low dimensional embedding approach is utilized to extract the characteristics of the data for dimension reduction, capturing the multi cluster structure and fully exploring the potential manifold structure. Experimental results on the UCI data sets show that the improved method can effectively improve the clustering performance, and also decrease the computational complexity when coping with large scale datasets. Additionally, the proposed method is utilized to make clustering analysis for the energy consumption data of the Jiangnan university, which can efficiently partition different types of people, providing basis for further analysis of hidden knowledge in the energy consumption data.Finally, to further improve the clustering performance of the energy consumption data, an improved ant colony clustering algorithm based on the spectral analysis theory is proposed in this paper. For our method, the eigenvector of the Laplace matrix can be extracted by the spectral clustering algorithm, then constructing the initial pheromone matrix of the ant colony. Also, some steps of the original ant colony clustering algorithm, like updating the pheromone and picking up the data objects, are improved with the guidance of the objective function to develop a more effective hybrid spectral clustering method. The experiment shows that this algorithm can find the optimal solution region efficiently, and then the algorithm is applied to the energy consumption supervision database of Jiangnan university and realize the the clustering of node energy consumption. Different periodic energy consumption mode is established according to the clustering results. The mode is used to detect abnormal energy consumption point.