Research On Distributed Bi-Layer Clustering Analysis And Optimal Scheduling Strategy For Residents' Demand Response

Compared with traditional energy,renewable energy has the advantages of more reserves and less pollution,and therefore has been developed rapidly in recent years.However,the renewable energy has difficulties in consumption because of its dispersibility and instability.In smart grid,applying demand response to fully dispatch demand side resources and respond to system requirements is one of the important means to solve the problem of renewable energy consumption.Nevertheless,the promotion of smart meters,communication network and intelligent computing technology has led to the exponential growth of monitoring data,which has formed the big data.Consequently,challenges have emerged for utilities to analyze and interpret big data at demand side.The application of effective data mining techniques is of great significance to analyze the electricity consumption behavior of customers,adopt appropriate demand response strategies and consume renewable energy.To solve the difficulties in analyzing big power data,this paper focus on extracting practical characteristics by data mining techniques,establishing distributed clustering analysis and multi-period optimal scheduling strategy of load aggregators based on non-cooperative game,etc.The proposed methods and approaches provide fundamental information for customer segmentation in demand response and renewable energy consumption.The details are as follows:Firstly,a comprehensive feature set of load curves is constructed for demand response analysis,i.e.the time-domain,frequency-domain,fluctuation and stability features.In addition,the steps of calculation methods are given in detail.To extract typical features,reduce dimension and improve the efficiency and effectiveness of clustering algorithm,the improved Analytic Hierarchy Process(AHP)-entropy weighting method is utilized to conduct feature selection.To avoid the subjective weighting of time-domain features,the evaluation rule that calculates the contribution degree of features are designed,and the weights are determined and updated objectively and adaptively by entropy-weighting method.In the case study,feature selection and Affinity Propagation(AP)clustering algorithm is performed on public dataset.Consumption behavior analysis based on the proposed features and clustering results proves the validity of the proposed features,which is practical in analyzing demand response potential of different customers,and has potential in big power data analysis.Secondly,an efficient clustering algorithm combination is proposed based on a distributed bi-layer clustering framework.The effective feature extraction lays the foundation for constructing distributed clustering framework.Based on the principle of distributed algorithm,the physical structure and calculation steps of the proposed bi-layer clustering framework are elaborated.The improved Kmeans and AP clustering algorithm are applied in local and global modeling respectively,and the traditional algorithms are modified by evaluation indices to improve the performance of algorithm,i.e.the sum of squares due to error(SSE)index and Davies-Bouldin(DB)index.The performance tests on the simulation data and electricity consumption behavior analysis on the big dataset are conducted to verify the effectiveness of the proposed models and approaches.Finally,considering the decentralization of load distribution,it is difficult for utilities to execute demand response directly at the demand side.Therefore,a multi-period optimal scheduling strategy among aggregators based on non-cooperative game is proposed.The factors affecting customer participation in demand response are analyzed.The flexibility index is therefore calculated according to the proposed features,and the quantitative flexibility index is obtained,which reflects the aggregator scheduling ability in demand response.In day-ahead scheduling,considering the flexibility of aggregator's in demand response,incentive price is designed to encourage aggregator to distributed optimize scheduling plan,and Nash equilibrium is sought through non-cooperative game theory to maximize aggregator's scheduling benefits.In contrast,the real-time scheduling considers uncertainties of energy outputs and customer consumption,and scheduling strategy of aggregators is determined by achieving maximum real-time scheduling profits,while uncertainties of customer consumption caused by satisfying comfort or saving electricity costs are simulated by truncation distribution.The proposed model is validated by the modified IEEE 33-bus system.The results indicate the proposed model can not only meet the demand of customers while maximizing the benefits of aggregators,but also realize the consumption of renewable energy and ensure the safe and stable operation of grids.