Studies On Analysis Methods Of Demand Response Characteristics

As the development of social economy and improvement of technology, the energy crisis and environmental press are becoming more and more serious, so modern electricity power system is encountering higher challenges than ever. Under such circumstances, the concept "smart grid" is proposed, which has been becoming the development and study hotspot of modern electricity industry. As one of the most important constituents of smart grid, the great potential of electricity demand side resources has been gradually recognized and unearthed. Particularly, electricity consumers’demand response (DR) directly indicates the main feature of smart grid:"smart interaction", and has obtained more and more attentions all over the world.Through price signal and incentive mechanism, DR programs encourage electricity consumers to adjust their own electricity consumption arrangement in response to the operation goals of power system, which would keep the power system operation safe and reliable. Various DR programs have been gradually proposed and put into practice around the world, some of which have resulted in significant effects. Since DR programs are specially designed for the electricity consumers, their response characteristics in response to specific DR programs directly determine whether the DR programs are reasonable and whether they can obtain the expected results. Therefore, this paper will make a systematical study of electricity consumers’demand response characteristics (DRCs) in response to the time of use (TOU) pricing program through clustering analysis and regression modeling.First, according to the characteristics of electricity consumers’load data and their treatment requirements, Self Organizing Map (SOM) clustering algorithm and Support Vector Machine (SVM) regression algorithm are selected from the views of both clustering analysis and regression modeling. The basic principles, specialties and applications of these two algorithms are then introduced.Secondly, a general analysis method of electricity consumers’DRCs is proposed through the Self Organizing Map (SOM) clustering algorithm in this paper. At first, DRCs of electricity consumers are summarized and defined as a series of quantitive index. Then different industrial types of electricity consumers’DRCs are abstracted according to their load data in the TOU pricing program. Through the SOM clustering algorithm these industrial electricity consumers are clustered into several different groups based on their DRCs. Finally, these clustering results are used to analyze every consumer group’s typical DRCs and assess their DR potential, which would provide theoretical references for the adjustment and improvement of TOU pricing program.Thirdly, a demand response model of electricity consumers is proposed through Support Vector Machine (SVM) regression algorithm. Based on the daily load data of high energy-consuming industrial consumers in the TOU pricing program, the main factors which affect consumers’daily load arrangement are mined and analyzed. Then the input and output attribution vectors are constructed to train and test the consumers’ demand response model based on SVM regression algorithm. Finally this well-trained demand response model is applied in their DRCs exact analysis and DR potential quantitive assessment.In all, the two DRCs study approaches above (SOM clustering analysis and SVM regression modeling) will help to systematically grasp electricity consumers’DRCs from both generally qualitative aspects and exactly quantitive views. These study approaches proposed in this paper will significantly promote the implementation and improvement of current TOU pricing program. What’s more, they will provide some new referential approaches for other DR programs studies.