Research On Key Technologies Of Demand Response

With the advancement of the reform of the electricity market,the commodity attributes of electricity have been restored.As electricity is in short supply during periods of peak electricity use or system failures,the price of electricity will rise sharply,adversely affecting the normal and orderly production of society.In order to cope with the problem of excessively high electricity prices under special circumstances,demand response has started to participate in the optimal allocation of market resources as a means of adjustment,that is,End users can change their electricity consumption habits as electricity prices change,or reduce electricity consumption based on incentives when electricity prices in wholesale markets are too high or system reliability is threatened.China's demand response is still in its infancy,and there are still many issues that need to be resolved from the pilot to the improvement.This paper focuses on the key technologies for demand response implementation,from semi-supervised clustering of improved ant colony algorithm combined with manifold learning dimension reduction,demand response potential assessment method based on two-stage clustering analysis,and load baseline design ideas based on consumer baseline load calculation bias.Research in three aspects to solve the problems of user cluster analysis,potential assessment and settlement in demand response,which has theoretical guidance significance and practical application value for the key technologies of developing demand response.The main work of this article includes:(1)In order to solve the problem that load clustering application scenarios need clustering results as similar as possible to initial cluster center,improved ant colony clustering algorithms are designed based on the ant colony clustering algorithm.The two factors that determine the clustering effect and the distance between the cluster center and the initial cluster center constitute the fitness index instead of the traditional mean square error update pheromone matrix.It is verified by example analysis that this algorithm can well solve the application scenarios where the clustering results are as similar to typical load categories as possible,and has good clustering performance.Finally,in order to improve the computational efficiency and performance of the algorithm,the manifold learning dimension reduction method is applied to the processing of power load data.The results of numerical examples prove that the manifold learning dimension reduction method can well remove the noise from the power load data and extract its typical features.(2)Aiming at the problem that there is no good quantitative assessment method of regional demand response potential at the current stage,a method of demand response potential assessment based on two-stage clustering analysis is designed.This method fully considers the user's suitability and the characteristics of the user's process / equipment,which are two main factors that affect the demand response potential.Finally,the data of 386 customer loads in city D were used to evaluate its demand response potential,and the results were compared with the proportion of demand response participation in the US power markets.The results were reasonable,indicating that the method has certain practical value.(3)By comparing the performance of the existing consumer baseline load,it is found that the bias of the consumer baseline load may be quite large when the accuracy of the consumer baseline load is not much different.Then,based on the clearing rules of the Guangdong spot market,the relevant constraints of demand response and consumer baseline load are added,and the analysis concludes that the appropriate high baseline bias is beneficial to the improvement of social welfare.In view of the user's ‘bonus-getting' behavior that may be faced with high bias from the consumer baseline load,the access capacity setting is discussed.The final example shows that formulating appropriate access capacity standards can curb users' ‘bonusgetting' behavior to a certain extent.