| Along with the goal of "double carbon",building a clean,low-carbon,safe and efficient energy system and a new power system with new energy as the mainstay is the future development direction.As a key technology to enhance the regulation flexibility of power system and improve the level of power supply guarantee for power users,establishing a more flexible demand-side resource management mode is an urgent need in the new form.Traditional demand response methods are mostly for large loads such as commercial and industrial,and it is difficult to build a "win-win"mechanism between supply-side participation in demand response costs and demand-side electricity costs.For a large number of distributed demand response resources,it is necessary to accurately identify residential customer groups with demand response potential and form a certain scale of demand response aggregation resources to participate in the interaction between supply and demand through flexible aggregation methods.Under the premise of minimizing the cost of demand response and electricity consumption,corresponding demand response operation strategies are established to maximize the task of peak reduction and valley filling,so that demand response resources can be fully and reasonably utilized.In this paper,we focus on the residential load,and conduct research from three aspects:resource aggregation,potential analysis,and optimization strategies to ensure reasonable operation of demand response:First,the response mechanism under the specific mode of demand response is analyzed,and on this basis,the characteristics of typical power-using equipment are discussed to verify the feasibility of customer load participation in demand response.Based on the characteristics of demand response technology,we apply relevant machine learning algorithms to select a more appropriate method for demand response from the algorithm principle,so as to efficiently complete the task of demand response.Second,a two-stage clustering model and a potential analysis model based on LSTM data error correction are constructed.The strong input-output mapping capability of LSTM neural network is used to perform error repair on the original load data and improve the data errors caused by voltage fluctuations.The aggregation of demand response resources with similar electricity consumption habits is completed by K-Means and SOM two-stage clustering,and the CEEMD modal decomposition algorithm is used to improve the disadvantages of redundancy and inconspicuous characteristics of the original load data.In addition,the qualitative analysis method and quantitative analysis model of demand response potential are established,and the influence factor of customer responsiveness is considered to develop demand response potential analysis for customer categories.The aggregation potential calculation of 45 household loads is completed by simulating residential customer data in a certain area,and the effectiveness of this demand response aggregation potential analysis method is verified by comprehensive comparison.Finally,a demand response operation optimization strategy based on time-of-use tariff and incentive mechanism is proposed.The supply-side demand response cost and customer electricity cost are considered,and the improved genetic algorithm NSGA II is used to achieve a reasonable allocation of demand response resources with the goal of minimizing the cost of both sides and completing the demand response task.And through the simulation of a regional residential customer data set,different customers save ￥4.2 to ￥4.8 in electricity cost and $0.4 to $1.1 in demand response cost,and the comprehensive comparison verifies the superiority and necessity of this demand response operation strategy. |
PDF Full Text Request