Demand response of residential air conditioning load based on user behavior

Abstract

Residential side demand response is an important supplementary means to maintain the supply-demand balance of source-load in the power system. However， the uncertainty of user behavior makes it difficult to accurately control demand response. With residential air conditioning load being a research object， a complete distributed control model of demand response is established. A solution to unknown and variable air conditioning load is proposed. Firstly， the load control in demand response is modeled as a multi-period stochastic process， and the behavior of users’ participation in demand response is abstracted as a Markov chain. Afterward， a thermodynamic model of the user is established based on linear regression and the Gaussian process to generalize and define residential comfort and calculate Markov transfer probability. For the problems of high cost and poor privacy of centralized control， a distributed control algorithm is designed with peak-shaving demand response as a classical object. Finally， the algorithm is validated using actual load data， achieving an excellent peak-shaving effect.

Keywords

• demand response
• gaussian process
• machine learning
• distributed algorithm