In order to encourage users to implement friendly interaction with the power grid and reduce energy consumption and power load

the way to equitably share the interactive benefits of the power grid company to all participating users according to a certain subsidy proportion is studied. In order to solve the combination explosion problem of the traditional Shapley value method

a compensation method for sharing the interactive benefits of users based on the Shapley value sampling estimation is proposed. Under the constraint of meeting the balance of payments

the method can reduce sample size by stratified random sampling. In order to fix the sample allocation amount of each layer

the advantages and disadvantages of random allocation method

average allocation method and Neyman optimal allocation method are comprehensively compared. In order to solve the problem that the standard deviation of samples at each layer of participants is unknown in Neyman optimal allocation method

an iterative ε(t) estimation optimal allocation method based on reinforcement learning algorithm is proposed. Numerical examples show that the proposed method has all the characteristics of Shapley value method

can accurately estimate the allocation results of Shapley value method

and can achieve fair and reasonable allocation

while effectively reducing the calculation time.