In the power system

the rapid integration of new energy and extreme weather leads to frequent extreme scenarios of severe fluctuations in power demand

which has a great impact on the safe operation of the power system. In order to solve this problem

this paper proposes a method of constructing extreme scenarios based on improved diffusion model. By simulating various possible extreme scenarios in power system

it can provide data basis for formulating effective strategies and countermeasures

and reduce the cost of trial and error. The definition of extreme and normal states of trough

flat and peak periods is proposed. Based on Markov chain

the state transition probability matrix of each period is constructed and the scene state set is generated. The improved diffusion model is used to generate different forms of data in each period

and the data sampling group is constructed. The scene construction is realized by randomly sampling from the data sampling group and filling it into the scene state set. Through the example analysis

the reliability and efficiency of the proposed method in the generation of scenes covering extreme forms are verified.